Oxygen-vacancy-regulated molybdenum oxide nanosheets for a multi-phosphate discriminative colorimetric sensor array

Graphical abstract

Highlights

• •Novel oxygen-vacancy-engineered MoOx nanosheet array enables precise discrimination of ten phosphate species.
• •UV irradiation-induced gradient oxygen vacancies in MoOx avoid multiple probes and simplify sensing array fabrication.
• •Oxygen vacancy-mediated sensing mechanism of MoOx nanosheets for multi-phosphate detection is elucidated.
• •Effective phosphate recognition is realized in serum and other complex matrices.

Abstract

The detection and identification of various phosphates are crucial in disease diagnostics. However, traditional detection techniques often require the development of multi-ion systems, and the complexity of sensor unit fabrication has limited the broad applications of sensor arrays. In this study, we report for the first-time the design of a colorimetric sensor using oxygen vacancy-regulated MoO_x(2 ≤ x ≤ 3) nanosheets. By precisely modulating the concentration of introduced oxygen vacancies, the plasmonic resonance absorption peak of the MoO_x nanosheets exhibited tunable shifts, thereby influencing their optical properties. The addition of phosphate ions with different configurations influenced the light absorption range and the plasmonic peak values of the oxygen vacancy-rich MoO_x sensor array to various extents. Through statistical analysis methods, this sensing array effectively identified ten types of phosphates (i.e., ATP, ADP, AMP, PPi, Pi, UTP, GTP, GDP, GMP, and CTP). The sensor demonstrated excellent detection and discrimination capabilities in both quantitative identification and the recognition of mixed phosphates. The practical application of this sensor array was validated by precisely detecting phosphates in complex systems such as serum. Compared to the previous reports, the fabrication process of the oxygen vacancy-regulated MoO_x nanosheets is simple, the discrimination detection effect is good, and it provides a new approach and method for the design and fabrication of highly efficient sensor arrays.

Introduction

Phosphates are essential compounds with numerous important functions in biological systems and various industrial applications. In biological processes, phosphates are involved in energy metabolism, signal transduction, transmission of genetic information, and the maintenance of cellular structures [[1], [2], [3]]. For instance, adenosine triphosphate (ATP) serves as the primary energy currency in cells, driving numerous biochemical reactions, whereas other nucleotides including guanosine triphosphate (GTP), cytidine triphosphate (CTP) and uridine triphosphate (UTP) are essential for processes such as lipid metabolism, DNA synthesis, and cellular signals [4]. Moreover, phosphates are vital components of nucleic acids (DNA and RNA) and cellular membranes, emphasizing their critical role in maintaining life [5,6]. Despite their significance, abnormal phosphates levels in biological fluids can have detrimental effects. In humans, imbalances in phosphate levels are associated with diseases such as hypophosphatemia, rickets, and chronic renal disease [[7], [8], [9]]. Therefore, precise detection and quantification of phosphates are essential for medical diagnostics and environmental monitoring.

Traditional methods for phosphate detection, including chromatography [10], electrochemistry [11,12], and fluorescence [[13], [14], [15]], are often limited by their ability to detect only a particular type of phosphate at a time. This limitation is particularly challenging in complex biological samples containing numerous phosphates [16]. To tackle this issue, sensor arrays fashioned like human olfactory system have emerged as an appealing alternative. These arrays, often referred to as “electronic tongues”, can simultaneously detect and distinguish various phosphates by producing unique recognition patterns [[17], [18], [19]]. For instance, Guan et al. prepared bare gold nanoparticle metal ion arrays for the detection of phosphates and monitoring of the enzymatic hydrolysis process involving nucleotide phosphates [20]. Likewise, Sun et al. designed a fluorescent carbon dot metal ion as a three channel fluorescence sensor array to identify five phosphate anions, including ATP, adenosine diphosphate (ADP), adenosine monophosphate (AMP), pyrophosphate (PPi), and phosphate (Pi) [21]. This technique not only improve our understanding of phosphate dynamics in biological systems but also provide valuable tools for environmental monitoring, industrial quality control, and medical diagnostics. Beside these five commonly studied phosphates, other biologically relevant phosphate-containing compounds, such as guanosine diphosphate (GDP) and guanosine monophosphate (GMP), are essential for controlling physiological processes within biological fluids [22,23]. To address the growing demand for extensive phosphate analysis, there is a pressing need for highly efficient detection approaches that can simultaneously identify and quantify various types of phosphates, including ATP, ADP, AMP, PPi, Pi, GTP, GDP, GMP, CTP, and UTP.

As the core of a sensor array, the sensing unit typically incorporates multi-channel sensing materials designed to differentiate subtle variations caused by structurally similar analytes in complex environments. However, it is noteworthy that traditional sensor arrays often involve complex synthesis processes for fabricating various sensing units, limiting their use for the detection of certain phosphate species. Most of the current phosphate sensor arrays predominantly rely on multi-component fluorescent systems, such as rare-earth-doped nanomaterials or surface-functionalized quantum dots. The design of specific ligands or energy-level-modulated structures for different phosphate anions further increases synthetic complexity and environmental sensitivity, resulting in high costs, poor portability, and difficulties in achieving on-site instant detection, which hinders large-scale deployment [24]. In colorimetric sensor arrays, 3,3′,5,5′-Tetramethylbenzidine (TMB) is typically employed as a chromogenic substrate to generate optical signal variations [25]. Simultaneous detection of multiple targets is achieved through either combining TMB with other chromogenic substrates or coordinating it with various transition metal ions [26,27]. However, such systems are usually designed based on gold nanoparticles, metal–organic frameworks (MOFs), or nanozyme platforms, where phosphate recognition is realized by regulating their catalytic activity toward TMB. These methods rely heavily on externally added H₂O₂ and chromogenic reagents, leading to cumbersome detection procedures that are unsuitable for real-time analysis [28].

In contrast, inorganic nanomaterials generally exhibit superior chemical stability and biocompatibility, while their surfaces can be readily modified and functionalized to enable multicomponent detection. As a representative transition metal oxide, MoO₃, which is abundant, cost-effective, and environmentally friendly, offers a wide range of applications in biosensors [29,30]. More importantly, MoO₃ exhibits a localized surface plasmon resonance (LSPR) effect, and the incorporation of oxygen vacancies through UV irradiation induces a visible color transition from colorless to blue, enabling the construction of an intrinsic colorimetric sensing platform without the need for external chromogenic substrates [31]. Oxygen vacancies in MoO₃ act as adsorption sites for negatively charged phosphate ions through electrostatic interactions, altering the range of light absorption [32]. Additionally, phosphate adsorption induces a redistribution of electrons and enables charge compensation to the unsaturated coordination atoms surrounding the oxygen vacancies, thereby enhancing the concentration of oxygen vacancies and LSPR absorption [33,34]. Thus, oxygen-vacancy engineering in MoO_x offers a feasible route for fabricating colorimetric sensor arrays, in which UV-induced modulation of vacancy levels adjusts the optical response upon phosphate adsorption. Oxygen-vacancy-rich MoO_x nanomaterials can therefore form sensor arrays in a simple manner, requiring neither metal-ion dopants during synthesis nor exogenous metal coordination for signal generation.

Considering the significance and challenges of phosphate anion identification, and to verify the feasibility of the above multi-channel sensor array strategy based on the regulation of oxygen vacancies in MoO_x(2 ≤ x ≤ 3) nanosheets, herein, a MoO_x nanosheets system mediated by oxygen vacancies is designed as a colorimetric sensor array for the detection and identification of various phosphate anions (including ATP, ADP, AMP, PPi, Pi, GTP, GDP, GMP, CTP, and UTP). The abundance of oxygen vacancies in MoO_x nanosheets facilitates interaction with phosphates, and phosphate species with distinct molecular structures induce characteristic changes in the optical absorption of MoO_x, including variations in the visible absorption range and modulation of LSPR peak intensity. These multichannel spectral differences enable precise detection and discrimination of the ten phosphate species. Furthermore, the ability of the sensor arrays to detect and quantify mixed phosphates, including PPi, along with its proven applicability in recognizing phosphates in complex media like serum and cell lysates, is highlighted. This marks the first case of developing a sensor array based on the modulation of oxygen vacancies in MoO_x nanosheets for the simultaneous detection of ten phosphate types (Scheme 1). Notably, this sensing concept relies on the distinct adsorption strengths and vacancy modulation capabilities of different phosphates on MoO_x, thereby generating analyte-specific optical signatures. The design strategy is believed to be applicable to additional analytes with comparable chemical structures or properties, providing a universal and scalable approach for multi-analyte recognition.

Scheme 1.

Design and application of MoO_x nanosheets colorimetric sensor array.

Experimental

Chemical reagents

Molybdenum trioxide (MoO₃) powder (99%) and N-methyl-2-pyrrolidone (NMP, 98%) were purchased from Aladdin (Shanghai, China), whereas adenosine phosphates (ATP, ADP, AMP), pyrophosphate (PPi), inorganic phosphate (Pi), and nucleoside triphosphates and diphosphates (UTP, GTP, GDP, GMP, CTP) were acquired from Macklin. All chemicals employed in this study were of high analytical purity and utilized without additional purification. Ultrapure water (18 MΩ) was utilized for the preparation of all stock and buffer solutions.

Characterizations

The UV − vis absorption spectra were obtained using an Evolution-220 spectrophotometer. Transmission electron microscopy (TEM) examination was performed using a JEOL JEM-2100 microscope at an acceleration voltage of 200 kV. The zeta potential was measured using a Brookhaven Omni instrument with electrophoretic light scattering. The sample underwent freeze-drying with a SCIENTZ-30ND freeze drier and then crushed into a fine powder for structural analysis by X-ray diffraction (XRD) using a Nikaku SmartLab XRD system, utilizing Cu Kα radiation (45 kV, 200 mA) throughout a 2θ range of 5° to 80°. Fourier transform infrared spectroscopy (FTIR) analysis was conducted with a Thermo Fisher Scientific Nicolet iS50 to examine the vibrational properties of chemical bonds. Raman spectra were acquired using a Renishaw inVia confocal micro-Raman spectrometer (Renishaw plc, UK). X-ray photoelectron spectroscopy (XPS) analysis was conducted utilizing a Thermo Scientific Escalab 250Xi (Waltham, MA, USA). Electron Paramagnetic Resonance (EPR) measurements were performed using a Bruker EMXplus EPR instrument.

Preparation of the sensor array and sensing of phosphates.

MoO_x nanosheets were prepared and purified according to the previous report [35]. A dual-channel sensor array was designed using two MoO_x nanosheet working solutions. The detailed procedures are as follow. Two aliquots of homogeneous MoO_x nanosheet dispersions (neutral pH 7.0 ± 0.2, concentration 3 mg/mL) were prepared. Ten phosphate species were selected as model analytes, including ATP, ADP, AMP, PPi, Pi, GTP, GDP, GMP, CTP, and UTP. Each phosphate standard was precisely weighed, dissolved in ultrapure water, and diluted to a specific volume. Fresh working solutions were prepared by diluting the stock solutions with ultrapure water immediately before use. For each measurement, 990 μL of the MoO_x working solution was added to a quartz cuvette and equilibrated for 5 min at room temperature and considered as the blank system. Subsequently, 10 μL of the phosphate standard solution was added and gently vortexed for 5 s to ensure homogeneous mixing. Detection was performed in dual-channel mode. (1) Non-irradiated channel: after mixing, the sample was incubated at room temperature for 10 min without UV exposure. The absorbance was then recorded at 425 nm. (2) UV-irradiated channel: after mixing, the sample was exposed to a 5 W UV lamp (central wavelength 365 ± 5 nm, full width at half maximum 10–12 nm) for 6 s to generate oxygen vacancies (the vertical distance between the UV light source probe and the sample surface was fixed at 0.5 cm, with an irradiation area of 0.36 cm²). The sample was then incubated at room temperature for 10 min, followed by absorbance measurement at 690 nm.

To assess the reproducibility and stability of the sensing system, each experimental trial was performed in quadruplicate. The absorbance response of the array to phosphates was represented as A-A₀, where A and A₀ denotes the normalized absorbance intensity in the presence and absence of phosphates, respectively. Data analysis was conducted with linear discriminant analysis (LDA) or hierarchical cluster analysis (HCA) conducted in PyCharm, where absorbance data acquired at four wavelengths (425, 690, 590, and 740 nm) were subjected to dimensionality reduction to facilitate pattern recognition of phosphate species. Model performance was evaluated using precision, recall, F1-score, and overall accuracy, accompanied by visualization of the classification outcomes through confusion matrices. The ellipses represent the 90% confidence intervals for the overall mean of each category in the LDA plots.

Analysis of real samples

To evaluate the capability of the sensor array for detecting and discriminating various physiological phosphate species in complex biological matrices, performance tests were conducted in serum and cell lysate environments. Murine breast cancer cells (4T1) were suspended in ultrapure water at a concentration of approximately 2.4 × 10⁶ cells/mL, followed by swelling at 4 °C for 12 h to induce cell lysis. The resulting cell lysate was directly mixed with the MoO_x working solution and incubated for 10 min, after which different phosphate anions were added. The subsequent detection procedure was consistent with that described in Section “Preparation of the Sensor Array and Sensing of Phosphates”. Fetal bovine serum (FBS) was selected as the model for complex biological matrices and diluted 10-fold with ultrapure water. Phosphate solutions of varying concentrations were prepared using the diluted FBS and incubated with the MoO_x working solution (with the final FBS concentration maintained at 0.1%) for 10 min; the detection protocol was identical to that detailed in Section “Preparation of the Sensor Array and Sensing of Phosphates”. Additionally, the biocompatibility of the MoO_x nanosheet working solution was evaluated via the MTT assay using 4T1 cells as the model cell line.

Results and discussion

To further elucidate the changes in MoO_x before and after the addition of phosphates, a series of characterizations were conducted to determine the mechanism underlying the sensing array's ability to detect and differentiate phosphates. X-ray diffraction (XRD) analysis was performed to investigate the structural evolution of MoO_x nanosheets. Compared to the α-MoO₃ standard card (#05-0508), all MoO_x nanosheet samples exhibited significantly broadened diffraction peaks, and a distinct shift of the peaks to lower angles was also observed (Fig. 1A). Specifically, the diffraction peak corresponding to the (020) crystal plane shifted from the standard position of 12.67° to 10.60°. This shift directly indicates an increase in the lattice spacing and the loss of long-range crystalline order, which confirms the severe collapse of the MoO₃ crystal structure and the formation of an amorphous phase [36]. To further clarify the effect of phosphate on the MoO_x structure, quantitative microstrain analysis was conducted on the characteristic (021) diffraction peak at approximately 26°. The results reveal a clear positive correlation between phosphate concentration and the microstrain of the MoO_x system. With the gradual increase in phosphate concentration, the microstrain increased continuously from 37,353 ppm in the low-concentration phosphate sample to 263,103 ppm in the medium–high concentration phosphate sample (Table S1). Similarly, the grain size of MoO_x decreased gradually as the phosphate concentration increased. This relationship between phosphate concentration, microstrain and grain size confirms that the introduction of phosphate effectively promotes the formation of oxygen vacancies in the MoO_x. As intrinsic lattice defects, oxygen vacancies further intensified lattice distortion, which is consistent with the observed peak broadening and lattice spacing changes.

Fig. 1.

(A) XRD patterns of MoO₃, MoO_x nanosheets, MoO_x with low-concentration phosphate adsorption (LPi-MoO_x), and MoO_x with high-concentration phosphate adsorption (HPi-MoO_x). (B) TEM image and EDS elemental maps of MoO_x nanosheets. (C) TEM image and EDS elemental maps of MoO_x with phosphate adsorption (Pi-MoO_x). (D) Raman spectra of MoO_x and Pi-MoO_x. (E) EPR spectra of Pi-MoO_x. (F) XPS survey spectra of high-resolution Mo 3d XPS spectra. (G) XPS survey spectra of Pi-MoO_x and (H) high-resolution P 2p XPS spectrum and. (I) Selective analysis of MoO_x nanosheets (all substances at 1 mmol/L concentration). (J) Zeta potential of MoO_x nanosheets and doped with ten types of phosphates.

Transmission electron microscopy (TEM) analysis was done to further explore the structure of MoO_x in the absence and presence of phosphate anions. The TEM images clearly show that MoO_x exhibits a well-dispersed nanosheet structure without lattice fringes, indicating a certain degree of amorphization of the samples (Fig. 1B and Fig. S1A, B). Meanwhile, the small size of the exfoliated nanosheets exposes abundant surface-active sites, providing sufficient binding interfaces for phosphate adsorption. After phosphate adsorption, the MoO_x nanosheets became more disordered and displayed a more loosely connected and better-dispersed morphology (Fig. 1C, S1C). The energy-dispersive X-ray spectroscopy (EDS) elemental mapping analysis was done to investigate the elemental distribution in the Pi-MoO_x. The results reveal uniform distributions of Mo, O, and P elements. To probe long-term structural changes, the morphology of MoO_x nanosheets was evaluated at 48 h post-phosphate adsorption. HRTEM images of these samples reveal the emergence of extremely small nanostructures (∼1 nm) within the MoO_x (Fig. S1D, E, and F). These observations confirm that phosphate adsorption promotes the generation of oxygen vacancies in the MoO_x lattice. The presence of a large number of oxygen vacancies further disrupts the crystal structure of MoO_x, leading to a continuous decrease in its nanoscale size. The Raman spectra of MoO_x and Pi-MoO_x exhibited sharp characteristic bands at 288 cm⁻¹ (Mo≡O), 668 cm⁻¹ (Mo₃–O), 826 cm⁻¹ (Mo₂–O), and 999 cm⁻¹ (Mo=O) (Fig. 1D), confirming that the orthorhombic α-MoO₃ structure is well preserved [37]. Notably, after phosphate adsorption, the Mo₂–O vibration shifts from 826 to 824 cm⁻¹, which is likely attributed to the generation of oxygen vacancies that weaken local bonding [36,38]. In addition, Fourier-transform infrared (FTIR) revealed no additional peaks before and after the addition of Pi (Fig. S2), indicating that no chemical bond breakage or formation occurred.

To further investigate and validate the mechanism, X-ray photoelectron spectroscopy (XPS) and Electron paramagnetic resonance (EPR) were used to analyze the chemical states of surface atoms and the presence of phosphate adsorption. EPR measurements confirmed that oxygen vacancies were the primary factors involved in phosphate detection and recognition. As revealed in Fig. 1E, the EPR spectra consistently exhibited a peak at 2.003, indicating the presence of ionized oxygen vacancies. By adding Pi, there is a gradual increase in the surface oxygen vacancies and a noticeable enhancement in the signal intensity. The chemical state of Mo on the surfaces of Pi-MoO_x nanosheets was further analyzed by examining the Mo 3d peaks. The typical four-peak of Mo 3d spectrum could be well fitted into two spin–orbit doublets (Fig. 1F). The peaks at 235.6 eV and 232.5 eV correspond to Mo⁶⁺, while the weaker peaks at 234.6 eV and 231.5 eV are attributed to the Mo⁵⁺[39,40]. The presence of Mo⁵⁺ ions indicate the existence of oxygen vacancies on the surface of MoO_x nanosheets. With the addition of phosphates, the Mo⁶⁺ intensity decreased while the Mo⁵⁺ intensity increased, suggesting that phosphate adsorption enhanced the oxygen vacancy concentration in MoO_x nanosheets. In addition, there was a strong correlation between Pi adsorption and the valence state of MoO_x, because the content of Mo⁵⁺ increased with the increase in Pi concentration. As shown in Fig. 1G, the XPS spectrum of Pi-MoO_x revealed distinct signals for O and Mo, with a peak at 133 eV which corresponds to the P 2p (Fig. 1H). Additionally, we investigated the specific reactivity of MoO_x towards phosphates among numerous possible interfering agents. Fig. 1I illustrates that these interfering compounds had minimal impact on the solution's absorbance or resulted in a reduction of the absorption peaks, but the groups containing phosphates revealed a substantial increase. Furthermore, electrophoretic light scattering (ELS) experiments indicate the zeta potentials of MoO_x after adsorption of various phosphates (Fig. 1J). The MoO_x often possesses several oxygen vacancies, which is due to the removal of a neutral oxygen atom from oxygen-containing molecules, leading to the addition of two extra electrons [41]. MoO_x exhibited a surface potential of ‒11.7 mV, but the MoO_x systems adsorbed with 10 different phosphates displayed various levels of enhanced negative surface potentials.

The sensing mechanism underlying phosphate adsorption and oxygen vacancy modulation was further elucidated via theoretical calculation results. Using the DFT + U approach (PBEsol functional, U_eff = 4.06 eV), the band gap of pristine MoO₃ was calculated to be 1.47 eV (Fig. 2A). The valence band maximum of MoO₃ was dominated by O, while the conduction band minimum was governed by Mo. After phosphate molecule adsorption, the Fermi level of MoO₃ shifted toward the conduction band (Fig. 2B), leading to a significant reduction in band gap, enhanced electrical conductivity, facilitated electron transition from the valence band to the conduction band, and expanded light absorption range. Furthermore, as illustrated in Fig. 2C, phosphate adsorption on MoO_x in the presence of oxygen vacancies was a spontaneous exothermic process with an adsorption energy of −4.35 eV. Consequently, the introduction of phosphates into the MoO_x system extended the UV light absorption range to the visible region, accompanied by a color change of the solution from pale yellow to yellow (Fig. 2D). UV irradiation was capable of introducing additional oxygen vacancies into the MoO_x nanosheet system [31], and these vacancies generated a new defect energy level (ΔE) at 0.82 eV, below the CBM. The presence of this defect level resulted in a remarkably enhanced red-light absorption (Fig. S3). Calculation of oxygen vacancies formation energy revealed that the formation energy is −1.33 eV for phosphate-adsorbed MoO_x (Fig. 2E), which is lower than that of the MoO₃ without phosphate adsorption, confirming that phosphate adsorption favored oxygen vacancy generation. After UV irradiation, oxygen-vacancy-rich MoO_x (MoO_x-OVs) revealed an LSPR peak with strong red-light absorption. Phosphate adsorption further promoted oxygen vacancies formation, leading to a higher absorption peak and a corresponding color change of the solution from light blue to dark blue (Fig. 2F). Based on these distinct color variations, a dual-channel detection system regulated by phosphate adsorption and oxygen vacancies modulation was established.

Fig. 2.

(A) The optimization model and density of states diagram of MoO₃. (B) The optimization model and density of states diagram of MoO₃ adsorbed by phosphoric acid. (C) Phosphoric acid adsorption energies before and after introducing oxygen vacancies. (D) UV–vis absorption spectra and digital photographs of MoO_x before and after phosphate adsorption. (E) Oxygen-vacancy formation energies in the presence and absence of phosphoric acid adsorption. (F) UV–vis absorption spectra and digital photographs of MoO_x after UV irradiation, before and after phosphate adsorption.

A sensing array composed of two colorimetric modes with different oxygen vacancy concentrations was established to identify 10 different phosphates under optimal conditions. Based on the colorimetric responses induced by phosphate adsorption and oxygen vacancy modulation, we added 10 different concentrations of phosphate salts to two groups of MoO_x nanosheets under irradiation and without irradiation. These exhibited distinct responses at around 425 nm in the unlit system and around 690 nm after irradiation (Fig. S4). Notably, due to differences in molecular structure and the number of phosphate groups among the ten phosphate species, their adsorption behaviors on MoO_x nanosheets varied significantly (Fig. S5). This differential adsorption further translated into distinct colorimetric response patterns, with each phosphate species exhibiting a unique linear response range and detection performance (Table S2). Specifically, the limits of detection (LOD = 3σ/S) were 3.5 μM for ATP, 26.7 μM for PPi, and 18.0 μM for Pi.

As revealed in Fig. 3A, the addition of different phosphates induces distinct optical responses that can be directly observed through optical photographs, particularly for inorganic phosphates (PPi and Pi), which reveal significant color changes. To quantify these optical responses more precisely, the absorbance was measured through a UV–vis spectrophotometer, as shown in Fig. 3B and 3C. The adsorption of phosphates onto MoO₃ affects its band structure, leading to changes in its UV light absorption. The absorbance at 425 nm was precisely analyzed, as phosphates readily intercalate into MoO_x nanosheets, causing shifts in the UV absorption band. This feature allows Pi to be readily differentiated from other phosphates in later identification steps. The structural variations among nucleotide phosphates also have different effects on UV absorbance, serving as a critical feature for differentiation. After the incorporation of oxygen vacancies, the LSPR effect of MoO_x nanomaterials resulted in a characteristic absorption peak at around 690 nm, as previously identified. The incorporation of phosphates increases the concentration of oxygen vacancies in the oxide, which in turn affects the intensity of this characteristic peak. These distinct peak responses reflect the unique reaction patterns of specific analytes. As shown in Fig. 3D, the radar chart displays the differential responses of individual channels to the various phosphates. The ratio of A/A₀ at 690 nm exceeds 1 for all 10 phosphates, signifying that each one enhances the LSPR impact of the oxygen vacancy-rich MoO_x nanosheets to various extents.

Fig. 3.

(A) Color images of MoO_x nanosheets doped with 10 different types of phosphates before and after UV irradiation. (B) UV–vis absorption spectra and a line chart of absorbance values at 425 nm for MoO_x doped with 10 different types of phosphates without light irradiation. (C) UV–vis absorption spectra and a line chart of absorbance values at 690 nm for MoO_x doped with 10 different types of phosphates after light irradiation. (D) Cross-reactive response of 10 different phosphates on the MoO_x system. (E) Fingerprint maps of the MoO_x system doped with 10 different types of phosphates. (F) Heatmaps of the MoO_x system doped with 10 different types of phosphates under a four-channel configuration.

Due to their strong absorption of ultraviolet light, differences in the structures of inorganic phosphates and nucleotide phosphates also affect the UV absorption of MoO_x nanosheets doped with these phosphates to some extent. Based on the absorbance values at 425 and 690 nm, a unique fingerprint map was established for each phosphate, as shown in Fig. 3E. To more clearly present the identified results, the corresponding absorbance values were converted into a heatmap, which exhibited excellent distinguishing capability, as shown in Fig. 3F. In addition to the absorbance values of 425 and 690 nm, the heatmap also revealed absorbance values at 590 and 740 nm, forming a four-channel system to further enhance the accuracy of phosphate identification. The heatmap illustrates that the introduction of different phosphates leads to unique colorimetric response patterns, emphasizing the efficiency of this approach for phosphate differentiation.

In addition, the phosphate colorimetric sensor based on MoO_x nanosheets has excellent stability. As shown in Fig. S6, MoO_x nanosheets maintained good stability in detecting phosphates within the pH range of 5–9. As the pH increased, the absorption peak at 690 nm slightly decreased. This is due to the fact that MoO₃ is an acidic oxide, which can form molybdate and lose its photochromic properties in alkaline solutions [42]. To systematically evaluate the long-term stability of the MoO_x nanosheet sensing system, we monitored the evolution of its optical properties and phosphate-binding capacity over different storage periods (0–210 days) in a room-temperature and light-shielded environment (Fig. S7). The results show that after 7 months of storage, the absorbance of MoO_x nanosheets at characteristic wavelengths (425 nm, 690 nm) does not change significantly, and the response signal to phosphate remains over 95% of that of fresh samples, with RSD% ≤ 1.5% for three parallel detections, confirming excellent structural and optical stability. This provides key support for practical application and commercialization. The phosphate colorimetric sensor based on MoO_x nanosheets exhibits good stability over a wide pH range and long-term storage, demonstrating its reliable detection performance in practical applications.

Compared to the previously reported fluorescent and colorimetric phosphate sensor arrays, the oxygen-vacancy-regulated MoO_x colorimetric array developed in this work shows clear advantages in material cost, system simplicity, and discrimination capacity (Table S3). By controlling oxygen-vacancy levels, we obtained differential spectral features, which allow accurate discrimination of 10 phosphate species, far exceeding the 3–5 analytes, typically distinguished by conventional colorimetric arrays. Although the target concentrations in our system are relatively high, each phosphate species exhibits distinct response signatures. The oxygen-vacancy-regulated MoO_x colorimetric array provides a practical and efficient alternative to conventional fluorescent/colorimetric sensor arrays for phosphate detection.

Further, linear discriminant analysis (LDA) was employed to process these data to better quantify the experimental results. By projecting the data onto a two-dimensional space through LDA, the between-class variance was maximized while the within-class variance was minimized, resulting in linear separation of the data. Utilizing absorbance values at 425 nm (indicating variations in UV absorption) and 690 nm (indicating variations in the plasmonic peak) as classification features could facilitate classification; however, the proximity of the categories hindered the differentiation of multiple concentrations and varied application contexts (Fig. S8A). By incorporating additional absorbance values at 590 and 740 nm (representing changes in the shape of the plasmonic peak) after illumination, the classification accuracy was enhanced. A training matrix was constructed with four discriminant elements, ten types of phosphates, and four replicates.

LDA converted the response data into standardized scores and separated the dataset into two principal factors, representing the degree of contribution to overall classification. At a phosphate concentration of 800 μM, the sensor array revealed excellent discrimination for the ten phosphates, with 90% confidence intervals that did not overlap and were precisely positioned (Fig. 4A). At 500 μM, effective discrimination was still possible, although the confidence intervals were broader (Fig. 4B). At 300 μM, the phosphates could still be distinguished, but some confidence interval ellipses overlapped (Fig. 4C). The MoO_x sensor array demonstrated effective discriminating of the 10 phosphates within the 300–800 μM range. At both higher and reduced phosphate concentrations, a minimum of eight phosphates could still be identified. Moreover, ATP and its hydrolysis products (ADP, AMP, PPi, and Pi), which play crucial roles in significant physiological processes in the human body, may be accurately differentiated within the 50–1000 μM range (Fig. S8B and Fig. S8C). Notably, variations in phosphorus content across nucleotide phosphates caused monophosphate nucleotides (AMP, GMP), diphosphate nucleotides (ADP, GDP), and triphosphate nucleotides (ATP, GTP, UTP, CTP) to cluster within their respective nucleotide categories.

Fig. 4.

LDA plot of ten phosphates at (A) 800, (B) 500, (C) 300 μM. LDA plot of mixtures (D) of binary and (E) (F) of ternary phosphates. (G) LDA plot of various concentrations of Pi. (H) LDA plot of various concentrations of PPi. (I) the linear regression of Factor 1 versus the logarithm of various concentrations of PPi. (J) Heatmaps of the MoO_x system doped with ten types of phosphates (100, 300, 500, 800 μM) under a four-channel configuration.

Next, the discrimination ability of the colorimetric sensor array for phosphate binary mixtures was further evaluated. To assess the array’s potential for identifying mixtures, experiments were performed with GTP and GMP at different molar ratios, attaining a total concentration of 600 μM. Fig. 4D clearly demonstrates the ability to differentiate between 600 μM GTP and GMP alone, as well as their combinations. The response signals for GTP and GMP were significantly dissimilar from those of the mixtures, and the certainty interval ellipse distance for samples containing GTP was closer. It is presumed that GTP has a stronger binding effect on metal ions, making signal changes more pronounced even with a small amount added to the system. Notably, ternary mixtures containing ADP, GMP, and PPi could be accurately identified (Fig. 4E). Similarly, ternary mixtures of UTP, AMP, and Pi could also be effectively distinguished. Based on the distribution of discrimination factors, mixtures containing 70% of a particular phosphate were found to be close to the 100% corresponding monophosphate, indicating that the sensor could significantly distinguish the dominant phosphate in the mixture and classify it accordingly.

To verify the quantitative recognition ability of the sensor array, Pi and PPi were selected as examples, given their roles as extracellular signaling agents and energy sources in biological processes. A fingerprint map was generated for various concentrations of Pi (Fig. S8D), and effective discrimination was achieved for Pi concentrations in the range of 50 to 1000 μM (Fig. 4G). As shown in the LDA plot, various concentrations of PPi (200–800 μM) were clearly separated, well distinguished and precisely classified (Fig. 4H). It was also observed that the concentrations in the LDA plots are not randomly arranged but regularly distributed from left to right. Furthermore, the projections of each group on the fractional graph (Factor 1) X-axis correlated well with the phosphate concentration (Fig. 4I).

To verify the model’s robustness and generalizability, a dual validation strategy was adopted. A stratified random train-test split yielded 100% precision, recall, F1-score, and overall accuracy for the independent testing set (Table S4-S9). The corresponding confusion matrix, generated from this validation, showed all diagonal values as 1 and off-diagonal values as 0, confirming complete classification of the 10 phosphate species with no missed detection or misclassification (Figs. S9-S11). Supplementary stratified 4-fold cross-validation confirmed stable performance with an average overall accuracy of 95.00% (Table S10-S11). Together, these results demonstrate the model’s high robustness and reliability, reflecting both its optimal recognition capability and consistent generalization across diverse sample partitions. To assess reproducibility, we calculated the coefficient of variation (CV%) for 4 repeated measurements (n = 4) across 10 phosphate classes. The average intra-class CV% of 4 wavelength-specific features was 3.2% (range: 2.0%–4.5%), and the overall CV% after mean normalization was 3.56%, both below the 5% standard threshold. All class-specific intra-class CV% values were < 3.8% (Table S12-S13), confirming the system’s excellent reproducibility and data reliability for quantitative analysis. In addition, the heatmap in Fig. 4J shows that lower colors represent lower absorbance values, which correspond to higher levels of phosphates in MoO_x. This heatmap shows how the addition of different phosphates produces different colorimetric response patterns.

To evaluate the biological application potential of the sensor, we supplemented systematic biocompatibility tests. As an oxide of an essential trace element in the human body, MoO_x has good dispersibility and chemical stability, and is a well-recognized low-toxicity biocompatible nanomaterial widely used in the biomedical field [[43], [44], [45]]. Quantitative evaluation was performed using the MTT assay with 4T1 cells as the model. After co-culturing 4T1 cells with MoO_x nanosheets at different concentrations (i.e., 0, 12.5, 25, 50, 100, 200 μg/mL) for 24 h, absorbance measurement at 450 nm showed that the cell viability remained above 90% at all concentrations, reaching 91.5% even at the high concentration of 200 μg/mL, confirming no significant cytotoxicity (Fig. 5A). The excellent biocompatibility of MoO_x demonstrate its potential for disease diagnosis and medical monitoring applications (Fig. 5B).

Fig. 5.

(A) Viability of 4T1 cells in culture medium with different MoO_x contents. (B) Schematic illustration of the application of the MoO_x sensor array. (C) LDA plot of five phosphates at 800, (D) 500, (E) 300 μM in serum. (F) LDA plot of five phosphates at 400, (G) 100 μM in cell lysate. (H) LDA plot of various concentrations of ATP. (I) The linear regression of Factor 1 versus the logarithm of various concentrations of ATP. (J) HCA plot of serum samples spiked with different concentrations of ATP.

The practicality of the designed sensor array for utilization in complex biological systems was established. Low phosphate concentrations in fetal bovine serum are difficult to identify because of the presence of different proteins, electrolytes, and endogenous phosphates. To evaluate the sensor array's performance in practical applications, we analyzed fetal bovine serum. These samples were first 100-fold diluted, then spiked with ATP, ADP, AMP, PPi, and Pi. As seen in Fig. 5C, D, and E, each phosphate had a unique reaction pattern. LDA revealed that the five phosphates were precisely classified into their respective clusters, with no obvious overlap, demonstrating the precision and specificity of the sensing method. Furthermore, different concentrations of GTP, GDP, GMP, UTP, CTP, five types of phosphates, also showed good discriminatory effects in serum (Fig. S12). Furthermore, the detection performance of the sensor for five common phosphate species including ATP was evaluated in a cell lysate environment derived from swollen 4T1 cells. The cell lysate matrix contained complex biological components such as intracellular proteins, nucleic acids, and endogenous small-molecule metabolites, which could potentially interfere with phosphate recognition. Despite such matrix complexity, LDA plots of the five phosphate species exhibited clear separation at concentrations of 400 μM and 100 μM (Fig. 5F and 5G), confirming the sensor’s robust anti-interference capability and reliable discrimination performance in complex biological matrices.

To verify the practical application reliability of the MoO_x nanosensor in complex biological matrices, spike-recovery experiments were performed on five target phosphates (ATP, ADP, AMP, PPi, Pi) in cell lysate and fetal bovine serum at 100 μM and 400 μM with 3 parallel replicates per group (Table 1). Quantitative performance was evaluated by recovery rate, bias, and relative standard deviation (RSD%). The results revealed that the recovery rates of the five phosphates ranged from 95.74% to 104.21% in the FBS matrix and 97.45% to 101.7% in the cell lysate matrix, all within the ideal quantitative range of 95%–105%. The absolute values of bias for all analytes were ≤ 4.21%, with ATP ≤ 1.63% and PPi ≤ 1.29%, and the RSD% of 3 parallel experiments ranged from 0.19% to 4.81% (all < 5%). These data confirm that the sensor exhibits excellent accuracy and reproducibility in complex biological matrices, with effective resistance to matrix interference for precise phosphate quantification. This feature provides key data support for the application of the sensor in biomedical research and clinical real-sample detection.

Table 1.

Determination of phosphates in real samples.

Sample		Added scalar quantity (μM)	Amount of measurement (μM)	Recovery rate (%)	bias (%)	RSD (%) (n = 3)
FBS	ATP	100	100.61	100.61	0.61	1.23
		400	393.49	98.37	−1.63	0.67
	ADP	100	101.95	101.95	1.95	0.32
		400	399.71	99.93	−0.07	0.31
	AMP	100	104.21	104.21	4.21	1.78
		400	402.62	100.66	0.66	0.73
	PPi	100	98.71	98.71	−1.29	0.67
		400	403.20	100.8	0.80	2.74
	Pi	100	95.74	95.74	−4.26	3.12
		400	388.53	97.13	−2.87	4.81
cell lysate	ATP	100	100.89	100.89	−0.89	0.88
		400	401.62	100.40	0.40	1.47
	ADP	100	99.54	99.54	−0.46	0.19
		400	404.17	101.04	−1.04	1.19
	AMP	100	98.68	98.68	−1.32	3.29
		400	406.78	101.7	1.70	0.71
	PPi	100	100.74	100.74	0.74	1.73
		400	400.12	100.03	0.03	1.45
	Pi	100	99.82	99.82	−0.18	0.77
		400	389.79	97.45	−2.55	0.86

Consider that ATP serves as the primary energy source for various metabolic activities and physiological functions in living organisms, the discrimination and detection of its different concentrations by biosensor arrays hold significant importance. Utilizing LDA on serum samples infused with varying ATP concentrations, ATP levels between 300–700 μM may be clearly identified, exhibiting a distribution pattern that advances from left to right with increasing concentration (Fig. 5H). According to relevant literature reports [16], the ATP concentration of emergency patients at admission ranges from 250 to 700 μM, which aligns closely with the 300–700 μM detection scope of our sensor array, confirming its clinical applicability. By extracting Factor 1 to construct an ATP concentration-Factor 1 plot, a good linear relationship was observed within the 300–700 μM range (Fig. 5I). Hierarchical cluster analysis (HCA) utilizes the whole information derived from the multivariate response signal [46]. Utilizing the clustering of the array response data, the HCA created a dendrogram which clearly illustrated the clustering of serum and varying amounts of ATP (Fig. 5J). When LDA and HCA were applied to the MoO_x-based sensing data, both analytical approaches achieved remarkable performance in detecting and identifying ATP across a range of concentrations, underscoring the performance of the MoO_x sensor array in phosphate target analysis.

Conclusion

In summary, this study proposes for the first time a multi-channel colorimetric sensing method based on oxygen-vacancies modulation to broaden the scope of phosphate detection and recognition within sensing systems and to further reduce the fabrication processes and costs related to the sensor arrays. A colorimetric sensor array based on MoO_x nanosheets was designed and fabricated, which performed effectively not only in detecting and distinguishing ten different types of phosphate anions, but also in quantitative analysis. Further research indicates the optical properties of MoO_x nanosheets, which are affected by the quantity of oxygen vacancies, shift to various degrees in plasmonic resonance absorption intensity and UV–visible absorption range with the addition of phosphates. Furthermore, this array approach has high sensitivity, excellent interference resistance, and promise in complex fluids such as serum. The sensor array is prepared by regulating oxygen vacancies and achieving high-sensitivity colorimetric sensing, providing a new technique for detecting and identifying phosphates and other chemically/property related analytes. Meanwhile, it provides new design concepts and technical approaches for the development of transition metal oxides in sensing applications.

Compliance with ethics requirements

This article does not contain any studies with human or animal subjects.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: