Cetyltrimethylammonium chloride (CTAC), triethanolamine (TEA), 1H,1H,2H, 2H-perfluorodecanethiol, pyridine, benzyl chloride, and cyclohexane were purchased from Sigma-Aldrich Corp. Sodium chloride, sodium hydroxide, tetraethyl orthosilicate (TEOS), methanol ethanol, bovine serum albumin (BSA) proline, mannitol, lysine, and sucrose were purchased from Adamas Reagent Ltd. All the above chemicals were used as received without further purification. ITO slides were purchased from Luoyang Guluo Glass Co., Ltd. These ITO slides were first treated with NaOH solution (2 M) at room temperature for 12 h to clean the organic residues and then stored in deionized water for the subsequent experiments. Gold sputtering material (99.99%) was purchased from Hefei Kejing Auto-instrument Co., Ltd, China. Deionized water (18.2 MΩ cm) was obtained by a Milli-Q system (Millipore, United States) and used for all experiments in this work.

Ion desorption efficiency was gauged by the chemical thermometer (benzyl pyridinium (BP)). 100 μL of Benzyl chloride was mixed with 2 mL pyridine and stirred in a 50 mL round-bottomed flask at 60°C for 5 h. Then the excess pyridine was removed by rotary evaporation. The end-product was redissolved in 1:1 (v/v) MeOH: H₂O to prepare a 0.1 mM [BP]⁺ solution.

When the internal energy of [BP]⁺ (parent ions) is larger than the dissociation reaction critical energy, the parent ions would undergo bond cleavage: B P +   →     B P − p y r i d i n e   + + p y r i d i n e m / z 170   m / z 91  

The total intensity of BP ions includes the intensity of parent ions and [ BP—pyridine]⁺ (fragment ions). A higher total ion intensity indicates a higher desorption efficiency. The survival yield (SY) of parent ions is used to gauge the degree of parent ions fragmentation from internal-energy transfer. S Y = I m / z   170   /   I m / z 170 + I m / z 91 × 100 %

Scanning electron microscopy (SEM) images of chips were obtained from an S4800 field emission scanning electron microscope (Hitachi, Japan). Elemental mapping and energy-dispersive X-ray spectroscopy (EDX)of chips were obtained from a GeminiSEM450 field emission scanning electron microscope (Zeiss, Germany). Transmission electron microscopy (TEM) and selected area electron diffraction (SAED) were obtained from a JEM-2100F transmission electron microscope (JEOL, Japan). Digital images were captured by iQOO three plus (VIVO, China). Microscope images were captured by an FV3000 confocal laser scanning microscope (Olympus, Japan). The contact angle was detected by DSAeco device (KRUSS GmbH, Germany) using 5 μL of water. Inductively coupled plasma optical emission spectrometer (ICP-OES) data was collected by Agilent 5,100 (Agilent Technologies Inc, United States).

All of the research protocols in this study were approved by the institutional ethics committees of Tianjin Third Central Hospital Affiliated with Nankai University and East China Normal University, School of Chemistry and Molecular Engineering. Written informed consent from patients had been obtained since the project started. Urines samples were donated by patients from Tianjin Third Central Hospital Affiliated with Nankai University and stored in tubes at −80°C before use. All kidney stone (KS) patients recruited were eligible and precisely diagnosed based on ultrasound and computed tomography. The urine samples of healthy control (HC) were collected from individuals who were not diagnosed with KS through health examinations. Patients with other medical conditions (such as active bleeding) were excluded. To distribute the bio-fluids on the surface of the prepared chips, GeSim Nano-Plotter TM 2.1 was applied. Each sample spot volume was 1 µL by running the designed printing procedure in the non-contract model. After drying at room temperature, bio-sample on the as-printed chips was detected by LDI MS without ant treatment. 1 µL of standard analytes and prepared mixtures were dissolved in deionized water and spotted on the prepared chips. 1 µL of urine samples were directly spotted on the prepared chips.

The initiator/gold/mesoporous silica membrane chips (IGMSM chips, meanwhile, denoted mesoporous silica membrane chips as MSM chips and gold/mesoporous silica membrane chips as GMSM chips.) were fabricated through a three-step process on ITO slides including mesoporous silica membranes synthesis, gold sputtering, and initiator loading, facile for ion signal production in LDI-MS (Scheme 1A). The prepared chips with optimized parameters have a uniform and suitable hydrophobicity structure to achieve high reproducibility in practical use (Figure 1A). Meanwhile, the chip is compatible with microarray printing, enabling high-throughput metabolic analysis with bio-samples (Figure 1B). The designed chip can simplify the sample pretreatment process and achieve laboratory automation to realize ideal clinical use. Specifically, the mesoporous silica membranes were evenly distributed on the ITO slides using a biphase stratification growth method reported in the previous research (Liu et al., 2017) to provide substrate for supporting and heat insulation. From the top-view SEM (Figure 1C), the uniform large-domain mesoporous silica membrane has an average pore size of ∼11 nm. Then the gold nanoparticles (AuNPs) were deposited on the silica layer by sputtering as the matrix for LDI-MS. Compared to the MSM chip, gold islands were clearly observed on the top of the GMSM chip from the top-view image of SEM, indicative of the successful sputtering of Au nanoparticles (Figure 1D). The cross-section image of SEM (Figure 1E; Supplementary Figure S1) also showed the silica layer and gold layer on the surface of the ITO slide with a thickness of ∼17 nm and ∼11 nm, respectively. The HRTEM image showed clear fringes with an interplanar distance of 0.203 nm (Figure 1F), which can be ascribed to the (200) plane of Au crystal (JCPDS 00–004–0,784). The SAED pattern gave diffraction spots that correspond to the (111), (200), (220), and (311) lattice planes from Au crystal (Figure 1F). It is well known that AuNPs can be functionalized with a variety of ligands by performing gold - thiol reactions with straightforward synthetic protocols (Häkkinen, 2012; Nyamekye et al., 2021; Zhang et al., 2021). For IGMSM chips, we finally functionalized the surface of AuNPs with 1H,1H,2H, 2H-perfluorodecanethiol chains via gold-sulfur coupling, leading to the formation of a highly ordered perfluorinated nanostructured monolayer. No obvious difference can be seen from the SEM images of the chip before and after initiator loading (Figure 1D; Supplementary Figure S2). The contact angle increased from 11.5 to 91.8 through the distribution of the initiators on the gold surface (Figure 1A). The as-made initiators not only improve the non-thermally desorption but also adjust the hydrophobicity of the chip surface for sample droplets spread. From the overlaid element mapping image of the chip, we can observe a clear distribution boundary between the region of non-initiator and initiator (Figure 1G). The EDX spectra of the initiator and non-initiator region further confirm the presence of fluorine from the initiator (Figure 1H; Supplementary Figure S3; Supplementary Figure S4).

We controlled the conditions of mesoporous silica membrane synthesis, gold sputtering, and initiator loading to optimize the structure parameters of IGMSM chips for excellent LDI performance. For the silica membrane, two parameters are possible to be adjusted including membrane thickness and pore size. Of note, the influence of membrane thickness on thermal conductivity has been investigated in the literature, suggesting that thermal conductivity is independent of thickness. (Coquil et al., 2009). On contrast, mesoporous materials with high porosity exhibit excellent thermal insulation for the desorption process in LDI MS (Nayak and Knapp, 2007; Xiao et al., 2009), and it has been confirmed that the porosity of the mesoporous silica membrane in this study is positively correlated with pore size (Yan et al., 2019). Hence, we tuned the pore size of the mesoporous silica membranes by varying the ratio of TEOS and cyclohexane (v/v %) to get the optimal porosity for thermal insulation. Briefly, the pore size is inversely correlated with TEOS amount. With a smaller content of the TEOS (5%), the mesoporous silica membranes have an average pore size of ∼11 nm (Figure 2A; Supplementary Figure S5), larger than the average pore size of ∼8 nm for 10% (v/v) (Figure 2A). While further increasing the content of TEOS to 30% (v/v), the resultant pore diameters decrease to hardly seen on the chip (Figure 2A). We opted to further decrease the TEOS content below 5%, while the pore became very un-uniform (Supplementary Figure S6). To investigate the impact of the pore size on LDI efficiency, we loaded the three silica chip substrates (denoted chip_5%/10%/30%) and bare ITO slide with the same amount of gold and initiator for comparison (Supplementary Figure S7; Supplementary Table S1). Then, we conducted LDI-MS detection of three metabolites (proline, lysine, and mannitol) (triplicate results in Figure 2D and typical spectra in Supplementary Figure S8), and chip_5% afforded the highest peak intensity and signal-to-noise ratio (p < 0.05), indicating that the pore size is positively correlated with the LDI efficiency. Furthermore, the peaks of three amino acids on bare ITO slide were hardly observed (Supplementary Figure S8), suggesting that the mesoporous silica membrane modification enhanced the LDI efficiency, possibly due to better laser energy transfer through the excellent thermal insulation. The detailed thermal-driven desorption process on the mesoporous silica membrane will be discussed later.

The area density of gold on-chip decides the average interparticle distance for plasmon coupling, which is involved in the process of hot carrier production, local heating, and photodesorption for plasmonics enhanced LDI MS. Therefore, we optimized the gold density on the optimal mesoporous silica chip by changing the sputtering time of 30s/45s/60s (denoted chip_{g30s/g45s/g60s}). Through the SEM images of the chips with different sputtering times, we observed the increased content of gold and more narrow gaps with longer sputtering time (Figure 2B). We recorded corresponding ICP-OES of the three plasmonic chips (Supplementary Figure S9), in which the gold content increased with the increase of the sputtering time (5.62 μg/cm² for chip_g30s, 6.32 μg/cm² for chip_g45s, and 6.65 μg/cm² for chip_g60s, respectively). From the LDI-MS detection of proline, lysine, and mannitol, we demonstrated that the optimized chip_g45s with a gold area density of ∼6.32 μg/cm² afforded specific nanogaps and surface plasmon resonance for highly efficient analysis (triplicate results in Figure 2E and typical spectra in Supplementary Figure S10). The weak ionization efficiency in chip_g30s was mainly attributed to the insufficient Au content on the surface. While overloading of gold on the silica membranes (chip_g60s) also affected their LDI performance, possibly due to the undesired plasmon coupling arising from the overaggregation of gold nanoparticles (Sun et al., 2018; Shu et al., 2020). Furthermore, the specific nanogaps and nanocrevices of gold on-chip_g45s may selectively trap small metabolite molecules and transfer the laser energy, toward advanced metabolic analysis of complex biosamples in real cases.

Controlling the surface wettability of the chip is essential for the microscale sample analysis, preventing the lateral spreading of the droplet and localizing the analytes with the desired enrichment. Wettability-controlled IGMSM chips were constructed by adjusting the loading time of the highly hydrophobic initiator (1H,1H,2H, 2H-perfluorodecanethiol) on the optimal gold/mesoporous silica chip, from 0s, 30s, 60s–90s (denoted as chip_{i0s/i30s/i60s/i90s}). This perfluoro-initiator trapped in the gold islands can be released when heated by the laser, promoting the generation of intact molecular ions while producing lower background signals (Northen et al., 2007; Kurczy et al., 2015; Palermo et al., 2018). In general, the hydrophobicity increased with the increasing of the initiator amount on the chip. The contact angles of the chip_{i0s/i30s/i60s/i90s} were measured to be 11.5, 57.2, 91.8, and 111.1°, respectively (Figure 2C; Supplementary Figure S11). We used these chips to conduct LDI-MS detection of proline, lysine, and mannitol (triplicate results in Figure 2F, typical spectra in Supplementary Figure S12), and the chip_i60s with the contact angle of 91.8 provided the highest peak intensity than the others. Insufficient loading of the initiator remained a hydrophilic surface, leading to the excessive spread of a droplet of aqueous sample solution on the chip and poor reproducibility (Supplementary Figure S13). Overloading the initiator on the IGMSM chips also reduced the analytical efficiency, owing to strong hydrophobicity causing thick samples layer to prevent initiator-assisted desorption of analytes (Woo et al., 2008) (Supplementary Figure S14) and the contact between analytes and gold matrix. Through the three-step optimization, the limit of detection (LOD) of the optimized IGMSM chips in detecting metabolites was down to ∼2 pmol (Supplementary Figure S15) and the relative standard deviations was down to 4.3% (4.4% for proline, 8.7% for lysine and 4.3% for mannitol) when coupled with the routine LDI TOF MS. Hence, compared with previously reported matices and chips, the optimized IGMSM chips with low thermal conductivity, surface plasmon resonance effect, and well wettability enabled the metabolic detection of analytes with desired sensitivity and reproducibility (Supplementary Table S1). Meanwhile, the background signal of the bare IGMSM chip was very low, so it would not interfere with the detection of the analytes (Supplementary Figure S16).

To better interpret the enhanced LDI performance, we investigated the detailed desorption mechanisms of the IGMSM chip, using BP salts as the “chemical thermometer” (Tang et al., 2009). The mechanistic studies of gold and initiator have been reported by many researchers (Palermo et al., 2018), while the effect of different mesoporous silica membrane substrates on the desorption efficiency remains to be elucidated. With this regard, we chose the IGMSM chips with three different mesoporous silica membranes (chip_5%/10%/30%) to reveal the possible desorption mechanisms in our design. The detailed definition, calculation of desorption efficiency, and SY are provided in the part 2.3. We plotted desorption efficiency and SY of chips against different porosity of silica membranes. Briefly, the parent ion [BP]⁺ can be dissociated to fragment ion [BP - pyridine]⁺ in the LDI process, and the desorbed BP (a summation of the parent and fragment BP ions) is positively correlated with the desorption efficiency. A plot of the ion intensity of desorbed BP against the pore size of the chip is shown in Figure 3A (red). Compared with bare ITO slide with the same matrix formula, the chip_5%/10%/30% with mesoporous silica membrane as the substrate all exhibited a higher ion desorption efficiency under the same laser intensity, in which the chip_5% with the largest pore size afforded the best ion desorption efficiency. To rationalize the effect of different pore sizes on the ion desorption efficiency, the extent of internal energy transfers to BP ions generated by LDI was determined from the SY of the parent BP ions. A plot of the SY against the pore size is illustrated in Figure 3A (blue), in which the SY of parent BP ions decreased with the increase of the silica substrate pore size. Of note, the lower SY equals higher internal energy transfer. Hence, the desorption efficiency (chip_5% > chip_10% > chip_30% > ITO) in general exhibits a consistent trend to the extent of internal energy transfer (chip_5% > chip_10% > chip_30% > ITO), suggesting that increasing the extent of internal energy transfer in the LDI process enhanced the ion desorption efficiency. This phenomenon can be explained by an enhanced thermal desorption mechanism in the chip_5%, indicating that the increase of the pore size of the silica substrate is positively associated with thermal-driven desorption.

It is reported that the initiator owns a significant effect on the non-thermal driven desorption, as it can perform a phase-transition when heated by the laser, facilitating the desorption of the intact analytes (Northen et al., 2007; Palermo et al., 2018). Therefore, the non-thermal driven desorption would produce more parent BP ions rather than the fragment ion (Tang et al., 2009; Jiang et al., 2021). As shown in Supplementary Figure S17, the survival yield of parent BP is much higher in the IGMSM chip compared to the GMSM chip without the initiator, indicative of the enhanced non-thermal driven desorption through the addition of the initiator. More interestingly, the ion intensity of parent BP desorbed from the different silica substrates with the same amount of gold and initiator are in the order of chip_5% > chip_10% > chip_30% > ITO, suggesting that an enhanced non-thermally driven desorption also existed when the pore size increased (Figure 3B). We then proposed a possible enhanced desorption mechanism for the IGMSM chip: 1) in rapid laser-induced heating of the IGMSM chip, the mesoporous silica substrate with high thermal insulation well reserved the energy, causing a higher chamber temperature in the substrate layer; 2) the trapped heat energy not only bring in a better internal energy transfer for the enhanced thermal desorption but also generated more violent phase-transition of the gold (Supplementary Figure S18) and initiator for enhanced non-thermal desorption (Figure 3B; Figure 3C).

To verify the feasibility in real-case bio-fluids, we first tested the salt tolerance and protein endurance of the optimized IGMSM chip for low-abundance metabolite detection. A mixture containing four small metabolites (proline, lysine, mannitol, and sucrose, 1 mg/mL each), salts (0.5 M NaCl), and proteins (5 mg/mL bovine serum albumin (BSA)) were used to simulate the real sample environment. The characteristic peaks of proline ([M + Na]⁺ [M + 2Na - H]⁺), lysine ([M + Na]⁺ [M + 2Na - H]⁺), mannitol ([M + Na]⁺), and sucrose ([M + Na]⁺) can be detected on the optimized IGMSM chip, highlighting its selectivity and sensitivity for analysis of small metabolites in complex biofluids (Figure 3D). We then achieved the direct metabolic fingerprinting of 1 μL of urine without any pre-treatments. We collected 50 independent metabolic patterns for one KS patient’s urine sample and plotted the heat map, showing that the metabolite signals were distributed vertically and uniformly in a given m/z range (Figure 3E). Notably, the features of m/z 135.93 and m/z 151.91 were normally distributed (p > 0.05, Supplementary Figure S19) at a 5% significance level for one urine sample with (Liu et al., 2017) independent patterns respectively, validating the reproducibility of the metabolic pattern extraction (Huang et al., 2020). The above results indicated the reliability and potency of the urine metabolic patterns obtained with IGMSM chip-assisted LDI MS for diagnostic applications.

We differentiated 44 KS patients from 45 HCs by on-chip metabolic analysis of urine (Figure 4A; Supplementary Table S1), and there was no significant difference in age distribution between the two cohorts (p = 0.3654). The typical mass spectra of KS patients and HCs are shown in Figure 4B, in which multiple peaks with distinct differences can be observed in both cases. We recorded the metabolic m/z signals of KS patients and HCs in the low mass range (m/z of 100–1,000) from urine by the IGMSM chip. We extracted urine metabolic fingerprints (UMFs) from the above 89 KS patients and HCs by preprocessing the original mass spectra. Specifically, the 89 UMFs within m/z range from 100–1,000 signals were further organized as the blueprint in Figure 4C, serving as the database for building a diagnostic model. To better elucidate these data, OPLS-DA was applied and displayed the clear group separation based on the UMFs (Supplementary Figure S20, R²Y (cum) = 0.883, Q² (cum) = 0.816, p < 0.005). Then, we built a diagnostic model and obtained an AUC value of 0.997 with a sensitivity/specificity of 0.9822/1.0000 for the train cohort. (Figure 4D). To elucidate the diagnostic performance of the established model, we conducted blind tests of 30 subjects (KS/HC, 15/15, Supplementary Table S1) with an AUC value of 0.912 with a sensitivity/specificity of 0.8667/0.8267 (Figure 4D). The classification results were quantitatively summarized in a confusion matrix (Figure 4E), and an accuracy of 82.67% was obtained for classifying KS patients versus HCs. Permutation tests with 200 iterations were performed for validating the supervised model to avoid overfitting (Supplementary Figure S20). Therefore, the established KS diagnostic model by urine metabolic fingerprint can successfully differentiate KS patients and HCs, proving the established on-chip in vitro metabolic screening enjoys high throughput and considerable accuracy.

KS is a common disease in urology and lays a heavy economic burden on patients worldwide (Lieske et al., 2006; Rule et al., 2020; Abufaraj et al., 2021). The diagnosis of the KS is mainly based on ultrasonography (Pearle Margaret et al., 2014; Bultitude et al., 2016), and the etiology of the disease remains unclear (Alelign and Petros, 2018; Howles and Thakker, 2020; Halbritter, 2021). Thus, how to prevent stone formation and recurrence, as well as predicting its potential risk, remains a big challenge (Hyams and Matlaga, 2015). Urine tests promise easy and accurate diagnosis of KS for the simple measurement using urine samples and low costs for point-of-care testing, which is non-invasive and facile for universal applications (Yang et al., 2020; Ding et al., 2022). Furthermore, the biological predictors from urine can better interpret the underlying molecular mechanisms for etiology (Maiuolo et al., 2016; Noone et al., 2018; Ståhl et al., 2019). Considering the metabolic disturbances in KS as demonstrated by previous reports (Khan et al., 2016; Sorokin et al., 2017), our results offer new insights for in vitro diagnosis of KS by direct metabolic fingerprinting of urine. Though many studies also attempted to construct the MS-based platform for the KS differentiation (Liu et al., 2012; Chao et al., 2018; Wang et al., 2019), our established on-chip LDI MS platform is highlighted due to the minimum sample pre-treatment and the convenience for large-scale screening. In traditional MS-based metabolomics, tedious sample pre-treatments (at least 0.5–1 h for each sample) are indispensable to address the low abundance of metabolites and interference of salts and proteins in bio-samples. These time-consuming procedures may cause sample information loss and high overhead, hindering the universal application of conventional MS techniques in clinics. By comparison, our approach directly profiled urine metabolites (∼1 μL) with the enhanced LDI efficacy in seconds, exhibiting high efficiency and feasibility for real clinic use.