Sensitive, high-throughput, metabolic analysis by molecular sensors on the membrane surface of mother yeast cells

MOMS fabrication

To create MOMS, yeast cells were initially treated with sulfo-NHS-LC-biotin to biotinylate the proteins on the cell wall. This was followed by the attachment of streptavidin and biotin-bearing DNA aptamers. The charged sulfonyl group ensured that the biotinylating reagent remained impermeable to the cell membrane⁴¹, providing adequate grafting sites exclusively on the cell surface. The aptamer sequences were explicitly designed to target a range of molecules, including ATP, glucose, vanillin, and zinc ion (Zn²⁺)^42–45, imparting flexibility to the MOMS for capturing various secreted metabolites and proteins. The MOMS coating was visualized by anchoring Cy5-labeled aptamers (excitation: 646 nm, emission: 664 nm) on the yeast cells. We stained cell walls with Alexa Fluor 488-Concanavalin A (ConA, excitation: 495 nm, emission: 520 nm). Confocal laser scanning microscopy (CLSM) images demonstrated that the aptamers were localized exclusively on the cell surface, with no detectable impact on yeast cell physiology or secretion profiles (Supplementary Fig. 3). We confirmed that the intrinsic functions of the cells were preserved. After the MOMS coating procedure, the viability of yeast cells was assessed by feeding them fluorescein diacetate (FDA), which is converted into a fluorescent signal by esterase activity in live cells⁴⁶. A high viability fraction (>93%) was observed through CLSM imaging and flow cytometry analysis (Supplementary Fig. 4). Notably, yeast cells functionalized with the MOMS retained their ability to proliferate and secrete molecules normally, comparable to native cells. We confirmed that the MOMS modification did not adversely affect cellular health or functionality (Supplementary Fig. 5).

Selective dense MOMS coating on mother yeasts

MOMS were coated on yeast surfaces via functional biotin groups binding to biotin-labeled aptamers as a part of the streptavidin sandwich (Fig. 2a). The selective coating of MOMS on mother yeast cells was also validated by Cy5-labeled aptamer coating followed by Alexa Fluro 488-labeled ConA dying, confirming exclusive anchoring to mother cells during proliferation. The MOMS coating remained confined to the original mother cells as daughter cells budded with newly synthesized membranes. This ensured a high sensor density and produced distinct fluorescence signals, differentiating mother and daughter cells (Fig. 2b).

To evaluate MOMS density on the yeast surface, cells at a fixed concentration (2.0 × 10⁷ cells/mL) were incubated with varying aptamer concentrations in a 100 µL medium, followed by flow cytometry analysis (Fig. 2c) and plate-reader calibration (Supplementary Fig. 6)⁴⁷. Increasing the aptamer concentration from 0.5 µM to 1.5 µM led to a proportional rise in fluorescence intensity, indicating a higher sensor density. A saturation point was reached at approximately 1.4 × 10⁷ molecules per cell, beyond which fluorescence remained constant, suggesting full occupancy of available binding sites on the yeast surface. The stability of MOMS coatings on mother cells was further validated via flow cytometry (Fig. 2d). Mother cells coated with fluorophore-labeled MOMS were distinguished from daughter cells by appropriate gating strategies (Supplementary Fig. 7). MOMS selectively coated mother yeast cells without transferring to daughter cells during budding, maintaining this exclusivity over three days and more than 30 generations⁴⁸. Many mother cells retained dense MOMS coatings, as indicated by Cy5 fluorescence signals, while daughter cells remained uncoated.

Notably, after prolonged yeast budding (12 hours), a small percentage of MOMS ( ~16.4%) detached, causing a slight decrease in fluorescence signal. One possible reason is that parts of the mother cells began the budding process before MOMS coating; the MOMS-coated buds then gradually grow and detach from the mother cell surface due to cytokinesis (Supplementary Fig. 8). However, the majority (~83.6%) remained on mother cells, ensuring high stability for long-term secretion measurements. A high MOMS density (~1.0 × 10⁷ molecules per cell) was maintained on mother cells even after 72 hours, despite a reduced ratio of mother cells in the population due to extensive proliferation (Fig. 2e). CLSM images of mother yeast cells coated with Cy5 labeled aptamers also revealed that multiple generations of budding did not significantly affect the overall MOMS density on yeast cell surfaces (Supplementary Fig. 9). It was observed that, although the proportion of mother cells in the population decreased significantly over time, the density of MOMS on the surfaces of mother yeast cells exhibited only a slight reduction, ensuring reliable long-term capturing of secreted molecules. The MOMS coating did not adversely affect yeast metabolic activity or proliferation, preserving the cells’ natural physiology to ensure their secretion profiling capability.

Sensitive molecular measurements via MOMS

By grafting dense MOMS onto yeast cells, we developed a highly sensitive assay capable of detecting target compounds through fluorescence signaling generated by structure-switching aptamer sensors. Initially, the fluorescent probe-labeled aptamer was quenched by a complementary DNA (cDNA) labeled with a quenching probe. Upon binding to target compounds, the aptamer underwent a structural change, dissociating from cDNA and emitting a fluorescence signal⁴⁹. With dense sensor coatings, the MOMS assay enables the detection of low-abundance metabolic secretions, overcoming limitations in current sensing technologies (Fig. 2f).

To validate assay specificity, vanillin-targeting aptamers were tested against structurally related compounds, including vanillyl alcohol, vanillic acid, and ferulic acid. MOMS generated a 3.86-fold fluorescence signal in response to vanillin, while the other compounds produced minimal signals (1.06–1.18-fold), resulting in a specificity ratio of 3.45-fold (Supplementary Fig. 10a). Similarly, for glucose specificity testing, MOMS exposed to 1 M glucose exhibited a strong fluorescence increase (~3-fold), whereas fructose and galactose triggered only minimal responses, confirming the aptamer’s high molecular recognition specificity (Supplementary Fig. 10b). In the case of ATP detection, ATP-specific MOMS produced a robust fluorescence increase (~4-fold) upon exposure to 1 mM ATP, while structurally similar nucleotides—GTP, UTP, and CTP—elicited negligible fluorescence signals (Supplementary Fig. 10c), further demonstrating the platform’s excellent selectivity.

The dense MOMS coatings enabled the detection of vanillin at concentrations as low as 100 nM, demonstrating a low detection limit. Similarly, detection limits for ATP and glucose approached 100 nM and 1 µM, respectively (Fig. 2g). Compared to conventional droplet assays employing free-floating aptamer sensors, the MOMS system significantly enhanced local sensor density, achieving ~1.4 × 10⁷ MOMS units per yeast cell. This enhancement led to a more than 10-fold increase in sensitivity compared to droplet-based living cell sensors and a more than 100-fold increase compared to aptamer sensors in droplets. To confirm the sensitivity improvement provided by surface-immobilized MOMS, we tested ATP aptamer sensors of the same structure in a freely dissociated state. These sensors (100 nM or 2 µM) were co-incubated with uncoated yeast cells (2.0 × 10⁷ cells/mL), a blocking agent (0.1 mg/mL herring sperm DNA), and varying concentrations of ATP. Fluorescence signals were measured using a microplate reader (Supplementary Fig. 11).

In contrast to the distinct fluorescence increases observed with MOMS, freely suspended aptamers exhibited fluctuating signal differences between ATP concentrations ranging from 100 nM to 10 µM. This lack of sensitivity in suspended aptamer sensors is likely due to non-specific interactions between the aptamers and cell surfaces, which generate high background signals. In contrast, MOMS immobilization on cell surfaces minimizes non-specific binding, enhancing stability and significantly improving sensitivity. Notably, cell-free droplet systems—comprising enriched cellular extracts with functional components such as tRNAs, nucleotides, and cofactors⁵⁰—enable mutation screening without membrane structures, thereby reducing background noise and improving detection sensitivity. However, their encapsulation efficiency for individual mutant DNA templates remains relatively low (~5–20%)^51–53. Moreover, most cell-free platforms are highly miniaturized and not yet scalable for industrial biomanufacturing⁵⁴. Even when promising biosynthetic pathways are identified in vitro, they must ultimately be reintroduced into living cells for functional validation⁵⁵.

The sensitive detection capabilities of MOMS enabled the visualization of secretions from single Saccharomyces cerevisiae cells, as secreted compounds were directly captured by MOMS anchored to yeast cell surfaces (Supplementary Fig. 12a). ATP secretions over time were visualized using CLSM, imaging at 5-minute intervals over 40 minutes (Supplementary Fig. 12b). Upon capturing secreted ATP, MOMS aptamers underwent structural changes that triggered Cy5 fluorescence signals (excitation at 646 nm; emission at 664 nm). Yeast cells dispersed in a diluted culture medium (~5.0 × 10⁵ cells/mL) minimized cross-contamination through diffusion, ensuring that their own MOMS immediately captured secreted ATP molecules upon release. Under normal culturing conditions, even low-abundance ATP secretions by single yeast cells over 15 minutes were distinguishable from those observed at 0, 1, 5, and 10 minutes (Fig. 2h). Flow cytometry further validated the temporal increase in ATP secretion across individual yeast cells (Supplementary Fig. 12c). The high-sensitivity and temporal resolution of MOMS provided insights into secretion dynamics, enabling the visualization of individual yeast cell metabolic activities with unparalleled precision.

Sensitive multiplexed assay via combinational MOMS

By coating yeast cells with different MOMS, each featuring distinct aptamer sequences, we detected multiple target metabolites simultaneously for multiplexed assays (Fig. 2i). To evaluate the sensing capability, MOMS-coated yeast cells were incubated with vanillin, ATP, glucose, or Zn²⁺ under various conditions over time (Fig. 2j, Supplementary Fig. 13). For the vanillin assay, fluorescence signals significantly increased within 1 hour of incubation and saturated after 2 hours. At a vanillin concentration of 1 mM, a 3.82-fold signal amplification was observed, with a dynamic detection range of 100 nM−1 mM. For ATP detection, fluorescence signals exhibited a 9.81-fold amplification at 4 mM, with saturation achieved within 2 hours and a dynamic range of 100 nM−4 mM. The glucose assay showed gradual signal plateauing after 2 hours with a fluorescence amplification of 3.00-fold at 1.5 M and a dynamic detection range of 0−2 M. The Zn²⁺ assays plateaued after 3 hours with fluorescence amplification of 2.78-fold at 2 mM. These results demonstrate the MOMS system’s versatility for high-sensitivity detection of a wide range of small molecules.

Three distinct types of MOMS were simultaneously grafted onto the surfaces of the same yeast cells, each functionalized with aptamers specific to vanillin, ATP, and glucose, enabling a multiplexed assay (Fig. 2k, Supplementary Fig. 14). The combinatorial MOMS system produced distinct fluorescence signals for each molecule: ATP (Cy5: excitation 646 nm, emission 664 nm), vanillin (FITC: excitation 490 nm, emission 525 nm), and glucose (Cy3: excitation 550 nm, emission 570 nm). When exposed to a mixed solution containing all three targets, MOMS-coated yeast cells emitted clear, distinguishable fluorescence signals for each compound. Flow cytometry analysis further demonstrated the system’s capacity to differentiate yeast populations based on fluorescence profiles, effectively isolating cells with multiple signals from negative controls with low fluorescence levels. Heatmap analysis of various target compound mixtures also affirmed the MOMS assay’s accuracy, achieving 100% specificity in multiplexed measurements (Fig. 2l).

High-throughput screening using MOMS

Our high-throughput screening process consisted of two steps: capturing and screening (Fig. 3a). Step 1: Capturing: We employed droplet microfluidics to encapsulate individual MOMS-grafted yeast cells. Yeast cells coated with MOMS were suspended in YPD medium containing ferulic acid as a vanillin substrate and encapsulated in ~33-pL water-in-oil droplets stabilized with a biocompatible surfactant. A cell concentration of ~1.5 × 10⁶ cells/mL ensured a single-cell encapsulation rate of <5%, with a co-encapsulation probability of <0.13% (Poisson distribution). Yeast cells were incubated in droplets for approximately 48 hours, during which mother cells proliferated into daughter cells (Supplementary Fig. 15). MOMS on mother cells maintained a high sensor density, unaffected by budding, while secreted metabolites were efficiently captured within the droplets, generating strong fluorescence signals that reflected secretion profiles. Step 2: Screening: Droplets were disrupted to extract mother yeast cells for screening via flow cytometry. Unlike conventional droplet-based secretion assays, which are limited by throughput constraints (~10–200 Hz, at a cell density λ of 0.05), our MOMS-based approach surpassed droplet encapsulation limitations, achieving throughput exceeding 10³–10⁴Hz—a 10–100-fold improvement (Fig. 3b).

To demonstrate the case, we engineered S. cerevisiae strains for the production of vanillin. The synthesis of vanillin is catalyzed by 4CL and ECH, while native pathways convert vanillin into vanillyl alcohol or vanillic acid (Supplementary Fig. 16)^56–58. To increase vanillin secretion, we generated a series of engineered yeast strains containing plasmids with the 4cl and ech genes, including VAN-1 (gre2 deleted), VAN-2 (adh6 deleted), and VAN-3 (adh6 and gre2 deleted). These deletions partially obstructed the vanillin reduction pathway. Bulk culturing with 1.0 g/L ferulic acid for 48 hours yielded vanillin at concentrations of 9.6 mg/L (VAN-1), 12.3 mg/L (VAN-2), and 22.9 mg/L (VAN-3). During droplet incubation, signal amplification rose from 1.11 to 1.56 for VAN-1, 1.12 to 1.96 for VAN-2, and 1.13 to 2.81 for VAN-3 over a period of 20 hours (Fig. 3c). Vanillin detection covered a wide dynamic range (1–1000 μM) after 48 hours of droplet incubation, confirming that MOMS enables accurate quantification without signal saturation during extended cell proliferation (Supplementary Fig. 17).

High-throughput screening of large populations of single yeast cells (1.2 × 10⁷ cells) was demonstrated using co-anchored MOMS with distinct aptamers for multiplexed detection. Vanillin secretion (FITC: excitation 490 nm, emission 525 nm) and ATP secretion (Cy5: excitation 646 nm, emission 664 nm) were measured simultaneously. Flow cytometry successfully distinguished VAN-3 from the negative control (NC), which consisted of wild-type (WT) cells coated with DNA duplexes that targeted neither vanillin nor ATP (Fig. 3d).

The system clustered VAN-1, VAN-2, VAN-3, WT, and NC based on secretion profiles, with VAN-3 showing the highest vanillin secretion, followed by VAN-2 and VAN-1, all of which outperformed WT. ATP secretion profiles were consistent across strains but significantly higher than NC, demonstrating the MOMS platform’s multiplexed detection capabilities (Fig. 3e). Additionally, a 1:1 mixture of WT and VAN-3 strains was screened using flow cytometry. Three distinct clusters were observed (Fig. 3f): VAN-3 mother cells (high fluorescence, right), WT mother cells (quenched fluorescence, middle), and daughter cells (low fluorescence, left). Mixed samples with varying WT:VAN-3 ratios (1:2, 1:1, 2:1, 5:1, 1:0) produced distinct fluorescence peaks, enabling precise differentiation and efficient sorting of VAN-3 cells from mixtures (Fig. 3g).

High-speed analysis for rapid directed evolution using MOMS

High-speed sorting of secreted strains for rapid directed evolution has been demonstrated. Single yeast secretions were effectively measured using MOMS grafted onto the surfaces of mother yeast cells. After droplet incubation to capture the secretions, the droplets were removed, and yeast secretions were directly screened by analyzing MOMS’ fluorescence signals on the cell surfaces via flow cytometry at a speed of approximately 3.0 × 10³ cells/s. This method achieved a screening speed over 30 times faster than traditional droplet screening techniques (~10–200 Hz), despite a low droplet encapsulation rate of less than 5%.

To validate this method, we screened a mixed sample containing 0.5% VAN-3 and 99.5% WT yeast strains to isolate secretory populations (Fig. 4a). MOMS-grafted yeast cells were incubated in droplets for 48 hours to capture vanillin secretions. Following droplet removal, fluorescent yeast cells were sorted at a rate of about 3.0 × 10³ cells/s. Leveraging MOMS’ high-speed capabilities, 1.2 × 10⁶ yeast cells were screened and sorted within 7 minutes, successfully isolating ~100 vanillin-secreting strains. Vanillin production by randomly selected yeast strains from pre-sorting and post-sorting populations was validated using an absorbance-based assay measuring optical density at 438 nm (OD₄₃₈)⁵⁹. The secretory population (VAN-3) was enriched from 0.5% to around 90% after sorting, demonstrating exceptional sorting efficiency.

To further explore the utility of this strategy in directed evolution, we constructed a sizeable mutant library (~2.2 × 10⁶ variants) using error-prone PCR to evolve enoyl-CoA hydratase/aldolase, encoded by the ech gene, for enhanced vanillin productivity. The library plasmids were transduced into yeast strain D-3, which carries deletions in gre2 and adh6 genes, and subsequently coated with vanillin-targeted MOMS. Following a 48-hour droplet incubation of MOMS-coated yeast cells, the top 0.05% of cells—ranked by fluorescence intensity as an indicator of vanillin secretion—were sorted from a total of 2.2 × 10⁶ cells in around 12 minutes (Fig. 4b). This high-speed sorting process, facilitated by MOMS, efficiently eliminated inactive mutants.

After sorting, approximately 301 candidates were isolated and cultured on SD agar plates. From this group, 132 clones were randomly chosen and cultured individually to confirm vanillin productivity using the absorbance-based method. The results showed that the sorted pools achieved over 68% accuracy in isolating rare secretory phenotypes (Supplementary Fig. 18). Among these, the top 10 variants, ranked by their vanillin production activity, demonstrated a mean production level 1.93-fold higher than that of variants randomly selected from the unsorted parent population.

The vanillin productivity of the sorted variants was validated through batch fermentation using high-performance liquid chromatography (HPLC) (Fig. 4c). Among them, the S-10 strain showed the highest yield, producing 86.4 mg/L vanillyl alcohol and 27.7 mg/L vanillin (totaling 113.0 mg/L vanillin equivalents) after 48 hours—a 2.7-fold increase over the control strain VAN-3, which produced 42.6 mg/L vanillyl alcohol (equivalent to 42.0 mg/L vanillin). Notably, VAN-3 secreted detectable vanillin within the first 24 hours, but levels dropped significantly by 36 hours—likely due to partial conversion to vanillyl alcohol—and became nearly undetectable by 48 hours, when vanillyl alcohol reached 42.6 mg/L. A time-course comparison across strains is shown in Supplementary Fig. 19. Similar vanillin depletion patterns were observed in ECH mutant strains S-1, S-8, and S-10, though the rate of decline varied. This suggests vanillin was converted to vanillyl alcohol by endogenous alcohol dehydrogenases (ADHs), aldo–keto reductases (AKRs), and aldehyde reductases (ALDRs), consistent with previous reports^56–58. Sequencing of these vanillin-secreting variants revealed three unique, full-length active sequences without internal stop codons. Compared to the original ECH, the ECH variant from the S-10 strain contained four distinct mutations—I90N, Y169C, N212Q, and P213L (Fig. 4d). Additionally, mutations D80V and Y75H were identified in the ECH variants from the S-8 and S-1 strains, respectively (Supplementary Fig. 20).

To evaluate enzyme activity, wild-type and mutant ech genes were expressed in Escherichia coli BL21 (DE3). Cultures were incubated with 200 µM feruloyl-CoA and 7.5 µM β-NAD at 37 °C for 24 hours, and vanillin levels in the supernatant were quantified by an absorbance-based method. All three mutants produced significantly more vanillin than the wild type, with the I90N/Y169C/N212Q/P213L variant yielding the highest level (~26 mg/L), representing a ~ 11-fold increase (Supplementary Fig. 21). To further quantify catalytic efficiency, wild-type and mutant ECH enzymes were purified via nickel-affinity chromatography (Supplementary Fig. 22a). Equal amounts of enzyme (1 mg/mL) were incubated with 300 µM feruloyl-CoA and 7.5 µM β-NAD at 30 °C for 12 hours. Vanillin production was again quantified (Supplementary Fig. 22b), and relative enzyme activities were calculated (Supplementary Fig. 22c)⁶⁰. All mutants showed consistently higher activity than the wild type, with the I90N/Y169C/N212Q/P213L variant displaying a ~2.3-fold enhancement, confirming improved catalytic performance.

In addition to evolving ECH enzymes, we used the MOMS platform to engineer transporter proteins that enhance extracellular secretion in yeast. A mutant library of the vanillin transporter PP_0179⁶¹ was generated in the VAN-3 strain via error-prone PCR, yielding over 2 × 10⁷ variants. MOMS enabled single-cell screening by capturing and quantifying vanillin secreted from each mutant. The top 0.05% of high-secreting strains were sorted (Fig. 4e), plated, and 108 colonies were randomly selected for validation (Supplementary Fig. 23). Over 78% of these outperformed the wild-type PP_0179, confirming the platform’s screening precision. Notably, fluorescence signal variations due to cell size and budding (Supplementary Fig. 24) had minimal impact on sorting accuracy, as a stringent gating strategy (Supplementary Fig. 25a) maintained consistent cell size distributions before and after sorting (Supplementary Fig. 25b). The ten top-performing variants were further evaluated in batch fermentation, with four showing significantly enhanced vanillin secretion (Fig. 4f). The most effective variant, derived from the T-4 strain, carried the K40R and V79D mutations, which exhibited a synergistic enhancement in transporter efficiency (Fig. 4g). Other key mutations—including G45C in T-1 strain, K48M in T-2 strain, and the combination T15I/G45R/Q98H in T-3 strain—are shown in Supplementary Fig. 26. These findings underscore the versatility of MOMS for practical applications in metabolic engineering, including the directed evolution of transporter proteins.

High-flexibility of MOMS for protein analysis

MOMS is a highly versatile platform that accommodates a wide range of sensor types via biotin–streptavidin interactions, enabling the detection of diverse molecular targets. It is compatible with high-affinity aptamers for both metabolite (Supplementary Table 4) and protein detection (Supplementary Table 5). For example, using aptamers against human interferon gamma (IFN-γ), aptamer-MOMS can detect IFN-γ over a dynamic range of 1–1000 nM with an LOD of approximately 1 nM (Fig. 5a). Similarly, glycosylated human serum albumin (gHSA)-specific aptamers enable aptamer-MOMS to selectively recognize glycosylated pharmaceutical proteins (Fig. 5b). For targets lacking validated aptamers, emerging structure-switching aptamer technologies offer promising alternatives^62,63.

In addition to aptamer-based detection, the MOMS system can be adapted into an antibody-based format (antibody-MOMS) by surface-coating with streptavidin-modified antibodies, enabling detection of a broader range of protein targets. For example, yeast cells coated with streptavidin-conjugated IFN-γ antibodies achieved uniform surface coverage at concentrations above 5 μg/mL (Fig. 5c), enabling detection with high sensitivity (LOD~1 nM) and a wide dynamic range (1–1000 nM) (Fig. 5d). This modular architecture allows MOMS to be readily customized for various analytes—from small molecules to large proteins—by simply substituting the sensing elements.

Notably, beyond S. cerevisiae, the MOMS platform is also adaptable to other budding fungi, such as Pichia pastoris, for secretion profiling applications (Supplementary Fig. 27). A key advantage of MOMS lies in its utilization of the asymmetric budding process in yeast, which ensures that surface-bound sensors remain localized on mother cells during division. This preserves consistent sensor density across generations and supports robust, fluorescence-based detection of extracellular secretions. In contrast, symmetrically dividing organisms like E. coli distribute surface sensors evenly between daughter cells, leading to signal dilution and reduced detection stability, making them unsuitable for MOMS. Supplementary Table 6 summarizes fungal species that are compatible or incompatible with the MOMS platform.

Chemicals and reagents

Unless otherwise specified, all commercially purchased reagents were of analytical reagent grade. Biotinamidohexanoic acid 3-sulfo-N-hydroxysuccinimide ester sodium salt (Sulfo-NHS-LC-Biotin), streptavidin, MgCl₂, NaCl, ZnCl₂, concentrated HCl, Na₂HPO₄, imidazole, and bovine serum albumin (BSA) were procured from Shanghai Aladdin Biochemical Technology Co., Ltd., China. Feruloyl-CoA and β-NAD were provided by Shanghai Yuanye Bio-Technology Co., Ltd, China. Ferulic acid (HPLC), vanillin (HPLC), vanillyl alcohol (HPLC), vanillic acid (HPLC), D-glucose, thiobarbituric acid (TBA), fluorescein diacetate (FDA), isopropyl-β-D-thiogalactopyranoside (IPTG), and glycosylated human serum albumin (gHSA) were sourced from Sigma, USA. All the antibodies and recombinant IFN-γ protein were procured from Biolegend, USA. Alexa Fluor 488-Concanavalin A (ConA) was obtained from Biotium, USA. Dulbecco’s Phosphate-Buffered Saline (DPBS), tris buffer (1 M, pH 7.0 or 8.0), herring sperm DNA, and dimethyl sulfoxide (DMSO) were purchased from Thermo Fisher Scientific, USA. Components of all culture mediums and agar were obtained from Sangon Biotech Co., Ltd., China, except for the minimum SD medium, Dropout (DO) Supplement-Leu and DO Supplement-Ura, which were purchased from Clontech, USA. DNA primers and aptamers, purified by High-Performance Liquid Chromatography (HPLC), were directly ordered from BGI, China.

MOMS fabrication

The fabrication of MOMS involved three main steps: (1) Cell surface biotinylation, (2) preparation of the aptamer-streptavidin complex, and (3) coating the aptamer onto biotinylated yeast cells.

In the first step, biotin was covalently conjugated to proteins on yeast cell surfaces. Yeast cells (2.0 × 10⁶) were suspended in 90 μL of DPBS supplemented with 5 mM MgCl₂, followed by 10 μL of 10 mM sulfo-NHS-LC-biotin (dissolved in DMSO). The mixture was incubated at room temperature (~25 °C) with shaking at 150 rpm for 30 minutes. Biotinylated yeast cells were then washed three times with 150 μL of DPBS containing 5 mM MgCl₂. In the second step, the aptamer-streptavidin complex was prepared by mixing biotin-labeled aptamers with streptavidin in specific molar ratios (aptamer to streptavidin: 3:1). Aptamer solutions (0.5, 1.0, 1.5, or 3.0 μL of 100 μM) were combined with 0.63, 1.25, 2.50, or 5.00 μL of 20 μM streptavidin solution in 99 μL of DPBS containing 5 mM MgCl₂. The mixture was incubated at room temperature for 5−10 minutes and stabilized by adding 1 μL of 10 mg/mL Herring Sperm DNA. In the third step, biotinylated yeast cells were resuspended in 100 μL of the prepared aptamer-streptavidin solution and incubated at room temperature with shaking at 150 rpm for 30 minutes. After three washes with 150 μL of DPBS containing 5 mM MgCl₂, the yeast cells were ready for use.

Double-stranded DNA aptamer sensors were incorporated into MOMS to measure the metabolic compounds. Biotin- and fluorescent-labeled aptamers (e.g., biotin-Xapt-Cy5, biotin-Xapt-Cy3, or biotin-Xapt-FITC; X denotes the target molecule) were first hybridized with complementary DNA (cDNA) labeled with a quenching probe (e.g., BHQ3, for Cy5 quenching; BHQ2, for Cy3 quenching; BHQ1, for FITC quenching) to form a quenched DNA duplex. Specifically, the biotin-labeled aptamer was incubated with 2.5−4.0 equivalents of its quenching cDNA in 50 μL DPBS with 5 mM MgCl₂ at room temperature for over 2 hours. This duplex formation quenches the aptamer fluorescence until activation. The prepared DNA duplex was subsequently incubated with 3 equivalents of streptavidin (relative to the aptamer concentration) in 100 μL DPBS with 5 mM MgCl₂ and 0.1 mg/mL Herring Sperm DNA to generate a DNA-streptavidin complex. This complex was finally coated onto biotinylated yeast cells as described above. These functionalized yeast cells serve as versatile platforms for biochemical and fluorescence-based assays.

The aptamer sensors used in this study to target different molecules were prepared as follows: Vanillin sensors: 3 μL of 100 μM biotin-VANapt-Cy5 or biotin-VANapt-FITC was hybridized with 7.5 μL of 100 μM BHQ3-VANQ or BHQ1-VANQ. ATP aptamer sensors: 3 μL of 100 μM biotin-ATPapt-Cy5 was hybridized with 7.5 μL of 100 μM BHQ3-ATPQ. Glucose sensors: 3 μL of 100 μM biotin-GLUapt-Cy5 or biotin-GLUapt-Cy3 was hybridized with 12 μL of 100 μM BHQ3-GLUQ or BHQ2-GLUQ. Zn²⁺ aptamer sensors: 3 μL of 100 μM biotin-ZINapt-Cy5 was hybridized with 12 μL of 100 μM BHQ3-ZINQ. IFN-γ sensors: 3 μL of 100 μM biotin-IFNapt-Cy5 was hybridized with 7.5 μL of 100 μM BHQ3-IFNQ. gHSA sensors: 3 μL of 100 μM biotin-HSAapt-Cy5 was hybridized with 7.5 μL of 100 μM BHQ3-HSAQ. All sequences of aptamers used in this work are provided in Supplementary Table 7.

Flow cytometry methodology

An optimized gating strategy was applied to all flow cytometry experiments. To distinguish the target yeast population from debris, an initial gate was defined based on forward scatter area (FSC-A) and side scatter area (SSC-A) signals (Supplementary Fig. 7a). To further differentiate mother yeast cells coated with fluorophore-labeled MOMS from daughter cells, a secondary gate was established using fluorescence-negative control samples (Supplementary Fig. 7b). For sorting secretory strains, a stringent gating strategy was employed to minimize false-positive signals arising from variations in cell size or budding. Specifically, only the top 0.05% of mutant yeast cells exhibiting the highest fluorescence intensities were sorted (Supplementary Fig. 25).

Visualization and quantification of MOMS coating

To validate the MOMS coating, yeast cells were labeled with fluorescently tagged single-stranded DNA. Specifically, 3 μM biotin-VANapt-Cy5 (excitation: 646 nm, emission: 664 nm) was incubated with a suspension of 1.0 × 10⁶ biotinylated yeast cells (wild-type S. cerevisiae strain BY4742) for fluorescent labeling. After incubation, the labeled yeast cells were treated with 2.5 μM Alexa Fluor 488-ConA (excitation: 495 nm, emission: 520 nm) in 100 μL DPBS at 37 °C with shaking at 150 rpm for 30 minutes. The yeast cells were subsequently washed once and resuspended in either DPBS or YPD medium for imaging using a confocal laser scanning microscope (CLSM; TCS SP8, Leica, Germany).

To systematically quantify the density of the MOMS coating, 2.0 × 10⁶ yeast cells were incubated with varying concentrations of biotin-VANapt-Cy5 (0.5–3.0 μM) and analyzed using a flow cytometer (FACSVerse™, BD Biosciences, USA). Yeast cells without fluorescently labeled DNA coating served as negative controls. For each sample, 3.0 × 10⁵ yeast cells were analyzed. To estimate the number of sensor molecules on each yeast cell, fluorescence signals from solutions with known biotin-VANapt-Cy5 concentrations were calibrated using a microplate reader (SpectraMax M5e, Molecular Devices, China) to generate a standard curve. The MOMS density per yeast cell can be calculated by determining the remaining biotin-VANapt-Cy5 concentration in the incubation solution post-coating (Supplementary Fig. 6)⁴⁷.

Viability and growth test

To assess the biocompatibility of MOMS, yeast viability and growth were evaluated. For the viability test, yeast cells coated with fluorescently labeled single-stranded DNA (biotin-VANapt-Cy5) were suspended in 199 μL of DPBS and supplemented with 1 μL of 5 mg/mL FDA (dissolved in acetone). The suspension was incubated at 30 °C while shaking at 150 rpm for 15 minutes. After incubation, yeast cells were analyzed using CLSM imaging and flow cytometry. For the growth test, yeast cells with or without MOMS coatings were suspended in 5 mL of SD medium with an initial optical density at 600 nm (OD₆₀₀) of approximately 0.2. Cultures were incubated at 30 °C while shaking at 250 rpm for 72 hours, and cell density (OD₆₀₀) was measured at 12-hour intervals using a microplate reader.

MOMS sensitivity and specificity evaluation

To calibrate the sensing sensitivity of MOMS, yeast cells coated with MOMS were prepared by mixing 2.0 × 10⁶ biotinylated yeast cells with 3 μM MOMS targeting specific molecules in 100 μL of DNA-streptavidin solution. The mixture was incubated at 30 °C for 30 minutes. The coated yeast cells were then incubated in solutions containing known concentrations of target molecules (vanillin, ATP, glucose, and Zn²⁺) to determine their limits of detection (LODs). To minimize non-specific binding of quenching cDNA detached from the aptamer to the yeast cell surface, 0.1 mg/mL Herring Sperm DNA was included in the buffer solution as a blocking agent. Tailored buffer solutions were used to optimize sensing conditions for different target molecules: Vanillin and ATP sensing: 10 mM Tris, 150 mM NaCl, and 5 mM MgCl₂; Glucose and Zn²⁺ sensing: DPBS supplemented with 5 mM MgCl₂; IFN-γ and gHSA sensing: DPBS. Following incubation at 30 °C with shaking at 150 rpm for a specified duration, yeast cells were washed and analyzed via flow cytometry. The specificity of the MOMS system was evaluated by exposing MOMS-coated yeast cells to structurally related analogs. For vanillin sensors, cells were incubated with 1 mM solutions of vanillin, vanillyl alcohol, vanillic acid, or ferulic acid. For ATP sensors, cells were treated with 1 mM of each nucleotide—ATP, CTP, GTP, and UTP. For glucose sensors, cells were exposed to 1 M concentrations of glucose, galactose, or fructose. Fluorescence responses were then measured and compared to determine the selectivity of each sensor for its intended target.

MOMS multiplexed sensing experiment protocol

For multiplexed sensing, three types of MOMS with distinct aptamer sequences were simultaneously coated onto the same yeast cells. The MOMS specifically targeted ATP, vanillin, and glucose and were fabricated according to established protocols, with each labeled using distinct fluorescent dyes: Cy5 for ATP (excitation: 646 nm, emission: 664 nm), FITC for vanillin (excitation: 490 nm, emission: 525 nm), and Cy3 for glucose (excitation: 555 nm, emission: 570 nm). These MOMS (1.5 μM each) were co-incubated with 2.0 × 10⁶ biotinylated yeast cells in a DNA-streptavidin solution containing 1.5 μM streptavidin and 0.1 mg/mL Herring Sperm DNA, resulting in a total reaction volume of 200 μL. After multiplex MOMS coating, the yeast cells were incubated in solutions containing various target molecules (0.5 mM ATP, 1.0 mM vanillin, and/or 1.0 M glucose) in a buffer composed of 10 mM Tris, 150 mM NaCl, 5 mM MgCl₂, and 0.1 mg/mL Herring Sperm DNA. The incubation took place at 30 °C with shaking at 150 rpm for 1 hour. Following the incubation, the MOMS-coated yeast cells were washed and analyzed using flow cytometry to assess their sensing capacity for multiplexed assays.

Rapid secretion monitoring

To validate the sensitive rapid secretion analysis using MOMS, MOMS-coated yeast cells were prepared by incubating 2.0 × 10⁶ biotinylated yeast cells with 3 μM ATP-targeted MOMS in 100 μL of DNA-streptavidin solution at 30 °C for 30 minutes. The resulting MOMS-coated yeast cells were diluted to a concentration of 5.0 × 10⁵ cells/mL and monitored using CLSM for 40 minutes, with images captured at 1-minute intervals. The solution was maintained in a stable position on an optical table at 30 °C during observations to ensure consistent tracking of the same yeast cells over time. Flow cytometry was employed to validate single yeast cell secretion over varying periods. A total of 2.0 × 10⁶ MOMS-coated yeast cells were incubated in YPD medium for varying durations (e.g., 20, 30, 40, 60, or 80 minutes). After incubation, the yeast cells were washed with buffer (DPBS supplemented with 5 mM MgCl₂) and subsequently analyzed using a flow cytometer to profile single-cell secretion dynamics across the various incubation times.

Culturing and transformation procedures for E. coli and yeast

The strains used in this study are listed in Supplementary Table 8. The construction and amplification of E. coli–S. cerevisiae shuttling plasmids were carried out in E. coli DH5α chemically competent cells (Sangon Biotech, China). E. coli cultures were grown overnight in Lysogeny Broth (LB) medium (5 g/L yeast extract, 10 g/L tryptone, and 10 g/L sodium chloride) at 37 °C with shaking at 220 rpm. When necessary, ampicillin (100 mg/L) was added for plasmid propagation. E. coli transformations were performed using a standard chemical transformation protocol following the manufacturer’s instructions. Transformants were plated on LB agar containing 100 mg/L ampicillin and incubated overnight at 37 °C. Single colonies were inoculated into 5 mL of LB medium and cultured overnight. Plasmids were extracted from E. coli cultures (OD₆₀₀ = 2.0) using the TIANprep Mini Plasmid Kit (TIANGEN, China) and verified via Sanger sequencing (Sangon Biotech, China).

For yeast strain development, S. cerevisiae BY4742 was obtained from Beijing Zoman Biotechnology Co., Ltd., China, and propagated in yeast extract peptone dextrose (YPD) medium (10 g/L yeast extract, 20 g/L tryptone, 20 g/L glucose) at 30 °C with shaking at 250 rpm. For plasmid maintenance, yeast strains were cultured in SD medium supplemented with either DO supplement-Leu or DO supplement-Ura. Chemically competent yeast cells were prepared using the S.c. EasyComp™ Transformation Kit (Thermo Fisher Scientific, USA) according to standard protocols. Plasmids were then introduced via chemical transformation, and the engineered yeast cells were plated onto SD agar and incubated at 30 °C for 2–3 days until single colonies formed. These colonies were subsequently cultured overnight in 5 mL of SD medium to generate concentrated cell suspensions (OD₆₀₀ = 2.0). The molecular profiles of the transformed yeast strains were confirmed by colony PCR, verifying successful plasmid integration.

Molecular cloning procedures

The plasmids used in this study are listed in Supplementary Table 9. The E. coli–S. cerevisiae shuttle vectors pYCP (VectorBuilder, China) and pML104 (Addgene, USA) were employed for gene expression and gene deletion in yeast strains, respectively. Recombinant plasmids were constructed using the Gibson assembly method following standard protocols⁶⁶. PCR amplifications were performed using Q5® High-Fidelity DNA Polymerase (NEB, USA), with all primers listed in Supplementary Table 10. For the construction of a plasmid enabling vanillin production in yeast, the target genes 4cl (from Petroselinum crispum) and ech (from Pseudomonas putida KT2440) were PCR-amplified using specific primers. The resulting DNA fragments were purified via gel extraction and subcloned into the pYCP vector, generating the recombinant plasmid pVAN. Gene deletions in yeast strain BY4742 were performed using CRISPR/Cas9 gene editing methods⁶⁷.

Microfluidic device fabrication

The fabrication of microfluidic devices followed established soft lithography protocols⁶⁸. Briefly, a silicon wafer was coated with a layer of SU-8 2025 photoresist (MicroChem Corp., USA) via spin coating. The photoresist was patterned using UV photolithography with a photomask, followed by development in SU-8 developer. After cleaning and drying of the wafer, Polydimethylsiloxane (PDMS, Sylgard 184 silicone elastomer kit, Dow Corning, USA) was poured over the wafer and cured at an elevated temperature to form a solid replica. The cured PDMS layer, containing the microchannel patterns, was carefully peeled off and bonded to a glass slide pre-coated with a thin PDMS layer using plasma bonding. To restore the hydrophobicity of the PDMS microchannels, the assembled device was heated at 95 °C for over 2 hours.

High-throughput single yeast secretion analysis

To evaluate the screening capability of MOMS based on single yeast cell secretion levels, yeast strains with varying efficiencies of vanillin secretion were created by partially inhibiting the native vanillin reduction pathway. Specifically, three gene-deleted yeast strains were developed: D-1 (gre2 deleted), D-2 (adh6 deleted), and D-3 (gre2 and adh6 deleted). The plasmid pVAN was introduced into these strains, resulting in the transformed variants VAN-1 (gre2 deleted), VAN-2 (adh6 deleted), and VAN-3 (gre2 and adh6 deleted). For the control experiment, the blank plasmid pYCP was introduced into D-3, generating the wild-type (WT) strain.

The analysis process consists of two steps: capture and screening. Step 1: Capture: Single yeast cells were encapsulated in water-in-oil droplets (40-µm diameter) using a microfluidic droplet generator (Supplementary Fig. 28), with flow rates of 600 μL/h for the cell suspension and 5.0 × 10³μL/h for fluorocarbon oil (HFE7500 containing 0.5% v/v surfactant, Suzhou Cchip Scientific, China), controlled by syringe pumps (Harvard Apparatus, USA). A total of 1.0 × 10⁶ yeast cells (WT, VAN-1, VAN-2, or VAN-3) coated with vanillin-targeted MOMS were suspended in 660 μL YPD medium supplemented with 1.0 g/L ferulic acid to induce vanillin production. For the negative control, WT cells coated with a non-targeting DNA duplex (Supplementary Table 7) were processed under identical conditions to establish a fluorescence baseline for signal amplification calculations. Following droplet encapsulation, the samples were incubated at 30 °C for 12–36 hours, allowing each yeast cell to secrete vanillin, which the MOMS captured on its surface. Step 2: Screening: After incubation, the droplets were broken by adding a de-emulsifier (1H,1H,2H,2H-perfluoro-1-octanol, J&K Scientific, China) to the oil at a 1:1 ratio, releasing the MOMS-coated yeast cells. The cells were collected and analyzed via flow cytometry for high-throughput screening at a rate of ~4.4 × 10³ cells per second.

To simultaneously analyze vanillin and ATP secretion, yeast cells were coated with two types of MOMS: FITC-labeled MOMS for vanillin and Cy5-labeled MOMS for ATP. Specifically, 1.5 μM vanillin-targeted MOMS and 1.5 μM ATP-targeted MOMS were mixed with 1.5 × 10⁷ biotinylated yeast cells (WT, VAN-1, VAN-2, or VAN-3) in 1.5 mL DNA-streptavidin solution containing 0.5 μM streptavidin, followed by incubation at 30 °C for 30 minutes. For the negative control, WT cells were coated with DNA duplexes labeled with the same fluorescent probes but targeting neither vanillin nor ATP (Supplementary Table 7). These coated cells (1.5 × 10⁷) were suspended in 10 mL YPD medium containing 1.0 g/L ferulic acid and encapsulated in water-in-oil droplets using microfluidic droplet generators. The droplets were then incubated at 30 °C for 48 hours, allowing yeast cells to secrete and capture their own vanillin and ATP via MOMS. Following incubation, the droplets were broken, and yeast cells were analyzed via flow cytometry for high-throughput multiplexed single-cell secretion profiling.

High-speed sorting

To evaluate the efficiency of high-speed sorting using MOMS, a yeast cell mixture containing approximately 0.5% VAN-3 strain and 99.5% WT strain was encapsulated in droplets with YPD medium supplemented with 1.0 g/L ferulic acid to induce vanillin production. The droplets were then incubated at 30 °C for 48 hours, allowing MOMS-coated yeast cells to capture their secretions on their surfaces within the droplets. Following incubation, the droplets were broken to recover yeast cells for high-speed sorting. A fluorescence-activated cell sorter (FACSMelody, BD Biosciences, USA) operating at ~3.0 × 10³ cells per second was used to sort the brightest 0.05% of cells, yielding approximately 100 yeast cells from >10⁶ cells. The sorted cells were then plated onto YPD agar plates for further culturing and analysis. To assess vanillin production in the sorted population, 10 clones were randomly selected from both the post-sorting and pre-sorting populations, cultured in SD medium, and re-plated onto the SD agar plate. Vanillin production was quantified using an absorbance-based method, measuring optical density at 438 nm (OD₄₃₈) with a microplate reader.

Rapid directed evolution experiment

A mutant library (~2.2 × 10⁶ variants) was constructed to enable the directed evolution of ECH, derived from the wild-type enzyme. The target gene segment was PCR-amplified using an error-prone approach with the QuickMutation™ Random Mutagenesis Kit (Beyotime, China) at a medium mutation rate (4.5–9 mutations per kilobase). The amplified products were purified and cloned into the pVAN plasmid backbone using the Gibson assembly method. The newly generated plasmid library was then transformed into E. coli Trans1-T1 chemically competent cells (TransGen Biotech, China) for plasmid amplification. Subsequently, the library was transduced into yeast strain D-3 (adh6 and gre2 deleted). After transformation, the yeast mutant libraries were cultured in SD medium for ~18 hours until reaching the stationary phase, then glycerol-stocked for long-term storage.

To screen for highly secretory mutants during directed evolution, yeast cells from the mutant libraries were coated with MOMS and encapsulated in droplets containing YPD medium supplemented with ferulic acid. After a 48-hour incubation, single-cell vanillin secretion was measured via fluorescence. The top 0.05% of yeast cells with the highest fluorescence intensity were sorted and recovered onto YPD agar plates.

To validate the sorted strains, 132 clones were randomly selected from 301 sorted strains (~44%) on the agar plates and analyzed for vanillin production using an absorbance-based method. To further validate the vanillin production efficiency of the sorted variants, the top 10 secretory strains were re-plated onto SD agar plates. For each strain, three single colonies were selected and subjected to batch fermentation in 250 mL flasks. Cell growth was assessed by measuring OD₆₀₀ using a microplate reader, and supernatant concentrations were analyzed via HPLC after a 48-h incubation. The molecular profiles of these top vanillin-producing strains were determined via Sanger sequencing (Sangon Biotech) to identify critical mutation sites.

Vanillin productivity measurement by absorbance detection

The vanillin productivity of yeast strains was assessed using an absorbance-based method with a microplate reader (Molecular Devices). Vanillin (dissolved in SD medium) was mixed with 250 μL of HCl solution (20:1 in water) and 100 μL of TBA solution (0.5% w/w in water) to a total volume of 500 μL. The mixture was reacted at 55 °C for 10 minutes, followed by incubation at room temperature for 20 minutes. The OD₄₃₈ was measured, and vanillin concentrations were quantified using a standard curve of OD₄₃₈ versus vanillin concentration (Supplementary Fig. S29)⁵⁹. To evaluate vanillin production in different yeast strains, single yeast clones were picked from YPD or SD agar plates and inoculated into 5 mL of YPD medium. For each strain, three colonies were selected as biological triplicates. Cultures were incubated at 30 °C with shaking at 250 rpm for 16-20 hours. The precultures were then transferred into 5 mL of YPD medium supplemented with 1.0 g/L ferulic acid and cultivated for 48 hours under the same conditions. Supernatants were collected by centrifugation, and vanillin concentrations were determined by measuring OD₄₃₈ with a microplate reader.

Validation of vanillin production by HPLC

The production of vanillin by yeast strains was validated using HPLC. Step 1: Batch fermentation: Yeast batch fermentation commenced with the selection of top vanillin-producing single clones from SD agar plates, which were then inoculated into 5 mL of SD medium. For each strain, three colonies were picked to serve as biological triplicates. The cultures were incubated at 30 °C while shaking at 250 rpm for 16–20 hours. The precultures were subsequently transferred to 5 mL of fresh SD medium, maintaining an initial OD₆₀₀ of 0.2, and incubated under the same conditions for an additional 16–20 hours. Following this, the yeast cells were transferred to 250 mL flasks containing 30 mL of SD medium, also adjusted to an initial OD₆₀₀ of 0.2. After 12 hours of incubation at 30 °C with shaking at 200 rpm, ferulic acid was added as a substrate at a final concentration of 3.0 g/L. The cultures were then further incubated at 30 °C with shaking at 250 rpm for 48 hours.

Step 2: HPLC Analysis: Fermentation supernatants were analyzed for vanillin, vanillyl alcohol and vanillic acid using an Agilent 1290 HPLC system (Agilent, USA). Separation was carried out on an InfinityLab Poroshell 120 EC-C18 column (50 mm × 2.1 mm, 1.9 μm particle size) with a mobile phase consisting of solvent A (pure acetonitrile) and solvent B (0.1% formic acid in water, v/v). The gradient elution program was as follows: 0–7.5 min: 10% A, 90% B at 0.1 mL/min; 7.5–16.0 min: 100% A at 0.1 mL/min; 16.0–19.0 min: 100% A at 0.3 mL/min; 19.0–25.0 min: 10% A, 90% B at 0.1 mL/min. The column temperature was kept at 30 °C. Detection wavelengths were set as follows: 259 nm for vanillyl alcohol, 280 nm for vanillic acid, and 285 nm for vanillin.

Statistics & reproducibility

Unless otherwise stated, all experiments in this study were independently performed three times to ensure reproducibility, and data are presented as mean ± standard deviation (mean ± s.d.). For CLSM and fluorescence microscopy, representative images from three biological replicates were shown. In each experiment, yeast cells were randomly selected for MOMS coating, screening, and sorting to ensure objective analysis. No statistical method was used to pre-determine sample size, no data were excluded from the analyses, and the investigators were not blinded to allocation during experiments and outcome assessment. Flow cytometry data were analyzed by FlowJo (V10) software, signal magnification was calculated based on changes in mean fluorescence intensity (MFI). To quantify the fluorescent intensity of yeast cells in CLSM images, ImageJ (1.54p) software was used.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Peer review information

Nature Communications thanks Jifeng Yuan and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.