# Fileset

[biosensors-13-00152.pdf](https://mdr.nims.go.jp/filesets/a320f3a2-1a7e-48dc-bdb4-388a0bddffff/download)

## Creator

[Kosuke Minami](https://orcid.org/0000-0003-4145-1118), Hisami Kobayashi, Masaaki Matoba, Yuko Kamiya, Subrata Maji, Takahiro Nemoto, Masanori Tohno, Ryoh Nakakubo, [Genki Yoshikawa](https://orcid.org/0000-0002-9136-8964)

## Rights

[Creative Commons BY Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)

## Other metadata

[Measurement of Volatile Fatty Acids in Silage through Odors with Nanomechanical Sensors](https://mdr.nims.go.jp/datasets/324270e9-72ce-4c10-af93-b6921189804e)

## Fulltext

Measurement of Volatile Fatty Acids in Silage through Odors with Nanomechanical SensorsCitation: Minami, K.; Kobayashi, H.;Matoba, M.; Kamiya, Y.; Maji, S.;Nemoto, T.; Tohno, M.; Nakakubo, R.;Yoshikawa, G. Measurement ofVolatile Fatty Acids in Silage throughOdors with Nanomechanical Sensors.Biosensors 2023, 13, 152. https://doi.org/10.3390/bios13020152Received: 16 December 2022Revised: 10 January 2023Accepted: 16 January 2023Published: 18 January 2023Copyright: © 2023 by the authors.Licensee MDPI, Basel, Switzerland.This article is an open access articledistributed under the terms andconditions of the Creative CommonsAttribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).biosensorsArticleMeasurement of Volatile Fatty Acids in Silage through Odorswith Nanomechanical SensorsKosuke Minami 1 , Hisami Kobayashi 2, Masaaki Matoba 1, Yuko Kamiya 2, Subrata Maji 1, Takahiro Nemoto 1,Masanori Tohno 2,3,*, Ryoh Nakakubo 4,* and Genki Yoshikawa 1,5,*1 Center for Functional Sensor & Actuator (CFSN), Research Center for Functional Materials,National Institute for Materials Science (NIMS), 1-1 Namiki, Tsukuba 305-0044, Ibaraki, Japan2 Institute of Livestock and Grassland Science, National Agriculture and Food Research Organization (NARO),768 Senbonmatsu, Nasushiobara 329-2793, Tochigi, Japan3 Research Center of Genetic Resources, National Agriculture and Food Research Organization (NARO),2-1-2 Kannondai, Tsukuba 305-8602, Ibaraki, Japan4 Institute of Livestock and Grassland Science, National Agriculture and Food Research Organization (NARO),2 Ikenodai, Tsukuba 305-0901, Ibaraki, Japan5 Materials Science and Engineering, Graduate School of Pure and Applied Science, University of Tsukuba,1-1-1 Tennodai, Tsukuba 305-857, Ibaraki1, Japan* Correspondence: tohno@affrc.go.jp (M.T.); ryohnakakubo@affrc.go.jp (R.N.);yoshikawa.genki@nims.go.jp (G.Y.)Abstract: The measurement of volatile fatty acids (VFAs) is of great importance in the fields of foodand agriculture. There are various methods to measure VFAs, but most methods require specificequipment, making on-site measurements difficult. In this work, we demonstrate the measure-ments of VFAs in a model sample, silage, through its vapor using an array of nanomechanicalsensors—Membrane-type Surface stress Sensors (MSS). Focusing on relatively slow desorption be-haviors of VFAs predicted with the sorption kinetics of nanomechanical sensing and the dissociationnature of VFAs, the VFAs can be efficiently measured by using features extracted from the decaycurves of the sensing response, resulting in sufficient discrimination of the silage samples. Since thepresent sensing system does not require expensive, bulky setup and pre-treatment of samples, it hasa great potential for practical applications including on-site measurements.Keywords: volatile fatty acids (VFAs); nanomechanical sensors; Membrane-type Surface stress Sensor(MSS); electronic nose (e-nose); silage1. IntroductionSilage is a type of roughage, which is made of fermented grass or crops to enable long-term storage and a stable supply of feed. The fermentation quality of silage affects not onlythe feeding system and milk production but also diseases in dairy cows [1]. For example,butyric acid, which is mainly produced by aerobic fermentation, can cause reduced drymatter intake [2], leading to energy metabolism disorder, such as ketosis [3,4]. Therefore,examining the fermentation quality of silage is of great importance to determine whetherand how to feed silage. There are two major approaches to examining the quality of silagefermentation: with chemical analysis and with sensory evaluation using human olfaction.Although the sensory evaluation can be performed in the field, its accuracy is limited.Chemical analysis, on the other hand, generally provides high accuracy by quantifyingthe fermentation products. However, chemical analysis requires expensive instrumentsand pretreatment procedures, resulting in time consumption and delayed results [5]. Inaddition, in the case of wrapped silage, the aerobic fermentation proceeds from the momentthat the wrap is opened. Therefore, there is a demand for a method to measure the qualityof silage fermentation accurately on site.Biosensors 2023, 13, 152. https://doi.org/10.3390/bios13020152 https://www.mdpi.com/journal/biosensorshttps://doi.org/10.3390/bios13020152https://doi.org/10.3390/bios13020152https://creativecommons.org/https://creativecommons.org/licenses/by/4.0/https://creativecommons.org/licenses/by/4.0/https://www.mdpi.com/journal/biosensorshttps://www.mdpi.comhttps://orcid.org/0000-0003-4145-1118https://orcid.org/0000-0002-9136-8964https://doi.org/10.3390/bios13020152https://www.mdpi.com/journal/biosensorshttps://www.mdpi.com/article/10.3390/bios13020152?type=check_update&version=1Biosensors 2023, 13, 152 2 of 13As sensory evaluations are made on odors of silage, odor is one of the importantaspects to determine the quality of silage fermentation. To measure such odors, gaschromatography (GC) is the most common approach, while it also has limitations foron-site measurements. The concept of artificial olfaction—also known as electronic nose(e-nose)—has been proposed using an array of chemical sensors [6] consisting of multiplesensors with different chemical properties. Using this artificial olfaction, a wide rangeof studies have been reported on the identification, discrimination, and quantification oftarget odors. Since the size reduction of sensing elements through recent nanotechnologyhas made the artificial olfaction system small and mobile, potential applications includingon-site measurements have been proposed in various fields, such as food, agriculture,environment, medicine, and healthcare [7–14].In this study, we demonstrate that a nanomechanical sensor array can be utilized fordifferentiating the silage samples through their odors, especially volatile fatty acids (VFAs),which are one of the key indicators for the silage quality [15]. We used a nanomechanicalMembrane-type Surface stress Sensor (MSS) as a platform of artificial olfaction [14,16,17].By focusing on the relatively slow desorption behaviors of VFAs predicted according to thesorption kinetics of nanomechanical sensing [18,19] and dissociation nature of VFAs, wefound that the trends of VFAs in moisture-rich odors can be clearly seen in the decay curveof the signals, regardless of the hydrophilicity or hydrophobicity of the receptor materials.Since the MSS array has various advantages on the on-site measurement [14], MSS-basedartificial olfaction provides a promising evaluation platform for the silage quality.2. Materials and Methods2.1. MaterialsPolystyrene (PS), polymethyl methacrylate (PMMA), and poly(2,6-diphenyl-1,4-pheny-lene oxide) (known as Tenax) were purchased from Sigma-Aldrich Inc. and GL Science andused for receptor materials. N,N-dimethylformamide (DMF) was purchased from KantoChemical Co., Inc. and used as a solvent to prepare solutions of receptor materials for inkjetspotting. Acetic acid (ethanoic acid), propionic acid (propanoic acid), butyric acid (butanoicacid), valeric acid (pentanoic acid), and caproic acid (hexanoic acid) were purchased fromSigma-Aldrich Inc. (Tokyo, Japan), Tokyo Chemical Industry (Tokyo, Japan), FUJIFILMWako Pure Chemical Corporation (Osaka, Japan), Kanto Chemical Co., Inc. (Tokyo, Japan).All chemicals were used as received. MilliQ water (Merck MilliPore, Tokyo, Japan) wasused as water vapor.2.2. Plant Materials and Preparation of SilageThe whole-plant corn (Zea mays L.) at the yellow ripe stage, grown in a field at theNational Institute of Livestock and Grassland Science (Nasushiobara, Tochigi, Japan), wereharvested and chopped into a 32 mm theoretical length size using a forage harvester(Model-790MD; New Holland, New Holland, PA). The plant material was then ensiled inan underground silo. After ensiling, the silage samples with different fermentation qualitywere chosen and designated as S1-1, S1-2, and S1-3 (the upper part of the silo) and S2-1,S2-2, and S2-3 (the bottom part of the silo), respectively.2.3. Chemical Analysis of Silage SamplesSilage homogenates were prepared as previously described [20] with some modifica-tions. Briefly, 100 mL of distilled water was added to 10 g (fresh weight) of silage material.The resulting solution was homogenized by using a laboratory homogenizer (Pro·mediaSH-IIM, Elmex). After 5 min of extraction, the sample was filtered through 5A filter pa-per (Advantec). The resulting eluate was treated with Amberlite (Amberlite IR120H+,Tokyo Chemical Industry) and centrifuged at 20,000 × g for 5 min. The supernatant wasfiltered through a membrane filter (pore size 0.45 µm; Advantec) and analyzed with a high-performance liquid chromatograph (HPLC; JASCO Corporation) equipped with ShodexRaspak KC-811 column (8 mm × 300 mm; Showa Denko) and a UV spectrometer (detectionBiosensors 2023, 13, 152 3 of 13wavelength was 450 nm). The column was maintained at 60 ◦C. The flow rate of the mobilephase (3 mmol L−1 of HClO4 aq.) was 1.2 mL min−1. BTB solution (0.2 mmol L−1 ofbromothymol blue, 8 mmol L−1 of Na2HPO4, and 2 mmol L−1 of NaOH) was used as thereaction mixture.The water content is 100 less the proportion of dry matter content (%). Dry matterweight of silage was determined with heating material at 60 ◦C for 48 h. The pH values ofthe silage extract were determined using a pH meter (Seven Excellence; Mettler-Toledo)attached to an electrode (InLab Expert Pro-ISM; Mettler-Toledo).2.4. Fabrication of MSSThe construction of the MSS chip and its working principle has been previouslyreported (Figure 1a) [16,17]. Briefly, MSS consists of a silicon membrane suspended byfour bridges, composing a full Wheatstone bridge (Figure 1a). In each bridge, piezore-sistors were embedded through boron doping. Receptor materials (i.e., PS, PMMA, andTenax) were coated on each membrane. When the target gas is introduced to MSS, thereceptor layer deforms from sorption of target analytes, such as VFAs, generating surfacestress [14]. The surface stress on the membrane is transduced to the four piezoresistivebridges as amplified uniaxial stress, resulting in the changes in the electrical resistanceof the piezoresistors embedded in the bridges. The MSS chips used in this study werepurchased from NanoWorld AG, Switzerland and provided from Asahi Kasei Co., Ltd.Each receptor material was deposited directly on the membrane of MSS by using inkjetspotter (LaboJet-500SP, MICROJET Corporation, Nagano, Japan) with a nozzle (IJHBS-300,MICROJET Corporation). In the present study, we used commercially available polymersas receptor materials, which exhibit different chemical selectivity. Each receptor materialwas dissolved in DMF (1 mg mL−1), and the solutions were deposited onto each channel ofthe MSS. A stage of the inkjet spotter was heated at 80 ◦C during the deposition to promoteevaporation of DMF.Biosensors 2023, 13, x FOR PEER REVIEW 3 of 13  paper (Advantec). The resulting eluate was treated with Amberlite (Amberlite IR120H+, Tokyo Chemical Industry) and centrifuged at 20,000 × g for 5 min. The supernatant was filtered through a membrane filter (pore size 0.45 μm; Advantec) and analyzed with a high-performance liquid chromatograph (HPLC; JASCO Corporation) equipped with Shodex Raspak KC-811 column (8 mm × 300 mm; Showa Denko) and a UV spectrometer (detection wavelength was 450 nm). The column was maintained at 60 °C. The flow rate of the mobile phase (3 mmol L–1 of HClO4 aq.) was 1.2 mL min–1. BTB solution (0.2 mmol L–1 of bromothymol blue, 8 mmol L–1 of Na2HPO4, and 2 mmol L–1 of NaOH) was used as the reaction mixture. The water content is 100 less the proportion of dry matter content (%). Dry matter weight of silage was determined with heating material at 60 °C for 48 h. The pH values of the silage extract were determined using a pH meter (Seven Excellence; Mettler-Toledo) attached to an electrode (InLab Expert Pro-ISM; Mettler-Toledo). 2.4. Fabrication of MSS The construction of the MSS chip and its working principle has been previously re-ported (Figure 1a) [16,17]. Briefly, MSS consists of a silicon membrane suspended by four bridges, composing a full Wheatstone bridge (Figure 1a). In each bridge, piezoresistors were embedded through boron doping. Receptor materials (i.e., PS, PMMA, and Tenax) were coated on each membrane. When the target gas is introduced to MSS, the receptor layer deforms from sorption of target analytes, such as VFAs, generating surface stress [14]. The surface stress on the membrane is transduced to the four piezoresistive bridges as amplified uniaxial stress, resulting in the changes in the electrical resistance of the pie-zoresistors embedded in the bridges. The MSS chips used in this study were purchased from NanoWorld AG, Switzerland and provided from Asahi Kasei Co. Ltd. Each receptor material was deposited directly on the membrane of MSS by using inkjet spotter (LaboJet-500SP, MICROJET Corporation, Nagano, Japan) with a nozzle (IJHBS-300, MICROJET Corporation). In the present study, we used commercially available polymers as receptor materials, which exhibit different chemical selectivity. Each receptor material was dis-solved in DMF (1 mg mL–1), and the solutions were deposited onto each channel of the MSS. A stage of the inkjet spotter was heated at 80 °C during the deposition to promote evaporation of DMF.  Figure 1. Schematic illustrations of the MSS and its sensing system. (a) Configuration of the MSS. (b) Schematic illustration of the measurement system. 2.5. Sensing To estimate the diffusion time constant τs of each receptor material to each solvent vapor, we used a sensing measurement system according to our previous work [19] as shown in Figure 1b. The coated MSS chips were placed in a Teflon chamber, which was placed in an incubator with a controlled temperature of 30.00 ± 0.02 °C. The chamber was connected to a gas flow system: a purging line, an injection line, a mixing chamber, two mass flow controllers (MFCs), and a vial for solutions of VFAs (Figure 1b). The vapor of Figure 1. Schematic illustrations of the MSS and its sensing system. (a) Configuration of the MSS.(b) Schematic illustration of the measurement system.2.5. SensingTo estimate the diffusion time constant τs of each receptor material to each solventvapor, we used a sensing measurement system according to our previous work [19] asshown in Figure 1b. The coated MSS chips were placed in a Teflon chamber, which wasplaced in an incubator with a controlled temperature of 30.00 ± 0.02 ◦C. The chamber wasconnected to a gas flow system: a purging line, an injection line, a mixing chamber, twomass flow controllers (MFCs), and a vial for solutions of VFAs (Figure 1b). The vapor ofVFAs was produced by bubbling carrier gas. As carrier and purging gases, pure nitrogengas was used. The duration and the concentrations of the five different VFA vapors wereprecisely controlled using MFC-1 (injection line) at Pa/Po of 0.1, where Pa and Po denotethe partial vapor pressure and saturated vapor pressure of the VFAs, respectively. Beforemeasuring MSS signals, pure nitrogen gas was introduced into the chamber for at least1 min to promote the desorption of molecules adsorbed in the previous measurement.Biosensors 2023, 13, 152 4 of 13Subsequently, MFC-1 was switched on/off every 10 s with a controlled total flow rate of100 mL min−1 using MFC-2 (purging line) for four injection-purge cycles. Sensing signalsof MSS were measured with a bridge voltage of –1.0 V and recorded with a sampling rateof 20 Hz. The data collection program was designed using LabVIEW (NI Corporation).To measure the silage samples, we used the MSS Standard Measurement module [21]produced by the industry–academia–government collaboration framework called “MSSAlliance” and “MSS Forum” [22–24]. The coated MSS chips were placed in a Teflon chamberin the MSS module. The chamber was connected to a gas flow system: a switching valveconnected with injection and purging gas lines, a flow meter, and an aspiration pump.The sample and purge gas flows were controlled by the pump with a flow rate adjustedto 30 mL min−1. Each silage sample was placed in 20 mL vial and was connected to theinjection line. Before measuring MSS signals, pure nitrogen gas was introduced into theMSS module for at least 2 min. Subsequently, the switching valve was switched to thesampling line for 2 min and then switched back to the purging line for 8 min. Sensingsignals of MSS were measured at the bridge voltage of –1.0 V and recorded at a samplingrate of 100 Hz.2.6. VFA Profile in Headspace GasTo measure the VFAs in the headspace gases of silage samples and aqueous solutionsof VFAs, PTR-TOF-MS (PTR-TOF 6000 X2, Ionicon Analytik GmbH) equipped with StaticHeadspace Autosampler was used. The measurement setup was set according to the manu-facture’s protocol. The ion source of PTR was operated at a current of 4 mA and a voltage of145 V with the source-out voltage maintained at 78.56 V. The source valve operating was setat 51%. The voltage, pressure, and temperature of drift tube were maintained at 557 V, 2.8mbar, and 70 ◦C, respectively. The E/N value, where E and N are the applied electric fieldand the number density of the gas in the drift tube (1 Td = 10−17 V cm2), respectively [25],was approximately 101 Td. The mass spectrum was recorded in the mass range of m/z9–400, and mass calibration was performed using two ion peaks, which are known exactmasses, i.e., hydronium ion isotope (H318O+; m/z = 21.022) and diiodobenzene fragment(C6H4IH+; m/z = 203.943). The count rate of primary ion H3O+, which is calculated fromthe count rate at m/z = 21.022 multiplied by 500, was ca. 1.0 × 108 count per second (cps) inthis work. The raw mass spectrum obtained from PTR-MS was analyzed using the PTR-MSViewer ver. 3.3.9.1.2.7. Curve Fitting of the Signal ResponseTo estimate the diffusion time constant τs, we used an analytical solution of nanome-chanical sensing based on the sorption kinetics and viscoelastic behaviors as follows [19]:σ(t) =0, t < t0−σsat. + σsat.αe−t−t0τs2(n−1)∑i=0(−eTτs)i+ σsat.(1 − α)e−t−t0τr2(n−1)∑i=0(−e−Tτr)i, t0 + 2(n − 1)T ≤ t < t0 + (2n − 1)Tσsat.αe−t−t0τs2n−1∑i=0(−eTτs)i+ σsat.(1 − α)e−t−t0τr2n−1∑i=0(−e−Tτr)i, t0 + (2n − 1)T ≤ t < t0 + 2nT, (1)where σsat. and α denote the amplitude of the signal and fitting parameter (see Ref. [19]for more detail); τr is the time constant of the stress relaxation for viscoelastic behaviorsof receptor materials; T and n are the duration and the number of injection/purge cycles.The analytical solution derived for multiple injection/purge cycles (Equation (1)) allows usto extract fitting parameters of the current measurements more accurately than previoussingle injection–purge model [18,19]. To extract values of receptor material properties anddiffusion time constant of gases from the experimental data, we used least squares methodswith trust region reflective algorithm using Python 3 with SciPy module according toour previous work [19]; σsat., τs, τr, EU/ER, and t0 were extracted using Equation (1) (seeRef. [19] for more detail).Biosensors 2023, 13, 152 5 of 132.8. Pattern Recognition AnalysisTo discriminate the silage samples, principal component analysis (PCA) was used forreducing dimensionality of the dataset. By projecting the data onto a lower-dimensionalspace, one can visually recognize each silage sample according to the cluster separation.The following parameters Sij were extracted from each MSS response as features for PCA:Sij = Sj(ti) - Sj(t0) with i = 1, 2, 3, 4, 5 (2)where Sj(t) and t0 denote a signal output of the jth receptor material at time t and the timewhen the signal response starts to rise (Figure 2). In this study, we chose four time pointsfor ti; t1 = t0 + 2 [s], t2 = t0 + 10 [s], t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s]. Threesets of the five parameters were extracted. PCA was adopted using scikit-learn packagesfor Python.Biosensors 2023, 13, x FOR PEER REVIEW 5 of 13  analytical solution derived for multiple injection/purge cycles (Equation (1)) allows us to extract fitting parameters of the current measurements more accurately than previous sin-gle injection–purge model [18,19]. To extract values of receptor material properties and diffusion time constant of gases from the experimental data, we used least squares meth-ods with trust region reflective algorithm using Python 3 with SciPy module according to our previous work [19]; σsat., τs, τr, EU/ER, and t0 were extracted using Equation (1) (see Ref. [19] for more detail). 2.8. Pattern Recognition Analysis To discriminate the silage samples, principal component analysis (PCA) was used for reducing dimensionality of the dataset. By projecting the data onto a lower-dimensional space, one can visually recognize each silage sample according to the cluster separation. The following parameters Sij were extracted from each MSS response as features for PCA: Sij = Sj(ti) – Sj(t0) with i = 1, 2, 3, 4, 5 (2) where Sj(t) and t0 denote a signal output of the jth receptor material at time t and the time when the signal response starts to rise (Figure 2). In this study, we chose four time points for ti; t1 = t0 + 2 [s], t2 = t0 + 10 [s], t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s]. Three sets of the five parameters were extracted. PCA was adopted using scikit-learn packages for Python.  Figure 2. A schematic diagram of the five parameters extracted from each response curve. 3. Results 3.1. Silage Samples Preparation and Their Headspace Gas Concentrations During silage fermentation, several compounds are produced such as VFAs and non-volatile lactic acid [1,2]. VFAs, such as acetic acid and butyric acid, are one of the key indicators to determine the fermentation quality of silage [2]; in general, anaerobic lactic acid fermentation results in good fermentation quality, whereas aerobic butyric acid fer-mentation (alteration) results in poor fermentation quality. To demonstrate the identifica-tion of silage with different fermentation qualities with an MSS array, we prepared corn silage in a silo and collected three silage samples from two different positions: the upper and the bottom. The results of the chemical analysis are listed in Table S1. Various VFAs including propionic acid were measured, while valeric acid was not detected with chem-ical analysis. The concentrations of VFAs in the headspace gas of each silage sample were measured with PTR-TOF-MS (Table 1). The headspace gases of silage samples used in the present study contained not only acetic acid, propionic acid, and butyric acid but also trace amount of valeric acid. Comparing the two different silage samples, the silage col-lected from the bottom (S2) yields higher VFAs than the silage collected from the upper (S1). Specifically, for the major VFAs in silage such as acetic acid and butyric acid, the concentrations in S2 were approximately 4.2 and 1.7 times higher than those in S1. These silage samples, in addition to water vapor, were used for the discrimination of silage sam-ples through their odors using an MSS array.   Figure 2. A schematic diagram of the five parameters extracted from each response curve.3. Results3.1. Silage Samples Preparation and Their Headspace Gas ConcentrationsDuring silage fermentation, several compounds are produced such as VFAs andnon-volatile lactic acid [1,2]. VFAs, such as acetic acid and butyric acid, are one of thekey indicators to determine the fermentation quality of silage [2]; in general, anaerobiclactic acid fermentation results in good fermentation quality, whereas aerobic butyricacid fermentation (alteration) results in poor fermentation quality. To demonstrate theidentification of silage with different fermentation qualities with an MSS array, we preparedcorn silage in a silo and collected three silage samples from two different positions: theupper and the bottom. The results of the chemical analysis are listed in Table S1. VariousVFAs including propionic acid were measured, while valeric acid was not detected withchemical analysis. The concentrations of VFAs in the headspace gas of each silage samplewere measured with PTR-TOF-MS (Table 1). The headspace gases of silage samples usedin the present study contained not only acetic acid, propionic acid, and butyric acid butalso trace amount of valeric acid. Comparing the two different silage samples, the silagecollected from the bottom (S2) yields higher VFAs than the silage collected from the upper(S1). Specifically, for the major VFAs in silage such as acetic acid and butyric acid, theconcentrations in S2 were approximately 4.2 and 1.7 times higher than those in S1. Thesesilage samples, in addition to water vapor, were used for the discrimination of silagesamples through their odors using an MSS array.3.2. MSS Responses to Headspace Gases of Silage SamplesTo investigate the response patterns to the VFA-rich silage samples, we preparedan MSS array coated with three different receptor materials: PS (hydrophobic material),PMMA (hydrophilic material, sensitive to water), and Tenax (hydrophobic material, widelyused as a trapping material) [14,19,21]. Using the MSS array, two sets of silage samples weremeasured. Each silage sample was measured three times. All signal responses are shown inFigure 2 (see also Figure S1). As can be seen from Figure 2 and Figure S1, signal responsesobtained from each sample are highly reproducible. The MSS coated with the hydrophilicBiosensors 2023, 13, 152 6 of 13receptor, i.e., PMMA, results in similar responses for all silage samples because of thedominant influences of water vapor on the sensing signals, since the major component insilage headspace gases is water vapor. In contrast, the MSS coated with the hydrophobicmaterials, i.e., PS and Tenax, yielded different signal patterns for each silage sample owingto its low affinity to water vapor. In the cases of PS- and Tenax-coated MSS (Figure 3a,c),the rising curves just after the sample gas injection at 0 s showed similar trends for allsamples, while significant differences were observed for each sample from about 10 s afterthe injection. Unlike the hydrophobic receptors, PMMA-coated MSS showed no significantdifferences in the rising curve (Figure 3b). In contrast to the rising curves, significantdifferences are observed in the decay curves not only for the hydrophobic receptors butalso for the hydrophilic receptor (Figure 3).Table 1. Organic acids in silage samples examined with chemical analysis and their vapor con-centrations measured with PTR-TOF-MS. Vapor concentrations are shown in the unit of parts permillion (ppm).Silage Plant Acetic Acid Propionic Acid Butyric Acid Valeric AcidS1-1 Corn(Upper) a15.5 ± 4.4 3.72 ± 0.42 9.84 ± 1.11 0.64 ± 0.05S1-2 31.7 ± 3.2 5.97 ± 0.41 8.78 ± 0.69 0.64 ± 0.01S1-3 26.1 ± 2.0 5.75 ± 0.42 9.27 ± 0.53 0.74 ± 0.05S2-1 Corn(Bottom) a108 ± 5 8.89 ± 0.60 16.8 ± 0.9 2.09 ± 0.12S2-2 103 ± 11 7.21 ± 0.66 15.2 ± 1.2 1.56 ± 0.14S2-3 98.8 ± 7.6 7.27 ± 0.49 15.4 ± 1.2 1.66 ± 0.11a Sampling position of silo.Biosensors 2023, 13, x FOR PEER REVIEW 6 of 13  Table 1. Organic acids in silage samples examined with chemical analysis and their vapor concen-trations measured with PTR-TOF-MS. Vapor concentrations are shown in the unit of parts per mil-lion (ppm). Silage Plant Acetic Acid Propionic Acid Butyric Acid Valeric Acid S1-1 Corn (Upper) a 15.5 ± 4.4 3.72 ± 0.42 9.84 ± 1.11 0.64 ± 0.05 S1-2 31.7 ± 3.2 5.97 ± 0.41 8.78 ± 0.69 0.64 ± 0.01 S1-3 26.1 ± 2.0 5.75 ± 0.42 9.27 ± 0.53 0.74 ± 0.05 S2-1 Corn (Bottom) a 108 ± 5 8.89 ± 0.60 16.8 ± 0.9 2.09 ± 0.12 S2-2 103 ± 11 7.21 ± 0.66 15.2 ± 1.2 1.56 ± 0.14 S2-3 98.8 ± 7.6 7.27 ± 0.49 15.4 ± 1.2 1.66 ± 0.11 a Sampling position of silo. 3.2. MSS Responses to Headspace Gases of Silage Samples To investigate the response patterns to the VFA-rich silage samples, we prepared an MSS array coated with three different receptor materials: PS (hydrophobic material), PMMA (hydrophilic material, sensitive to water), and Tenax (hydrophobic material, widely used as a trapping material) [14,19,21]. Using the MSS array, two sets of silage samples were measured. Each silage sample was measured three times. All signal re-sponses are shown in Figure 2 (see also Figure S1). As can be seen from Figures 2 and S1, signal responses obtained from each sample are highly reproducible. The MSS coated with the hydrophilic receptor, i.e., PMMA, results in similar responses for all silage samples because of the dominant influences of water vapor on the sensing signals, since the major component in silage headspace gases is water vapor. In contrast, the MSS coated with the hydrophobic materials, i.e., PS and Tenax, yielded different signal patterns for each silage sample owing to its low affinity to water vapor. In the cases of PS- and Tenax-coated MSS (Figure 3a,c), the rising curves just after the sample gas injection at 0 s showed similar trends for all samples, while significant differences were observed for each sample from about 10 s after the injection. Unlike the hydrophobic receptors, PMMA-coated MSS showed no significant differences in the rising curve (Figure 3b). In contrast to the rising curves, significant differences are observed in the decay curves not only for the hydro-phobic receptors but also for the hydrophilic receptor (Figure 3).  Figure 3. Signal responses to silage samples. Each panel from the top to the bottom shows the signal response measured with MSS coated with PS (a), PMMA (b), and Tenax (c), respectively. Magnified signal responses around t0 = 0 [s], t1 = 2 [s], t2 = 10 [s], t3 = 120 [s], t4 = 122 [s], and t5 = 130 [s] are shown Figure 3. Signal responses to silage samples. Each panel from the top to the bottom shows the signalresponse measured with MSS coated with PS (a), PMMA (b), and Tenax (c), respectively. Magnifiedsignal responses around t0 = 0 [s], t1 = 2 [s], t2 = 10 [s], t3 = 120 [s], t4 = 122 [s], and t5 = 130 [s]are shown on the right. Red, blue, and black lines correspond to silage 1 (S1), silage 2 (S2), andwater, respectively.Biosensors 2023, 13, 152 7 of 13From each response curve shown in Figure 3, five output values at t1 to t5 wereextracted as the features of the sensing signals (Figure 2). The extracted features aresummarized in Tables S2–S4. As can be seen in Figure 4, the values extracted at t2 to t5for PS and Tenax and the values extracted at t4 and t5 for PMMA yield clear differencesfor each sample. Although other values are rather scattered, the relative deviations of allsignal outputs obtained from each sample compared to the signal intensity (i.e., outputvalue at t = t3) were less than 1% as summarized in Tables S2–S4, indicating that eachreceptor shows high stability and reproducibility. Using the feature set in Figure 4, PCAwas conducted as shown in Figure 5. Scree plots are shown in Figure 4c. Each silagesample and the water vapor are clearly distinguished with well-separated clusters in theprincipal component space (Figure 5a,b). It should be noted that PC 2 does not contributeto identifying the silage samples. This may reflect a baseline drift of the sensing responseswith time, as features with small differences, such as the rising curve of PMMA, contributemore significantly because of the standardization of the feature set.Biosensors 2023, 13, x FOR PEER REVIEW 7 of 13  on the right. Red, blue, and black lines correspond to silage 1 (S1), silage 2 (S2), and water, respec-tively. From each response curve shown in Figure 3, five output values at t1 to t5 were ex-tracted as the features of the sensing signals (Figure 2). The extracted features are summa-rized in Tables S2–S4. As can be seen in Figure 4, the values extracted at t2 to t5 for PS and Tenax and the values extracted at t4 and t5 for PMMA yield clear differences for each sam-ple. Although other values are rather scattered, the relative deviations of all signal outputs obtained from each sample compared to the signal intensity (i.e., output value at t = t3) were less than 1% as summarized in Tables S2–S4, indicating that each receptor shows high stability and reproducibility. Using the feature set in Figure 4, PCA was conducted as shown in Figure 5. Scree plots are shown in Figure 4c. Each silage sample and the water vapor are clearly distinguished with well-separated clusters in the principal component space (Figure 5a,b). It should be noted that PC 2 does not contribute to identifying the silage samples. This may reflect a baseline drift of the sensing responses with time, as features with small differences, such as the rising curve of PMMA, contribute more sig-nificantly because of the standardization of the feature set.  Figure 4. Features extracted from each signal response. Each panel from the top to the bottom shows the plots of output extracted from the signal responses of PS (a), PMMA (b), and Tenax (c), respec-tively. Each panel from the left to the right shows the output at t1 = t0 + 2 [s], t2 = t0 + 10 [s], t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s], respectively. Figure 4. Features extracted from each signal response. Each panel from the top to the bottomshows the plots of output extracted from the signal responses of PS (a), PMMA (b), and Tenax (c),respectively. Each panel from the left to the right shows the output at t1 = t0 + 2 [s], t2 = t0 + 10 [s],t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s], respectively.Biosensors 2023, 13, 152 8 of 13Biosensors 2023, 13, x FOR PEER REVIEW 8 of 13   Figure 5. Principal component analysis (PCA). (a,b) PCA score plots of two sets of silage samples listed in Table 1 with principal component (PC) 1–2 (a) and PC 1–3 (b). Red, blue, and black corre-spond to silage 1 (S1), silage 2 (S2), and water, respectively. (c), Scree plot of eigenvalues. 3.3. MSS Responses to Each VFA Vapor To investigate the detailed mechanism of the responses of each receptor material to VFAs, we measured MSS responses to the vapors of the aqueous solutions of VFAs. Fig-ures S2–S4 show the signal responses to the series of VFAs with different concentrations in water, which cover the vapor concentrations of VFAs in silage samples (see also Figure S5 for the vapor concentrations measured with PTR-TOF-MS). From the signal responses to aqueous VFAs, the corresponding signal outputs at different time points ti (i = 1–5) were plotted as a function of the VFA concentrations in water (Figure 6). On the one hand, out-puts at t2 to t5 of hydrophobic PS increase monotonically with increasing concentrations of VFAs, while the outputs at the beginning of the rising curves (i.e., t = t1) show a decreas-ing trend with respect to the concentrations of VFAs (Figure 5a). On the other hand, most of the outputs during rising curves of hydrophilic PMMA (i.e., t = t1, t2, and t3) decrease with increasing VFA concentrations (Figure 6b). This trend probably reflects the decrease of water vapor according to Raoult’s law—the decrease in the mole fraction of water as-sociated with the increase in the concentration of VFAs in water. Compared to the rising curves, the outputs of all receptor materials in the decay curves at t = t4 and t5 exhibit a linear correlation with the VFA concentrations.  Figure 6. Signal output at ti of each receptor material to VFAs as a function of VFA concentration in water. (a) PS; (b) PMMA; (c) Tenax. Acetic acid (blue), propionic acid (red), butyric acid (green), valeric acid (purple), and caproic acid (orange) are shown. Each panel from the top to the bottom Figure 5. Principal component analysis (PCA). (a,b) PCA score plots of two sets of silage sampleslisted in Table 1 with principal component (PC) 1–2 (a) and PC 1–3 (b). Red, blue, and blackcorrespond to silage 1 (S1), silage 2 (S2), and water, respectively. (c), Scree plot of eigenvalues.3.3. MSS Responses to Each VFA VaporTo investigate the detailed mechanism of the responses of each receptor materialto VFAs, we measured MSS responses to the vapors of the aqueous solutions of VFAs.Figures S2–S4 show the signal responses to the series of VFAs with different concentrationsin water, which cover the vapor concentrations of VFAs in silage samples (see also FigureS5 for the vapor concentrations measured with PTR-TOF-MS). From the signal responsesto aqueous VFAs, the corresponding signal outputs at different time points ti (i = 1–5)were plotted as a function of the VFA concentrations in water (Figure 6). On the one hand,outputs at t2 to t5 of hydrophobic PS increase monotonically with increasing concentrationsof VFAs, while the outputs at the beginning of the rising curves (i.e., t = t1) show adecreasing trend with respect to the concentrations of VFAs (Figure 5a). On the otherhand, most of the outputs during rising curves of hydrophilic PMMA (i.e., t = t1, t2, andt3) decrease with increasing VFA concentrations (Figure 6b). This trend probably reflectsthe decrease of water vapor according to Raoult’s law—the decrease in the mole fractionof water associated with the increase in the concentration of VFAs in water. Compared tothe rising curves, the outputs of all receptor materials in the decay curves at t = t4 and t5exhibit a linear correlation with the VFA concentrations.One of the important aspects for explaining such complicated responses is the absorp-tion/desorption behaviors of the target gases. In nanomechanical sensing, an analyticalsolution was derived based on the first-order sorption kinetics with viscoelastic behaviorsof receptor materials in the form of Equation (1) [18,19]. According to the models, one of theparameters (i.e., diffusion time constant τs) reflects the absorption/desorption behaviorsof the target analyte in nanomechanical sensing [18,19,26]. Considering the first-ordersorption kinetics of the single injection–purge model (n = 1), the concentration of the targetanalyte in a receptor layer C(t) is given with [19]C(t) =0 , t < t0KpCg(1 − e−t−t0τs), t0 ≤ t < t3KpCg(1 − e−Tτs)e−t−t3τs , t3 ≤ t, (3)where Kp and Cg are the partition coefficient and gas concentration of the target analyte.When the diffusion time constant τs is relatively small and/or the duration T is long enough,e−T/τs in Equation (3) becomes approximately 0, resulting in the symmetric responsebetween the rising and decay curves.To estimate the diffusion time constants τs of VFAs and water, we measured the purevapors of VFAs (i.e., acetic acid, propionic acid, butyric acid, valeric acid, and caproic acid)as well as water. Figure S6 shows the signal responses to the pure vapors of VFAs dilutedwith pure nitrogen without water vapor. The signal responses to water vapor are alsopresented in Figure S6 for comparison. Using these signal responses, we apply the fittingof Equation (1) to each response and obtain the diffusion time constant τs of the series ofBiosensors 2023, 13, 152 9 of 13VFAs as shown in Figure 7. Some of the diffusion time constants τs of VFAs are largerthan that of water, indicating the slow desorption of VFAs. In such VFA vapors (e.g., aceticacid), asymmetric behavior is observed between the rising and decay curves, which can beexplained by Equation (3), resulting that each cycle of the injection and purge graduallyincreases (Figure S6). It indicates that a part of the absorbed VFAs remains in the receptorlayer during the purge process because the slow desorption requires more time than thepurging period (10 s).Biosensors 2023, 13, x FOR PEER REVIEW 8 of 13   Figure 5. Principal component analysis (PCA). (a,b) PCA score plots of two sets of silage samples listed in Table 1 with principal component (PC) 1–2 (a) and PC 1–3 (b). Red, blue, and black corre-spond to silage 1 (S1), silage 2 (S2), and water, respectively. (c), Scree plot of eigenvalues. 3.3. MSS Responses to Each VFA Vapor To investigate the detailed mechanism of the responses of each receptor material to VFAs, we measured MSS responses to the vapors of the aqueous solutions of VFAs. Fig-ures S2–S4 show the signal responses to the series of VFAs with different concentrations in water, which cover the vapor concentrations of VFAs in silage samples (see also Figure S5 for the vapor concentrations measured with PTR-TOF-MS). From the signal responses to aqueous VFAs, the corresponding signal outputs at different time points ti (i = 1–5) were plotted as a function of the VFA concentrations in water (Figure 6). On the one hand, out-puts at t2 to t5 of hydrophobic PS increase monotonically with increasing concentrations of VFAs, while the outputs at the beginning of the rising curves (i.e., t = t1) show a decreas-ing trend with respect to the concentrations of VFAs (Figure 5a). On the other hand, most of the outputs during rising curves of hydrophilic PMMA (i.e., t = t1, t2, and t3) decrease with increasing VFA concentrations (Figure 6b). This trend probably reflects the decrease of water vapor according to Raoult’s law—the decrease in the mole fraction of water as-sociated with the increase in the concentration of VFAs in water. Compared to the rising curves, the outputs of all receptor materials in the decay curves at t = t4 and t5 exhibit a linear correlation with the VFA concentrations.  Figure 6. Signal output at ti of each receptor material to VFAs as a function of VFA concentration in water. (a) PS; (b) PMMA; (c) Tenax. Acetic acid (blue), propionic acid (red), butyric acid (green), valeric acid (purple), and caproic acid (orange) are shown. Each panel from the top to the bottom Figure 6. Signal output at ti of each receptor material to VFAs as a function of VFA concentration inwater. (a) PS; (b) PMMA; (c) Tenax. Acetic acid (blue), propionic acid (red), butyric acid (green), valericacid (purple), and caproic acid (orange) are shown. Each panel from the top to the bottom shows theoutputs at t1 = t0 + 2 [s], t2 = t0 + 10 [s], t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s], respectively.Biosensors 2023, 13, x FOR PEER REVIEW 9 of 13  shows the outputs at t1 = t0 + 2 [s], t2 = t0 + 10 [s], t3 = t0 + 120 [s], t4 = t3 + 2 [s], and t5 = t3 + 10 [s], respectively. One of the important aspects for explaining such complicated responses is the ab-sorption/desorption behaviors of the target gases. In nanomechanical sensing, an analyti-cal solution was derived based on the first-order sorption kinetics with viscoelastic behav-iors of receptor materials in the form of Equation (1) [18,19]. According to the models, one of the parameters (i.e., diffusion time constant τs) reflects the absorption/desorption be-haviors of the target analyte in nanomechanical sensing [18,19,26]. Considering the first-order sorption kinetics of the single injection–purge model (n = 1), the concentration of the target analyte in a receptor layer C(t) is given with [19] 𝐶(𝑡) = ⎩⎪⎨⎪⎧0 , 𝑡 < 𝑡0𝐾 𝐶 1 − 𝑒 0 , 𝑡0 ≤ 𝑡 < 𝑡3𝐾 𝐶 1 − 𝑒 𝑒 3 , 𝑡3 ≤ 𝑡 , (3) where Kp and Cg are the partition coefficient and gas concentration of the target analyte. When the diffusion time constant τs is relatively small and/or the duration T is long enough, 𝑒 ⁄  in Equation (3) becomes approximately 0, resulting in the symmetric re-sponse between the rising and decay curves. To estimate the diffusion time constants τs of VFAs and water, we measured the pure vapors of VFAs (i.e., acetic acid, propionic acid, butyric acid, valeric acid, and caproic acid) as well as water. Figure S6 shows the signal responses to the pure vapors of VFAs diluted with pure nitrogen without water vapor. The signal responses to water vapor are also presented in Figure S6 for comparison. Using these signal responses, we apply the fitting of Equation (1) to each response and obtain the diffusion time constant τs of the series of VFAs as shown in Figure 7. Some of the diffusion time constants τs of VFAs are larger than that of water, indicating the slow desorption of VFAs. In such VFA vapors (e.g., acetic acid), asymmetric behavior is observed between the rising and decay curves, which can be explained by Equation (3), resulting that each cycle of the injection and purge gradually increases (Figure S6). It indicates that a part of the absorbed VFAs remains in the receptor layer during the purge process because the slow desorption requires more time than the purging period (10 s).  Figure 7. Estimated diffusion time constants τs of VFAs to each receptor material: PS (a); PMMA (b); Tenax (c). 4. Discussion In the present study, we demonstrated the discrimination of different silage samples through their odors (i.e., headspace gases) measured with a nanomechanical sensor array. We focused on VFAs, such as acetic acid and butyric acid, produced during silage fermen-tation because these VFAs are one of the key indicators to assess the fermentation quality of silage [2]. In the silage samples used in the present study, the series of VFAs, including Figure 7. Estimated diffusion time constants τs of VFAs to each receptor material: PS (a); PMMA (b);Tenax (c).4. DiscussionIn the present study, we demonstrated the discrimination of different silage samplesthrough their odors (i.e., headspace gases) measured with a nanomechanical sensor array.We focused on VFAs, such as acetic acid and butyric acid, produced during silage fermen-tation because these VFAs are one of the key indicators to assess the fermentation qualityof silage [2]. In the silage samples used in the present study, the series of VFAs, includingpropionic acid and valeric acid, were detected in their odors with PTR-TOF-MS (Table 1),although valeric acid was not detected through chemical analysis (Table S1). The concentra-Biosensors 2023, 13, 152 10 of 13tions of VFAs in the odors of one silage group were significantly different from those in theother group of silage, while the concentrations of VFAs in the same group were at similarlevels. When these silage samples including water vapor were measured with an MSS arraycoated with hydrophobic PS, hydrophilic PMMA, and popular trapping material Tenax,the signal responses showed different trends (Figures 3 and 4). By extracting five featuresfrom the responses of the MSS coated with each receptor material, the silage samples andwater were clearly discriminated with PCA (Figure 5).In the field of artificial olfaction including MSS-based systems, various feature ex-traction methods have been proposed [27,28]. We have also reported that it is possible toidentify and quantify various analytes using features extracted from their outputs withina couple of seconds of the rising and decay curves [28–31]. In the present silage samples,however, the features extracted from the beginning of the rising curves (e.g., at 2 s aftersample injection, t1) cannot be effectively used for the discrimination of samples proba-bly because of less correlation with their concentrations of VFAs (Figure 4). In contrastto the rising curves, the features extracted from the decay curves (i.e., t4 and t5) clearlyshow the significant differences reflecting their concentrations of VFAs in silage samples(Figure 4). While the hydrophobic receptors (i.e., PS and Tenax) yield clear differences att2–t5 (Figure 4a,c), the hydrophilic PMMA does not show the significant differences duringthe sampling process (i.e., rising curve at t1–t3) (Figure 4b). Since the major component ofthe silage headspace gases is water vapor, it is suggested that the rising curves of PMMAare strongly influenced by the water vapor concentrations rather than the differences in theconcentrations of VFAs in silage vapors.To evaluate the trends of silage odors in the signal responses, we also measured thesignal responses of aqueous solutions of VFAs varying their concentrations. It is noteworthythat the output of PS at t1 and the outputs of PMMA at t1–t3 tend to be negatively correlatedwith their concentrations (Figure 6a,b), probably reflecting the decrease in water vaporwith increasing VFAs in water, according to Raoult’s law. This trend was also observedin the signal responses of silage samples although it was not a linear correlation with theconcentrations of VFAs because mole fraction of water in the silage samples can be variedby not only VFAs but also non-volatile compounds such as non-volatile lactic acid. Incontrast to the sample injection period (i.e., rising curve), the outputs at t4 and t5 showedstrong correlation with each VFA concentration (Figure 6). The reason for the differentoutput trends in the rising and decay curves in Figure 6 can be attributed partially to theslow desorption of VFAs. According to the estimated diffusion time constant τs (Figure 7),some of the VFAs obtained relatively large time constants than that of water, resulting inthe asymmetric responses of slow desorption VFAs.In contrast to the above-mentioned VFAs, other VFA vapors obtained relatively smallτs (Figure 7), resulting in the symmetric responses (Figure S6). However, the responses tothe aqueous solutions of these VFAs also exhibit asymmetric behavior (Figure 6). Since theasymmetric responses occur only in the presence of water vapor, VFAs in the receptor layerwould dissociate into ionic forms in the presence of water. The sensing responses to VFAsin the presence of water can include the following reactions:H2O (g) � H2O (s), (4)R–CO2H (g) � R–CO2H (s), (5)R–CO2H (s) + H2O (s) � R–CO−2 (s) + H3O+ (s), (6)where g and s in parenthesis denote the molecule in the gas phase and in the solid phase(i.e., in the receptor layer), respectively. The potential scheme is as follows (Figure 8):(i) during injection process, the sorption of water and VFAs represented by Equations (4)and (5) contributes to the deformation of a receptor layer, while the dissociation processin Equation (6) does not affect the deformation (i.e., does not affect the signal response);(ii) during the purge process, the concentrations of VFAs and water in the receptor layerBiosensors 2023, 13, 152 11 of 13decrease because of the dissociation process in Equation (6), resulting in the slow desorptionof water and VFAs, i.e., a delay in the desorption.In conclusion, we have demonstrated that an MSS array can discriminate silagesamples through their odors in terms of VFAs as indicators. Focusing on the relativelyslow desorption behaviors and dissociation nature of VFAs, the resultant large diffusiontime constants can be effectively utilized for the discrimination of silage samples. In otherwords, we have shown that the sensing signal outputs of MSS in the relatively late periodin the purge process can be utilized for detecting, identifying, and monitoring varioustargets in complex mixtures of odors, such as VFAs in silage. This approach should beapplicable to other targets such as agricultural and biological aqueous samples becausesuch samples tend to contain gas species with relatively slow desorption behavior. Further,since nanomechanical sensors including MSS can utilize almost any material as a receptorlayer, one can select/design/synthesize receptor materials that exhibit distinct features in acertain period during adsorption and/or desorption.Biosensors 2023, 13, x FOR PEER REVIEW 11 of 13  Equation (6) does not affect the deformation (i.e., does not affect the signal response); (ii) during the purge process, the concentrations of VFAs and water in the receptor layer de-crease because of the dissociation process in Equation (6), resulting in the slow desorption of water and VFAs, i.e., a delay in the desorption.  Figure 8. Plausible mechanism of VFA detection. (a) Working principle of MSS and their related reactions. (b) Signal-related reactions in the absorption and desorption processes. In conclusion, we have demonstrated that an MSS array can discriminate silage sam-ples through their odors in terms of VFAs as indicators. Focusing on the relatively slow desorption behaviors and dissociation nature of VFAs, the resultant large diffusion time constants can be effectively utilized for the discrimination of silage samples. In other words, we have shown that the sensing signal outputs of MSS in the relatively late period in the purge process can be utilized for detecting, identifying, and monitoring various targets in complex mixtures of odors, such as VFAs in silage. This approach should be applicable to other targets such as agricultural and biological aqueous samples because such samples tend to contain gas species with relatively slow desorption behavior. Further, since nanomechanical sensors including MSS can utilize almost any material as a receptor layer, one can select/design/synthesize receptor materials that exhibit distinct features in a certain period during adsorption and/or desorption. Although the precise determination of the fermentation quality of silage requires not only VFAs but also non-volatile lactic acid, total nitrogen, volatile basic nitrogen, and wa-ter content [2], the amounts of VFAs in silage can provide additional information on feed design to maintain the conditions of dairy cows; the amount of butyric acid is related to the probability of ketosis occurrence as well as the activity of Clostridium. Moreover, us-ing precisely selected receptor materials, it has a possibility to measure the volatile basic nitrogen through their odors. The decrease of moisture according to Raoult’s law can also be detected by using hydrophilic receptors, as can be seen in Figure 4b. Therefore, the artificial olfaction system may allow us to evaluate non-volatile components such as lactic acid indirectly by measuring water contents. Machine learning may also support such ap-proaches by highlighting subtle features that may not necessarily appear at specific time points. It should be emphasized that the artificial olfaction system does not require expen-sive instruments and pre-treatment, such as water extraction, which are used for the con-ventional chemical analysis, and hence, it has a great potential for on-site evaluation of silage fermentation quality. Figure 8. Plausible mechanism of VFA detection. (a) Working principle of MSS and their relatedreactions. (b) Signal-related reactions in the absorption and desorption processes.Although the precise determination of the fermentation quality of silage requires notonly VFAs but also non-volatile lactic acid, total nitrogen, volatile basic nitrogen, and watercontent [2], the amounts of VFAs in silage can provide additional information on feeddesign to maintain the conditions of dairy cows; the amount of butyric acid is relatedto the probability of ketosis occurrence as well as the activity of Clostridium. Moreover,using precisely selected receptor materials, it has a possibility to measure the volatile basicnitrogen through their odors. The decrease of moisture according to Raoult’s law canalso be detected by using hydrophilic receptors, as can be seen in Figure 4b. Therefore,the artificial olfaction system may allow us to evaluate non-volatile components such aslactic acid indirectly by measuring water contents. Machine learning may also supportsuch approaches by highlighting subtle features that may not necessarily appear at specifictime points. It should be emphasized that the artificial olfaction system does not requireexpensive instruments and pre-treatment, such as water extraction, which are used for theconventional chemical analysis, and hence, it has a great potential for on-site evaluation ofsilage fermentation quality.Biosensors 2023, 13, 152 12 of 135. PatentsK.M., H.K., T.N., M.T., R.N., and G.Y. are inventors on WIPO patent application num-ber WO2021/200262, submitted by National Agriculture and Food Research Organization(NARO) and National Institute for Materials Science (NIMS).Supplementary Materials: The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/bios13020152/s1, Figure S1: Raw signal responses; Figures S2–S4: Si-gnal responses of aqueous solutions of VFAs to each receptor material; Figure S5: Vapor concentrationof VFAs in water; Figure S6: Signal responses of pure VFAs; Table S1: Results of chemical analysis ofsilage samples; Tables S2–S4, Summary of the extracted features from the responses to silage samples.Author Contributions: Conceptualization, R.N. and G.Y.; methodology, K.M., R.N. and G.Y.; soft-ware, G.Y.; validation, K.M.; formal analysis, K.M., H.K., M.M. and T.N.; investigation, K.M.; re-sources, S.M. and Y.K.; data curation, K.M. and H.K.; writing—original draft preparation, K.M.;writing—review and editing, H.K., M.M., Y.K., S.M., T.N., M.T., R.N. and G.Y.; visualization, K.M.;supervision, M.T., R.N. and G.Y.; project administration, G.Y.; funding acquisition, K.M. and G.Y. Allauthors have read and agreed to the published version of the manuscript.Funding: This study was financially supported by a Grant-in-Aid for Scientific Research (A), JSPS,MEXT, Japan (No. 18H04168); a Grant-in-Aid for Scientific Research (C), JSPS, MEXT, Japan (No.22K05324); a Grant-in-Aid for Challenging Research (Pioneering) (No. 20K20554); the Public/PrivateR&D Investment Strategic Expansion Program (PRISM), Cabinet Office, Japan; and Center forFunctional Sensor & Actuator (CFSN), NIMS.Data Availability Statement: Not applicable.Acknowledgments: We would like to sincerely express our heartfelt gratitude to Megumi Ezure andNoriko Tsukie for their expert advice, technical assistance, and insightful discussions. We would liketo thank Asahi Kasei Co., Ltd. for providing MSS chips.Conflicts of Interest: M.M., Y.K., and S.M. declare no conflict of interest.References1. Driehuis, F.; Wilkinson, J.M.; Jiang, Y.; Ogunade, I.; Adesogan, A.T. Silage review: Animal and human health risks from silage.J. Dairy Sci. 2018, 101, 4093–4110. [CrossRef] [PubMed]2. Kung, L., Jr.; Shaver, R.D.; Grant, R.J.; Schmidt, R.J. Silage review: Interpretation of chemical, microbial, and organolepticcomponents of silages. J. Dairy Sci. 2018, 101, 4020–4033. [CrossRef]3. Tveit, B.; Lingaas, F.; Svendsen, M.; Sjaastad, Ø.V. Etiology of Acetonemia in Norwegian Cattle. 1. Effect of Ketogenic Silage,Season, Energy Level, and Genetic Factors. J. Dairy Sci. 1992, 75, 2421–2432. [CrossRef]4. Lingaas, F.; Tveit, B. Etiology of acetonemia in Norwegian cattle. 2. Effect of butyric acid, valeric acid, and putrescine. J. Dairy Sci.1992, 75, 2433–2439. [CrossRef]5. Baath, G.S.; Baath, H.K.; Gowda, P.H.; Thomas, J.P.; Northup, B.K.; Rao, S.C.; Singh, H. Predicting Forage Quality of Warm-SeasonLegumes by Near Infrared Spectroscopy Coupled with Machine Learning Techniques. Sensors 2020, 20, 867. [CrossRef]6. Persaud, K.; Dodd, G. Analysis of discrimination mechanisms in the mammalian olfactory system using a model nose. Nature1982, 299, 352–355. [CrossRef]7. Wilson, A.D.; Baietto, M. Applications and advances in electronic-nose technologies. Sensors 2009, 9, 5099–5148. [CrossRef][PubMed]8. Baldwin, E.A.; Bai, J.; Plotto, A.; Dea, S. Electronic noses and tongues: Applications for the food and pharmaceutical industries.Sensors 2011, 11, 4744–4766. [CrossRef] [PubMed]9. Wilson, A.D. Review of Electronic-nose Technologies and Algorithms to Detect Hazardous Chemicals in the Environment. Proc.Technol. 2012, 1, 453–463. [CrossRef]10. Gutierrez, J.; Horrillo, M.C. Advances in artificial olfaction: Sensors and applications. Talanta 2014, 124, 95–105. [CrossRef][PubMed]11. Kou, L.; Zhang, D.; Liu, D. A Novel Medical E-Nose Signal Analysis System. Sensors 2017, 17, 402. [CrossRef] [PubMed]12. Kim, C.; Raja, I.S.; Lee, J.M.; Lee, J.H.; Kang, M.S.; Lee, S.H.; Oh, J.W.; Han, D.W. Recent Trends in Exhaled Breath DiagnosisUsing an Artificial Olfactory System. Biosensors 2021, 11, 337. [CrossRef] [PubMed]13. Shooshtari, M.; Salehi, A. An electronic nose based on carbon nanotub-titanium dioxide hybrid nanostructures for detection anddiscrimination of volatile organic compounds. Sens. Actuators B Chem. 2022, 357, 131418. [CrossRef]14. Minami, K.; Imamura, G.; Tamura, R.; Shiba, K.; Yoshikawa, G. Recent Advances in Nanomechanical Membrane-Type SurfaceStress Sensors towards Artificial Olfaction. Biosensors 2022, 12, 762. [CrossRef]https://www.mdpi.com/article/10.3390/bios13020152/s1https://www.mdpi.com/article/10.3390/bios13020152/s1http://doi.org/10.3168/jds.2017-13836http://www.ncbi.nlm.nih.gov/pubmed/29685279http://doi.org/10.3168/jds.2017-13909http://doi.org/10.3168/jds.S0022-0302(92)78003-5http://doi.org/10.3168/jds.S0022-0302(92)78004-7http://doi.org/10.3390/s20030867http://doi.org/10.1038/299352a0http://doi.org/10.3390/s90705099http://www.ncbi.nlm.nih.gov/pubmed/22346690http://doi.org/10.3390/s110504744http://www.ncbi.nlm.nih.gov/pubmed/22163873http://doi.org/10.1016/j.protcy.2012.02.101http://doi.org/10.1016/j.talanta.2014.02.016http://www.ncbi.nlm.nih.gov/pubmed/24767451http://doi.org/10.3390/s17040402http://www.ncbi.nlm.nih.gov/pubmed/28379168http://doi.org/10.3390/bios11090337http://www.ncbi.nlm.nih.gov/pubmed/34562928http://doi.org/10.1016/j.snb.2022.131418http://doi.org/10.3390/bios12090762Biosensors 2023, 13, 152 13 of 1315. Okubo, M.; Sato, K.; Matsuda, S.; Masuko, T.; Souma, K. Data on chemical compositions and fermentation quality of silages madefrom low-market-value vegetables supplemented with potato protein concentrate, a byproduct of starch production. Data Brief2018, 21, 1829–1832. [CrossRef]16. Yoshikawa, G.; Akiyama, T.; Gautsch, S.; Vettiger, P.; Rohrer, H. Nanomechanical membrane-type surface stress sensor. Nano Lett.2011, 11, 1044–1048. [CrossRef]17. Yoshikawa, G.; Akiyama, T.; Loizeau, F.; Shiba, K.; Gautsch, S.; Nakayama, T.; Vettiger, P.; de Rooij, N.F.; Aono, M. Twodimensional array of piezoresistive nanomechanical Membrane-type Surface Stress Sensor (MSS) with improved sensitivity.Sensors 2012, 12, 15873–15887. [CrossRef]18. Wenzel, M.J.; Josse, F.; Heinrich, S.M.; Yaz, E.; Datskos, P.G. Sorption-induced static bending of microcantilevers coated withviscoelastic material. J. Appl. Phys. 2008, 103, 064913. [CrossRef]19. Minami, K.; Shiba, K.; Yoshikawa, G. Sorption-induced static mode nanomechanical sensing with viscoelastic receptor layers formultistep injection-purge cycles. J. Appl. Phys. 2021, 129, 124503. [CrossRef]20. Tohno, M.; Kobayashi, H.; Tajima, K.; Uegaki, R. Strain-dependent effects of inoculation of Lactobacillus plantarum subsp.plantarum on fermentation quality of paddy rice (Oryza sativa L. subsp. japonica) silage. FEMS Microbiol. Lett. 2012, 337, 112–119.[CrossRef] [PubMed]21. Inada, K.; Kojima, H.; Cho-Isoda, Y.; Tamura, R.; Imamura, G.; Minami, K.; Nemoto, T.; Yoshikawa, G. Statistical Evaluation ofTotal Expiratory Breath Samples Collected throughout a Year: Reproducibility and Applicability toward Olfactory Sensor-BasedBreath Diagnostics. Sensors 2021, 21, 4742. [CrossRef] [PubMed]22. MSS Alliance Launched to Set De Facto Standard for Odor-Sensing Systems. 2015. Available online: https://www.nims.go.jp/eng/news/press/2015/10/201510130.html (accessed on 16 December 2022).23. “MSS Forum” Launched to Promote the Establishment of a De Facto Standard for Olfactory IoT Sensing Systems. 2017. Availableonline: https://www.nims.go.jp/eng/news/press/2017/201710160.html (accessed on 16 December 2022).24. MSS Forum. Available online: https://mss-forum.com/en/ (accessed on 16 December 2022).25. Malaskova, M.; Olivenza-Leon, D.; Chellayah, P.D.; Martini, J.; Lederer, W.; Ruzsanyi, V.; Unterkofler, K.; Mochalski, P.; Mark,T.D.; Watts, P.; et al. Studies pertaining to the monitoring of volatile halogenated anaesthetics in breath by proton transfer reactionmass spectrometry. J. Breath Res. 2020, 14, 026004. [CrossRef] [PubMed]26. Hu, Z.; Thundat, T.; Warmack, R.J. Investigation of adsorption and absorption-induced stresses using microcantilever sensors.J. Appl. Phys. 2001, 90, 427–431. [CrossRef]27. Yan, J.; Guo, X.; Duan, S.; Jia, P.; Wang, L.; Peng, C.; Zhang, S. Electronic Nose Feature Extraction Methods: A Review. Sensors2015, 15, 27804–27831. [CrossRef] [PubMed]28. Minami, K.; Shiba, K.; Yoshikawa, G. Discrimination of structurally similar odorous molecules with various concentrations byusing a nanomechanical sensor. Anal. Methods 2018, 10, 3720–3726. [CrossRef]29. Shiba, K.; Tamura, R.; Imamura, G.; Yoshikawa, G. Data-driven nanomechanical sensing: Specific information extraction from acomplex system. Sci. Rep. 2017, 7, 3661. [CrossRef]30. Shiba, K.; Tamura, R.; Sugiyama, T.; Kameyama, Y.; Koda, K.; Sakon, E.; Minami, K.; Ngo, H.T.; Imamura, G.; Tsuda, K.; et al.Functional Nanoparticles-Coated Nanomechanical Sensor Arrays for Machine Learning-Based Quantitative Odor Analysis. ACSSens. 2018, 3, 1592–1600. [CrossRef]31. Xu, H.; Kitai, K.; Minami, K.; Nakatsu, M.; Yoshikawa, G.; Tsuda, K.; Shiba, K.; Tamura, R. Determination of quasi-primary odorsby endpoint detection. Sci. Rep. 2021, 11, 12070. [CrossRef]Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individualauthor(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury topeople or property resulting from any ideas, methods, instructions or products referred to in the content.http://doi.org/10.1016/j.dib.2018.11.043http://doi.org/10.1021/nl103901ahttp://doi.org/10.3390/s121115873http://doi.org/10.1063/1.2902500http://doi.org/10.1063/5.0039045http://doi.org/10.1111/1574-6968.12014http://www.ncbi.nlm.nih.gov/pubmed/23003205http://doi.org/10.3390/s21144742http://www.ncbi.nlm.nih.gov/pubmed/34300482https://www.nims.go.jp/eng/news/press/2015/10/201510130.htmlhttps://www.nims.go.jp/eng/news/press/2015/10/201510130.htmlhttps://www.nims.go.jp/eng/news/press/2017/201710160.htmlhttps://mss-forum.com/en/http://doi.org/10.1088/1752-7163/ab5e30http://www.ncbi.nlm.nih.gov/pubmed/31796655http://doi.org/10.1063/1.1378333http://doi.org/10.3390/s151127804http://www.ncbi.nlm.nih.gov/pubmed/26540056http://doi.org/10.1039/C8AY01224Ehttp://doi.org/10.1038/s41598-017-03875-7http://doi.org/10.1021/acssensors.8b00450http://doi.org/10.1038/s41598-021-91210-6 Introduction  Materials and Methods  Materials  Plant Materials and Preparation of Silage  Chemical Analysis of Silage Samples  Fabrication of MSS  Sensing  VFA Profile in Headspace Gas  Curve Fitting of the Signal Response  Pattern Recognition Analysis  Results  Silage Samples Preparation and Their Headspace Gas Concentrations  MSS Responses to Headspace Gases of Silage Samples  MSS Responses to Each VFA Vapor  Discussion  Patents  References