Advancing Breath‐Based Diagnostics: 3D Mesh SERS Sensor Via Dielectrophoretic Alignment of Solution‐Processed Au Nanoparticle‐Decorated TiO2 Nanowires

Surface enhanced Raman spectroscopy (SERS) is becoming an attractive analytical technique for the next generation of breath diagnostics. However, current SERS substrates present challenges related to fabrication cost, complexity, signal uniformity, and reproducibility. Here, a low‐cost, label‐free SERS sensor based on fully solution‐processed decoration of TiO2 nanowires is demonstrated (NW) with plasmonic Au nanoparticles (NP) followed by the dielectrophoretic self‐assembly into a 3D mesh with high signal to noise ratio. The sensor performance is tested using 4‐aminothiophenol (4‐ATP) as a model analyte in gas phase, at concentrations down to 10 ppbv, and in solution, with limit of detection ≈2.4 pM. Finally, to explore the sensor capability for breath‐based diagnostics, a proof‐of‐concept experiment is performed with exhaled breath condensates (EBCs). The possibility to discriminate EBCs of individuals with upper respiratory tract infection (URTI) from healthy ones is demonstrated. Multiple SERS spectra (n≈50) from each sample are analyzed using orthogonal partial least squares discriminant analysis (OPLS‐DA), which identifies spectral features representative of URTI in up to 80% of the infection‐related spectra. These results demonstrate the applicability and potential of 1D nanomaterials together with state‐of‐the‐art solution‐processed techniques for the development of low‐cost and compact SERS breath‐based diagnostic platforms for clinical point‐of‐care applications.


Introduction
Breath analysis is an emerging strategy for the evaluation of the physiological state of the human body and the early diagnosis of DOI: 10.1002/adsr.202300161disease. [1,2]Breath-based diagnostics can enable rapid and cost-effective screening, which has already been proven invaluable during the recent pandemic. [3,4]Additionally, the need for non-invasive diagnostic tools in oncology, able to detect cancer early, on a large scale, and at a low cost remains unmet.The molecular profile of breath has been linked to the associated type of cancer and its stage and/or grade, [5] making it an ideal sample for fast and cost-effective cancer screening.However, the low abundance of diagnostically indicative molecules in breath necessitates high-performance analytical techniques, such as GC-MS, [6] which rely on costly equipment and trained personnel, rendering it prohibitive for largescale breath-based diagnostics.
An emerging diagnostic approach is offered through surface enhanced Raman spectroscopy (SERS).This optical spectroscopic technique has been established as a versatile tool serving as an ultrasensitive and non-destructive analytical method for the rapid detection of chemicals and biological molecules. [7]ERS is based on the Raman effect, which upon laser illumination provides a characteristic optical spectrum identifying the substance under examination.In SERS, the Raman signal of a molecule is amplified mainly via the excitation of localized surface plasmon resonance (LSPR) on metal nanostructures.Metallic nanoparticles (NPs) can amplify the Raman scattered photons by up to 15 orders of magnitude, [8] with the maximum enhancement reported to stem from low-order nanoparticle aggregates. [9]hese NP clusters produce areas of intense electric field, often referred to as 'hot-spots', that greatly increase the plasmonic enhancement.In this way, SERS boosts the spectral signal of any molecule found in a hot-spot and reveals its unique chemical fingerprint with single-molecule sensitivity. [8]ERS-based chemical detection offers high sensitivity, molecular specificity, and high multiplexing capacity, while it requires relatively low-cost optical components that are amenable to miniaturization; thus, it is an attractive and feasible choice for breath diagnostics.SERS detection of volatile organic compounds (VOCs) present in breath has been demonstrated as a promising tool not only for the early diagnosis of cancer [5,[10][11][12][13][14][15] but also for mass screening for pandemic surveillance, [16] while SERS has also been applied to breath analysis for the detection of VOCs linked to metabolism. [17]evertheless, significant challenges keep the commercial development of SERS-based detection methods. [18,19]To date, the majority of reported SERS substrates are based on the bottom-up assembly approach, as it offers a simple, fast, and cost-effective solution for SERS substrate preparation. [20]This approach can be further categorized into colloidal and non-colloidal substrates.Colloidal substrates allow the NPs and analyte molecules to mix in an aqueous phase, resulting in NP aggregation and analyte adsorption in the hot-spot locations. [21][24] This technique offers a simplistic, quick, and low-cost SERS substrate fabrication, suitable for feasibility studies in a lab setting.However, it produces an uncontrolled distribution of NPs, including areas of low aggregation and over-aggregation, that lead to low uniformity, and low reproducibility, thus hampering the sensor's performance and amenability to commercial deployment.On the other hand, non-colloidal substrates, where NPs are deposited and/or selfassembled to form nanostructures on a solid surface followed by the analyte exposure and adsorption, offer higher uniformity and reproducibility. [21]However, non-colloidal substrate fabrication can become complex, costly, and time-consuming.In both categories, the controllable aggregation of NPs is important for the formation of hot-spots and the production of an effective SERS sensor. [18,25]n effective alternative for controlling the state of aggregation of plasmonic NPs is by binding them onto nanowires (NWs).Assembly of NPs on NWs has attracted interest for SERS diagnostic applications, due to the high enhancement factors and low detection limits reported, yielding various types of NPdecorated NWs.For example, vertical gold (Au) NWs followed by Au NP deposition were developed for gastric and breast cancer diagnosis. [26]A similar approach was followed for the fabrication of dense vertical Si NWs coated with Au-silver (Ag) bimetallic NPs for the detection of trace analytes. [27]Vertically grown Si NWs coated with Ag NPs or Au NPs have also been reported for biological detection, [28,29] food testing, [30] biosensing, [31] and medical [32] applications.Other groups, reported the Au decoration of SnO 2 NWs and ZnO nanoarrays for chemical analysis applications. [33,34]The reports to date highlight the use of 1D nanostructures for the formation of dense configurations of plasmonic NPs for high-performance SERS applications.
The controllable assembly of SERS-active NWs into a dense 3D mesh can potentially enable the detection of trace analytes with high Raman signal intensity and improved detection limit.Several assembly techniques, each of them carrying their own particularities, have been developed for the assembly of 1D nanostructures into 2D and 3D geometries.The main challenge in self-assembly is the alignment of 1D nanostructures on predefined substrate areas.Many reported assembly techniques, including Langmuir Blodgett, [35] Blown-Bubble films, [36] flowdirected assembly, [37] electrostatic interactions, [38] and mechanical shear forces [39] can align NWs along one direction but provide no control on the NWs' lateral positioning.Dielectrophoresis (DEP), [40] the controlled positioning of micro-and nano-particles via alternating electric fields, allows both alignment and position control for micro-and nano-particles.DEP is the most widely studied, experimented, and analyzed technique for the self-assembly of NWs on pre-defined locations, providing controllable deposition densities that make it suitable for SERS sensing applications. [41]ere, we present a state-of-the-art, low-cost sensor for SERSbased breath diagnostics, based on DEP alignment of NWs decorated with plasmonic NPs.As illustrated in Figure 1, the device architecture features TiO 2 NWs, decorated with SERS-active Au NPs using a solution-processed technique aligned between two electrodes.DEP was used to assemble the NWs into a 3D mesh of roughly parallel TiO 2 NWs-Au NPs, enabling the substantial increase of the locally available SERS-active surface of the device, in a microscopically defined area.The importance of the presented SERS sensor is twofold.First, a fully solution-processed technique for the rapid Au NP decoration of the TiO 2 NWs was developed; this method is quick, time-and cost-effective, and can be conducted at room temperature, obviating the use of cleanroom-reliant micro-and nano-fabrication techniques.Second, DEP-based self-assembly was employed to shape the NWs into higher-order structures; DEP offers unique features, including NW alignment on predefined locations, preferential orientation, and controllable formation of a dense plasmonic NP assembly. [41]This novel material architecture increases the overall hot-spot availability to analytes (airborne or aqueous), and combines strong SERS signal amplification with increased sample capture, allowing the detection of analytes at trace concentrations.
The performance of the SERS sensor was investigated using 4-aminothiophenol (4-ATP) as a model analyte in gas and liquid phase.This analyte is used in many studies, as it has a high affinity for the plasmonic substrates and a strong SERS signal.In addition, numerical simulations were conducted in COM-SOL to evaluate the plasmonic properties of the material.Finally, a proof-of-concept study in breath diagnosis was conducted, demonstrating the applicability and potential of the sensor for the development of the next generation of exhaled breath screening assays.

Synthesis of Au Nanoparticle-Decorated TiO 2 Nanowires
The list of materials used, the details of the synthesis procedure, and the characterization tools are provided in the Experimental Section.The completed device is composed of TiO 2 NWs decorated with Au NPs, aligned along two electrodes via DEP, as shown in Figure 2a.
A fully solution-processed technique was used for the Au NP decoration of the TiO 2 NW.Sodium borohydride (NaBH 4 ) was used as a mild reducing agent to obtain Au NPs via the  reduction of chloroauric acid (HAuCl 4 ).TiO 2 NWs were dispersed in anisole to which HAuCl 4 was added, followed by NaBH 4 .The reduction of gold ions on the surface of the TiO 2 NWs begins with the formation of Au seeds, which then grow via Ostwald ripening over time. [42][45] Electron microscopy verified the presence of Au NPs on the NWs (Figure 2b,c).Experimentally, the formation of Au NPs can be observed by the change in color from yellow-gold to light purple.The color change indicates immediate nucleation and rapid growth of Au NPs at room temperature.The NP size can be tuned by changing the NaBH 4 to HAuCl 4 concentrations (Figures S1,S2, Table S1, Supporting Information).The final nanoparticle size distribution was determined via TEM analysis and is shown in Figure 2d, with a mean size of 40 nm and a mode of 30-40 nm.The nanowire distribution was similarly determined via TEM analysis and is shown in Figure 2e, with a mean diameter of 70 nm.
The presence of plasmonic NPs on the surface of TiO 2 NWs is further probed optically with UV-vis absorption spectroscopy showing a plasmonic peak 596 nm (Figure 2f).The broad absorption band demonstrates the size variability of the grown Au NPs due to the NW's non-uniform, porous surface, and the various small-order NP aggregates formed on the wire.

COMSOL Simulations
The local electric field distribution and absorbance cross-section of Au NPs attached to a single TiO 2 NW were simulated using COMSOL Multiphysics (v5.6) as described in the Experimental Section.Au NPs were simulated in close proximity and in direct physical contact to investigate the effects of NP geometry to the plasmonic enhancement.
Figure 3a,b shows the calculated electric field enhancement in a logarithmic scale (log 10 (E/E 0 )) in the z-x plane upon excitation with 785 nm light oscillating in the z direction, with Au NP interparticle distances of 5 nm and 0 nm.The electric field around the contact area of the two Au NPs is enhanced up to almost three orders of magnitude due to the coupling effect of surface plasmons on the Au NPs.Specifically, the maximum log 10 electric field enhancement for 785 nm excitation was 1.42 for the NPs with 5 nm distance, and 2.70 for the two NPs in contact.In both cases the enhancement is more pronounced in the areas of close proximity of the NPs, emphasizing the significance of the inter-particle distance for the Raman enhancement of the SERS sensor.
Figure 3C illustrates the calculated absorbance spectrum for the Au NPs.The peak around 320 nm is due to the absorbance of the TiO 2 NW, while the two peaks in the optical range correspond to the plasmonic resonances of the Au NPs.The variation in the inter-particle distance strongly affects the absorbance spectrum.The absorbance associated with the NPs in physical contact exhibits a substantial bathochromic shift due to the morphological change of the plasmonic structure (the NPs appear as a single elongated nanostructure in the z-direction).As expected, when two NPs are in contact and aligned to the electric field polarization direction, they show additional resonances related to the aggregate, hence, their absorbance is enhanced and shifted towards higher wavelengths. [46]The simulated results of the Au NP absorption explain the broad absorbance spectrum shown in Figure 2e.While the mean size of the plasmonic nanoparticles was measured ≈30-40 nm (Figure 2d) it is evident that the absorbance profile is dominated by the larger nanoparticles (around 60 nm diameter) and their low-order aggregates.Figure S3 (Supporting Information) shows the absorption spectrum of 60 nm Au NPs compared to 30 nm ones.Corroborating the experimental measurements, the plasmonic activity of the larger Au NPs is markedly higher than that of the smaller NPs and provides greater contribution in the Raman spectra.

Dielectrophoretic Alignment of NWs
The TiO 2 NWs-Au NPs assembly and alignment were performed via DEP on a Si/SiO 2 substrate with platinum (Pt) electrodes separated by a 10 μm gap.This approach allows the positioning of NWs on predefined substrate areas, with preferential orientation along the inter-electrode gap, with controllable density of the deposited nanostructures, as well as length selection relative to the electrode gap. [41]EP refers to the translational motion of particles induced via the polarization effects induced by non-uniform electric fields, as defined by Herbert Pohl. [47]When a particle is found in a region with a high electric field, it becomes polarized as its inherent charges are displaced due to the field.Hence, the particle moves in the direction of the denser electric field and is positioned across the electrodes by means of a dielectrophoretic force (F DEP ).For asymmetric particles such as NWs, the F DEP is maximized when the NWs are oriented across the electrode gap and perpendicular to the electrode edges.If the F DEP acting on the nanostructures exceeds the random effects of Brownian motion, the nanostructures are positioned across the electrode gap; otherwise, the nanostructures escape.The magnitude of F DEP depends on the applied voltage (V), the frequency of the electric field (Hz), the material and solvent dielectric properties (), as well as the geometry of the micro-electrodes. [40,48]e can model the NWs as cylindrical particles, such that the F DEP can be approximated to that of a long prolate ellipsoid, [40,41,[49][50][51] given by: where, r is the radius of the NW, L is its length,  m is the permittivity of the fluid medium, and ∇E 2 is the gradient of the local applied electric field squared.K f is the Clausius-Mossotti factor calculated for an ellipsoid, [41] given by εp − εm εm , where the relative permittivity is expressed by the complex combination of conductivity (), permittivity (), and angular velocity of the electric field () as εp =  − j  p  for the particle and εm =  − j  m  for the medium, and j = √ (−1) .According to the equation, the NW deposition depends on the value of K f (hence, the electric field frequency) and strongly on the electric field gradient.Effectively, for the NWs to be positioned in the region of the highest electric field, i.e. in the electrode gap, the value of K f must be positive; that means the particle must be more polarizable than the medium for that specific frequency, inducing what is referred to as positive DEP (pDEP).Otherwise, the particle is repelled from the regions of high electric field (negative DEP; nDEP).
The DEP assembly offers a high degree of selectivity for NW subpopulations as it exerts the strongest force on NWs that polarize most strongly.This is especially useful for isolating specific subpopulations within polydisperse nanostructures such as the Au NP-decorated TiO 2 NWs, whose lengths vary from 0.5 to 100 μm and diameters from 50 to 100 nm.The optical microscope image (see Figure 2a) demonstrates the controllable alignment of the NWs across the electrode gap during the DEP process.Considering the rectangular shape of the Pt electrodes -separated by a 10 μm gap -the great majority of the TiO 2 NWs are oriented at a 90°angle toward the electrode edges (see Figure 2b).Also, DEP serves as a length selection tool by providing the strongest DEP interactions for the NWs with length sizes similar to the electrode gap or slightly longer.This enables DEP to align only selected subpopulations within polydisperse NW mixtures onto patterned electrodes with varying electrode gap sizes.
For this study, a sinusoidal DEP signal, with amplitude at 5 V pp AC and frequency at 10 kHz, was used as it ensures a positive value of F DEP and induces the assembly of TiO 2 NWs-Au NPs by means of pDEP.It has been reported that varying frequencies induce the preferential collection of NWs based on their conductivity properties, [41] but this is out of the scope of this study.
As seen in Figure 2b), potential hot-spots are formed either by adjacent Au NPs attached to the same TiO 2 NWs, or by neighboring Au NPs attached to different NWs in close proximity.Effectively, the DEP alignment of TiO 2 NWs-Au NPs creates a NW mesh which subsequently forms a 3D configuration of plasmonic NPs providing SERS signal across the electrode gap area.The NW density can be controlled by adjusting two parameters: the applied voltage or the NW concentration in the formulation.By increasing the applied voltage, stronger F DEP is exerted across the electrode gap, resulting in higher NW density and thus improving the deposition yield. [41,52,53]Alternatively, increasing the NW dispersion concentration results in denser TiO 2 NWs-Au NPs alignment across the electrodes.The NW deposition density is a critical parameter for obtaining a dense NW mesh that offers high SERS signal intensity at low analyte concentrations.Assembly of NWs in low densities produced markedly lower SERS signal (Figure S4, Supporting Information).

Nanowire SERS Sensor Performance
The Raman signal evaluation of the complete TiO 2 NWs-Au NPs DEP-aligned sensor was performed against drop-casted TiO 2 NWs-Au NPs on a glass substrate.The methodology, including analyte exposure, Raman scanning, and data processing is described in the Experimental Section. Figure 4a,b show the DEPaligned and drop-casted sensor configurations, respectively.The highlighted areas (in white) denote the sample areas scanned during Raman mapping.Prior to Raman scanning, both configurations were individually exposed to gas phase 4-ATP at a relatively high concentration (100 ppm v ) to allow the comparison of their Raman sensing performance.As shown in Figure 4c, the DEP-aligned sensor provides much higher SERS signal intensity in the inter-electrode area, due to the DEP-induced assembly of dense TiO 2 NWs-Au NPs.This can be compared to the lower and spatially disparate SERS signal intensity of the drop-casted sensor (Figure 4d) which stems from an uneven distribution of nanostructures, as seen in Figure 4b.With the use of DEP as an alignment technique, a controllable and high-density NW mesh is formed, which enables the controllable distribution of the Raman signal on pre-defined substrate areas.
Figure 4e,f show the average Raman spectrum for the scanned areas of the DEP-aligned and drop-casted SERS sensors, respectively.The main spectral peaks of 4-ATP deposited on the surface of TiO 2 NWs-Au NPs are observed to slightly shift from the intrinsic Raman spectrum of 4-ATP due to the association of the molecules with the metal particles via gold monosulfide (Au-S) bonds. [58,62]or the DEP-aligned sensor, the characteristic Raman peaks of 4-ATP at 1074 cm −1 and 1581 cm −1 (corresponding to a 1 vibrational modes of 4-ATP [54][55][56][57] ) appear as reported in the literature.For the drop-casted case, strong enhancement was observed at 1166 cm −1 (b 2 mode of 4-ATP [54][55][56][57] ) which corresponds to C-H bending. [58,59]Even though the drop-casted SERS sensor provided a dense TiO 2 NWs-Au NPs distribution around the rim, the characteristic peaks of 4-ATP were weaker due to the nonuniform formation of hot-spots.It is expected that the uncontrolled interparticle distances between the Au NPs from different NWs limit the possible hot-spot formation over the Raman mapping region, which results in a weaker SERS signal.Additionally, the overabundance of plasmonic NPs may negatively impact SERS intensity due to the re-absorbance of emitted photons. [9]s reported in the literature, [60,61] the b 2 modes of the molecule are intensified as the temperature increases.It is evident that the high-density deposition of the unstructured nanomaterials in the drop-casted sensor leads to the dissipation of the laser power into heat, diminishing the Raman intensity, and enhancing the b 2 mode, as seen in Figure 4f.The DEP-aligned SERS sensor enables the uniform and dense formation of hot-spots in the interelectrode area, as evidenced by the high 4-ATP SERS signal.This is further illustrated in Figure 4g which shows the intensity of the characteristic Raman peak of the DEP-aligned SERS sensor projected along the x-axis as a function of position in y (across the electrode gap); the high Raman intensity is confined only to the DEP-aligned substrate area.On the contrary, the dropcasted configuration (shown in Figure 4d) when projected along the same direction shows the uncontrolled NW aggregation (Figure 4h).
The performance of the device (DEP-aligned SERS sensor) was then investigated by using gas phase 4-ATP at lower concentrations; the results are summarized in Figure 5. Devices were individually exposed to 4-ATP, with different gas concentrations ranging from 100 ppm v down to 10 parts-per-billion (ppb v ) in decades.Four devices were exposed individually per concentration and no device was reused.A systematic Raman mapping was performed, consisting of 600 point-spectra, and spanning the area containing the DEP-aligned NWs.The mean Raman spectra were calculated from the 2D Raman mapping performed on each device and are shown in Figure 5a.In Figure 5b the average Raman intensity of the 1074 cm −1 band is shown for each Raman map, after baseline subtraction.As expected, higher gas concentrations (100 ppm v ) provided stronger SERS signals, whereas lower gas concentrations provided weaker intensities.Since each gas measurement was tested on a different device, certain variability in the DEP-aligned NW and 3D configuration density of plasmonic NPs was expected.Two sources of variability were identified, inter-device and intra-device variability.Inter-device variability may stem from differences during fabrication, resulting in NW assemblies with variable density, or from small variations in focusing during each scan and can be seen by the differences of the means between measurements.Intra-device variability is due mostly to non-random spatial variations in the scan; it stems from the inclusion of areas of high NW density (i.e., the electrode gap) as well as areas of low NW density (i.e., the platinum electrodes) in the same scan (see Figure 4g).This intra-device variability is visualized as the standard deviation around each of the means, shown with dotted lines.
In an attempt to circumvent the inter-device variability, we followed a spectral ratiometric approach as presented in the literature. [63,64]This method relies on analysis of a whole spectrum (or a broad spectral region) instead of on individual peaks.Each spectrum was unmixed using non-negative least squares (nn-LS) regression, based on three reference spectra, corresponding to 4-ATP, a pure sample of TiO 2 NWs, and the silicon-silicon oxide wafer.The scores on the first two are shown in Figure 5c,d), and their ratio in Figure 5e) on a base 10 logarithmic scale.It was observed that the scores corresponding to the 4-ATP reference increased with concentration, whereas the ones of the NWs remained around a steady low value.The ratio of the 4-ATP to NW signal was found to increase monotonically, and the increase was approximately 3-fold per decade of concentration above 1 ppm.For lower concentrations, the ratio obtained was ≈1.The negative control, with no 4-ATP exposure, featured a negative log ratio as its scores on the 4-ATP reference were smaller than the corresponding scores on TiO 2 .The concentration dependence of the measured intensity may be attributed to the percentage of plasmonic hot-spots occupied by the analyte molecules.At high concentrations, the surface coverage follows a Langmuir isotherm curve, and as more of the sensor surface is covered, the signal grows proportionally.At lower concentrations, the hot-spot occupancy follows small-number statistics due to the single-molecule sensitivity of SERS: a few hot-spots filled by molecules provide sufficient signal for analyte identification, but the signal does not increase in a quantitative way.
To investigate the limit of detection (LoD) of the sensor, we exposed the device to liquid samples of 4-ATP dissolved in ethanol, as we could more finely control the analyte concentration in this way compared to the gas phase exposure.The results are summarized in Figure 6, following the same format as for the gas measurements.The 4-ATP spectrum was detectable down to a concentration of 24 pM and fell below the detection threshold close to 2.4 pM.
It is possible that the detection limit for both gas and liquid phase detection can be further lowered by increasing the number of Au NPs on the surface of the TiO 2 NWs, the NW density across the electrode channel, or the acquisition apparatus.However, this approach is not likely to improve the limit of quantitation; quantitation could benefit from additional components, such as a pre-concentrator.
Comparable sensors, using SERS detection with 4-ATP, have been reported in the literature.Gas phase 4-ATP detection has been reported with dendritic Ag nanocrystals and Ag NPs embedded in silica aerogels, with concentrations of 100 and 50 ppb, respectively. [11,65]Liquid-phase detection of 4-ATP has been reported with a variety of plasmonic substrates such as Au nanorods, [58] Au nanorings, [59] Au nanospheres, [66] Ag-coated Si NWs, [32] and Ag/ZnO nanotubes [67] However, these studies used relatively high concentrations of the analyte, several orders of magnitude higher than the ones tested here.Limits of detection were reported for PdAg NPs (10 nM), [68] Ag-coated silicon nanocones (5 nM), [69] graphene oxide with Ag NPs (100 pM), [70] Au core-Ag shell NPs (6 pM), [71] and Ag nano-cubes on a hollow Co−Ni layered double hydroxide (1 pM).The LoD reported here for 4-ATP, in the low pM range, is on par with the lowest reported, without the need for complex multistep chemistry or a preconcentrator.
Commercially available VOC sensors, based on technologies other than SERS, have markedly lower sensitivity and little to no specificity.The performance of the DEP-aligned SERS sensor was compared to a commercial gas sensor (the experimental setup is described in the Experimental Section).Figure 7 demonstrates the steady state resistance values of the commercial sensor when exposed to 4-ATP against the corresponding concentrations.There is a downward trend in sensor resistance with increasing concentration, the higher the concentration of the analyte, the larger the drop in resistance.The sensor showed poor performance, for concentrations below 5 ppm.When all data points were fitted using a linear equation, the correlation coefficient was 0.160, whereas excluding the two smallest concentrations, the fit improved dramatically with correlation coefficient 0.919.This suggests that the commercial sensor cannot be used for concentrations below 5 ppm.For higher concentrations, we can calculate the sensitivity of the sensor to 4-ATP from the slope of the linear fit to be 5.2 kΩ ppm −1 under the conditions of this measurement.However, the extrapolated data cannot be trusted for concentrations below 5 ppm.This comparison highlights the superiority of SERS for low gas phase analyte concentrations compared to off-the-shelf sensors.

NW SERS Breath Analysis
Inspired by the promising performance of the sensor using a test analyte, we set forth to test whether it could be used for breathbased analysis.Specifically, we selected to analyze exhaled breath condensates (EBCs), as they can be readily acquired, have recognized diagnostic potential, [72] and provide liquid-phase samples which are easier to handle.We performed a preliminary pilot study based on EBC samples collected from two volunteers over four different days; on one of the days, one volunteer was diagnosed with an upper respiratory tract infection (URTI).The sample collection procedure is described in the Experimental Section, and an illustration of the methodology is provided in the Supporting Information Figure S5 (Supporting Information).
Freshly prepared devices (DEP-aligned Au NP-decorated TiO 2 NWs) were exposed to the breath condensates, and multiple Raman spectra were acquired point-wise from within the electrode gap region.Around 50 points were collected from each device, all within the inter-electrode region.SERS representative spectra are shown in Figure S6 (a) (Supporting Information).The spectra were processed and classified using orthogonal partial least squares discriminant analysis (OPLS-DA) as described in the Experimental section.PLS-DA is a supervised classification technique able to discern the discriminating differences between different classes of multivariate data.Unlike other analysis methods, such as PCA, that aim to describe the total variance within a dataset, PLS-DA aims to identify features that maximize the intergroup distance, even if they do not contribute significantly to the total sample variance.These inter-class differences are calculated as latent variables (LVs), which are vectors in the space spanning the multivariate data that maximize the interclass distance.The LVs serve as a new coordinate system on which the samples are projected.The samples are then assigned classification probabilities and classes based on their coordinates (scores) in this new space.
OPLS-DA orthogonalizes the LVs, which results in the discriminating features being expressed in directions along the axes of the LVs.This is particularly useful for identifying specific spectral peaks related to the physical spectra. [25]igure 8. Breath-based URTI detection.a) 80% of spectra collected from EBCs of an individual suffering from URTI were classified as distinct from the healthy samples.b) Sample distribution in the OPLS space shows the original category of each sample with symbols (URTI: cross -×, n = 48; healthy: circle -∘, n = 180) and the assigned (predicted) class with colors (URTI: brown; healthy: copper).c) The calculated LVs (loadings) spanning the directions of interclass variance.LV1, separating URTI samples from healthy ones, shows Raman peaks that may correspond to diagnostically relevant molecules.
In our case, two were assigned to the spectra, 'URTI' or 'Healthy'.The classification accuracy of the OPLS-DA model is shown in Figure 8a.The reported results follow a strict classification criterion, i.e., a sample is considered a member of a specific class only if the predicted probability for that class is greater than 50%.Out of the spectra collected from the URTI sample, the OPLS model classified 80% as distinct from their healthy counterparts.We consider this to be a high percentage with potential diagnostic significance.Certain overlap was expected, as EBCs contain a variety of molecules, many unrelated to disease, which would be classified as 'Healthy'.By examining the scores of the samples in the LV space, as shown in Figure 8b, we can see that LV1, capturing 1.96% of the total dataset variance, separates the URTI samples from the healthy EBCs.The calculated LVs are shown in Figure 8c.Specific peaks that are defined by LV1 as associated with URTI are 997, 1091, 1269, and 1448 cm −1 .These peaks were also identified in the difference between the mean spectra of the two populations, as shown in Figure S6 b) (Sup-porting Information).The observed peaks were found to match peaks related to SERS detection of SARS-CoV-2. [73]

Conclusion
In conclusion, we demonstrated an effective label-free SERS sensor based on the DEP alignment of TiO 2 NWs decorated with plasmonic Au NPs for breath-based diagnostics.A fully solutionprocessed technique was developed for the plasmonic decoration of the NWs, followed by their DEP alignment into a 3D mesh forming the SERS sensor.Our approach is quick, time-and costeffective, and can be conducted at room temperature, forming a controllable and dense SERS hot-spot assembly.These features significantly contribute to the realization of a SERS substrate with high commercialization potential.The performance of the DEPaligned SERS sensor proved superior to the conventional dropcasted technique and a commercial gas sensor when using 4-ATP as an analyte and comparable to the best SERS sensors reported in the literature.Based on the Raman mapping results, we attribute the superior performance of the sensor to the DEP assembly of the NWs.The sensor was found to be effective in detecting trace quantities of the analyte, down to the ppb range in gas phase and low pM for liquid samples.The DEP-aligned SERS sensor was tested using EBC samples producing the successful discrimination of URTI against healthy samples.These results indicate that it may serve as a state-of-the-art biosensor for rapid breathbased disease screening.The proposed label-free DEP-aligned SERS sensor offers reliable and highly sensitive SERS signal on pre-defined areas that can be useful for the development of future breath-based point-of-care diagnostic platforms.
Si/SiO 2 Cleaning Procedure: The Si/SiO 2 substrates were cleaned using solvent-based and plasma cleaning methods.The substrates were ultrasonically cleaned in isopropanol (IPA) for 10 mins, followed by a plasma cleaning (PE-25, Plasma Etch) at 400 W at 25 cc min −1 air gas for 30 mins.
Synthesis of Gold Nanoparticles: The total reaction volume was maintained at 4 mL in anisole.Anisole solvent was used as it offers better dispersion of the NWs for more uniform deposition, due to its very long static relaxation time (0.4 s) and low dielectric constant (4.33). [52]TiO 2 NWs (2 mg) were dispersed in 2 mL of anisole prior to the chemical synthesis.HAuCl 4 (2 mL of 4 mM) were added to the NW solution and mixed well with a vortex (Vortex Mixer -ISG).Afterward, 40 μL of 5 mM of NaBH 4 was added using a 10-100 μm micropipette, while being vigorously mixed using a vortex at a high spinning rate (≈2500 rpm) for 1 min.The color change of the solution occurred immediately (to light purple).
Nanowire Self-Assembly: A Si/SiO 2 substrate was placed on an inclined surface of ≈30 degrees (versus horizontal plane) to allow the NW solution to flow along the substrate, avoiding the use of pumping motors and micro-fluidic channels.The inclined position of the substrate provides a gravity-assisted flow of the NW solution, perpendicular to the pre-patterned DEP micro-electrodes, assisting the NW lateral alignment across the electrode gap and the removal of the weakly-interacting NWs. [41]The TiO 2 NW-Au NP solution of 5 μL was drop-casted using micropipette (10 μL volume) on an inclined substrate in order to provide a gravity-assisted slow flow of the NW formulation, with the electrodes being perpendicular to the NW flow.The NWs were aligned under the influence of an alternative voltage potential of 5 V pp and sinusoidal frequency of 10 kHz, supplied by a portable USB-controlled function generator (Analog Discovery 2, Diligent).The pre-set DEP sinusoidal signal (5 V pp , 10 kHz) was applied prior to the NW droplet being placed on the substrate.After the DEP assembly process, the remaining solution was removed by rinsing with ethanol, to remove weakly attached NWs on the Pt electrodes.The substrate was then dried under a drying gun.
4-ATP Sample Preparation: Different molar (M) amounts of 4-ATP were diluted in 1 mL of ethanol.The sample was sonicated in an ultrasonic bath for its dilution.
Absorption Spectra Characterization: The extinction spectrum of the solution was acquired using a UV-vis spectrophotometer (Thermo Scientific, Multiskan SkyHigh), scanning wavelengths ranging from 300 to 800 nm.SERS Measurements: SERS measurements were performed using a 785 nm diode laser (Ramulaser™, SterllarNet Inc.), with a spectral range from 124 to 2805 cm −1 .The Raman spectra were collected under 3000 ms time integration at 240 mW output power in a 210 μm spot diameter.A custom MATLAB script was used for acquiring the Raman spectra.
Raman Mapping: The Raman spectral imaging (point-by-point mapping) of the NW SERS sensor was performed using a motorized stage.Two-dimensional Raman mappings were collected.The Raman spectra were acquired as listed in the SERS measurements section.The Raman mapping for the DEP-aligned SERS sensor was collected for a 150×1000 μm 2 area within the electrode-gap region with a step distance of 5 μm in the vertical direction (y) and a 50 μm in the horizontal direction (x).The Raman mapping for the drop-casted SERS sensor was collected for a 1000×1000 μm 2 area with a step distance of 50 μm in both directions (x and y).A custom MATLAB script was used for controlling the motorized stage.
Raman Spectral Preprocessing: Raw (unprocessed) data were collected for each experiment.The collected SERS spectra included wavenumbers from 124 to 2805 cm −1 .Before each Raman experiment, the dark current was acquired and subsequently subtracted from the raw Raman data.For the data preprocessing (incl.baseline subtraction, smoothing, normalization), PLS Toolbox software (v9.0,Eigenvector Research, Inc), incorporated within the MATLAB computational environment, was used.The baseline subtraction was based on a Whittaker filter ( = 200), and the smoothing of the Raman spectra was performed using the Savitzky-Golay polynomial filter (2nd order, width = 15).
Gas Exposure: A glass petri dish (STERIPLAN Petri Dish 60×15 mm) was used with an internal volume of 2.49×10 −5 m 3 .The NW SERS sensor and a blank microscope slide were placed inside the petri dish.Liquid Exposure: A glass petri dish (STERIPLAN Petri Dish 60×15 mm) was used with an internal volume of 2.49×10 −5 m 3 .The NW SERS sensor was placed inside the petri dish.4-ATP was dissolved in ethanol (at different concentrations for each experiment) and 150 μL were drop-casted on the surface of the NW SERS sensor.The petri dish was heated on a hot plate (ISG, Hotplate & Magnetic Stirrer Pro #153-006) at 40 °C for 15 mins, to allow the SERS sensor to dry up completely.The substrate was equilibrated at room temperature for 10 mins before being Raman scanned.
Commercial Sensor: The commercial gas sensor (BME688, Bosch Sensortec GmbH) was fitted on a SparkFun board (190961, SparkFun Electronics).A microcontroller (ESP32-PICO-/ V4.1, Espressif Systems) was used for the data acquisition.The microcontroller was programmed using Arduino IDE 2.1 (Arduino Software 2.1.0)and API from Bosch Sensortec (BME68x-Sensor-API, GitHub).A glass beaker (sealed by a glass lid) with an internal volume of 2.16×10 −3 m 3 was used as a gas container with the sensor being enclosed inside.Prior to the gas measurements, 0.75 μL of ethanol was drop-casted at the bottom of the gas container for establishing the equilibrium state of the sensor (the liquid evaporation was performed at room temperature).Once the sensor reached equilibrium resistance value, the gas container was ventilated to allow the recovery of the sensor.The same procedure was followed for various concentrations of 4-ATP in the same amount of ethanol (0.75 μL) that result in different concentrations in ppm inside the glass container.Pure ethanol was measured at the beginning, middle and end of the measurement as a control.
Informed Consent Statement: Informed consent was obtained from all volunteers involved in the study.
EBC Sample Collection: The custom-made EBC collecting system consisted of a 1 L glass beaker containing ice (= cooling device) with a PTFE tubing immersed in it.The flexible plastic tube directs the exhaled breath through the cooling device.The EBC samples -mostly composed via the condensation of water vapor -were collected in a plastic conical tube in ambient conditions.1-2 minutes of tidal breathing yielded 150-200 μL of sample.
Multivariate Analysis and Machine Learning Classification: Spectral analysis was performed in Matlab (R2022b) using the PLS_Toolbox, version 9.1, by EigenVector Research Inc. Non-negative Least Squares (nn-LS) regression was performed using reference spectra from known pure substances.Only wavenumbers between 550 and 1700 cm −1 were included.The reference spectra were subjected to baseline subtraction via the Whittaker filter ( = 200), smoothed via a Savitzky-Golay polynomial filter (2nd order, width = 15), and normalized by max.Sample spectra were subjected to baseline subtraction and smoothing, but not normalized.
OPLS was performed using 228 point spectra assigned to 2 classes (URTI: 48, healthy: 180).The spectra were collected from 4 different devices on four different days.Only wavenumbers between 600 and 1700 cm −1 were included.All spectra were subjected to baseline subtraction via the Whittaker filter ( = 200), smoothed via a Savitzky-Golay polynomial filter (2 nd order, width = 15), and mean-centered.The model was calibrated following the SIMPLS algorithm and using pre-assigned classes (Healthy and URTI) for each sample.3 samples were excluded outliers (URTI: 2, healthy:1) based on high Hotelling T 2 values during model calibration.
Statistical Analysis: Statistical analysis was performed using MATLAB.PLS Toolbox (v9.0, Eigenvector Research, Inc), was used for spectral preprocessing, nn-LS, and PLS-DA.For the model analyte analysis (4-ATP in gas and liquid), four replicates were measured for each concentration, each consisting of 600 Raman point spectra.For the EBC analysis, four samples were acquired from volunteers, representing 3 healthy samples and one sample with an undiagnosed URTI.48 and 180 Raman spectra were acquired from the URTI and healthy counterparts, respectively.For both analyses, individual and mean spectra were presented, with the number of samples stated in the figure captions.
Optical Characterization: The optical imaging of the NW SERS sensor was performed using an optical microscope (Olympus; CX41) equipped with stereo microscope lighting (Olympus; JK 1500 LCD).
SEM Characterization: The surface morphology of the TiO 2 NWs-Au NPs was performed using scanning electron microscopy (Tescan Vega LSU).The electron accelerating voltage was adjusted at 30 kV.
TEM Characterization: A copper grid (Electron Microscopy Sciences, CF300-Cu) was used for transferring the TiO 2 NWs-Au NPs to the TEM equipment.The copper grid was coated with poly-L-lysine solution (0.1% (w/v) in H 2 O; CAS: 25988-63-0) and washed with DI-H 2 O, prior to the deposition of the TiO 2 NWs-Au NPs.The structure of the TiO 2 NWs-Au NPs was visualized using transmission electron microscopy (Thermo Scientific, Talos L120C).The electron accelerating voltage was adjusted at 120 kV.
COMSOL Simulation: For the study of the plasmonic electric field of Au NPs, the finite element method was performed using COMSOL Multiphysics tool (v5.6) using the Electromagnetic Waves, Frequency domain (ewfd) interface under the Wave Optics module.In all simulations, the Au NPs diameter was set at 60 nm.The diameter and length of the TiO 2 NW were set at 70 nm and 150 nm, respectively.The material dimensions were set according to TEM images.The outer space of the main structure was a perfectly matched layer (PML) which works as simulation domain boundaries.PML sets the vacuum conditions in the model as it absorbs energy without inducing any reflections.Maxwell's wave equations were solved by performing a parametric sweep study in a frequency range of 200 nm to 90 nm, and 785 nm wavelength was used as excitation wavelength for the enhancement factor calculations.A plane incident wave propagates in z-x axis direction and direction of the polarization was in z-axis.The incident electric field intensity (E 0 ) of 1 V m −1 was used in all calculation studies.

Figure 1 .
Figure 1.Schematic illustration of TiO 2 NWs-Au NPs solution-processed synthesis and self-assembly.Top row, left to right: Au NPs are nucleated and grown on TiO 2 NWs.Bottom row, left to right: DEP is used to align the NWs across a set of parallel electrodes.After exposure to a gas phase analyte, the SERS signal is obtained.

Figure 2 .
Figure 2. Photophysical characterization of the Au NP decorated TiO 2 NWs.a) Optical microscope image and b) SEM image of DEP-aligned TiO 2 NWs-Au NPs across 100 μm Pt electrodes with a 10 μm gap.c) TEM image of TiO 2 NWs-Au NPs.d) Size distribution histogram of Au NPs attached on the TiO 2 NWs.e) Diameter distribution of TiO 2 NWs.f) UV-vis absorbance spectrum of TiO 2 NWs in anisole -before and after Au NP decoration.

Figure 3 .
Figure 3. COMSOL simulation of local electric field.a) Electric field enhancement shown as log 10 (E/E 0 ), for two 60 nm (diameter) Au NPs attached on a TiO 2 NW with inter-particle distances of 5 nm.b) Electric field enhancement shown for Au NPs in direct contact (d = 0 nm).c) Calculated absorbance spectrum of Au NPs attached to a TiO 2 NW for the geometries in (a) and (b).

Figure 4 .
Figure 4. Spatial sensor characterization.Optical image of a) TiO 2 NWs-Au NPs DEP-aligned across 10 μm gap Pt electrodes and b) TiO 2 NWs-Au NPs drop-casted on a glass substrate.Raman maps showing the intensity at 1074 cm −1 of c) the DEP-aligned and d) drop-casted TiO 2 NWs-Au NPs after 4-ATP exposure at 100 ppm v .Average Raman spectra of e) the DEP-aligned sensor (600 point spectra) and f) the drop-casted sensor (400 point spectra).Spatial distribution analysis of g) the DEP-aligned sensor and h) the drop-casted sensor, showing the intensity projections along the y-direction, with individual intensities (dots), mean intensities (of 20 point spectra) (crosses), and standard deviations (whiskers).

Figure 5 .
Figure 5. Detection of gas phase 4-ATP.a) Average Raman spectra acquired with the DEP-aligned NW SERS sensor at various 4-ATP concentrations (10 ppb v -100 ppm v , 4 replicates of 600 point spectra each).b) Intensity of the characteristic 1074 cm −1 peak.Each dot is the mean of 600 point spectra, and the diamond the average of 4 replicates.The dotted lines show the standard deviation around each average.c,d) Scores on 4-ATP and TiO 2 references calculated via nn-LS regression.e) Log 10 mean ratios of 4-ATP score vs TiO 2 score for the different tested concentrations.

Figure 6 .
Figure 6.Detection of 4-ATP dissolved in ethanol.a) Average Raman spectra acquired with the DEP-aligned NW SERS sensor at various 4-ATP concentrations (2.4 pM -2.4 nM, 4 replicates of 600 point spectra each).b) Intensity of the characteristic 1074 cm −1 peak.Each dot is the mean of 600 point spectra, and the diamond the average of 4 replicates.The dotted lines show the standard deviation around each average.c,d) Scores on 4-ATP and TiO 2 references calculated via nn-LS regression.e) Log 10 mean ratios of 4-ATP score vs TiO 2 score for the different tested concentrations.
4-ATP was dissolved in ethanol (at various concentrations for each experiment) and 150 μL were drop-casted on the blank microscope slide.DI-H 2 O (150 μL) was placed on the surface of the NW SERS sensor.The petri dish was heated on a hot plate (ISG, Hotplate & Magnetic Stirrer Pro #153-006) at 40 °C for 15 mins to evaporate 4-ATP into the gas phase.Then the temperature was raised at 65 °C to allow the evaporation of the DI-H 2 O droplet and the SERS sensor to dry up completely.The substrate was equilibrated at room temperature for 10 mins before Raman scanning.