1 Introduction

In the last decades, the constant and rapid expansion of the human activities in terms of industry, agriculture, transports and worldwide urbanization, in order to satisfy the ever-increasing living standards, has led to a consequent uncontrolled increase of the environmental pollution.

The presence of contaminants in air, water and soil highly affects our ecosystem and, in turn, the human health, being responsible for different chronic and mortal diseases. Indeed, the World Health Organization (WHO) reported that in 2012 the global deaths due to the environment were estimated to be about 12.6 million, representing 23% of all deaths [1], while the updated statistics show that in 2016 the environment deaths are increased, representing now 24.3% of the global ones.

Among the different contamination sources, an existing burning issue is represented by the so-called emerging pollutants (EP) mostly affecting the aquatic environment. These chemicals are still not included in the current worldwide monitoring programs and their toxic effects on the environment, as well as on the human health, are not well understood, thus being potentially detrimental. In this regard, within the Water Framework Directive adopted by the European Union in 2000, a Surface Water Watch List has been introduced in order to monitor this kind of emerging substances selected on the basis of their potential risk, but for which the collected data are not sufficient to establish an actual risk [2].

The common protocols to detect EP in water imply the use of research techniques like liquid or gas chromatography and mass spectrometry, generally coupled with different preconcentration methods that display a high sensitivity level. However, this kind of techniques allows to perform only off-site measurements and in the current scenario, with an intensive increase of environmental pollution, new sampling strategies that in turn allow to perform in-situ monitoring and rapid identification are highly desirable.

In this regard, Surface Enhanced Raman Spectroscopy (SERS) could be a valid candidate. Indeed, SERS spectroscopy allows to perform on site sampling within a fast time scale investigation. The SERS intensification of the Raman scattered light by molecules that are in proximity or adsorbed on metallic or semiconducting nanostructures, depends on chemical and electromagnetic effects, being the latter dominant when metallic nanostructures are involved. The trigger of the increased scattered fields is given by the excitation of the localized surface plasmon resonances (LSPR). A critical parameter is the so-called Enhancement Factor (EF), theoretically defined as the ratio between the SERS intensity and the Raman one of the same analyte under investigation [3].

The strengthening of SERS as a routine analytical method requires to find the best compromise between different issues: higher EFs, low-cost substrates manufacturing, stable and reproducible performances and environmental friendly chemical routes. In this regard, a key-role is played by the proper design of the nanostructured SERS substrates acting in particular on the nature of the involved metals as well as on the shape and architecture of the nanostructures.

The current research strategy is focused on the developments of different synthesis routes through which anisotropic shapes [4,5,6] like nanocubes, nanorods, nanoplates and nanostars can be obtained [7]. These kinds of nanostructures display a great number of tips and edges where, thanks to the lightning rod effects [8], the electromagnetic filed is highly concentrated, thus acting as intrinsic hot spots.

Among these different nanostructures, gold nanostars (GNS) with their elongated branches and numerous tips generally provide a huge number of hot spots and thus they are considered excellent SERS-active substrates [9,10,11]. Moreover, their SERS performances could be further improved by adding a covering layer of silver, as it has been recently demonstrated [12].

In our work, we present a detailed and critical study on the performance of silver coated GNS colloids e solid chips, in view of potential large-scale application as SERS sensors for in-situ detection of emerging pollutants in water and wastewater. At first, we deeply investigate the response of colloidal GNS by varying the thickness of the silver shell in order to set the best synthesis parameters to achieve the highest enhancement. The experimental results were corroborated by full field electrodynamics modeling. In a second instance, the optimized GNS@Ag were grafted on glass slides to prepare SERS chips, using simple alkoxysilane chemistry. We carefully investigate their enhancing and reproducible performances using two well-known Raman reporters, namely Rhodamine 6G (R6G) and 7-mercapto-4-methylcoumarine (MMC) on a wide population of substrates synthesized within a year, to a get a reliable statistic. The GNS@Ag300 chips were subsequently used to detect two analytes, representative of the class of water pollutants, namely norfloxacin and thiram .

2 Experimental

2.1 Materials and synthesis processes

Referring to [13], the colloidal suspensions of nano-objects were obtained through a fast and low-cost seed-growth procedure: the starting seeds were prepared by adding 600 µL of an iced-cooled aqueous solution of NaBH4 (0.01 M) to an AuCl4- solution which was prepared by mixing 5 mL of Triton-X-100 aqueous solution (0.2 M) and 5 mL of HAuCl4 aqueous solution (4.5 × 10−4 M).

The GNS growth solution was prepared starting from 50 mL of a 0.2 M water solution of Triton-X-100 and under magnetic stirring adding the following constituents: 2500 µL of AgNO3 in water (0.004 M), 50 mL of aqueous HAuCl4 (4.5 × 10−4 M), 1700 µL of an aqueous solution of L-Ascorbic Acid (0.0788 M) and 120 µL of the previously described seed solution.

The Ag coating was performed by adding, under magnetic stirring, a variable volume (75 ÷ 600 µL) of AgNO3 (0.1 M) and an equivalent volume of Ascorbic Acid (0.1 M) to 30 mL of GNS solution. NH4OH was added to initiate the reduction of silver by ascorbic acid.

The stock solutions of norfloxacin and thiram were prepared in water with 10−3 and 3.12 × 10−5 molar concentrations, respectively, and subsequently further diluted in water to perform the SERS measurements.

As regards the synthesis of solid SERS substrates, they were prepared according to a reported method [14], where a particular care for the cleaning procedure of glass substrates is foreseen. The clean slides were functionalized by immersing them for 5 min in a 10% (v/v) solution of APTES in ethanol at 60 °C. The nano-objects were then fixed on the solid substrates by immersing for 15 h the glass slides in a GNS@Ag colloidal suspension kept at the controlled pH value of about 3.

2.2 Methods

UV–Vis-NIR absorption measurements were performed with a Varian Cary 6000i spectrophotometer in the range between 300 and 1800 nm, both for glass chips and colloidal suspensions.

Transmission Electron Microscopy (TEM) images of colloidal suspensions were collected by means of a Jeol JEM-1200 EX II instrument, depositing a 10 µL sample of 1:100 (GNS) and 1:200 (GNS@Ag) diluted solutions, on nickel grids (300 mesh) coated with a Parlodion membrane.

Zeta potential measurements of GNS and GNS@Ag were performed with a Zetasizer Nano-ZS90 (source: polarized He–Ne laser, 30 mW output power, vertically polarized) on 1 mL of colloidal suspension.

Static contact angle determinations were made with a KSV CAM200 instrument, using the water sessile drop method.

Scanning Electron Microscopy (SEM) micrographs were taken from a Tescan Mira XMU variable pressure Field Emission Scanning Electron Microscope–FEG SEM (Tescan USA Inc., USA), placed at the Arvedi Laboratory, CISRiC, Pavia. Slides were mounted onto aluminium stubs using double sided carbon adhesive tape and then were made electrically conductive by coating them with a thin layer of Pt/Pd (nm) while in vacuum. Images were obtained at 25 kV using In-Beam Secondary electron detector to have higher spatial resolution.

Raman and SERS measurements were performed at room temperature using an automated and integrated confocal microRaman spectrometer, XploRA Plus HORIBA Scientific, equipped with an Olympus microscope BX43. Three different laser sources are available: 532 nm, 100 mW, 638 nm, 90 mW, and 785 nm, 100 mW. Neutral filters with different optical density allow to set the incident laser power. The spectrometer is equipped with a motorized xy stage on which the investigated samples are positioned. Spectral resolution is about 1 cm−1. An Open Electrode CCD camera, with a multistage Peltier air-cooling system, is used as detector.

The measurements were performed with the 638 nm laser source, using two different objectives: for colloidal solutions a 10x magnification, while for glassy chips a 50x one, leading to spot sizes of 100 µm2 and 4 µm2, respectively. The laser power density on the samples was kept at 2 × 104 W/cm2 for the colloidal samples and 2 × 103 W/cm2 for the solid chips. The spectra were collected with an integration time of 5 s for the colloidal solutions and 10 s for the solid substrates, while a number of accumulations equals to 10 was used in both cases.

We performed EF measurements using what we call a “sandwich-like configuration” with a well-known Raman reporter, namely Rhodamine 6G (R6G), as probe molecule [15]: we deposited an 80 mL drop both on a SERS chip to collect the SERS response and on a blank glass slide to record the Raman signal and we spread it by covering the SERS and blank substrates with a cover glass slide of the same dimension.

Thanks to this approach we can obtain an almost homogeneous liquid film that allows to perform both SERS and Raman inspection in the same experimental geometry, namely to use of the same objective. Thus, the area irradiated in both Raman and SERS measurements is the same and the only operative parameter that changes is the height of the scattering volume: all the molecules in the interspace between the two glass slides contribute to the Raman signal, while only the molecules adsorbed or in the proximity of the metallic nanostructures give rise to the enhanced one.

In the first case, the Raman height is equal to the distance from the focus between the cover glass side and the liquid film to the focus between the liquid film and the nanostructures layer. We measured this Raman height using a graduated micrometric wheel and it turn out to be about 20 μm. In the second case, the SERS height is considered equal to the distance beyond which the SERS signal drops to zero and in our previous work [16], we estimated it to be about 10 nm.

3 Results and discussion

A representative TEM image of pristine GNS is reported in Fig. 1a. In general, the seed-growth method used to synthesize gold nanostars allowed us to obtain quite regular six- and five-branched objects with a maximum tip to tip distance in the range between 60 and 90 nm that display an absorption spectrum which is dominated by two main LSPR features: a higher order resonance close to 900 nm and the second one close to 1500 nm which was identified as the fundamental resonant mode [17] (Fig. 1d—black spectrum).

Fig. 1
figure 1

a–c TEM images of pristine GNS, GNS@Ag300 and GNS@Ag600 and d their UV–Vis-NIR absorption spectra

Full size image

The addition of silver and ascorbic acid leads to the formation of an Ag coating that, for small volumes, mainly involves the GNS cores leaving the tip of the branches partially uncovered (see Fig. 1b for GNS@Ag300), while for further amounts the coating itself starts to cover also the branches, till the highest concentration, i.e., sample GNS@Ag600 (Fig. 1c), for which almost every object is completely enveloped by the silver shell and no tips are seen at all.

The increase of the silver shell thickness causes a marked blue-shifting of the LSPR band initially peaked at 1500 nm and its subsequent complete disappearance (Fig. 1d). Meanwhile, a new LSPR peak arises in the range between 400 and 550 nm that also blueshifts as silver and ascorbic acid content further increases, until it becomes predominant for the highest concentration. In this case the absorption spectrum is characteristic for nano-objects consisting mainly of silver, as it is pointed out by the TEM images (Fig. 1c).

In order to study the enhancing properties of GNS@Ag as a function of the silver shell thickness, we investigated the SERS response of our colloidal solutions using Rhodamine 6G [18]. Figure 2 reports the SERS signals of 10−5 M R6G solution as a function of the amount of added silver. As it is possible to observe, the GNS@Ag300 sample is the one that displays the highest intensity signal. Indeed, for the starting GNS colloid the signal intensity is about 6 times lower, while for concentration higher than 300 µL of AgNO3 the R6G signal intensity starts to lower again.

These findings together with the TEM results seem to assert that the most performing nanostructure is the one in which the branches are almost completely covered by the silver shell and only the final part of the tips sticks out (Fig. 1b).

Further increasing the silver coating thickness, the SERS intensity starts to lower till the minimum value achieved for the GNS@Ag600 sample, i.e., when the GNS are completely embedded.

Fig. 2
figure 2

SERS spectra of 10−5 M R6G solution collected for each investigated colloidal sample

Full size image

The observed experimental results were corroborated by theoretical simulations based on the boundary element method which was used to model a six branched gold nanostar surrounded by a silver shell (Fig. 3) of increasing thickness and immersed in water [19] (the model details are deeply described in [13]).

Fig. 3
figure 3

Comparison between the SERS integrated intensity of the R6G mode at 610 cm−1 (blue histograms) and the average of the fourth power of the field enhancement computed for the modeled configurations reported in the graph, namely pristine GNS, GNS@Ag300 and GNS@Ag600. In the lower side, the local field enhancement maps at λ = 633 nm for the aforementioned samples are shown (Color figure online)

Full size image

Indeed, by looking at the local field maps at λ = 633 nm presented in the lower side of Fig. 3, one notices that larger field enhancements are obtained when the tips protrude from the silver shell, but that the enhancement decays as soon as the nanostar branches are enclosed. To better highlight this aspect, we compared the trend of the integrated intensity of the R6G mode at 610 cm−1 and the computed average of the fourth power of the field enhancement (|E|4) for the three representative samples, namely GNS, GNS@Ag300 and GNS@Ag600. As it is possible to observe the best performance belongs to the structures in which the main resonance is not quenched by the silver encapsulation that experimentally corresponds to the configuration of the GNS@Ag300 sample. Thus, we selected the GNS@Ag300 colloid for the preparation of SERS chips through a simple self-assembled alkoxysilane-based monolayer approach.

This synthesis procedure is a well-established, fast, and low-cost method that allows to obtain a quite dense and homogeneous distribution of nano-objects with a pretty good reproducibility intra samples and inter samples, at least, within the same preparation batch. In view of a potential large-scale production, we tested the reproducibility of the samples in terms of goodness of the glass surface covering and, in turn, of the homogeneity of the SERS response, extending the study to a wide population of SERS sensors synthesized within a year. This allowed us to get a reliable statistic.

All the investigated sample displayed a wide LSPR absorption in the range between 300 and 1800 nm, dominated again by a strong band between 400 and 600 nm, superimposable to the modes observed for the nano-objects in the colloidal solutions. The 70% of the investigated population was characterized by a medium value of absorbance, namely 0.35–0.55, the 20% by high abs values between 0.55 and 0.65 and only the 10% by low abs values in the range 0.15–0.35.

When one is exploiting the SERS effect, the absorbance values of the SERS substrates are critical issues because a compromise must be established between two different physical aspects ruling on SERS sensing: higher abs values lead usually to higher SERS signals because of the more efficient energy transfer, but laser-induced thermal damage has to be avoided.

The synthesis of the GNS@Ag300 colloids was verified to be reproducible, giving a distribution of nano-objects characterized by the same UV–Vis absorption spectrum during the different preparations. Thus, we consequently think that the deviation from a medium absorbance value, proper of the best performing substrates, as it will be highlight in the following, relays on the different way in which the nano-objects anchor to the APTES-functionalized glass slides. The glass slides with low abs values are the result of a poor number of GNS@Ag anchored to the slides themselves, due to the precipitation of the starting colloidal solution, and thus leading to low quality SERS substrates. However, if these low abs value chips should be discarded to fully exploit the potentiality of the GNS@Ag substrates, the ones with high abs values could be still useful as sensing devices.

These high abs value chips are clearly characterized by an increased number of anchored GNS@Ag. Indeed, these substrates resulted in a more intense dark colour with respect to medium abs vale ones. However, more rigorous evidence was given by the SERS inspection. From the microscope camera it was immediately clear the presence of dark areas of micrometre dimensions, as illustrated in Fig. 4, that thanks to homogeneity tests of the SERS response, turn out to be huge aggregates with an abundant number of hot spots. In particular, we verified the homogeneity of our GNS@Ag300 substrate response, by collecting the SERS signal of 7-mercapto-4-methylcoumarine (MMC), a molecule which strongly interacts with the metallic nanostructures thanks to the formation of stable Au-S bonds. Thus, the chips were dipped in an ethanolic solution of MMC (10−5 M) for 1 h, washed and dried under N2 flux before performing SERS measurement.

The general measuring procedure consists in performing multiple acquisitions on the same sample and on samples coming from same and different preparation batches, to assess the intra- and inter- samples homogeneity. In the case of high abs sample, we also performed an automatic acquisition across these macro-aggregates. As an example, we reported in Fig. 4 the colour map acquired with a 50x objective and obtained by plotting the integrated intensity of the MMC mode at 1170 cm−1 as a function of the spot position. It is possible to appreciate a huge increase of the MMC signal in the aggregate area with respect to the surrounding where in principle a monolayer of covering should be present.

Fig. 4
figure 4

Representative image of a macro-aggregate on the high abs value chips with a superimposed colour map displaying the integrated intensity of the MMC mode at 1170 cm−1 across a portion of the reported aggregate (Color figure online)

Full size image

Moreover, we calculated the relative standard deviation (RSD) by referring to the integrated intensity of the investigated MMC mode, obtained by performing a fit with a single Lorentzian curve as illustrated in Fig. 5a, and we found out RSD \(\approx\) 30%. This RSD value is too high to assert a good homogeneity of a SERS substrate, but this is a natural consequence of the high hot spots’ concentration. Thus, this kind of substrates could be very useful to gain in sensitivity and in turn to reach low limit of detection thanks to this huge increase of the local field. However, in order to perform a calibration curve of the potential investigated pollutant and, thus, to be not only sensitive, but also quantitative, the homogeneity of the SERS response has to be a key factor for the substrates. For this reason, we used for our research purposes the medium abs value substrates. Indeed, by performing the homogeneity tests with the MMC reporter, as previously described, we again performed with a 50x objective five linear acquisitions collecting 10 spectra along each 60 μm scan, finding out in this case an RSD value of about 16%. Moreover, we carried out further sampling by collecting map** acquisition with a 5x objective spanning a wider area of about 1 mm2 and we found out an RSD value of about 10%. According to [13], this value states that our medium abs value substrates display a good signal homogeneity [20, 21]. Figure 5b reports, as an example, the colour map we used to illustrate the variation of the height of the MMC mode at 1170 cm−1 as a function of the spatial position.

Fig. 5
figure 5

a Representative SERS spectra of 10−5 M MMC collected on a GNS@Ag300 chip. Inset: example of best-fitting procedure performed with a Lorentzian curve on the 1170 cm−1 mode. b Colour map of the height of 1170 cm−1 mode as a function of the spatial position

Full size image

In a second instance, we explored the performances of the chosen SERS substrates by quantifying the enhancement factor. We used again as testing molecule Rhodamine 6G in 10−5 M and 10−3 M diluted solutions to perform SERS and Raman measures, respectively. The procedure we followed is described in detail in subsection 2.2. To calculate the EF, we applied the general formula [3]:

$$EF=\frac{{I}_\text{SERS}}{{I}_\text{RS}}\frac{{N}_\text{RS}}{{N}_\text{SERS}},$$

where ISERS and IRS are the integrated intensity of the 610 cm−1 R6G mode, while NRS = cRS V and NSERS = cSERS V, are the average number of molecules in the scattering volume (V) in Raman and SERS experiments, respectively. However, as already described in depth above, since we were able to use the same scattering geometry, the relevant parameter is only the height of the scattering volume, thus we operatively used the following expression:

$$EF=\frac{{I}_\text{SERS}}{{I}_\text{RS}}\frac{{c}_\text{RS}{H}_\text{RS}}{{c}_\text{SERS}{H}_\text{SERS}}.$$

in which cRS and cSERS are the molar concentrations of R6G solution, namely 10−3 and 10−5 M, respectively, while HRS is the Raman height, equal to about 20 μm and HSERS \(\approx\) 10 nm, the SERS one. Through this reasoning we estimated an EF in non-resonant conditions of about 107 which is in good agreement with the values reported in literature for solid chips synthesized by means of a fast and cheap procedure.

After having deeply verified the confidence of our SERS-active chips, we fruitfully exploited them to detect an antibiotic, norfloxacin, and a pesticide, thiram.

In particular, for both norfloxacin and thiram, our limits of detection (LODs) were largely beyond the maximum residual levels (MRLs) set by the European Union for these pollutants. In particular, the study allowed us to perform calibration curves with the lower detectable limit equals to 3 ppb for norfloxacin and 100 ppb for thiram. The detailed description of the results is reported in [13]. Here we want to focus the attention on the comparison between Raman and SERS spectra in order to obtain information about the way in which the investigated molecules interact with the metallic nanostructures. Indeed, knowing how the analyte behaves in presence of a SERS substrate, it is of mandatory importance in order to best optimize the performances of the sensors and in turn to get better control and to improve the sensitivity in view of routinary as well as in-situ detection.

As it is possible to observe in Fig. 6, the comparison between the Raman spectra of the powder and the SERS spectra of 10−5 M analyte solution for each investigated compound, highlights a difference in both the enhancement and position of some modes, as it generally occurs in SERS experiments.

The norfloxacin Raman and SERS spectra (Fig. 6a) are dominated by three main modes, one at 740 cm−1 due to the symmetric stretching vibration of the C–F group together with the ring deformation, the second one at 1390 cm−1 ascribed the symmetric stretching vibration of the carboxylic group and the last one at 1620 cm−1 due to the stretching vibration of the (C–C) aromatic ring [22, 23]. Upon normalization of the mode intensity at 1390 cm−1, it is clear that the enhancement mechanisms are mostly affecting the mode at lower energy. Indeed, by performing a best-fitting procedure on these modes, the resulting FWHMs are appreciably greater than those obtained for the Raman spectrum. On the contrary, this is not observed in the high energy modes. This could suggest that the C–F group mainly feels the perturbation induced by the proximity to the metal nanostructures than the remaining part of the molecule. This fact could be due to the interplay between distance and orientation with respect to the nanoparticles. Indeed, higher SERS enhancement is expected from closer vibrational units and for those modes displaying polarizability changes along the direction of the local electromagnetic field.

As regards thiram, one can clearly see from Fig. 6b that the majority of the modes is characterized by relevant changes in terms of peak energies and relative enhanced intensity ratio with respect to the Raman one and also by the appearance of new modes (i.e., 1510 cm−1) in the SERS spectrum with respect to its Raman response. Only the mode at 555 cm−1 which is assigned to the S–S stretching, seems to be less affected at least in the energy position. The SERS spectrum is characterized by a mode at about 430 cm−1 which is assigned to the CH3CN bending and the stretching of the C=S bond, the one at 923 cm−1 due to the CH3NC and C=S vibration, the intense peak at 1386 cm−1 is due to the symmetric CH3 deformation mode weakly coupled to the CN stretching and the new-born at 1510 cm−1 resulting from the CN stretching itself [24, 25].

The different way in which these modes are affected by the SERS substrate could rely on the way in which the molecule interacts with the metal surface. Indeed, unlike norfloxacin, thiram creates a strong covalent bond with the silver surface through the sulfur atoms (S–Ag bond) upon cleavage of the S–S bond. Thus, if we look at the relative intensity ratio between the 555 cm−1 mode and the one occurring at 1386 cm−1 both in Raman and SERS spectra, it is clear that the enhancement mechanisms slightly affect the intensity of the 555 cm−1 mode; however, this is consistent with the reduction of the S–S bonds as a result of the cleavage and the consequent adsorption on the Ag surface. Moreover, a further evidence of S–Ag bond creation is supported by the strong changes that affect the modes of the C=S and C–S bonds, as for instance the ones at 430 and 923 cm−1. Finally, the growth of the mode of the CN stretching at 1510 cm−1 seems to indicate that the thiram resonance structures bind perpendicularly to the metal surface [25].

Fig. 6
figure 6

Comparison between the Raman spectrum (dashed line) and the SERS one (solid line) of 10−5 M solution of a norfloxacin and b thiram. For both the investigated molecules, the intensity of Raman spectra is normalized to the one of the dominant SERS modes, namely the one at 1390 cm−1 for norfloxacin and the mode at 555 cm−1 for thiram

Full size image

Finally, by comparing the SERS activity of the two pollutants, it is also possible to appreciate how on equal SERS substrates, the enhancement yield in the same spectral regions is clearly different. In particular, to get the most from the SERS enhancement of our GNS@Ag substrates, one has to look at the low energy mode, namely 600–700 cm−1, for norfloxacin, while for thiram, at the high energy ones, thus about 1400 cm−1.

4 Conclusions

In this study we further investigated the GNS@Ag-based SERS sensing chips by deeply characterizing their properties in terms of enhancing capability and homogeneity of the SERS response, extending the analysis to a wide number of samples. We observed some issues in reproducibility, with about an overall 30% of samples having too low or too high absorbance values to be routinely used. Nevertheless, we demonstrated that 70% of our GNS@Ag chips displayed good average EF and can be used trusting in a good homogeneity of the SERS response.

Based on the goodness of these substrates, they were successfully used to detect norfloxacin and thiram with appreciable LODs which are in both cases lower than the MRLs set by the European Union, namely 3 ppb for norfloxacin and 100 ppb for thiram. On this basis, we carefully analysed the SERS modes of the two molecules and their relationship with normal Raman modes, obtaining some information useful in the design of SERS chips for pollutants detection in water.