Introduction

The discovery of antibiotics at the beginning of the twentieth century revolutionized medicine. Antibiotics act by either killing bacteria or inhibiting their growth. Many antibiotic classes have a broad antibacterial spectrum. However, because of their overuse in medical treatments and live-stock raising, the presence of antibiotics has been detected in waste and surface waters [1–3]. High antibiotic concentrations in the ecosystem can cause the development of antibacterial resistance [4–7] and consequently, the necessity of finding new drug formulations. Ciprofloxacin, a second-generation fluoroquinolone, is widely used in the treatment of infections, such as urinary tract, prostatitis, sinusitis, and bone and joint infections [8]. The drug is excreted in urine, with the parental drug representing more than 50 %. Additionally, for some prescribed treatments, dosages up to 1000 mg are administered [9]. As a result, the concentration in environmental samples can exceed the minimum inhibitory concentration (MIC) values of bacteria [2, 3]. Ciprofloxacin MICs range between 0.03 mg/l (0.024 μM) for Escherichia coli ATCC 25922 and 16 mg/l (48 μM) for Enterococcus faecalis clinical strain 55 [8]. Thus, it is important to monitor and reduce the antibiotic concentrations that enter the ecosystem. Drinking water treatment plants play an important role in removing unwanted chemicals. Nonetheless, because of strong fluctuations throughout the year, the successful removal of antibiotics must be constantly monitored.

Surface-enhanced Raman spectroscopy (SERS) was proven to be suitable for the detection of trace amounts of molecules [10–20]. The strong signal enhancements originate from localized “hot spots” at the tip of sharp nanostructures or in the gap between spherical nanoparticles [15, 21]. Reliable and homogenous SERS active substrate development is one of the main concerns of the SERS community. Despite the high variety of available planar substrates [22–25], metallic colloidal suspensions that are obtained by chemical reduction remain the easiest-to-access and most cost-effective structures. Their main drawbacks, such as aggregation time-dependent signals, precipitation [26], proper mixing with the analyte, and automatic measurement conditions, can be easily overcome using microfluidic setups [27–30]. Here, solution dosing is achieved by computer-controlled syringe pump systems. Mixing is performed by meander or zig-zag channels [29], while the aggregation time can be controlled by fixing the focus point during measurements. Additionally, because of automatic measurement conditions, large datasets can be easily recorded [31]. Lastly, memory effects encountered when applying continues flow microfluidic platforms [32] can be easily avoided by employing droplet-based microfluidic chips. Multiple strategies are available for the fabrication of these platforms. Although polymer-based chips are preferred due to their low cost, drawbacks associated with poor chemical compatibility with many organic solvents, unstable surface modification over time, and ability to absorb small molecules into the matrix inhibit their application [33]. Recently, the attention of the scientific community was redirected from polydimethylsiloxane (PDMS) as material to thiol-ene polymer-based materials [34]. Nevertheless, polymers are well known to provide high Raman spectral features, which could interfere with the fingerprint of the targeted molecules. Therefore, glass platforms are still preferred for Raman and SERS applications.

Throughout the last years, SERS advanced from being a technique mainly used to characterize plasmonic structure-molecule interactions [16] and to detect dye molecules in distilled water to quantification of biomolecules in clinical [18, 35] and environmental samples [36]. Because of the complexity of the chemical composition of samples, both clinical and environmental, various strategies, i.e., sample clean-up procedures or functionalization of plasmonic structures, have to be applied to reduce the matrix effects and to successfully detect the target molecule. However, these procedures increase the costs and the time required for performing the analytical measurement. SERS detection of ciprofloxacin was previously carried out using TiO2 nanoparticles [37], dendritic silver nanostructures [38], and core-shell structured magnetic molecularly imprinted polymers [39]. While in the first two cases the antibiotic was determined only in distilled water, in the latter case, diluted fetal bovine serum was also used as matrix. Here, the achieved limit of detection (LOD) was 10−7 M. However, no details regarding the extent of the dilution procedure is provided, while the preparation of the employed nanostructures is not straightforward.

The objective of the present study was to assess the potential of the LoC-SERS technique to detect ciprofloxacin spiked in river water samples at MIC concentrations. As sample clean-up procedure, a simple filtration is proposed, while as SERS, active substrates silver nanoparticles prepared at room temperature [40] are employed. Furthermore, due to the low solubility of the antibiotic in the neutral pH range, either acidic or basic ciprofloxacin solutions were prepared. Therefore, insights into the influence of pH on the SERS signal of the target molecule are also presented.

Materials and methods

Sample preparation

Ciprofloxacin (≥98 % HPLC), sodium hydroxide (NaOH), hydrochloric acid (HCl), silver nitrate (AgNO3), and hydroxylamine hydrochloride (H2NOH·HCl) were purchased from Sigma Aldrich and used as received. To prepare aqueous solutions, high purity water was used.

For cuvette-based SERS measurements, stock solutions of ciprofloxacin with a concentration of 1 mM were prepared by solving an appropriate amount of powder in HCl or NaOH solution with a concentration of 1, 0.1, and 0.01 M. Working solutions of the target analyte with a concentration of 0.1 mM were obtained by diluting the stock solutions with purified water in a 1-to-10 ratio. The pH was checked with a pH meter and adjusted by adding additional NaOH or HCl (see Electronic Supplementary Material (ESM) Table S1).

LoC-SERS measurements were performed on spiked river water sample from the Saale River (Jena, Thuringia, Germany). The river water was collected in two 50 ml falcon tubes on the same day, from the same spot. Prior to the measurements, the sample was filtered using a syringe filter with a pore diameter of 0.22 μm (Rotilabo®-syringe filter, PVDF, sterile). Filtration was necessary in order to remove large particles, such as sediments and soil. The river water sample was spiked with ciprofloxacin by mixing 100 parts of the complex matrix with 1 part of 10 or 1 mM ciprofloxacin solution prepared in 0.1 M NaOH, respectively. As a result, working solutions with a pH value of 11 with ciprofloxacin concentrations of 0.1 and 0.01 mM were prepared. Furthermore, it is important to note that the matrix was just slightly diluted, hence, conserving the complexity of the matrix. In order to assess the robustness of the LoC-SERS technique, the measurements for the detection of ciprofloxacin spiked in river water were carried out on two different days, using two different chips but the same working solutions.

The Ag colloidal solution was synthesized according to the protocol published by Leopold and Lendl [40]. For this, silver nitrate was reduced by hydroxylamine hydrochloride in the presence of sodium hydroxide. The molar ratio of AgNO3/H2NOH.HCl/NaOH was 1:1.5:3 in the final solution. As a result, a yellow-brown solution was obtained.

Instrumentation

For the pH measurements, a pH meter from Hanna Instruments (HI 9321), which was equipped with a combination microelectrode (HI 1083B), was used.

Raman and SERS spectra were acquired using a WITec Raman microscope, which was equipped with a diode laser (Fandango, Cobolt) that emitted at 514 nm with a maximum output power of 95 mW on the sample surface. The same objective (Zeiss, ×20, 0.4 N.A.) was applied to focus the laser beam on the surface of the sample and collect the backscattered photons. For detection, a thermoelectrically cooled CCD (1024 × 127) camera and two different gratings (600 and 1800 l/mm, which had spectral resolutions of ∼5 and 2 cm−1, respectively) were used.

Raman measurements on the ciprofloxacin powder were performed with 1800 l/mm grating, 5 s integration time, and 1 mW power on the sample. To record the Raman spectra of saturated aqueous solutions, the laser power was increased to 95 mW and the integration time was increased to 10 s.

Cuvette-based SERS measurements were performed with the identical grating as above, a laser power of 38 mW on the sample, and a 1-s integration time. Each spectrum is the result of 10 accumulations. For the measurements, 100 μl of colloid was mixed with 100 μl of analyte and 20 μl of H2O or KCl (0.3 and 1 M). After mixing, the cuvette was placed under the objective and the SERS spectra were recorded. For each mixture, four independent spectra acquisitions were performed using new colloid and analyte solutions.

For LoC-SERS measurements, a glass chip (Scheme 1) was used. The channel walls were functionalized with octadecyltrichlorosilane prior to the measurements to achieve hydrophobicity. As the carrier medium, mineral oil was selected. All of the analyte-containing solutions were in the aqueous phase. Hence, analyte- and colloid-containing droplets were formed in the immiscible continuous phase. A detailed description of the platform can be found elsewhere [41, 42]. To obtain differently concentrated solutions, the flow rates of the ciprofloxacin-containing syringes and of the one that contained the mixing medium were varied (Table 1). Mineral oil was pumped with 10 nl/s, whereas the colloidal solution was pumped with 9 nl/s. For each concentration step, 1200 spectra were recorded with a 1-s integration time and 38 mW power on the sample using the 600 l/mm grating.

Scheme 1
scheme 1

Picture of the droplet-based microfluidic chip for LOC-SERS measurements with the depiction of the droplets, inlets, and laser focus point. (Adapted from Ref. [38] with permission from the PCCP Owner Societies)

Full size image
Table 1 Flow rates and the corresponding measured ciprofloxacin concentrations
Full size table

Data processing

The acquired data were pre-processed and analyzed using the statistical programming language R [43]. The recorded Raman spectra and SERS spectra in the cuvette were background-corrected using the selective nonlinear iterative clipping (SNIP) [44] algorithm with 60 iterations. For better visualization, the signal was further normalized to the Raman mode at 1382 cm−1.

During the LoC-SERS measurements, the spectra were continuously acquired while maintaining the fixed focus point. Therefore, alternating Raman/SERS spectra that contained the signature of the mineral oil, analyte, or analyte and mineral oil were recorded. The first pre-processing step consisted of removing the mineral oil and mixed spectra, which was performed by manually setting a threshold value for the intensity of the Raman marker band of the oil at 2888 cm−1. The number of resulting spectra is listed in Table 1. After this step, the spectra were background-corrected and averaged. For the analysis, both uni- and multivariate statistics were applied. In the first case, selected Raman bands were integrated using Simpson’s rule. The second method combines the principal component analysis (PCA) with partial least-squares regression (PLSR) [45, 46]. Briefly, a PCA is performed in order to reduce the dimensionality of the dataset and, therewith, to reduce the computational costs significantly. The robustness of the LoC-SERS technique was assessed by training the PLS model by using the data recorded on the detection of ciprofloxacin spiked in river water sample on the first measurement day. The PLS model was subsequently used for predicting the concentrations of a second, independent data set, measured on a second day. For training the PLS model, only the first four PCA scores were applied, whereby four PLS latent variables were constructed. To evaluate the quality of the created PLS model, a ten times averaged 10-fold cross validation was performed. Therefore, the training dataset was split into 10 parts, the so-called folds, of which in every of the 10 cross-validation steps 1 was excluded from the training process and used for prediction. Afterwards, the root mean square error of prediction for cross validation (RMSEPCV) was calculated as the averaged RMSEP of the 10 validation steps. For the prediction, either on the training dataset or onto unknown data, all four PLS latent variables were used.

Results and discussion

The effect of pH on the SERS signal of ciprofloxacin

Before assessing the potential of the LoC-SERS technique to detect ciprofloxacin in spiked river water, a series of cuvette-based measurements was performed. Contradictory to the already reported publications [37, 39], the preparation of a homogenous solution with a concentration of 0.1 mM in a pH range of 4 to 10 was not successful at room temperature, which can be explained by considering the solubility properties of the antibiotic. According to Caço et al. [47], ciprofloxacin is slightly soluble in water. At 20 °C, the measured solubility after 6 h of equilibration time was 0.067 mg/ml (0.202 mM). To assess the potential of the LoC-SERS technique to detect the target molecule in the concentration range of 0.001 to 0.1 mM, stock solutions with ciprofloxacin amounts near the solubility limit had to be prepared. For this purpose, diluted HCl or NaOH was selected as a solvent instead of pure water.

In Fig. 1, the normalized Raman spectrum of ciprofloxacin powder and saturated aqueous solutions at pH 1 and pH 13 are plotted with the normalized SERS spectra, which were measured on solutions at a concentration of 0.1 mM (at pH 1 and 13). A comprehensive band assignment of the Raman modes of the antibiotic in powder form can be found in the literature [37, 48]. Briefly, the Raman bands that are ascribed to vibrations involving the piperazinyl ring are located at 1454 cm−1 (CH2 deformation), 1362 cm−1, and 1263 cm−1 (ring stretching). The most intense signal is centered at 1382 cm−1 because of the aromatic ring stretching mode that is coupled with ν s(COO−). The band at 1616 cm−1 is caused by ν as(C=C), that at 1541 cm−1 is caused by the stretching vibrations of the quinolone ring system, and that at 1591 cm−1 is caused by carbonyl stretching with ν as(COO−). The presence of vibrations characteristic of the carboxylate group is explained by the existence of ciprofloxacin as a zwitterion in the anhydrous form [37].

Fig. 1
figure 1

Normalized Raman spectrum of a ciprofloxacin powder and b saturated aqueous solutions at pH 1 and pH 13. The spectra were recorded with a 2 cm−1 resolution. c SERS spectra of ciprofloxacin solution with a concentration of 0.1 mM, pH 1 and 13 (∼5 cm−1 resolution)

Full size image

By comparing the spectrum that was measured on powder with those of saturated aqueous solutions at strongly acidic and strongly basic pH, one can observe significant spectral differences. First, the Raman mode at 1616 cm−1 is shifted by 13 cm−1 towards higher wavenumbers for the solution (see ESM Table S2 for the band positions). Furthermore, the relative intensity of the later band compared to that at 1382 cm−1 significantly depends on the chemical environment. At pH 1, there should only be the cationic form of the molecule, whereas at pH 13, the anionic form predominates. In the former case, the protonation occurs on the amine group of the piperazinyl ring, whereas in the latter case, the carboxylic acid moiety is deprotonated (see chemical structures in ESM Fig. S1). Therefore, at pH 13, the intensity of the Raman mode that is ascribed to the aromatic ring stretching mode coupled with ν s(COO−) at 1382 cm−1 increases because of the presence of deprotonated ciprofloxacin molecules. Further differences are observed for the Raman marker bands that are ascribed to the piperazinyl ring. Upon protonation, the spectral feature at 1362 cm−1 is downshifted by 17 cm−1 and its relative intensity is higher than the case at pH 13. A higher intensity is also observed for the Raman mode that is ascribed to the CH2 (1452 cm−1) deformation of the same ring.

To record the Raman spectra of the saturated aqueous solutions, a long acquisition time and a high laser power were used. However, the MICs for the ciprofloxacin are from 0.024 μM for E. coli ATCC 25922 to 48 μM for E. faecalis clinical strain 55 [8]. Therefore, plasmonic substrates are required to enhance the weak Raman effect. Here, a silver colloidal suspension, which was obtained through the reduction of silver nitrate by hydroxylamine hydrochloride in the presence of sodium hydroxide, was used as the SERS substrate. Ciprofloxacin solutions at a 0.1 mM concentration were prepared by solving the powder in diluted HCl or NaOH solutions. As a result, three acidic (pH 1, 2, and 3) and three basic (pH 11, 12, and 13) samples have been obtained. The SERS signal for all samples is presented in Fig. 2 (for normalized data, see ESM Fig. S2).

Fig. 2
figure 2

Mean SERS spectra and their double standard deviation of ciprofloxacin 0.1 mM solutions at different pHs (600 l/mm grating, spectral resolution ∼5 cm−1). The indicated band positions belong to the protonated molecule

Full size image

According to Fig. 2, the most significant difference among the recorded SERS spectra is represented by their intensity. Generally, the free electron pairs of the nucleophile oxygen and particularly those of nitrogen atoms present high affinity towards the silver surface. For low pH values, the free electron pairs interact with the protons of HCl. Hence, they can no longer optimally bind to the Ag surface and the intensity is considerably lower than those of the basic solutions. Additionally, Cl− is known to have high affinity towards Ag, which is proven by the presence of the Raman mode at ∼230 cm−1 and is ascribed to the Ag-Cl stretching vibration. Therefore, a competition between the target molecule and the ions is expected.

A further difference caused by the protonation state of ciprofloxacin can be observed by considering the intensity of the Raman mode at 1345 cm−1. This band was ascribed to piperazinyl ring stretching. When the molecule is protonated, the respective Raman band is clearly visible. Upon deprotonation, it becomes convoluted with the strong Raman band centered at 1382 cm−1. The identical behavior is observable in the 2800–3100 cm−1 spectral region. Namely, the Raman bands in this wavenumber region are significantly more intense for acidic pH than the recorded SERS spectra at basic pH values. These Raman modes are usually assigned to sp3 hybridized carbon atoms, which are localized on the piperazinyl and cyclopropane ring in the case of ciprofloxacin. Thus, one may conclude that either the two rings of ciprofloxacin cations are near the metallic surface or their polarizability tensor has such an orientation that its zz components are perpendicular to the surface. An additional hint concerning the orientation of ciprofloxacin on the metallic surface is given by the Raman band that is ascribed to the C=C stretching of the quinolone ring system at 1616 cm−1 in the Raman spectrum. This band is shifted by 8 cm−1 (see the high-resolution SERS spectra in Fig. 1c) when the SERS spectra are recorded under basic conditions compared to the band position in the Raman spectra of the saturated aqueous solutions. The shift may be caused by the interaction of the aromatic ring system with the Ag surface. Therefore, the deprotonated ciprofloxacin molecule probably adopts a flat orientation with interactions via the nitrogen atoms of the piperazinyl ring.

Optimization of the LoC-SERS measurements

The previous section shows that the highest signal intensities were recorded for the basic solutions. However, when experiments were conducted using the microfluidic setup, it was observed that the droplet formation was disrupted when a ciprofloxacin solution with a pH of 12 or 13 was pumped into the chip. To avoid memory effects that are characteristic to the flow through microfluidic platforms, the microchannels were functionalized with octadecyltrichlorosilane. The silane layer is not affected by the analyte-containing droplets at neutral pH levels. However, sodium hydroxide at a high concentration (0.1 and 0.01 M) can react with silane and inhibit the optimal droplet formation in the chip. Therefore, for experiments that were conducted with the LoC-SERS technique, a ciprofloxacin solution with pH 11 was selected.

Computer simulations have shown that strong signal enhancements cannot be obtained using a pure colloidal suspension in which only monomer metallic nanoparticles are present [21]. “Hot spots,” where two or more particles are separated by a small gap, must be created, which is usually achieved by adding different salts, for example, NaCl or KCl, to the colloid-analyte mixture. However, in the case of some particular molecules, the analyte itself can induce the aggregation of nanoparticles, which is also the case for the antibiotic under investigation. In order to demonstrate this, the SERS signal of ciprofloxacin solved in high purity water at pH 11 was recorded with no additional salt addition as well as in the presence of 0.3 M KCl and 1 M KCl. In Fig. S3 (see ESM), the integrated peak area of the Raman marker band located at ∼1380 cm−1 is represented for the three measurement conditions. According to these results, an optimal SERS signal is achieved when no additional salt is present in the mixture. This proves that ciprofloxacin at pH 11 induces the aggregation of the silver nanoparticles. The same behavior was found for a structurally related antibiotic, levofloxacin, at neutral pH [41]. The decrease of the peak area when KCl is present might be explained by an “over-aggregation” of the nanoparticles resulting in large clusters unable to support high electromagnetic enhancements. Therefore, all following measurements were carried out in the absence of any additional aggregation agent.

In the same graph (ESM Fig. S3), the influence of the measurement position (channel number, focus position depicted in Scheme 1), correlated with the time elapsed since mixing the nanoparticles with the analyte solution, on the peak area is also illustrated. Although the highest signal intensity was recorded short after mixing the nanoparticles with the analyte solution, the relative standard deviation of the signal was significant (28.7 %). The lowest relative standard deviation, 12 %, was obtained for the spectra acquired after 212 s of mixing time, and thus for all subsequent measurements, the focus point of the laser beam was fixed in the middle of the 5th channel.

River water as the matrix for ciprofloxacin detection

To show the potential of the LoC-SERS technique to detect ciprofloxacin in complex matrices, samples from a local river were collected, filtered, and spiked with the target molecule. Two stem solutions of ciprofloxacin with concentrations of 0.1 and 0.01 mM and a pH of 11 were filled in glass syringes. A third syringe contained river water with the identical pH. By varying the flow rates of the three syringes, 15 differently concentrated solutions have been obtained. In Fig. 3, the mean SERS spectra and their double standard deviation for the first eight concentration steps are plotted with the blank spectrum. Figure 4a presents the integrated peak area of the Raman mode at 1382 cm−1 against the concentration. The first conclusion is that for ciprofloxacin amounts higher than 10 μM, no further signal increment is present, which can be caused by the limited free binding sites on the surface of the Ag nanoparticles. Furthermore, the SERS spectrum of the blank sample shows broad bands in the 1100–1700 cm−1 spectral region. They have been associated with the presence of amorphous carbon as a result of the decomposition of organic materials. Although the signal of the blank is significant in the fingerprint region, one may observe by eye the characteristic Raman modes of the target molecule at the lowest concentration. According to the IUPAC definition that the limit of detection is equal to the signal of the blank plus three times its standard deviation, the limit of detection is below the lowest measured concentration (0.74 μM) (the threshold is illustrated in Fig. 4a). However, the linearity of the integrated peak area vs. concentration is weak. This might be the result of the convolution of the marker Raman band with the broad bands originating from the molecules of the complex matrix. Therefore, a univariate statistical method is not suitable to quantitatively determine ciprofloxacin in river water.

Fig. 3
figure 3

Mean SERS spectra and their double standard deviation of the spiked river water samples at different concentrations

Full size image
Fig. 4
figure 4

a Univariate statistical analysis: integrated peak area of the Raman mode at 1382 cm−1 against ciprofloxacin concentration in spiked river water samples. b PLS results, where one dataset (black scatter) was used for training the model used to predict a second independent set (red scatter)

Full size image

To improve the results, a multivariate statistical analysis was performed based on the combination of PCA and PLSR. Two independent datasets were recorded on two different days, with two chips and freshly prepared solutions. The first data set was used to train the PCA-PLS method, whereas the second dataset was treated as unknown. In Fig. 4b, the PLS prediction plot for the model is shown with black scattered points. The dotted diagonal line is depicted only as an eye guideline and does not represent a linear fit. To evaluate the created PLS model, a 10-fold cross validation was performed. The resulting RMSEPCV was 0.375 μM, whereas that for the prediction on all data was 0.374 μM. Both values were below the LOD, and a clear prediction for every concentration step was performed. The PLS model that was created with the first data set was used to predict the concentrations of the new one. The RMSEP value of prediction for an independent data set is 0.355 μM. Thus, the LoC-SERS technique is suitable for detecting ciprofloxacin in environmental samples in the concentration range that corresponds to its MIC values using two independent data sets. Concerning the application of the technique, a “yes-no” answer can be obtained based on the SERS results. Namely, if a SERS signal of ciprofloxacin is detected, one can conclude that the concentration in the sample is above the MIC value.

Conclusion

The effect of the pH value on the SERS signal of the ciprofloxacin solution has been assessed for the first time. To do so, three acidic (pH 1–3) and three basic (pH 11–13) solutions were prepared with a concentration of 0.1 mM. The presence of silver nanoparticles and the chemical state of the target molecule strongly affect both the absolute and relative intensities and the band positions of some marker Raman modes. First, the SERS signal is considerably stronger in the case of ciprofloxacin anions because of the higher chemical affinity and that lack of competition with chloride ions present at low pH. Second, the Raman bands that are ascribed to the C-H stretching vibrations of the piperazinyl ring have a higher intensity in the case of the protonated ciprofloxacin molecules. Third, at basic pH, the position of the band characteristic for the C=C vibration of the quinolone ring is shifted by 7 cm−1. Based on these three observations, the protonated ciprofloxacin molecule can interact with the surface of metallic nanoparticles via the piperazinyl ring, whereas the deprotonated molecule interacts via the quinolone ring.

The second part of the paper shows that the LoC-SERS technology is a suitable analytical tool to detect the target analyte in environmental samples at MIC values. The multivariate statistical analysis consisting of a combined PCA-PLSR algorithm shows that quantification is possible down to 0.74 μM with RMSEP values below the LOD.

In conclusion, in the present study, it was demonstrated that SERS combined with the droplet-based microfluidic platform and appropriate data analysis procedure has high potential for detecting ciprofloxacin in environmental samples. It was shown that a simple filtration step is sufficient and easy-to-produce silver nanoparticles can be employed as SERS active substrates. Furthermore, due to the continuous development of miniaturized and portable Raman setups, LoC-SERS has high potential for on-site applications.