2.1

Deep-UV Raman setup

This study is based on the prototype deep-UV Raman spectrometer described in Ref.³¹ The system is further reduced in dimensions and cost and is adapted for measurements of liquid samples in 180° backscattering geometry (Fig. 2). A commercial, passively q-switched microchip laser (Pulselas-P-236 by Alphalas) emitting 5 mW @λ_exc = 236 nm, with 10 KHz repetition rate is used as a light source. A single element (BS/filter in Fig. 2) reflects the incident light wavelength, while transmitting the Stokes line. Thereby, the laser line is removed. A single aluminum 90° off-axis parabolic mirror is used to focus the beam and collect the scattered light.³¹ The complete instrument with the compact light source fits on a 45 x 30 cm breadboard. The advantages of the chosen excitation wavelength include a spectral separation of the fluorescence and Raman signals and availability of the optical components.²³ In particular, commercial Rayleigh Raman filters (e.g., from Semrock) are available for the chosen wavelength and not for shorter excitation wavelength, e.g. Nd:YAG based 213 nm. The signal loss due to the absorption by optical components is negligible. The effective laser beam power at the sample position is reduced to less than 40% (from 5 to 3 mW). Due to its simplicity, the presented optical scheme is adaptable to further modifications without requiring the complete re-design of the set-up.

Figure 2.

Simplified optical scheme of the deep-UV Raman spectrometer. The optical components include UV-mirrors (M), a diverging lens (L), off axis parabolic mirrors (P) and a beam splitter/edge filter (BS/filter). The laser beam is shown in blue while the scattered light is shown in purple.

2.2

Raman spectroscopy in the visible energy range

Visible Raman spectra were measured by meas of a continuous wave laser (Ocean Optics, 100 mW nominal power, 40 mW effective power at the focal point, λ_exc = 532 nm) and a Raman probe (Bruker) connected by an optical fiber. All Raman spectra were acquired with the integration time of 20 seconds and 100 averages.

The dark spectrum (camera noise) was subtracted with no additional smoothing procedure. As a threshold to perform peak identification during the spectra evaluation, we assumed a signal to noise ratio equal to 3.

2.3

Ion Chromatography and UV absorption spectroscopy

As a validation method, Ion Chromatography (IC) was adopted to determine the nitrate content in the sample withdrawn from the river. The measurements were performed on a Metrohm IC 882 instrument with a ASUPP 5-250 separation column. Certified standard reference materials were used. UV absorption measurements were performed on a UV spectrometer (Varian UV scan 50 BIO).

2.4

Samples

Pharmaceuticals and benzotriazole were purchased from Sigma-Aldrich, specifically: Citalopram hydrobromide (C₂₀H₂₁FN₂O·HBr, molecular weight = 405.3 g/mol); Venlafaxine hydrochloride (C₁₇H₂₇NO₂·HCl, molecular weight = 313.8 g/mol); Carbamazepine (C₁₅H₁₂N₂O, molecular weight = 236.27 g/mol), benzotriazole (C₆H₅N₃, molecular weight = 119.1 g/mol). The structure of each molecule is reported in Fig. S1 Supplemental Material. Solutions were prepared using deionized water and diluted to required concentration. Nitrate solutions were prepared in deionized water and an environmental sample was taken from a local river (Chriesbach, Dübendorf Switzerland). Samples were stored at 5 degrees Celsius and were measured in quartz cuvettes.

3.1

Excitation and relaxation processes in the deep-UV spectral range

The interaction of a Raman-active molecule with the incident photon depends on the laser wavelength. The physical origin of such wavelength-dependent behaviour is briefly schematically represented in an extended Jablonsky diagram (Fig. 3). Upon non-resonant irradiation the excitation frequency v₀ is far from inducing any electronic transition, hence the molecule is excited from the ground state (GS) to a virtual state (Fig. 3A).²⁴ Radiation is emitted back at the same frequency v₀ (Rayleigh elastic scattering) and via inelastic scattering at a shifted frequency ν± v₀ (Stokes and Anti-Stokes inelastic scattering), which corresponds to the vibrational or rotational energies. In Fig. 3, for simplicity, only the the Stokes inelastic scattering (ν – ν₀) is represented. This phenomenon is known as Standard Raman (or non-Resonant Raman) and virtually occurs at any incident photon wavelength.³⁸

Figure 3.

Extended Jablonsky diagram showing main possible quantum electronic transitions occurring at different excitation wavelengths. (A) Standard Raman in the visible range, (B) Resonant Raman (C) Photon absorption and time delayed (clock) fluorescence emission (D) Photolysis generating new compound exciting new Raman emissions. The broken line separates the system before and after the photolysis. Second order processes and the probability of each electronic transition are omitted for simplicity.

When the photon energy of the excitation source is equal to the energy of an electronic transition, the molecule is excited from the GS to a real electronic state (Fig. 3B). In such resonance condition, the scattered signal can be enhanced of several order of magnitude (up to 10⁸ times higher) when compared to standard Raman scattering.²² The enhancement is due to the resonance-increased probability of excitation compared to an excitation to a virtual state.²⁵ The signal is also enhanced in quasi-resonance condition, and increases as the photon energy approaches the electronic transition. The emission of Stokes and anti-Stokes light is the fastest relaxation channel with competing emission originating from fluorescence (Fig. 3C). Fluorescence takes place on longer time scale than Raman scattering (nanosecond timescale) and often obstructs the weaker Raman signal.²² Additionally, if the molecule undergoes a process of photolysis, molecular bonds are broken (Fig. 3D). This leads to the formation of new compounds, which then act as Raman-active light scattering molecules.

The majority of molecular species exhibit absorption bands in the UV and deep-UV spectral range,²⁵ hence the scenario depicted in cases (B,C,D) is expected when deep-UV light is used to excite the specimen. We notice that the molecules under investigation do not exhibit absorption bands precisely at the excitation wavelength (236.5 nm; Figs.S5 and S6). Therefore, our method of choice is quasi-resonant deep-UV Raman Spectroscopy.

3.2

Raman intensity and self-absorption

The sensitivity of Raman analysis is mainly determined by the Raman scattering cross section, σ. The scattering is much stronger in resonant Raman, as σ increases proportionally to the square of the molar absorptivity, ε, of the absorber at the resonant frequency ν₀:³⁹

The values of σ can be as large as 10^–22cm^–2·mol^–1·sr⁻¹, which is roughly 10⁸ times higher than in non-resonant Raman (≈ 10⁻³⁰cm⁻²·mol⁻¹·sr⁻¹²⁴). Practically, effective sensitivity depends also on various parameters: the collection optics, scattering geometry, excitation wavelength, laser power, etc. In short-wavelength Raman the factor determining Raman sensitivity is the phenomenon of differential self-absorption. Briefly, the excitation light is attenuated by passing through the sample (self-absorption), affecting the collected Raman signal. The scattered light travelling back through the sample is also self-absorbed, but the single Raman lines are differently attenuated due to the dependency of ε (molar absorptivity) on the frequency of the scattered radiation (differential self-absorption).^39–41 Several theoretical and optical models have been proposed to describe the intensity of Raman bands accounting for experimental conditions and considering the differential self-absorption phenomenon (mostly analyte-related).^{35, 39–41} It is not the scope of this work to introduce a new model or discuss in detail the existing ones, but rather to evaluate, in a simplified way, the impact of differential self-absorption on the measured Raman spectra (we refer to it from here as self-absorption) by adapting the model introduced by Shiver et al.^{40, 41}

The intensity of a specific Raman band i, quantified in counts during the measurement process, can be generally expressed by:

where k is a coefficient accounting for the experimental conditions, i.e. laser Intensity (here defined as W/cm², illuminated volume V, collection aperture and detector efficiency, σ_i is the scattering cross section and N_i is the number of analyte molecules per unit of volume. In order to quantify I_i, the intensity of the Raman scattered light at a specific wavelenght, I_i,R, is estimated as follows. During the measurement process the Raman light is converted by the spectrometer, through a conversion factor k_s:

We define V as the interacting volume of the focused laser beam with the liquid sample. V can be roughly approximated by a portion L of a Gaussian beam propagating along the z direction (Fig. 4).

Figure 4.

Illuminated Gaussian-like volume in back-scattering geometry and self-absorption. Laser and scattered light propagates from the origin O to the point (x’, y’, z’). L is the depth of focus.

The following approximations are considered: 1) the scattered light, which may originate outside of the interacting volume is neglected and 2) the detector collects homogeneously the radiation scattered from different point inside the interacting volume. The infinitesimal scattering element dV at the position (x’,y’,z’), stems for a generic sampling point which is excited by the laser and which scatters light backwards. The laser light distribution function, I_λ is attenuated along the z direction according to the Beer Lambert’s law and expressed by:

where the nominal intensity of the laser is I₀ and α (mm⁻¹) is the absorption coefficient at the excitation wavelength λ. The Raman scattered light is also attenuated according to the Beer Lambert’s law, and the loss term is , where β is the absorption coefficient at the Raman band wavelength. The solid scattering angle Ω is relatively small in our setup and each scattering point dV scatters over a negligible angle dθ, hence cosθ ≈ 1. By defining I_Ω as its distribution function, the fraction of total scattered light is expressed by k_Ω = ∫ I_ΩdΩ over the solid angle Ω. The overall expression for I_i,R, accounting for all the introduced terms is:

The terms σ_i N_i do not depend on the spatial coordinates and P (Watt) is the laser power without any absorption loss at a surface centered at (x,y) along z direction, estimated as the integral of the laser nominal intensity P = ∫∫ I₀(x,y)dxdy. By incorporating the constant k_s of eq.(3), we obtain the expression for the detected intensity:

The integral is calculated over the length l, namely effective absorbance length. The solution and final expression for I_i is then:

This equation was originally formulated to evaluate self-absorption in mixtures containing resonantly active components.⁴¹ Here, we make use of it by considering the solution as a mixture of resonant (analyte) and non-resonant (water) components. In non-resonant condition (α+β) is small (<1). The term 1, so in eq.(7) the intensity I_i,R∝N_i, hence proportional to the analyte concentration. The length l, in absence of absorption, coincides with L as the collected light is scattered from the entire illuminated volume V. Conversely, in resonance (or pre-resonance) (α + β) > 1 and the exponential << 1, meaning that the beam is extensively absorbed along the probed volume. In this case, is ≈ 1 and the second bracket of eq.(7) is constant, as both N_i and the denominator are proportional to the concentration c (α = ε_α·c and ε_α is the molar absorption coefficient). A strong attenuation of the non-resonant component, i.e. a water Raman mode, is expected. In this respect we define the inhibition coefficient Ki as the ratio of the inhibited and non-inhibited water mode I_water ÷ I_0,water. This term, as derived from eq.(7), results in the following relation:

Water peak intensities are taken from Raman spectra while α and β are obtained from the UV absorption measurements through α = Abs÷10 mm·ln(10), where Abs is the measured Absorbance. The term l is estimated by graphically solving eq.(8). It is worth noticing that neglecting the effective laser intensity distribution and the spectrometer clipping effect represent a strong approximation.³⁵

4.1

Antidepressant agents

Solutions of CIT, VEN and CBZ were prepared at different concentrations and measured both by deep-UV and visible light (Fig. 5). Deep-UV spectrum of CIT presents two peaks: (1) 1236 cm⁻¹ and (2) 2232 cm⁻¹, assigned to the aromatic C-F stretching vibrations and the triple-bonded C-N stretching mode, respectively.⁴² The water-related mode appears at about 1620 cm⁻¹. The visible spectrum presents a huge fluorescence background, which prevents any reliable peak identification. The estimated LOD for the deep-UV Raman is 5 mg/l (12 μM). VEN deep-UV spectra have a slightly smaller signal to noise ratio than the CIT ones. A Raman mode is clearly identified at 1280 cm⁻¹ and assigned to a O-CH₃ stretching mode.⁴³ The visible range Raman spectrum presents a weaker fluorescence background than in case of CIT, however any clear peak identification is still perturbed by the overlap of the signals. The estimated LOD, according to a concentration series, is 6 mg/l (20 μM).

Figure 5.

Deep-UV and visible Raman spectra of CIT, VEN, CBZ at different concentrations. Main Raman modes are marked and assigned in the text. The left and right axes refer to the deep-UV and visible Raman measurements, respectively. The constant background is arbitrary and here used to increase readability of the spectra.

CBZ specrum is more complex, as several vibrational modes are detected. Main modes are assigned as follows: (1) (C-H) in plane bending mode at 1118 cm⁻¹, (2) (C-H) bending mode at 1160 cm⁻¹, (3) (C-C) stretching modes at 1219 cm⁻¹, (4) (C-H) bending and (C-C) stretching modes 1491 cm⁻¹, (5) (C-C) stretching mode 1566 cm⁻¹ and (6) (C-C) double bond stretching mode 1622 cm⁻¹.^{44, 45} The background intensity of the deep-UV spectra decrease more evidently relative to the other two pharmaceuticals, as the fluorescence observed in visible Raman is weaker. The estimated LOD of CBZ is 12.5 mg/l (50 μM).

As expected, all measured compounds present no fluorescence-related background in deep-UV Raman.^{18, 22} However, another factor, the analyte concentration, affects the background intensity, as shown prominently in the CBZ case. The dependence between the band intensity and concentration is not strictly linear for CIT and VEN, and is almost constant for CBZ (Fig. 6). At the same time, a strong attenuation of the O-H water modes at 3440 cm⁻¹ is observed (Figs. S2, S3 and S4, Supplemental Material). Addressing such effects to self-absorption, we apply eq.(8) to estimate the absorption length l. UV absorption at the laser wavelength and at λ = 257 nm (O-H water mode wavelength) was measured (Fig. S5, Supplemental Material) for the following solutions: CIT (100 mg/l), VEN (50 mg/l), CBZ (33 mg/l). The estimated l (mm) values and all the coefficients are listed in Table S1 (Supplemental Material). The strongest inhibition of the water mode is observed in CBZ, where the highest absorption is also measured. The estimated l = 9.8 mm is as long as the cuvette (10 mm). This physically means that the beam is extensively absorbed over the entire illuminated volume V⁴¹ explaining the observed non-linearity that originates from absorption (Fig. 6). CIT presents the lowest absorptivity, the highest k_i ratio and l = 6.7 mm.

Figure 6.

Intensity of CIT(2) @ 2232 cm⁻¹, VEN(1) @1280 cm⁻¹ and CBZ(5) @1566 cm⁻¹ Raman modes as a function of the concentration. The noise level is marked, (approximately 5 counts arbitrary unit).

4.2

Nitrate and nitrite

Compared to visible Raman spectra, the deep-UV Raman of a 100 mg/l NaNO₃ solution (Fig. 7 top), exhibits a clearly visible peak at 1050 cm⁻¹, which corresponds to the ν symmetric stretching mode. The related LOD, as demonstrated in our previous study,³¹ is approximately 2 mg/l and it is achievable by simply increasing the integration time from 20 to 60 seconds. Notably, a broader Raman mode at around 1330 cm⁻¹ is assigned to a vibration.^{30, 31} As discussed in our previous work, for λ_exc = 236.5 nm a quantum yield ϕ = 0.05 was estimated, hence a contribution of nitrite within the LOD is not surprising. The visible spectrum has lower signal to noise ratio. Above the fluorescence background only water-related bands at 1620 cm⁻¹ and about 2100 cm⁻¹ are detectable, with only a minor, barely visible nitrate band at 1050 cm⁻¹. Nitrate was clearly detectable also in the samples taken from an aquatic ecosystem (Fig. 7 bottom), including a peak indicative of photolysis. Using the intensity of the ν nitrate peak in the lab-made solution as a reference, an estimated 35 mg/l is present in the river. These results agree with the Ion Chromatography (IC) analysis adopted as a validation method. The latter estimated a 25 ± 0.04 mg/l concentration, with no presence of nitrite (at a concentration higher than 0.1 mg/l). The fluorescence background of visible Raman is especially strong in the river water spectrum, more than a factor of ten compared to the pure nitrate solution. We ascribe it to the presence of dissolved organic matter.⁴⁶ Therefore, the visible Raman, unlike its deep-UV alternative, is unsuitable for the detection of environmentally-relevant nitrate concentrations, especially in the samples taken directly from aquatic ecosystems. It can be noted, that the O-H water peak is slightly suppressed compared to the Raman spectra of the pharmaceuticals. By applying the same procedure as outlined above, the absorption length for the river sample (Table S1, Supplemental Material) was evaluated. Notably, the beam is absorbed within a limited length (l = 2.4 mm), hence the intensity of the nitrate mode has to scale linearly with the concentration, as predicted in the section 3.2. UV Absorption spectra are plot in Fig. S6 (Supplemental Material). The relatively high molar absorptivity a is assigned to the presence of organic matter.

Figure 7.

Deep-UV and visible Raman measurements of: (top) 100 mg/l NaNO₃ solution and (bottom) of the river water sample. The main features are marked by a red and black arrows for and , respectively.

4.3

Benzotriazole

Owing to its low sensitivity, visible Raman is not commonly used for detecting benzotriazole in solution at low concentrations (mg/l). On the other hand, the use of deep-UV light is hugely hindered by the ultraviolet absorption exhibited by this compound.⁸ In fact, the molecule presents two main absorption peaks at 259 and 278 nm.⁴⁷ Aware of such limitations, we first evaluated visible Raman LOD by performing a high concentration series (from 500 mg/l to 10 g/l (Fig. S7, Supplemental Material). Despite the presence of fluorescence background between 500 and 2500 cm⁻¹, main benzene ring modes are detectable at 783, 1010 and 1387 cm⁻¹. As expected, the minimal concentration for reliably identifying those vibrational modes is high, approximately 1 g/l. The deep-UV Raman concentration series were performed on a different range, from 0.5 to 1000 mg/l. As before, no fluorescence overlap was present, however, unexpectedly, no benzotriazole related Raman modes were identified at any concentrations (see Fig. S8 Supplemental Material). Furthermore, above 1 g/l all Raman modes disappear and almost no signal is acquired. This effect of attenuation is most evident on the main water O-H modes at 3420 cm⁻¹ (Fig. 8 top), with a strong attenuation in the deep-UV range and virtually none in the visible. This absorption effect is quantitatively and exponentially correlated to the analyte concentration (Fig. 8 bottom), with an appreciable drop above approximately 0.5 mg/l concentrations. Fitting the experimental data to the self absorption model discussed above yields the following function, which is based on eq.(8):

Figure 8.

(Top) Deep UV and visible measurement of deionized water and 10 mg/l benzotriazole concentration solution. (Bottom) Water Raman mode intensity (O-H stretching water mode at 3420 cm⁻¹) plotted with respect to benzotriazole concentration and fitted through eq.(9). Inset: corresponding spectra (0.5 mg/l, 5 mg/l, 25 mg/l, 50 mg/l, 100 mg/l, 250 mg/l, 500 mg/l, 1000 mg/l concentration).

where the constant A = I₀, x is the independent variable, C and B are free fit parameters. The intensity of the water peak is well fitted by the model (R² = 0.99), (Fig. 8 bottom), providing a value of C_fit = 0.088. This term represents the main calibration parameter for estimating the analyte concentration, while B accounts for intensity of the signal background. We notice that few data points (from 50 to 250 mg/l) lie below the fitting curve and this can be ascribed to a variation of the laser intensity during the measurement process. The penetration depth, calculated for a 50 mg/l solution, is equal to 7.8 mm (Table S1, Supplemental Material), indicating an extensive beam absorption.

5.1

Photolysis

Nitrate was successfully detected in pure water (2 mg/l) and in aquatic ecosystem, and the LOD may be improved as discussed above. In particular the deep-UV Raman analysis of nitrate in aquatic ecosystems is promising, as dissolved organic matter often generates an overlapping fluorescence background, which hinders the visible or near-UV Raman analysis. Nevertheless, the photolysis of nitrate ions upon UV light illumination affects the reliability of the analysis. The photolysis effect is diminished by the reduction of the exposure time of the sample to UV light, e.g. by means of a circulating liquid cell that allows for a “constant flow” mode measurement.

5.2

Absorbance

Benzotriazole vibrational modes were not detected using deep-UV light at any concentration between 0.5 mg/l to 1 g/l, due to the self-absorption phenomenon. This effect highlights the constraint for a potential use of deep-UV Raman spectroscopy as an environmental analysis method. If a strong UV absorber is present in the specimen, the light may be completely attenuated by the UV absorbing compounds. However, using the Raman intensity of water as internal standard it is possible to estimate the concentration of the UV-absorbers in the solution even in the absence of Raman signal from the analyte. Pure water can be used for reference. This can be achieved in one set-up, e.g by using a liquid cell connected to both the solution and the pure water. The measured Raman intensities of detected pollutants can then be corrected using the model introduced in the theory section. The intensity change on the water signal is a qualitative indicator of the overall water purity comparable to already established UV-absorption sensors with the advantage of providing fundamental chemical information not accessible otherwise. In particular for automatic sensing applications, this quality factor is of high relevance. This is most easily be explained using an example. The Raman setup will have difficulties in detecting nitrate, if the solution also contains benzotriazol. As benzotriazol (at these concentrations) does not show Raman lines, an automatic system will report ‘clean water’. The quality factor using the water Raman signal does not provide a chemical analysis, but ensures that the Raman signal is reliable in case of no changes. Small changes can be used to correct the Raman analysis, while strong changes indicate that the Raman analysis is not reliable anymore.

5.3

Estimation of minimal limit of detection

The above given concrete examples may generalized to estimate the theoretically minimal limit of detection. We have to consider: