SERS performance of silver nanoparticle/reduced graphene oxide-coated filter membranes

Abstract Polyvinylidene fluoride (PVDF), polyamide (Nylon) and anodic aluminum oxide (AAO) filter membranes were coated with in-situ synthesized silver nanoparticle/reduced graphene oxide (AgNP/rGO) nanocomposite by vacuum filtration, and surface-enhanced Raman scattering (SERS) performance of the prepared substrates was investigated using Rhodamine 6G (R6G) as a probe molecule. Analyte solutions were applied by drop casting and penetrated into the flexible PVDF and Nylon-based substrates within minutes, enabling very fast SERS measurements; however, this step took up to 2 hrs in the rigid AAO-based substrate. Scanning electron microscopy and Energy-dispersive X-ray spectroscopy mapping results showed that filter membrane surfaces were completely covered with AgNP/rGO nanocomposite at which AgNPs were uniformly distributed on rGO flakes; however, some large, partially reduced or unreduced GO flakes decorated with AgNP agglomerates (AgNP/GO) were also observed. Higher SERS signals were generally obtained from these AgNP/GO flakes compared to the AgNP/rGO coated regions, since they contain higher amount of oxygen-rich functional groups which enabled a higher chemical and electromagnetic enhancement in SERS signals. All the substrates could detect 10−7 M R6G even after being stored in a vacuum desiccator for 3 months, indicating that AgNP/rGO coated filter membranes are promising as stable SERS substrates thanks to their long shelf life.


Introduction
Environmental pollution, which threats human and animal health and ecosystem, is one of the major challenges that is globally struggling with. Organic pollutants, such as highly toxic and carcinogenic dyes and pesticides that are increasingly used in agriculture are among the major hazardous chemicals that contaminate natural waters and foods. [1][2][3][4][5] Therefore, constant monitoring of these contaminants is crucial. Surface-enhanced Raman scattering (SERS) offers the opportunity of fast and sensitive detection of trace amount contaminants for environmental monitoring and food safety as an alternative to commonly used mass spectroscopy and chromatography methods. [3] The application fields of the SERS are not limited to these, but its use in disease diagnosis (e.g., cancer, COVID-19) at molecular level has also been investigated. [6][7] In this technique, enhanced Raman signals is achieved by adsorption of analyte molecules onto a substrate comprising of colloids of noble metals (Ag, Au or Cu) or nanostructured surfaces of these metals. [8] When noble metal nanostructures interact with laser light, their free electrons oscillate. This resonant oscillation is called as Localized Surface Plasmon Resonance (LSPR), which leads to highly enhanced electromagnetic fields and consequently, enhanced Raman signals. [9] The largest enhancement is achieved at neighboring metal nanostructures or at the junctions of aggregates, called "hot spots." [10] Increasing the density of "hot spots" throughout the SERS substrate is critical in order to achieve high sensitivity and uniform SERS signals. [11] The "hot spot" density can be increased by adjusting the amount of nanoparticles and their aggregates, [1] designing periodic arrays [12] and forming roughened nanostructures. [13] Three-dimensional (3D) porous platforms allow adsorption of more nanoparticles onto porous structure due to their large specific surface area, allowing to create dense "hot spots." [11] They also allow fast penetration of analyte sample into the substrate during SERS analysis, providing fast and uniform measurements. [14] Several 3D materials, such as expanded graphite compressed into pellet form, [14] silica aerogel, [11] graphene-based aerogel, [15] foams, [16][17] metal-organic frameworks [18] and filter membranes [19] have been used as platforms to fabricate SERS substrates. Filterbased 3D substrates have attracted considerable attention in recent years due to their flexible nature, ability to be cut into desired size and shape, tunable pore size, various compositional types, and relatively low cost. [4,[19][20][21][22] Filter membrane platforms also make large-volume sample analysis possible for environmental monitoring and food analysis, providing pre-concentration of analyte through filtration. [19] Several methods have been reported to prepare filter-based SERS substrates. Spray coating [23] and in-situ growth of plasmonic nanoparticles on the filter matrix [24] encounter with uniformity problems and may require surface functionalization to improve uniformity. Dip-coating [25] requires concentrated nanoparticle dispersions and long process times (up to 24 hrs) to obtain a stable and uniform coating. [26] On the other hand, filtration of metal nanoparticle dispersions through filter membrane for direct assembly of nanoparticles onto filter matrix is a fast, simple and low-cost way of uniform distribution of metal nanoparticles and their agglomerates throughout [20][21] ; therefore, this approach is widely used to prepare filter-based SERS substrates. [19] As well as noble metal nanoparticles, graphene-based materials (i.e., single-layer graphene, few-layer graphene (3-5 layers), graphene oxide (GO), reduced graphene oxide [rGO]) have also been reported as active SERS substrates. [27,28] The signal enhancement in graphene-based materials is based-on chemical enhancement mechanism that occurs through charge-transfer between graphene-based material and analyte molecules. [29][30][31] However, the chemical enhancement provides limited contribution to SERS by itself compared to electromagnetic enhancement mechanism. Graphene-based materials/noble metal nanoparticles hybrids allow combining the chemical enhancement effect with electromagnetic enhancement, thus achieving high sensitivity. [29][30][31] Moreover, graphene-based materials can act as a protective layer against oxidation for plasmonic silver nanoparticles (AgNPs), which are commonly used in SERS applications [32][33] ; thus, the SERS substrates prepared by using AgNPs/graphene-based hybrids exhibit high stability. Nevertheless, the number of studies which investigate the performance of filter membranes decorated with noble metal nanoparticle/graphene-based material hybrid systems in SERS applications is very limited. In the only study on this subject, Qui et al. [34] fabricated flexible membranes of rGO supporting Ag meso-flowers and phenyl-modified graphitic carbon nitride nanosheets, which takes longer than 14 hrs to prepare, for self-cleaning SERS detection.
Herein, commercially available PVDF and Nylon-based flexible membranes, with a pore size of 0.10 and 0.45 mm, respectively, were coated with silver nanoparticles/reduced graphene oxide (AgNP/rGO) nanocomposite by vacuum filtration and their SERS performance was investigated by using Rhodamine 6G (R6G) as a probe molecule for the first time in the literature, to the best of our knowledge. 3D porous nature of these substrates enabled very fast SERS measurements due to penetration of the analyte solution into the substrate within a few minutes. 10 À7 M R6G was clearly detected on these substrates, both as-prepared and after 3 months of storage in a vacuum desiccator, indicating that these easy-to-prepare and low-cost AgNP/rGO coated PVDF and Nylon membranes are promising for routine and commercial SERS analyses thanks to their long shelf life. The rigid anodic aluminum oxide (AAO) filter membrane with a pore size of 0.02 mm was also coated with the same hybrid system and its SERS performance on R6G detection was examined for comparison. Although 10 À7 M R6G was also detected by this substrate, it took $2 hrs for the analyte drop to penetrate into the AAO-based substrate due to nano-porous structure of the membrane with no lateral pore overlap.
Polyamide (Nylon, 0.45 mm pore size, Lubitech Technologies Ltd., Shanghai, China), polyvinylidene fluoride (PVDF, 0.10 mm pore size, Durapore V R , Sigma-Aldrich, Missouri, USA) and AAO (0.02 mm pore size, Whatman Anodisc) filter membranes with a diameter of 47 mm were used as platform materials for SERS substrates without any pretreatment. The as-received Nylon filter membrane is supported by a polyester film to provide mechanical and chemical durability to the membrane. All the membranes used in the study were hydrophilic. SERS activity of the substrates were tested using R6G (C 28 H 31 N 2 O 3 Cl, 99%, Acros Organics) dye as a standard analyte.

Production of AgNPs/rGO nanocomposites
In-situ synthesis of AgNPs in aqueous GO dispersion was mainly performed by chemical reduction method described by Kasztelan et al. [35] Briefly, 100 mL of 0.05 mg/mL aqueous GO dispersion was prepared. Then, 3 mL of 0.04M AgNO 3 and 15 mL of 30 mM trisodium citrate aqueous solutions were added into the GO dispersion. The mixture was first stirred in magnetic stirrer for 5 min, then sonicated in an ultrasonic bath for 30 min. After that, 2.25 mL of 0.01M ascorbic acid was added into the mixture and mixed on a magnetic stirrer for 30 min. Addition of ascorbic acid reduced the AgNO 3 -GO aqueous mixture, leading to production of AgNPs over rGO flakes.

Preparation of AgNP/rGO nanocomposite coated filter membrane SERS substrates
Nylon, PVDF and AAO filter membranes were coated with AgNP/rGO dispersion by vacuum filtration. A total of 9 mL of dispersion was used for the coating of Nylon and PVDF membranes, while 3 mL of AgNP/rGO dispersion was filtered through the AAO membrane due to detachment of the coated film from the surface in case of using higher amount of dispersions ( Figure S1). After vacuum filtration, coated membranes were dried in air at room temperature. Each coated membrane filter was divided into four quarter pieces to be used a SERS substrate.

SERS measurements
R6G solutions with different concentrations (10 À3 to 10 À10 M) were prepared by serial dilution technique using deionized water. A total of 20 ml of the analyte solution was dropped onto the SERS substrate. The drop quickly penetrated into the Nylon and PVDF-based substrates within $1 min and $3 min, respectively. However, it took much longer time (20 min-2 hr) for the analyte drop to be penetrated into the AAO based substrate. Raman measurements were performed by Renishaw inVia Raman microscope using a laser excitation wavelength of 532 nm with 5 mW laser power and 2 s acquisition time under 50xL objective with a NA of 0.50. SERS spectra were recorded from several random points on the substrate. SERS mapping was performed using StreamHR imaging technique in Renishaw inVia system on a 20 mm Â 15 mm area with a step size of 1 mm, using 50xL objective, 532 nm laser, 1 mW laser power and 2 s exposure time. Stability of the SERS substrates was tested by re-performing SERS measurements after 3 months of storage in a vacuum desiccator.

Characterization
UV À visible absorption spectra of GO and AgNP/rGO dispersions were recorded by using Shimadzu UV-3600 spectrophotometer in the 200 À 800 nm wavelength range. Phase analysis of the AgNP/rGO nanocomposite was performed by using X-ray diffractometer (XRD, Rigaku Miniflex 600 with CuKa radiation) at a voltage and a current of 40 kV and 15 mA, respectively. The XRD data were recorded between 10 and 80 2h values with a rate of 2 /min. Raman measurements of these samples were performed by Renishaw inVia Raman microscope using a laser excitation wavelength of 532 nm with 5 mW laser power and 10 s acquisition time under 50xL objective. The morphology of the AgNP/rGO nanocomposites was characterized by using field emission gun scanning electron microscope (FEG-SEM, Zeiss Supra 50VP) and transmission electron microscope (Hitachi HT7800). The concentration of silver in AgNP/rGO dispersion was determined by inductively coupled plasma mass spectrometry (ICP-MS, Plasma Quant MS). The microstructures of the as-received and the AgNP/rGO coated filter membranes were investigated by FEG-SEM (Zeiss Supra 50VP). Energy-dispersive X-ray spectroscopy (EDX) mapping analyses of the SERS substrates were also performed.

Results and discussion
Figure 1(a) shows UV-vis spectra of the aqueous GO and AgNP/rGO dispersions. UV-vis spectrum of GO solution shows an intense absorption peak at $233 nm, which is attributed to p ! p Ã transitions of aromatic C-C bonds and a shoulder at $304 nm, which arises from the n ! p Ã transitions of aromatic C ¼ O bonds. [36,37] AgNP/rGO sample exhibited a maximum absorption peak centered at 415 nm, which is attributed to surface plasmon resonance of AgNPs, [38] confirming that AgNPs were successfully synthesized (Figure 1(a)). A relatively large FWHM value (117 nm) of this peak indicates a relatively high polydispersity of AgNPs. [39] The peak corresponding to C-C bonds redshifted from 233 to 261 nm, while the peak at 304 nm disappeared, indicating the reduction of GO into rGO by ascorbic acid and removal of oxygen-containing functional groups. [37] Inset in Figure 1(a) shows the photograph of aqueous GO (on the left) and diluted AgNP/rGO (on the right) dispersions. The brown color is specific to GO dispersion in water. For the AgNP/rGO dispersion, the change of color from brown to yellow confirms the in-situ formation of AgNPs within the GO dispersion, achieving AgNP/rGO nanocomposite powder. Figure 1(b) shows the Raman spectra of the GO and the AgNP/rGO samples. The peak at $1601 cm À1 , called the Gband, arises from the in-plane C-C bond stretching in graphitic materials and is common to all sp 2 -bonded carbon systems, while the peak at $1345 cm À1 for AgNP/rGO and at $1352 cm À1 for GO, called the D-band, originates from breathing modes of sp 2 atoms in rings and requires a disorder (defect or edge) for its activation. [40] Therefore, it is observed in the case of a disordered sample. The intensity ratio of the D-band to that of the G-band (I D /I G ) is used to characterize the defect content. The I D /I G ratio was determined as 0.92 for the GO sample, while this ratio increased to 1.02 for the AgNP/rGO sample, indicating a decrease in the size of sp 2 domains and an increase in their number, and the formation of new defects caused by the reduction process. [37,41] The band at $232 cm À1 is attributed to stretching vibrations of Ag-O bonds. [42,43] The significant increase in the Raman signals of AgNP/rGO also confirms the formation of AgNPs. [37] The Raman spectra also showed the 2D band and D þ G band at $2691 and $2927 cm À1 , respectively. The 2D-band is a second-order process related to a phonon near the K point in graphene and activated by double resonance process. [40] It requires scattering of two phonons with opposite wave vectors to ensure momentum conservation and its activation. [41] However, this two-phonon band is allowed in the second order Raman spectra of graphene and any other sp 2 carbon without any kind of disorder or defects. [41,44] This band is sensitive to the stacking order along the c-axis, thus it is used to evaluate the structural parameters of the c-axis orientation. [45]  Krishnamoorthy et al. [45] reported that harsh chemical oxidation process results in predominant structural changes in the graphite lattice due to formation of oxygenated functional groups at the basal plane and at the edges. The authors observed that oxidation of graphite resulted in a decrease in the 2D band intensity due to breaking of the stacking order as a result of oxidation reaction, and also in the appearance of D þ G band, which is the combined overtone of the D band and G band. [41] In the present study, the presence of AgNPs in the nanocomposite sample caused an enhancement in the intensity of the bands belonging to the rGO. XRD spectrum of GO shows a strong peak at 2h ¼ 12.3 , corresponding to an interlayer spacing (d-spacing) of 0.72 nm (Figure 1(c)). This peak disappeared in the XRD spectrum of the AgNP/rGO sample, and a broad peak at 2h ¼ 25.4 with a d-spacing of $0.36 nm was observed ((002) plane of graphene sheets), indicating the reduction of GO. [46] However, the slight difference between this value and the d-spacing of crystalline graphite (0.34 nm) is attributed to remaining small amount of oxygen-containing functional groups after reduction. [47][48][49] The XRD spectrum of the AgNP/rGO sample also revealed the characteristic (111), (200) and (220) diffraction peaks of metallic Ag at 2h of 38.7, 44.2 and 65.2 , respectively (inset in Figure 1(c)). [46,50] Transmission electron microscopy (TEM) images of the as-received GO (Figures 2(a) and (b)) and the synthesized Ag/rGO (Figure 2(c)) samples revealed that the large GO flakes were fragmented into smaller pieces with a lateral size of $2-4 mm during the AgNPs synthesis process, and the formed rGO flakes were uniformly decorated by AgNPs, as in agreement with the Raman results. Higher magnification TEM micrograph in Figure 2(d) shows that the synthesized AgNPs have a quasi-spherical shape and a wide particle size distribution, as in agreement with the UV-vis analysis result. The average particle size of AgNPs synthesized on rGO was determined as $19.0 ± 8.6 nm by measuring the size of 108 nanoparticles in the TEM image using ImageJ software. The corresponding statistical histogram of particle size distribution is shown in Figure 2(e). The concentration of silver in the AgNP/rGO dispersion was determined was determined as 7.34 mg/L by ICP-MS. Figure 3 shows the morphologies of Nylon, PVDF and AAO filter membranes with pore sizes of 0.45, 0.10 and 0.02 mm, before and after the vacuum filtration of AgNP/rGO dispersion. Compared with the Nylon and PVDF membranes, the inorganic AAO membrane has a uniform nanoporous structure with no lateral overlap of the pores (Figures 3(a-c)). It was observed that the filter membrane surfaces were completely covered with AgNP/rGO nanocomposite after vacuum filtration in such a way that the porous structure of the membranes was almost inevident (Figures 3(d-f)), except for the Nylon membrane, at which a few pores could still be seen at the surface due to presence of relatively large openings within the interconnected 3D network of pores (Figures 3(d) and (g)). The SEM micrographs also revealed some large flakes, protruding from the coated surface of the filter membranes (shown with yellow arrows in Figures 3(g-i)). High amount of nanoparticle agglomerates was also observed on these large flakes.
Elemental mappings of the protruding flakes on the coated membrane surfaces were performed by EDX analysis and revealed the presence of oxygen in these protruding large flakes, as well as carbon, indicating that some of the GO flakes remained without being fully reduced. The EDX mapping results of the protruding flakes on the AgNP/rGO coated PVDF membrane surface are shown in Figure 4, representatively. Relatively higher amounts of AgNPs and their large agglomerates were also noticed over these flakes (Figure 4(c)). This was attributed to the presence of more oxygen-containing functional groups on the surface of the GO flakes, causing more nanoparticles to be adsorbed onto the surface.
Higher SERS signals for the characteristic peaks of R6G were generally obtained from the protruding GO flakes (Point 1 in Figure 5) compared to the signals obtained from the AgNP/rGO coated surface around them (Point 2 in Figure 5). This was attributed to the presence of large AgNPs agglomerates on the surface of the GO flakes due to higher amount of oxygen-containing functional groups, providing a higher "hot spot" density, and therefore higher SERS signals. It has also been shown that GO provides a chemical enhancement in SERS signals through charge transfer between the oxygen-rich functional groups and the probe molecules, [51,52] which also helps to explain the higher  SERS signals on the GO flakes. Lower SERS signals that were obtained from the AgNP/rGO coated regions were attributed to a decrease in the amount of oxygen-containing functional groups as a result of the reduction process and to the more uniform distribution of AgNPs without being much agglomerated in these regions. This is in agreement with Fan et al. [52] who fabricated Ag octahedron@GO and Ag octahedron@rGO hybrids and investigated their single particle SERS performance using 4-MBT as a probe molecule by comparing with the SERS performance of pure Ag nanoparticles. They showed that the intensities of the 4-MBT peaks became weaker with increasing degrees of reduction and correlated the decrease in SERS signals with the decrease in the number of oxygen-rich species on rGO, confirming that the charge transfer between the oxygen-rich functional groups and probe molecules is the dominant mechanism contributing to the additional SERS enhancement observed in the hybrid nanostructures compared to pure Ag nanoparticles. Olar et al. [53] also reported that their GO/AgNP substrates provided a higher Raman signal enhancement compared to the rGO/AgNP ones; however, both could be used as highly efficient SERS substrates.
The uniformity of the prepared SERS substrates was evaluated by recording Raman signal intensity maps for 611 cm À1 peak of 10 À5 M R6G over an area of 20 mm Â 15 mm with a step size of 1 mm (Figures 6(a), (c) and (e)). The relative standard deviation (RSD) of 300 spectra for each substrate was calculated as 47%, 40% and 54% for AgNP/rGO coated Nylon, PVDF and AAO substrates, respectively (Figures 6(b), (d) and (f)). The quite high RSD values indicated a poor homogeneity of SERS signals, which can be attributed to inhomogeneous distribution of AgNPs throughout the substrate due to the presence of some unreduced or partially reduced GO flakes with large AgNPs agglomerates on their surface, as well as rGO flakes decorated with more homogenously distributed AgNPs, leading to a non-uniform "hot spot" density. Moreover, the degree of chemical enhancement varies depending on the amount of oxygen-containing functional groups that rGO and GO flakes have. The uniformity of the substrates can be improved by controlling the degree of reduction of GO flakes. Increasing the amount of AgNPs within the AgNP/rGO nanocomposites could also reduce the distance between AgNPs, encourage their agglomeration and increase the "hot spot" density, which could help to obtain higher and more uniform SERS signals.
The sensitivity of the AgNP/rGO coated Nylon, PVDF and AAO filter membranes were investigated by dropping 20 mL R6G onto the one-quarter of the 47 mm diameter substrate at different concentrations and recording the corresponding SERS spectra (Figure 7). Figure 7(a) shows the SERS spectra of R6G solutions with concentrations ranging from 10 À3 to 10 À10 M dropped onto AgNP/rGO coated Nylon membrane and the SERS spectra of the substrate (without R6G) taken from different points. The Nylon (polyamide) membrane used in this study is a supported, naturally hydrophilic membrane. The polyamide (nylon) was reported as a weak Raman scatterer; thus, its features are weaker than those of rGO. [54] Therefore, the Raman spectra of the AgNP/rGO coated Nylon membrane generally involved only the D and the G bands of the rGO flakes at $1356 and $1593 cm À1 , respectively (AgNP/rGO@Nylon_1 in Figure 7(a)). However, some spectra of the substrate also revealed peaks at 631, 858, 1292, 1614 and 1726 cm À1 (AgNP/rGO@Nylon_2 in Figure 7(a) and S2), which were attributed to polyester used as a membrane support. [54] For relatively high concentrations of R6G (10 À3 to 10 À5 M), neither rGO nor membrane support peaks were detected; only the characteristic peaks of R6G at 611 and 773 cm À1 arising from out-of plane bending of C-C-C ring and C-H, respectively, at 1126, 1183 and 1310 cm À1 due to in-plane bending of C-H and C-O-C stretchings, respectively, and at 1361, 1508, 1571 and 1651 cm À1 arising from the C-C stretching of the aromatic ring, were observed. [14,55,56] With further decrease in R6G concentration down to 10 À6 and 10 À7 M, the peaks of rGO and membrane support started to appear, as well as the characteristic peaks of R6G. For 10 À8 and 10 À9 M R6G concentrations, only the 611 cm À1 peak of R6G was detectable, albeit its very low intensity, which was attributed to the dominant background of rGO and the membrane support ( Figure S3(a)). It was reported that when GO is decorated with AgNPs, Raman scattering intensities of D and G band of GO increase dramatically, which is attributed to a short-range chemical effect between AgNPs and GO [57] and to electromagnetic enhancement of the AgNPs. [29] The enhanced SERS signal for GO may cause a large background that make target molecule sensing difficult. [29] A similar SERS behavior was observed for the AgNP/rGO coated PVDF and AAO filter membranes for R6G detection. The Raman analyses of the as-prepared PVDF and AAO-based SERS substrates without any R6G revealed only the D and the G bands of the rGO flakes at $1355 and $1593 cm À1 , respectively (Figure 7(b) and (c)). For the 10 À3 and 10 À4 M R6G concentrations, the characteristic peaks of R6G were dominated in the SERS spectra. rGO peaks were started to appear when the R6G concentration reduced to 10 À5 M, both for PVDF and AAO-based substrates. For lower concentrations of R6G (<10 À6 M), the D and the G bands of rGO started to suppressed the peaks of R6G; however, the 612 and 773 cm À1 peaks of R6G were still detectable even for the 10 À8 and 10 À9 M R6G on the AgNP/rGO coated PVDF substrate ( Figure S3(b)). On the other hand, limit of detection (LOD) was determined as 10 À7 M for the AgNP/rGO coated AAO substrate ( Figure  7(c)). Ding et al. [29] reported a lowest detectable concentration of 1 mM for R6G for their GO-Ag nanocomposite on silicon substrates. They reported that increasing the Ag loading and the dosage of GO-Ag nanocomposites always enhanced the adsorption ability to aromatic dyes. However, high Ag loading caused strong background response from GO (dominated D and G bands) during SERS measurements, while high composite dosage decreased the SERS signal of target molecules, since it led to a less amount of dye molecules available in the detection volume. [29] The authors also proposed that the coated GO-Ag formed a piled structure on Si and only the top layer of piled GO-Ag has significant contribution to the SERS signal. The use of 3D, porous, high surface area filter membranes with a relatively larger pore size such as Nylon and PVDF that we utilized as a platform for SERS analysis provides an advantage in terms of obtaining higher sensitivity, as it will allow more silver/ graphene-based hybrids and target molecules to be adsorbed to the surface.
The lower detection limit that we obtained for AAObased SERS substrate compared to Nylon and PVDF-based substrates was attributed to coating AAO membrane with a lower amount of AgNP/rGO (3 mL), which results in a lower amount of AgNPs and rGO or GO to provide SERS enhancement. Although graphene-based materials contribute to SERS enhancement through chemical enhancement, the electromagnetic enhancement provided by plasmonic AgNPs is the dominant mechanism in the total SERS enhancement. Therefore, the amount of AgNPs to be loaded should be sufficient to achieve a high "hot spot" density, but not so much as to form a thick silver film, in order to achieve a high SERS activity. [1,[20][21][22] Yu and White [21] filtered Ag colloid through 0.22 mm pore-sized Nylon and PVDF membranes with 13 mm diameter by using a syringe of the filter membrane. The authors reported that PVDF filter membrane was proven to be more effective than Nylon for melamine detection. The amount of silver colloid to be loaded through the membrane was determined as 1 mL in order to achieve a high surface density of Ag nanoclusters, thus a high SERS intensity, rather than a relatively thick Ag film. The SEM images revealed clusters of AgNPs that were trapped within the pores of the membrane. [21] Fateixa et al. [1] prepared Nylon filter membrane (47 mm diameter, 0.2 lm pore size) based SERS substrates by loading Ag colloids through vacuum filtering. They investigated the effect of variable AgNPs loading (2, 5, 10 and 20 mL) on the SERS activity of the substrates and found that SERS activity for crystal violet (CV) increased with increasing Ag colloid amount, confirming the dependence of SERS activity on the aggregation state of AgNPs in the nanocomposite membranes. However, there was a slight difference between the 10 and 20 mL AgNPs loading in terms of SERS intensity for CV; therefore, they used 10 mL of Ag colloid to prepare their substrates. Chin et al. [22] investigated the assembly of AgNPs in different size and shape with porous hydrophilic membranes with different pore sizes, such as PVDF, PTFE, polyamide (Nylon), cellulose (filter paper), and polycarbonate, and reported that the strongest SERS intensity for the three main peaks of R6G was obtained when their solvation-pretreated 25 mm diameter and 0.22 mm pore size hydrophilic PVDF membranes were decorated with 5 mL of multi-shaped colloidal AgNPs by vacuum filtration and 1 mM aqueous R6G was used as the analyte. The authors explained this phenomenon with an increase in the amount of AgNPs, thus decreasing the distance between AgNPs and increasing "hot spots." The morphology of the AgNPs-decorated PVDF membrane revealed uniformly dispersed AgNPs on the surface of the PVDF fibers rather than AgNPs clusters. Applying volumes of larger than 5 mL caused formation of multilayer Ag films which destroyed the plasmonic structure of the "hot spots" on the surface. The LOD was determined as 0.1 nM for R6G using 3 mL of aqueous sample in that study. [22] It should be noted that the AgNPs amount used in the preparation of SERS substrates in different studies cannot be compared by comparing the volume of the colloid used, since the concentration of AgNPs within a given volume may differ depending on the AgNP synthesis methods.
To investigate the stability of the AgNP/rGO coated Nylon, PVDF and AAO-based SERS substrates, they were stored in a vacuum desiccator at room temperature for 3 months, then the SERS measurements were performed under the same conditions using 10 À7 M R6G (Figure 8). Although the signal intensity showed a decrease for the Nylon and the AAO-based substrates to some extent, the peaks of R6G at $611 and $773 cm À1 were still clearly detectable even after 3 months. On the other hand, PVDFbased substrate provided signals as high as when it is as-prepared after 3 months. These results have shown that AgNP/rGO coated filter membranes are promising for routine, commercial SERS analyses thanks to their long shelf life. The high stability was attributed to rGO and GO sheets acting as a protective layer for AgNPs against oxidation. [58,59] Chin et al. [22] stored their multi-shaped AgNPsdecorated PVDF-based SERS substrate in air at room temperature and performed the SERS experiments. In the first week, the peaks intensities of the three main peaks of R6G (611, 1363 and 1511 cm À1 ) were maintained with $10% decrease in intensity compared to the freshly prepared SERS substrate; however, the signal intensities gradually weakened to $10% of the original intensity after 1 week. This peak intensity decrease was attributed to the poor stability of AgNPs due to their tendency to oxidation in air. [22]

Conclusions
Quasi-spherical AgNPs with an average particle size of $19.0 ± 8.6 nm were in-situ synthesized in aqueous GO dispersion by chemical reduction route, and vacuum filtered through PVDF, Nylon and AAO membranes with a pore size of 0.45, 0.10 and 0.02 mm, respectively. A total of 9 mL of AgNP/rGO nanocomposite dispersion was used for the coating of flexible Nylon and PVDF membranes, while only 3 mL of dispersion was filtered through the rigid AAO membrane due to detachment of the coated film from the surface in case of using higher amount of nanocomposite dispersions.
The SERS performance of the fabricated AgNPs/rGOcoated filter membranes was investigated by using R6G as a probe molecule. A total of 20 mL R6G solution with a concentration ranging from 10 À3 to 10 À10 M was drop casted onto the SERS substrates and the drop penetrated into the PVDF and Nylon based substrates within $1 min and $3 min, respectively, enabling very fast SERS measurements thanks to 3D highly porous structure of these filter membranes with relatively large interconnected pores. However, it reached up to 2 h for the analyte drop to be penetrated into the AAO-based substrate due to very small pore size of this filter membrane with no lateral pore overlap.
SEM and EDX analyses of the SERS substrates revealed the presence of some large un-reduced (or partially reduced) GO flakes decorated with higher amount of AgNP agglomerates that were protruding from the coated substrate surface and provided generally higher SERS signals for the characteristic peaks of R6G. This was attributed to the presence of higher amount oxygen-containing functional groups on these flakes, which provides a chemical enhancement in SERS signals through charge transfer between the oxygenrich functional groups and the probe molecules. Moreover, these functional groups enable adsorption of more nanoparticles onto the surface, providing higher hot spot density. Although the RSD values of 611 cm À1 peak intensities of 10 À5 M R6G over an area of 20 mm Â 15 mm of the substrates were relatively high, the SERS signal uniformity can be improved by controlling the degree of reduction of GO and the amount of the AgNPs in the nanocomposite.
A total of 10 À7 M R6G was clearly detected on all the three substrates, both as-prepared and after 3 months of storage in a vacuum desiccator, indicating that AgNP/rGO coated filter membranes are promising as easy to prepare, stable SERS substrates thanks to their long shelf life.

Disclosure statement
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Funding
This study is based in part on The Scientific and Technological Research Council of Turkey (TUBITAK) 2209-A Research Project Support Programme for Undergraduate Students (1919B012100874), the financial support of which is gratefully acknowledged.

Data availability
The data presented in this study are available on request from the corresponding author.