Highly e ﬀ ective hot spots for SERS signatures of live ﬁ broblasts †

Pre-formed silver – boron nanoparticles of 22 nm form pearl-like necklace nanostructures with interparticle junctions of less than 10 nm length in the matrix of polyethylene glycol (8000 Da). The silver necklace nanostructure is stable at 37 (cid:1) C or 70 (cid:1) C and also inside a live cell medium. A polyethylene glycol matrix with a shorter chain length (1000 Da) does not protect the nanoparticles against attraction, and random aggregates are formed. Silver necklace nanostructures exhibit strong Raman enhancement by more than (cid:3) 10 9 which is much higher than for silver – citrate or random silver – boron aggregates. The polymeric matrix of 8000 Da contributes strongly to the electromagnetic ﬁ eld enhancement and removes the chemical contribution to the surface Raman scattering increase. The stable interparticle junctions act as local hot spots for strong Raman scattering signals collected from live ﬁ broblasts and allow systematic in situ studies.


Introduction
Silver nanoparticles (NPs) are of highest interest in nanotechnology being among the most commercialized products in healthcare and medicine. The reasons for this interest are the unique antibacterial and optical properties of silver among all metals that are strongly dependent on the colloidal geometry and the dielectric function. This dependence is being exploited for the development of novel biosensors, 1 electro-optical devices, 2 but mostly in biological imaging and surface enhanced Raman scattering (SERS) spectroscopy. 3 In nanotechnological development of analytical tools, the SERS application meets technical demands, but it is also dependent on advanced nanoparticle design for strong enhancement.
Compared to conventional Raman spectroscopy, the SERS intensities achieved by silver nanoparticles 4,5 can be as high as 10 6 to 10 10 or even higher for single molecules 6,7 (three orders of magnitude higher compared to nano-gold) with near infra-red laser excitation. 8 This increase is due to the dominant inuence of the electromagnetic enhancement ($10 4 ) by the local surface plasmon resonance (LSPR) of silver.
Among other metals capable of excitation of surface plasmon oscillations, silver has the lowest imaginary part of the dielectric permittivity, which is responsible for the dissipation of the electric eld energy. 9 Hence the efficiency of surface plasmon excitations is highest for silver NPs. Nanoparticles with a large aspect ratio or those such as periodic arrays and networks exhibit maximum refractive index sensitivity and strongly contribute to the SERS increase. The enhancement of the electromagnetic (EM) eld of the silver SPR occurs at the edges of the nanoparticle surface and in the junctions (<10 nm) between the particles. 10 The coupling between two adjacent silver or gold nanoparticles forms an efficient "hot-spot" for SERS. The SERS enhancement factor at such hot spots can be as high as 10 14 , which is sufficient for single molecule detection. 11 Many efforts have been focused on the design of various hot spots such as silver nanocube dimers, 12 nanoparticle-nanowire couples 13 or nanorod dimers. 14 The different coupling geometries can be formed by nanofabrication techniques or self-assembly methods to achieve enhanced EM elds for SERS. 15,16 The main drawbacks are (i) the very small number of plasmonic structures, (ii) a single value of EF for each cluster geometry, which is difficult to assess, (iii) the heterogeneity of the colloidal solution and cluster precipitation and (iv) the stability against dissolution.
Alternatively the particle morphology and size distribution can be controlled by a plethora of lithographic approaches. [17][18][19][20][21] However they are time consuming, expensive and inefficient for large scale substrates. The gaps between uniform nanostructures are higher than the range of strong electromagnetic enhancement (i.e. >20 nm), and the biocompatibility is poor. 22 The highest SERS EF from the 4 nm silver islands coated with about 6 nm silica lm on the glass surface achieved by temperature-programmed desorption 23 does not exceed 10 8 . This value is several orders of magnitude less than from nonordered elongated colloidal silver.
For live cell studies hot spots of silver NPs can be achieved in the polymeric matrix with a variety of polymers that are nontoxic and biocompatible. 24,25 One thus achieves a high stability of NPs in the cell cytoplasm. However, very little is known about silver 'hot spots' optical properties and calculated SERS efficiency. Still, randomly aggregated gold NPs (without junctions) have been widely studied by SERS inside the cells. 26 As an initial approach of a long-term study our rst aim is to form hot spots of pre-formed silver NPs that enable high SERS efficiency inside live cells (e.g. broblasts). In addition, our goal is to dene appropriate conditions of this enhancement.
For hot spot formation pre-formed silver NPs are coated with polyethylene glycol (PEG) and the interparticle distance is controlled by the polymer chain length (e.g. 8000 Da or 1000 Da) (Scheme 1).
Distinct from previous studies, 27 silver NPs assemble into a necklace structure with 8000 Da PEG and can form hot spots. A short PEG chain length is ineffective against particle attraction and hot spots cannot be formed. It is remarkable that hot spots are only of several nm distance and the necklace assembly holds its structure at 37 C (or even at 70 C). The SERS enhancement factor (10 9 ) at such hot spots is several orders of magnitude higher than from gold or silver aggregates previously reported. 3,28 Because of its strong stability the silver-PEG necklace nanostructure can be used as a exible SERS platform for a meaningful collection of high quality Raman spectra inside live cells. The unique removal of the chemical contribution to the SERS increase is a prominent advantage of a polymer matrix in a hot spot nanostructure. As another advantage, hot spots can act as secondary sources of the uorescence increase. This property is highly demanding especially in two photon nonlinear spectroscopy of biomolecules. 29 Overall, our challenging work stimulates not only a systematic SERS study of organelles in live broblasts but also opens new pathways for hot spot formation in a exible polymeric system with a specic part of SERS activity.

Results and discussion
(A) Hot-spot formation and stabilization of silver with a necklace nanostructure Two types of silver nanoparticles (NPs) were freshly formed by the reduction of silver nitrate aqueous solution with ice-cold sodium borohydride in water (Fig. 1). Large silver NPs appear with a brownish-grey color and form elongated assemblies due to aggregation (TEM image in Fig. 1A). The NPs exhibit a strong absorption near 390 nm with an elevated baseline due to scattering (Fig. 1A). The average size of the NPs is about 100 nm AE 20 nm and the NPs have a negative zeta potential of À36 mV. The other type of silver NPs is formed with an excess of sodium borohydride and exhibits a strong narrow absorbance peak near 395 nm (Fig. 1B).
Silver NPs show a bright yellow color in water and have an average size of 30 nm AE 3 nm (TEM image in Fig. 1B). The negative zeta potential is around À63 mV, and no sediment appeared within several months. To remind, the high stability of smaller silver NPs is due to the hydroborate (BH 4À ) and borate (BO 2 2À or BO 3 3À ) ions in base-stabilized and partially hydrolyzed sodium borohydride. 33 This provides a substantial electrostatic barrier to aggregation. Despite the strong Coulombic repulsion the most stable arrangement of silver NPs can be disrupted by a change of the surrounding medium or thermal uctuations. Without change of the ionic strength silver NPs are added to polyethylene glycol (PEG). The latter acts as a exible dispersant and easily binds water molecules protecting NPs against attraction. The distance between neighboring NPs can be restricted due to the low congurational freedom and increased local concentration of the polymer chains resulting in the lowering of entropy. Upon heating the electrostatic protective layer of silver nanoparticles may be weakened partially due to the weaker hydroborate or borate ions on the surface of nanoparticles. However, a polymer matrix with a longer chain length may form an additional protective layer effectively contributing to the steric stabilization of nanoparticles. Later polyethylene glycol with long (PEG8, 8000 Da) and short (PEG1, 1000 Da) chains are used to control the distance between NPs resulting in hot spot or non-hot spot formation.
To test the thermal stability silver-PEG solutions were incubated overnight at room temperature followed by 30 min of Scheme 1 (A) Schematic illustration of "hot spot" formation of silver nanoparticles in the matrix of polyethylene glycol (8000 Da). In the interparticle junctions the local increase of the electromagnetic field due to the surface plasmon resonance will contribute to the strong surface Raman scattering enhancement (SERS) from an analyte. (B) "Non hot spots" of silver nanoparticles can be formed in the matrix of polyethylene glycol with a shorter chain length (1000 Da) and in this case Raman signals of an analyte cannot be intensified. Fig. 1 (A) UV-vis absorbance spectra of a silver colloidal solution after reduction of 1 Â 10 À3 mol L À1 silver nitrate with 2 Â 10 À3 mol L À1 sodium borohydride aqueous solution. The volume ratio of silver nitrate to sodium borohydride is 3 : 1. The two insets show a snapshot of the final freshly prepared silver colloidal solution and the corresponding TEM image (scale bar is 200 nm). (B) UV-vis absorbance spectra of silver colloidal solution after reduction of 1 Â 10 À3 mol L À1 silver nitrate with 1 Â 10 À3 mol L À1 sodium borohydride aqueous solution at a volume ratio of 1 : 3. The two insets show a snapshot of the final freshly prepared silver colloidal solution and the corresponding TEM image (scale bar is 100 nm). mild shaking in a thermomixer at 37 C or 70 C in a dark room. Before heat treatment bare silver NPs exhibit one distinct absorbance band at 380 nm that is characteristic for absorption of silver NPs of 22 nm AE 3 nm (Fig. 2). The peak at 260 nm appearing shortly aer injection of silver into the ice-cold sodium borohydride solution is due to the absorbance of small species at the initial stage of silver(I) reduction by borohydride. To remind, the formation of silver nanoparticles is due to aggregation induced growth, in which initially formed small borohydride-bound silver clusters grow by a mechanism that is largely aggregative in nature.
At 37 C the 255 nm peak rapidly grows in, the SPR peak originally centered at 380 nm then red shis to 385 nm and its intensity decreases threefold. The SPR red shi is due to an increased density of free electrons in the particle, 34 because the energy of SPR depends on the dielectric constants of both the nanoparticle and the surrounding medium. 35 The full width at half maximum, FWHM, increases from 45 nm to 50 nm (Fig. 2, red curve), indicating formation of larger NPs. The broadening and shi to lower energy occurs due to retardation and excitation of higher order multipoles, when the particles grow beyond a certain diameter. [36][37][38] At 70 C the absorbance peak is blue shied again to 380 nm, its intensity is slightly smaller (Fig. 2, blue curve) and the FWHM value gradually increases up to 55 nm. The TEM image shows spherical silver-PEG8 NPs that assemble into a pearlnecklace structure aer heat treatment at 70 C (inset of Fig. 2). Silver-PEG8 pearl-necklaces have a clear bright yellow color (another inset of Fig. 2) that is distinct from the brown-grey random silver-boron aggregates mentioned above (Fig. 1A). The stability of silver-PEG8 NPs is increased due to PEG. The latter at high molecular weight exhibits increased reactivity of binding water and effectively assembles particles in a necklace structure against attraction.
When silver NPs are coated with polyethylene glycol with a lower molecular weight (PEG1, MW ¼ 1000 Da) a narrow absorbance peak appears at 385 nm and does not shi at 37 C or 70 C (Fig. SI1 †). Silver-PEG1 colloidal solution is transparent with a clear yellow color and no sediment appears on the bottom aer heating. The FWHM value does not change much upon heating being only slightly increased from 43 nm to 46 nm (at 37 C or 70 C). However, the intensity of the absorbance peak is decreased by almost threefold aer heating of silver-PEG1 solution at 37 C or 70 C. The TEM image shows aggregated silver-PEG1 NPs that strongly merge aer heating. No assemblies with a necklace-structure are observed (inset of Fig. SI1 †). In addition, DLS diagrams of size distribution of colloidal silver solutions are shown in Fig. SI2. † Overall, silver-boron NPs in the matrix of a longer chain polyethylene glycol (8000 Da) lead to the controlled aggregation of NPs into a pearl-like necklace structure. In this way interparticle junctions are only of several nm size and are stable at 37 C or even 70 C. A strong local electromagnetic eld enhancement is achieved in the junctions of plasmonic NPs and can be used for Raman scattering enhancement. Later we use this approach to examine the Raman scattering of silver-PEG NPs and their surface enhancement.

(B) Surface-enhanced Raman scattering (SERS) efficiency of silver-PEG NPs
Rhodamine 6G (Rh6G) was used as a test dye to quantify the surface enhanced Raman scattering (SERS) efficiency of silver NPs. Prior to SERS measurements silver NPs were incubated in the presence of Rh6G in MQ water with 10 À4 mol L À1 NaCl for at least 3 hours to ensure the adsorption of dye molecules onto the particle surface. The presence of Cl À anions is necessary in order to assist adsorption of dye molecules, as these anions are specically adsorbed at Ag surfaces. 39 Low concentrations of NaCl promote dye chemisorption via a N-Ag covalent bond. 39 It does not induce coagulation of particles (no changes of the UVvis absorbance spectrum of silver-Cl À solution were observed).
The adsorption of dye molecules onto the silver NP surface is proved by the systematic disappearance of the characteristic Rh6G absorption near 525 nm (Fig. SI3 †). As the concentration of Rh6G is decreased to 10 À5 mol L À1 the absorbance peak becomes a shoulder of the main UV-vis band of silver near 393 nm. It disappears when the concentration of the dye is further decreased to 10 À10 mol L À1 .
Moreover, the Rh6G chemisorption on silver NPs is also proved by the electron transfer quenching of the uorescence of the silver-dye colloidal solution (Fig. SI4 †). As the concentration of the dye decreases from 10 À6 mol L À1 to 10 À10 mol L À1 , fewer dye molecules contribute to the emission and the uorescence intensity linearly decreases ( Fig. SI4A and B †). However, it is almost quenched in silver-boron colloidal solutions aer incubation with 10 À6 mol L À1 Rh6G (Fig. SI4C †). Random aggregates of silver-boron NPs quench the uorescence by one order of magnitude stronger than monodisperse NPs due to the higher contact surface area. Less effective quenchers of uorescence are silver-boron NPs that are coated with polyethylene glycol (Fig. SI4D †). Silver-PEG8 pearl-necklaces appear to decrease the dye uorescence only by ten times, that is two Fig. 2 UV-vis absorbance spectra of silver colloidal solution (1 Â 10 À3 mol L À1 silver nitrate with 1 Â 10 À3 mol L À1 sodium borohydride) coated by polyethylene glycol of 8000 Da (silver-PEG8 NPs) before and after heating at 37 C or 70 C. The two insets show a TEM image (scale bar is 200 nm) and a snapshot of silver-PEG8 NPs after heating at 70 C. The other two TEM images (scale bar is 50 nm) on the right show silver-PEG8 NPs compared with silver-PEG1 NPs (polyethylene glycol of 1000 Da). Silver-PEG8 NPs form a necklace structure, while PEG1 is inefficient against aggregation of nanoparticles.
times lower than that of merged silver-PEG1 NPs. Speculatively, this might have several possible reasons.
(a) Silver-PEG1 NPs have a large surface contact area and a thinner polymeric protective layer (eight times less molecular weight than PEG8); (b) silver-PEG8 NPs form plasmonic hot spots that might serve as secondary sources of uorescence excitation.
To increase the Raman cross-section, Rh6G aqueous solution was selected at a concentration of 10 À8 mol L À1 in order to ensure the necessary low threshold surface enhancement (10 6 ) that is accompanied by a sufficient suppression of uorescence. 40 The contribution of water can be neglected, as the Raman scattering from >10 5 molecules increases by 300-1500 fold. SERS for water is based exclusively on the EM eld increase, so that there is no charge-transfer induced surface enhancement. The effective number of the adsorbed dye is about one or two molecules per single silver NP with an average diameter of 22 nm and about four dye molecules per 30 nm NP. Detailed calculation can be found in the Experimental section.
The SERS spectra of bare silver-boron NPs or those coated by PEG8 with adsorbed Rh6G (10 À8 mol L À1 in water) are shown in Fig. 3. In the frequency region below 300 cm À1 small peaks are due to vibrations which involve chemical bonds of silver atoms or aggregates. The Ag-Cl silver-halide vibration occurs at 246 cm À1 and is not inuenced by the cations. 41 At a sufficiently negative surface potential the surface Ag-Cl vibration disappears due to the forced desorption of Cl À ions.
In Fig. 3A the N-H stretch (plane ring deformation) appears near 610 cm À1 and 774 cm À1 (out-of-plane bending motion of the hydrogen atoms of the xanthene skeleton 42 ). On bare silverboron aggregates the absolute intensity of the 774 cm À1 band is higher due to the conformational distortion by chemisorption of the molecule by the well-known vibronic coupling. 43 Moreover, this line is broadened due to the more than one vibration that contributes to this band. Similarly, the C-H in-plane bending coupled to the C-C stretching near 1181 cm À1 is stronger in the aggregates of silver-boron NPs and is in good agreement with the literature. 39 In addition, prominent peaks at 1308 cm À1 and 1360 cm À1 from the aromatic C-H stretching might be due to the stronger extinction of aggregated silver NPs (Fig. 1A).
Aer heating silver-PEG8 NPs preserve their stability, as agglomerates contribute strongly to the elastic Rayleigh scattering, and Raman signals are not superimposed (Fig. 3B). In contrast to bare silver-boron NPs, the N-H stretch vibrations appear with a twofold lower relative intensity with a red shi near 613 cm À1 and 775 cm À1 (or in the region from 774-777 cm À1 ). This indicates the inhibition of the charge transfer by polyethylene glycol, i.e. weaker chemisorption of the dye that is more pronounced at 37 C.
In a similar way the carbon-hydrogen vibration is red shied near 1191 cm À1 at 70 C and 1199 cm À1 at 37 C (Fig. 3B). One of the possible reasons could be the stretching ability of polyethylene glycol around silver NPs in water that might be higher at 37 C. Induced desorption of the C-H aromatic stretching is reected in red shied frequencies near 1312 cm À1 and 1364 cm À1 with relative intensities twice smaller than on bare silver-boron NPs (Fig. 3A).
Because of its high sensitivity to molecular distortion the vibrational mode of the carbon skeleton appears on bare silverboron aggregates as a prominent peak near 1509 cm À1 (Fig. 3A), but it is blue shied near 1502 cm À1 with PEG8 (Fig. 3B). This indicates that a stronger coupling between the anion and the Ag-Rh6G bonds is achieved on coated silver NPs. However, the C-C aromatic ring stretching near 1560 cm À1 (Fig. 3A) is strongly red shied to 1574 cm À1 on silver-PEG8 NPs, and the band at 1650 cm À1 is absent.
Overall, on silver NPs the most intense ring and trigonal "breathing" of pyridine at 992 cm À1 and at 1030 cm À1 appear only as small peaks near 1001 cm À1 and 1037 cm À1 (Fig. 3A). The characteristic ring vibrations involving considerable N-O stretching at 1252 cm À1 , 835 cm À1 and 541 cm À1 are not observed. Moreover, the pyridinium halide C 5 H 5 NH + Cl À that is located at 2380 cm À1 does not appear as well as the O-H vibrations of water near 2500-2900 cm À1 associated with the EM enhancement. However, on silver-PEG8 NPs the band around 2850 cm À1 appears as a small broad peak, indicating that polymeric matrix promotes the EM enhancement due to the hot spots. To remind, the SERS for water is based exclusively on the EM enhancement so that there is no charge transfer induced surface enhancement.

(C) Laser heating effect versus concentration and SERS factor of silver-PEG NPs
For a meaningful SERS study of live cells reported later the drastic heating effect of silver NPs must be eliminated as a potential drawback because of spectral uctuations. The local heating of an aqueous drop with silver nanoparticles with 2.55 mm radius due to the laser excitation (l exc ¼ 785 nm, power < 6 mW) is less than DT ¼ 3 K assuming a laser beam spot with the full width at half maximum (FWHM) of $1 mm (Fig. 4). However, the temperature rise depends on the concentration of silver NPs and the laser power.
As the typical concentration of silver NPs in a drop is in the order 10 16 per mL the temperature rises up to <0.1 K at 7.9 kW Fig. 3 (A) Surface enhanced Raman scattering (SERS) spectra of silver-boron nanoparticles incubated with 10 À10 mol L À1 aqueous rhodamine 6G solution compared to a normal Raman spectrum of bulk rhodamine 6G. (B) SERS spectra of a silver-PEG8 colloidal solution after heating at 37 C or 70 C. The spectra were collected with 785 nm excitation wavelength at a laser power of 10 mW and an integration time of 1 s with a spectral grating 300 gr mm À1 blazed at 750 nm. cm À2 (the lowest value to collect SERS spectra) to <1 K at 23.8-47.6 kW cm À2 . But the heating can reach several K at 158 kW cm À2 (the highest value for well-resolved SERS lines) comparable to that reported for gold nanoparticles. 42 The critical concentration of silver NPs, for which the temperature can reach 10-100 K, is in the order of 10 18 to 10 20 per mL, and for this SERS spectra exhibit a strongly increased underlying continuum with one or two broad merged spectral lines.
To understand the interaction of the incident near IR light with the silver nanoparticle surface, the 2D far-eld scattering diagram and a 3D plot of electromagnetic power loss density (or heat dissipation density) over the volume of an individual silver NP (average diameter 22 nm) are theoretically modelled (Fig. 5). In this model, the scattering of a plane wave with l exc ¼ 785 nm is computed for the optical frequency range (7.52 Â 10 14 Hz to 4.24 Â 10 14 Hz) in which silver as a material has negative complex permittivity. 9 Due to the symmetry of this model, only one quarter of the sphere is modelled. A region around the silver nanosphere is taken to be half (or 1/20) the wavelength in free space with an outside domain as an absorber of the scattered eld that is within the reactive near-eld of the scattering sphere (half-wavelength).
The far-eld scattering pattern shows the E-eld-plane (black) and the H-eld-plane (green) with a form of a dipole antenna (Fig. 5A). The scattered light has an E-eld parallel and an H-eld perpendicular to the plane of this pattern. To remind, the angular distribution of the scattered light is determined by the angle with the main polarizability axis. In this way, the scattered intensity is zero in the direction of this vector (forward scattering) and can reach 180 in the backward scattering with the E-far-eld values of about 2 Â 10 À8 V m À1 . It is noticeable that this value is slightly larger for 100 nm NPs being about 40 Â 10 À8 V m À1 (Fig. SI5A †).
The electromagnetic power density loss (P V , W cm À3 ) or resistive losses of total dissipation density over a silver nanosphere (22 nm) are shown as a 3D plot in Fig. 5B. To note, by the heat losses we derive how many electrons are passing through a silver NP, multiplied by the amount of energy each electron loses in the form of heat as it goes, giving an overall rate of heat production. In this way, the resistive heating power is directly proportional to the square of current multiplied by the resistance of silver. The total power dissipation density over silver nanoparticle volume is dened as where f is the frequency of the incident wave (Hz, near IR), 3 0 is the vacuum permittivity, 3 00 is the imaginary part of the relative permittivity of silver (dissipation factor) and E is the electric eld strength (V m À1 ). The bright areas show the resonance, where most of the losses take place over the volume of the NP (Fig. 5B). The highest value of this resonance appears as a red spot on the NP and has a value of 4413 W m À3 . The xyz shows the space distribution of the energy source coming from heat dissipation, i.e. Q(r,t), which is dened as Q(r,t) ¼ hj(r,t)E(r,t)i t where j(r,t) is the current density (A m À2 ) and E(r,t) is the stimulating electric eld calculated from a system of Maxwell's equations. An analytical solution of the heat dissipation is introduced as where u is the frequency of the incident near IR light, E 0 is the electric eld amplitude of the incident light (near IR, 785 nm), 3 0 and 3 NP are the dielectric constants of the medium and nanoparticle, and 3 00 is the imaginary part of the dielectric function of silver. The P V value of an individual silver NP nonlinearly drops from 4.4 mW cm À3 to 0.53 mW cm À3 with the increase of particle size from 22 nm to 100 nm (Fig. 6). The polynomial t (red) of the fourth order ts yielded the P V values rather precise (black) (Fig. 6). The detailed investigation in dependence of the NP concentration (including other conditions) and at different parameters of incident light is our ongoing research activity.
In this work the SERS spectra are collected at low power density so that the dynamic uctuations of particles in the surrounding water are mostly due to the Brownian motion with a negligible laser heating. The number of silver nanoparticles  per volume does not exceed the critical value and is in the order of 10 16 per mL. Water is a poor Raman scatterer and its maximum scattering intensity depends on the co-adsorption with the Cl À ions at a relatively high salt concentration (four orders of magnitudes higher than we use here).
The calculated SERS enhancement factors (EFs) are listed in Table 1. The classications of the main results are 'VS' (very strong) with EF $ 10 9 , 'M' (medium) with EF $ 10 7 and 'W' (weak) with EF $ 10 6 . The bare silver-boron NPs exhibit SERS signals of Rh6G with moderate enhancement comparable to those reported before. However, silver hot spots protected by polyethylene glycol with a longer chain length (PEG8) increase the SERS EF up to $10 9 . This value is higher than from random aggregates of silver-citrate NPs 4 and comparable with the hot spots on colloidal graphene oxide. 44 However, the SERS enhancement is still moderate on silver NPs in the polyethylene glycol matrix with shorter polymeric chains (PEG1).
Such high EF values with PEG8 can be achieved at the stable interparticle junctions of several nm distance, so-called 'hot spots' for local EM increase (Fig. 2). The controlled necklace nanoparticle arrangement and SERS activity are stable under physiological conditions, i.e. 37 C in aqueous medium of live broblasts. However, under equimolar conditions shorter PEG chains might not entirely coat the silver NPs due to their less coiled structure. At a temperature above room temperature, e.g. 37 C or even higher, at 70 C, silver-PEG1 NPs form random aggregates that merge at contact (Fig. SI1 †).

(D) SERS molecular mapping of live broblasts
As we have shown that we can achieve unprecedented enhancement of Raman signals, it appears promising to advance this technique for intracellular analysis. Fig. 7A shows optical microscopy phase contrast images of live NIH/3T3 broblasts with incubated silver-PEG8 NPs (the inset shows the image of the control cells without NPs). For SERS measurements the culture medium was removed and the Petri dish was rinsed three times with PBS solution.
Silver-PEG8 NPs were embedded into the cell interior by incubation 45,46 or by electro-permeabilization based on electroporation 47 of the cell membrane. Both methods are successful and no visible changes of the cell damage or apoptosis were observed aer overnight incubation. However, by the electropermeabilization approach NPs are more homogeneously distributed mostly in the area close to or surrounding the nucleus leaving the spread tails almost untouched. By the cellular uptake during incubation most NPs were observed throughout the cellular body. About 99% of the control cells and 95% of those with embedded NPs were stretched on the quartz plate of the Petri dish without visible signs of cell damage or apoptosis.
Silver-PEG8 NPs aer heating at 37 C survived the overnight incubation inside the cells. Strong SERS spectra could be collected from 15 cells each in two regions: I (close to the nucleus) and II (throughout the intracellular cytosol) (Fig. 7B).
The SERS bands from I and II are considerably different. Overall, the absolute intensities of the SERS peaks are higher in II region excluding the 1390 cm À1 band. This band was observed in I region and appears due to the DNA/RNA aromatic ring vibrations. 48 Low frequency SERS peaks (<605 cm À1 ) appear with higher intensity in the II region than in the I region ( Table  2). They are due to the presence of sugar molecules, proteins (S-S stretching disulde at 502 cm À1 ) and a minor component in the cytosolic side of the cell membrane (phosphatidylinositol at 602/605 cm À1 ). 50 Fig. 6 Dependence of the total power dissipation density, P V , at resonance over the volume of individual silver NPs on their average diameters from 22 nm to 100 nm. The P V values (black) are fitted to a polynomial fit (red) of the fourth order (y ¼ 0.6 + 4 Â 10 À2 x À 1 Â 10 3 x 2 + 1.5 Â 10 À5 x 3 À 6.6 Â 10 8 x 4 with adj. R 2 ¼ 0.99). The excitation wavelength is 785 nm.  Strong SERS peaks at 711 cm À1 and 857 cm À1 arise from phospholipids 50 and tyrosine 51 ( Table 2). The symmetric ring breathing of phenylalanine 52 with a twice increased intensity was detected at the higher frequency of 1006 cm À1 in II (at 1002 cm À1 close to the nucleus). Additional peaks from the DNA-backbone (DNA: P-O-P) 51 and the most prominent aromatic ring vibrations of nucleic acid DNA/RNA macromolecules 48 appear close to the nucleus. In the II region the DNAbackbone (DNA: O-P-O) band appearing at 1075 cm À1 and aromatic ring vibrations of DNA (1390 cm À1 ) with the guanine vibration as a shoulder at 1324 cm À1 can arise from mitochondria or the mitochondrial membrane. [51][52][53] In both I and II the deformation of hydrocarbon chains appeared at 1448 cm À1 with a series of small peaks due to the amide I vibration (1580-1700 cm À1 ). 51,54 Overall, the collected SERS signals are much stronger and the spectra are more informative than those reported in the literature. 49,50,54,55

Conclusions
The silver NPs (22 nm) prepared by the reduction of sodium borohydride can assemble into a necklace nanostructure when coated with polyethylene glycol of 8000 Da. The necklace silver-PEG8 NPs can be stable at 37 C (or 70 C) necessary for live cell studies by Raman spectroscopy. Such silver NPs can enhance Raman signals of rhodamine 6G up to 10 9 fold due to the local electromagnetic eld enhancement at the interparticle junctions.
By varying the chain length of polyethylene glycol the neighboring silver-boron NPs can be brought into contact (merge with PEG 1000 Da) or kept stably apart at a distance of several nm (with PEG 8000 Da). The polymeric coiled structure can be stretched in a controlled way at 37 C (or 70 C) and the chemisorption of dyes can be decreased, the charge transfer thus weakened, leaving only the electromagnetic part for the Raman scattering enhancement. Inside live NIH/3T3 broblasts the silver-PEG8 NPs can strongly enhance the characteristic molecular vibrations of the DNA backbone, DNA/RNA aromatic ring vibration, stretching in tyrosine, polypeptides, sugar molecules (amylose), and many others.
Our preliminary results introduce a novel analytical approach towards the engineering of effective SERS substrates with a selectable type of Raman scattering enhancement (electromagnetic or charge transfer). It can be used as an informative analytical tool to examine the mechanisms of interest inside live cells in a systematic manner.

Materials
Silver nitrate (AgNO 3 , analytical grade, 99.8%) was purchased from Serva (Germany). Sodium borohydride (NaBH 4 , 98%) and Symmetric stretch vibration of choline group, characteristic for phospholipids 857 On-plane ring breathing mode in tyrosine C-C 943 C-C skeletal stretch in protein 1002 Symmetric Amide I vibration polyethylene glycol (PEG-1, biology grade, MW z 1000 Da) were produced by Alfa Aesar (Germany). Polyethylene glycol (PEG-8, biology grade, MW z 8000 Da) was purchased from Aldrich. Sodium chloride (NaCl, $99.5%, biology grade, suitable for the cell culture medium) was produced by Sigma (Germany). DMEM, gentamicin, and glucose were purchased from Sigma. Calf serum was purchased from PAA GmbH (Austria). The water used in all experiments was prepared in a threestage Millipore Milli-Q Plus 185 purication system and had a resistivity higher than 18.2 MU Â cm À1 . For the experiments with living cells Milli-Q water was autoclaved.

Preparation of silver nanoparticles
Prior to hot spot preparation we formed silver NPs by reduction with sodium borohydride. Small silver nanoparticles (20-30 nm) have been produced by Creighton's procedure, 29 while larger particles (60-100 nm) have been formed by a modied Schneider approach. 30 This is explained by the relatively high reactivity of borohydride (as compared with citrate 31,32 and carbohydrates), its handiness (as compared with gaseous hydrogen and physical methods) and not too high toxicity (as opposed to hydrazine and hydroxylamine). 29 Silver colloidal solution was prepared by chemical reduction of silver nitrate using sodium borohydride as a reducing agent in aqueous solution without organic stabilizers. 24 The process was carried out in a 0.25 L Erlenmeyer ask prewashed in concentrated nitric acid. The remains of the acid were removed from the glass walls by abundant amounts of deionized water. A 1 Â 10 À3 mol L À1 concentration of AgNO 3 solution (room temperature) was mixed with fresh, ice-cold 1 Â 10 À3 mol L À1 or 2 Â 10 À3 mol L À1 sodium borohydride aqueous solution under vigorous stirring (300 or 600 rpm) at a nitrogen atmosphere. In the rst 20 s, the mixture turned bright yellow. Aer complete injection (less than 2 min) of silver nitrate solution, stirring was stopped immediately. The nal solution of the mixtures changed to brownish-grey (100 nm nanoparticles) or clear yellow color (20-30 nm nanoparticles).

Preparation of polyelectrolyte solutions
Polymer aqueous solutions were prepared with a typical concentration of 2 mg mL À1 followed by purication in a sealed semipermeable membrane (cellulose acetate) against MQ water (MW cutoff 20 000 Da) and lyophilized. PEG8 (pH ¼ 6.5) and PEG1 (pH ¼ 6.4) aqueous solutions were obtained under vigorous stirring at room temperature.

Treatment of silver nanoparticles with polyethylene glycol
For every treatment freshly prepared silver colloidal solution was used. To stabilize silver colloidal solution against unavoidable aggregation at 37 C, silver sols were coated by polyethylene glycol at a volume ratio of 1 : 2. The mixture was incubated during mild shaking in the Eppendorf Thermomixer compact (Hamburg, Germany) overnight, followed by triple washing with MQ water and centrifugation at 14 000 rpm at 4 C.
To test the colloidal stability, the "poly-silver" mixture was incubated at 37 C or 70 C for 60 min and allowed to cool to room temperature before characterization. The color of silver sols with polyethylene glycol was clear yellow without sedimentation at the bottom. The average pH value of silver sols with polyethylene glycol (1000 Da) was between 8 and 9 and for those with a longer chain length (8000 Da) around 9.

Preparation of rhodamine 6G-silver aqueous solutions for Raman measurements
The stock rhodamine 6G solution was prepared at a concentration of 10 À3 mol L À1 in MQ water (pH ¼ 8.2). Lower concentrations from 10 À4 mol L À1 to 10 À10 mol L À1 were obtained by successive dilution of the stock solution by factors of 10 and 100. The dye concentration was monitored via the uorescence intensity with a calibration curve. For Raman measurements fresh silver colloidal solution was added to a 10 À4 mol L À1 NaCl solution at a volume ratio 1 : 5 of salt to silver followed by the addition of fresh 1 mL rhodamine 6G solution. At electrolyte concentrations higher than 10 À3 M no reliable data could be obtained due to instability of the silver sol. The mixture was incubated for at least 3 hours at room temperature in darkness before the Raman experiment.
The concentrations of bulk rhodamine 6G solutions were from 10 À6 M to 10 À10 M, as at higher concentrations (10 À3 M to 10 À5 M) the uorescence is quenched due to the dimerization of dye molecules (i.e. methyl substituents of rhodamine 6G). The uorescence intensity of bulk rhodamine 6G at 10 À6 M to 10 À10 M concentration linearly decreased with the decreased concentration of dye molecules (Fig. SI3B †).

Intensity measurements of rhodamine 6G-silver solutions at very low concentrations
A repeated chemical cleaning with concentrated HNO 3 /HCl or concentrated H 2 O 2 /NH 3 was sufficient to remove adsorbed rhodamine 6G molecules in the dye solutions with concentrations from 10 À8 M to 10 À10 M. However, the cuvette required overnight boiling in order to remove dye molecules aer the measurements at higher concentrations (10 À6 to 10 À7 M). To get stable and reproducible signals and to avoid unnecessary adsorption and desorption processes at the cuvette walls the clean cuvettes were rinsed several times with portions of the sample solution. This procedure was repeated until equilibrium between the adsorbed and dissolved molecules was established. This was controlled by the uorescence intensity. In this way a linear relationship between concentration and uorescence intensity could be established in the range between 10 À6 M and 10 À10 M. Reliable data were obtained for each concentration of rhodamine 6G solution by several measurements from independent samples.

Cell culture, co-incubation and electroporation of cells with colloidal silver
Fibroblasts NIH/3T3 (purchased from DMSZ) were cultured in Dulbecco's modied Eagle's medium (DMEM) supplemented with 4.5 g L À1 glucose, 10 v% calf serum, and 10 À2 g L À1 gentamicin (antibiotic). Cells were seeded with 6 Â 10 3 cells cm À2 on the glass bottom dish (purchased from Greiner Bio For One) 8.8 cm 2 culture surface and incubated in an incubator (binder) with 5% CO 2 at 37 C overnight before Raman measurements.
Silver sols were delivered into the cellular interior by internalization with endocytosis and transported into the late endosomes and lysosomes via overnight incubation of 200 Â 10 À6 L (1.3 Â 10 5 cells) in a culture medium suspension with 400 Â 10 À6 L particle solution. Alternatively the delivery of silver nanoparticles was achieved by electro-permeabilization of the cell membranes in a BIO-RAD gene-pulser cuvette with 0.4 cm electrode gap inserted inside a home-made electric cell connected to the GHT-Bi500 electroporator (btech, France). 200 Â 10 À6 L (5.34 Â 10 4 cells) in a culture medium were mixed with 400 Â 10 À6 L of silver colloidal solution at a sequence of pulse voltages of 250 V for the positive and negative parts with a pulse length 500 ms within the sequence. Aer the co-incubation and electroporation the colloidal mixture with the cell suspension was mixed with 1 Â 10 À3 L of the culture medium and deposited onto the glass bottom of a Petri dish in the incubator. The inspection of the incubated cells was checked by optical microscopy aer 12 hours of aging.

Characterization
A. UV-vis absorption, zeta potential, dynamic light scattering, electron/optical microscopy and uorescence. To characterize the size distribution of silver NPs a Zeiss EM 912 Omega transmission electron microscope (TEM) and a high-performance particle sizer (Malvern Instruments) for dynamic lightscattering (DLS) measurements were employed. For TEM analysis a drop of the colloidal solution was applied onto the copper grids coated with a carbon lm and le to evaporate.
Surface plasmon resonance absorption and concentration of silver nanoparticles were monitored using a Varian CARY50 Conc UV-vis spectrophotometer in the wavelength range from 200 to 900 nm in a quartz cell with 10 mm path length. The z-potential of silver colloidal solutions was measured using a zeta sizer (Malvern Instruments). The uorescence measurements were performed using a FluoroMax-4 spectrouorometer (Horiba Jobin Yvon). A Nikon Eclipse TS100 routine inverted microscope was employed to collect the refracted light from the transparent live cells through the enhancement of their phase contrast. Observation of the morphology of the cell was conducted using a Leica TCS SP inverted confocal microscope system (Leica, Germany) in transmission mode equipped with a 40Â oil immersion objective having 1.25 numerical aperture and 0.1 mm working distance.
B. Calculation of the local temperature gradient dependence on the effective number of silver nanoparticles (plot in Fig. 4). We use the model that accounts for the inuence of absorption, size and concentration of nanoparticles in order to understand the temperature gradient increase around a heating sphere of silver nanoparticles. 56,57 For a colloidal drop silver nanoparticle is considered as a heating center with an average radius r 0 and the lling factor F s for a complete coverage of a drop with metallic nanoparticles (surface lling factor). The temperature gradient can be calculated as follows where r 0 is the radius of silver nanoparticles, n is the number of silver nanoparticles per droplet S i , S c is the surface area of a single drop and R 0 is the radius of a drop (2.55 mm in our experiments). Two surface lling factors are considered, F s 1 (small nanoparticles, tens of nm) and F s 2 (large nanoparticles, hundreds of nm). The absorbed energy, E, can be estimated as where A is the heating rate per unit volume per unit time and K is the thermal conductivity. Assuming that the surrounding medium is relatively unchanged, where a 1 and a 2 are absorption coefficients for small and large nanoparticles. The laser power density values were calculated by dividing the applied actual laser power over the surface area of a drop of the colloidal solution. The laser beam width is around one micron. C. Raman spectroscopy and microscopy. Raman and surface enhanced Raman scattering (SERS) spectra from the silver colloidal solutions and live cells were collected using a confocal Raman microscope (CRM200, WITec, Ulm, Germany) equipped with a piezo-scanner (P-500, Physik Instrumente, Karlsruhe, Germany) at a 785 nm excitation wavelength (Toptica Photonics AG, Graefelng, Germany). A linearly polarized diode laser beam was focused through the LWD 20Â, Nikon Fluor 60Â water immersion and Nikon 100Â oil immersion objectives with numerical apertures NA ¼ 0.40, 1.00 and 1.25, respectively.
The laser power at the silver colloidal solution was kept no higher than 10 mW, and at the live cell it did not exceed 2 mW as measured using a Newport optical power meter 1830-C. The scattered light was ltered with an analyzer (further polarizer) through the confocal microscope pinhole.
The spectra were collected with a 300 gr mm À1 grating blazed at 750 nm and recorded using a spectrograph (Acton, Princeton Instruments Inc., Trenton, NJ, USA) with a cooled CCD detector (PI-MAX, Princeton Instruments Inc., Trenton, NJ, USA) with an integration time 1 s of y accumulations. The signal to noise ratio was high enough to ensure the low value of the root mean square uctuations averaged in time (typically below 5%). Basic alignment was carried out by the Raman spectrum of a silicon wafer with a characteristic Si line at 520 cm À1 during integration times from 0.3 to 1 s.
The SERS measurements of live cells were conducted in vivo in a m-dish (35 mm, Ibidi, Munich, Germany) equipped with a heating stage and an external temperature probe (Bioscience Tools, USA) to maintain 37 C of the cells. The acquired Raman and SERS spectra were corrected for the baseline, background of the m-dish substrate and the SiO 2 485 cm À1 band.
Estimation of the SERS enhancement factor where EF is the average enhancement factor (averaged over all possible positions on the metallic surface and also from molecules randomly adsorbed on the surface as compared to the same number of non-adsorbed molecules), C norm and C SERS are concentrations of rhodamine 6G solutions for SERS and normal Raman measurements (i.e. number of molecules on the surface of a single nanoparticle and in bulk solution effectively excited by the laser beam); I SERS and I norm are corresponding intensities of the vibrational modes. The Raman peak of rhodamine at 1365 cm À1 was taken for the enhancement factor estimation. The molecular density of rhodamine 6G adsorbed on the silver nanoparticle surface depends on the average diameter of the particle. For example, let us assume a silver colloidal solution at 1.4 mmol L À1 concentration in a volume 1.2 mL incubated with 10 À8 mol L À1 rhodamine 6G solution.
where n Ag is the mole number of silver, C Ag is the concentration of silver nanoparticles (from UV-vis absorbance as estimated above) and V is the volume. n Ag ¼ 1.68 10 À6 mol.
where n Ag is the molecular density of silver, N Ag is the number of silver atoms in the volume, N A is Avogadro's number (6.02 Â 10 23 mol À1 ). The total average density of silver is 8.42 Â 10 20 L À1 and of rhodamine 6G molecules is 6.02 Â 10 15 L À1 . Assuming a covalent radius of silver of 145 AE 5 Â 10 À12 m and an average nanoparticle diameter of 30 Â 10 À9 m, the volume of a single nanoparticle is 1.41 Â 10 À23 m 3 and of an atom is 1.27 Â 10 À29 m 3 , so that one single silver nanoparticle consists of 1 Â 10 6 silver atoms. The silver nanoparticle density is 7 Â 10 14 L À1 and when it is divided by dye density, we have about 7 rhodamine 6G molecules adsorbed on one single silver nanoparticle. For bare silver-boron nanoparticles with diameters 22 nm and 100 nm, there are about 3 and 301 rhodamine 6G molecules per single silver nanoparticle, respectively. For silver-PEG nanoparticles with diameters 22 nm and 30 nm (average concentration is about 2 Â 10 À3 mol L À1 ) there are about one/ two and four rhodamine 6G molecules per single nanoparticle, respectively.
Theoretical modelling of the far-eld scattering pattern and the total power of heat density dissipation (P V ) by a Comsol Multiphysics soware tool (a) The scattering of a plane wave of light off of a silver nanosphere. In this calculation the far-eld scattering pattern of a plane wave of light with 785 nm wavelength is computed over a silver nanosphere of about 22 nm average diameter. Due to the symmetry of the nanosphere only one quarter of it is modelled. The wave vector, k, and the electric eld vector, E, are perpendicular to each other as indicated by arrows in Fig. 5B. The E vector is also perpendicular to the surface of the silver nanosphere.
A region of water around the sphere is also modelled. A perfectly matched layer (PML) domain is outside the water domain and acts as an absorber of the scattered eld. The PML is not within the reactive near eld of the scatterer and it is a half wavelength away. The far-eld radiation pattern is plotted in Fig. 5A and shows an electric eld, E-plane (black), and a magnetic eld, H-plane (green).
For the scattered eld, the plane wave travels in the positive x direction, with the electric eld (E-eld) polarized along the z-axis. The default boundary condition is a perfect electric conductor, which applies to all exterior boundaries including the boundaries perpendicular to the background E-eld polarization.
(b) The total power of heat density dissipation (P V , W m À3 ). This model denes the interaction of the incident electromagnetic plane wave of the wavelength 7.85 Â 10 À7 m with the silver sphere of 2.2 Â 10 À8 m radius by using specic domain properties described by the classical Maxwell's laws. It also uses the perfectly matched layer that has the equal size of the thickness of a water layer around the modelled silver sphere (3.92 Â 10 À7 m).
The electric eld strength (V m À1 ) with xyz spatial components has a shape function of quadratic curl E derived from the classical Maxwell's law as where H is the magnetic eld strength (T).
In the frequency domain of electromagnetic wave equation two interpolation functions are used. The rst one includes the real part of permittivity (3 real) and the second one uses its imaginary part (3 imaginary). The table below shows the calculated permittivity values of silver in the range from 397 nm to 705 nm that are used for our model.
Rumiana Dimova (MPI Theory). The help with the cell culture assistance by Christine Pilz (MPI Biomaterials) is greatly appreciated. The stipend from the Max-Planck society is acknowledged.