nnano_2019_supplementary_info.pdf

Full text

Articles
https://doi.org/10.1038/s41565-019-0535-6
High-speed imaging of surface-enhanced
Raman scattering fluctuations from individual
nanoparticles
Nathan C. Lindquist1,4, Carlos Diego L. de Albuquerque 2,3,4, Regivaldo G. Sobral-Filho2,3, Irina Paci2,3
and Alexandre G. Brolo 2,3*
1Department of Physics and Engineering, Bethel University, St Paul, MN, USA. 2Department of Chemistry, University of Victoria, Victoria, British Columbia,
Canada. 3Centre for Advanced Materials and Related Technologies (CAMTEC), University of Victoria, Victoria, British Columbia, Canada. 4These authors
contributed equally: Nathan C. Lindquist, Carlos Diego L. de Albuquerque. *e-mail: agbrolo@uvic.ca
SUPPLEMENTARY INFORMATION
In the format provided by the authors and unedited.
NATURE NANOTECHNOLOGY | www.nature.com/naturenanotechnology

S2
Supplementary Note 1. Tracking SERS intensity fluctuations (SIF) using the
Airyscan detector. While standard CCD and CMOS cameras aren’t able to provide
kHz frame rates, photomultiplier tubes (PMTs) are capable of very fast ~ns response
times and MHz acquisition rates while still being sensitive to single photons. Recently,
Zeiss Microscopy has shown a new approach for confocal laser scanning microscopies
(CLSM) with an “Airyscan” detector.1, 2 The new configuration is based on an array of
32 GaAsP PMTs as shown in Supplementary Fig. 1a. The Airyscan detector covers 1.25
Airy Units (AU) and each detector element is about 0.2 AU or 50 nm wide. This array
replaces the typical pinhole and point detector in a confocal microscope and collects an
image of the point-spread function (PSF). Combining information from all 32 detectors
elements allows the capture of sub-diffraction limited images with high sensitivity.1, 2
Supplementary Figure 1. The Airyscan detector. (a) The 32 PMT detectors are
arranged in a hexagonal array in an image plane of the microscope, covering roughly
1.25 AU. This translates to each detector element covering about 50 nm in sample
space. The nanoparticle samples were therefore small enough to be entirely within the
Airy spot and couldn’t be imaged directly. (b) The PSF of the microscope showing an
Airy spot. (c) Each channel collects a high-speed signal from the sample. Channel 1 is
the most intense since it is in the center, and can be used on its own as a simulated
confocal pinhole, but all channels together are combined to image the single

S3
nanoparticle. The exact centroid of the imaged spot can then be found by fitting with a
Gaussian function.
Supplementary Fig. 1a shows the projection of the light in the Airyscan detector. When
the beam is aligned, the maximum amount of light will be concentrated in the center
(channel 1) and it decreases from the middle, as can be observed in a simulated PSF in
Supplementary Fig. 1b. This feature implies that the signal will be stronger for channel
1 and very low in channel 32 (Figure S1c). In our experiments, the Airyscan detector
was used to track the trajectory of SERS intensity fluctuations (SIF) in a single
nanoshell. Notice that we first confirm that we have individual particles by collecting
hyperspectral dark field images of the particles immobilized on glass. Moreover a
regular SERS image is also obtained in the laser scanning microscope using a bandpass
Stokes filter to identify potential aggregates before collecting data from a single
particle. Considering the 543 nm laser, according to Abbe’s law, the lateral resolution
allowed is ~λ/(2NA), where NA = 1.4 is the numerical aperture of the microscope
objective. This corresponds to roughly ~200 nm and is larger than the size of a single
Si@Ag nanoshell (~110 nm). Illuminating this particle with a focused laser and
collecting all 32 channels very quickly (temporal resolution of 1.23 us), the SERS signal
will be projected in the Airyscan detector and SIF images could be recorded with
“pixel” resolution of ~50 nm (Supplementary Fig. 1d). Fitting this with a 2D Gaussian
function gives the location of the SERS centroid to within ~7 nm. The stochastic SERS
behavior can then be use to achieve super-resolution images3, 4, 5 as shown in the main
text. The fit precision is determined by the standard deviation of the fit location over a
single SIF event and is equivalent to the imaging resolution since each fit is separated in
time and is unique to a single event. In this way we can obtain ~7 nm resolution SERS
images at 800,000 fps. Figure 2d in the manuscript shows the standard deviation of the
fit location for several SIF events. Two other SIF events are also plotted, located <10
nm away. Of course, very weak SIFs with low signal to noise will have relatively poorer
fit precision and will provide lower resolution images. Because of this, a threshold was
chosen as described below for some of the data and only the strongest SIF events
formed the final images. Finally, while mechanical drift was typically minimal over the
course of the ~seconds long traces, any linear drift in the average of all the fit position
was subtracted away.

S4
Supplementary Note 2. Preprocessing the data using the CUSUM method.
Supplementary Fig. 2 shows a scheme for a typical data preprocessing using the
CUSUM algorithm6 before evaluating the time statistic of the SIFs. The signal obtained
from the 32 Airyscan detectors has a good S/N ratio (see Figure 2 in main text),
especially if all 32 detectors are averaged. Therefore, estimating a good threshold above
the noise that qualifies as a SIF event can easily be done as shown in Figure 3a in the
main text. However, when the S/N ratio is relatively poorer as from a single PMT
detector shown in Supplementary Fig. 2a, to properly digitize and extract the time
statistics it is crucial to remove the noise. The aim here was to use the CUSUM method6
to find when a signal in the noise fluctuates abruptly in time by looking for inflections
in the cumulative sum of the signal. A noisy signal with representative fluctuations is
shown in Supplementary Fig. 2b, along with the moving average (the baseline, shown
as a solid red line) and an upper and lower threshold. These thresholds are typically
chosen to be a few standard deviations above and below the average. The average is
removed from the signal and the cumulative sum is calculated. The local inflection
points in the cumulative sum are shown Supplementary Fig. 2c. Depending on the
strength (compared to the upper and lower thresholds) and sign (local minimum versus
local maximum) of these inflections, the signal is digitized to be ON or OFF. The
digitized is shown in Supplementary Fig. 2d. The ON / OFF statistics of the signal could
be calculated.

S5
Supplementary Figure 2. Signal digitization procedure. (a) A noisy signal. (b) Zoomed
in region of a the signal and some representative noisy fluctuations. The signal average
(solid red) and upper and lower thresholds (red dashed) are shown. (c) The cumulative
sum of the signal, minus the average, showing the inflection points. (d) The signal (SIF)
and its digitized version (d-SIF).
Supplementary Note 3. Nanoshells preparation, characterization and temperature
experiment. The nanoshells SiO2@Ag were prepared as reported in earlier
publications7, 8 with 90 nm SiO2 core and 15 nm Ag shell as shown in Figure 1c. Notice
that the surface of the nanoshells are very rough and those features should support
highly localized surface plasmons on the single nm scale.9, 10, 11 As previously reported7,
8, these nanoshells are very uniform in their spectral properties. The peak shown in the
Figure 1c in the main text represents the average of 100 nanoparticles and is without
shoulders, proving the high monodispersity of the nanoparticles and lack of
aggregations. That makes this a very suitable SERS substrate for single-nanoparticle
measurements. This is shown in Supplementary Fig. 3a. The core and shell sizes were
tuned according to our laser wavelengths: 543 and 633 nm, around 600 nm (Figure 1d).
After that, the nanoshells were incubated with 5,5’-dithiobis-(2-nitrobenzoic acid)

S6
(DTNB) molecules that break the S-S bond and bind to the silver surface by the thiol
terminal as a self-assembled monolayer (SAM) (Figure 1b in the main text). A SERS
spectrum can be then recorded to confirm the SAM formation over the surface (Figure
1b in the main text). Supplementary Fig. 3 shows the SERS spectrum with several
bands highlighted. The strong symmetric stretching of the nitro groups12, 13 at 1337 cm-1
(●) confirms that the molecule is perfectly binding to the nanoparticle. Other important
bands were also assigned: 847 cm-1 (▼) nitro scissoring vibration; 1059 cm-1 (■)
succinimidyl N-C-O; and 1558 cm-1 (▲) C-C stretching on aromatic ring12, 13. Since the
nanoshells are completely covered with the molecules, our experimental conditions are
favoring measurements of fluctuations in the SERS intensity due to surface
reconstruction. In other words, the hotspots are effectively finding the molecules rather
than the molecule finding the hotspot.
Supplementary Figure 3. Optical characterization. (a) Darkfield microscope image of
several isolated particles. (b) SERS spectrum of DTNB after the incubation with
SiO2@Ag. Spectrum obtained using a 50 × objective; the laser excitation and power
were 633 nm and ~100 µW; 10 s integration time and 1. The highlighted bands are at
847 cm-1 (▼), 1059 cm-1 (■), 1337 cm-1 (●) and 1558 cm-1 (▲). The spectrum of the
nanoparticles that were not incubated with DTNB (without DTNB) also present some
weak Raman background, due to stabilizing molecules that remained from the synthesis
procedure.

S7
Supplementary Figure 4: SIF trajectories in different spectral bands for excitation at
543 nm. (a) SIF trajectories for two spectral bands in the same particle, one band that
includes the SERS active region and one that doesn’t. The plot that includes the SERS
peaks (blue) shows significant activity whereas the second spectral region (orange) does
not. (b) Plot depicting the two spectral ranges for the panel (a) data. (c) Another set of
SIF trajectories for two different spectral bands, this time both regions containing SERS
active peaks. Both plots contain significant and synchronized activity. However, the
signal in one band is sometimes more or less intense than the signal in the other band,
indicating intensity fluctuations at different regions in the spectrum. This is a
characteristic of SM-SERS.14, 15, 16
Temperature experiments were carried out by mounting the glass slide with dried
nanoparticles on a temperature stage control LTS120 (from LINKAM Scientific) Peltier
system equipped with a supplied water circulator. The stage allows mounted a glass
slide of 40 x 40 mm which can be moved by 15 x 15 mm in X and Y directions and to
control the temperature from -40 to 120 °C. In our experiments, the temperatures chosen
were 0, 10, 20 and 50 °C, changed at a rate of 8 °C.min-1. Regarding local temperature
increases by laser illumination, metallic nanoshells are promising candidates for
photothermal cancer treatment57, 58, 59 and there are several temperature estimates in the
literature60, 61. For example, low power illumination of gold nanoshells at 4 W/cm2
power was clinically relevant in treating tumor cells in mice62, suggesting an increase in
local temperature of ~10° C to trigger cell hyperthermia63, while local temperature
increases of ~37 °C have also been reported64. In a different scenario, solar-powered

S8
steam production was observed through the use of gold nanoshells in suspension where
large amounts of energy are dissipated into the surrounding water to produce water
vapour in just a few seconds65. Since our laser is focused on a single nanoparticle,
corresponding to roughly ~1x105 W/cm2, and scaling by the relative thermal
conductivity of air (as is the case in our experiments) versus that of water, an increase of
~100° C seems feasible. Simulations on other nanoparticle shapes also show significant
temperature increases, this time ~35° C for 104 W/cm2 illumination in water66.
The enhancement factor (EF) for the SiO2@Ag nanoshells was estimated using finite
element numerical simulations done with COMSOLTM. These are shown in
Supplementary Fig. 4. Material constants for the silica core (n = 1.45), the glass
substrate (n = 1.52), and the silver shell (n = .06 + 3.5i) were taken from the COMSOL
database. Perfectly matched layers were used to isolate the nanoparticle and focused
light was incident from the substrate as in the experiments. The rough surface of the
SiO2@Ag nanoshell was generated randomly to qualitatively match that of the TEM
images. While these simulations are necessarily limited to larger-than-atomic-scale
dimensions with the smallest grid size being 1 nm, they still show areas on the rough
particle of large field enhancement. At 543 nm excitation, a typical field enhancement
on a rough protrusion was ~31 (field amplitude on the particle / incident field
amplitude). This corresponds to a SERS enhancement factor of ~314 as described in the
figure caption or roughly ~106. The average enhancement over the surface was ~104 and
is in line with previous research on metallic nanoshells.17 Note that the scale bar in
Supplementary Fig. 5 is saturated and that there are areas on the particle with very
localized field enhancements of 107 ~ 108.
Supplementary Figure 5. Finite element computer simulations to estimate the SERS
enhancement factor. (a) Another TEM image with more contrast to see the rough edges
of the SiO2@Ag nanoshells. (b) Cross-sectional map of the electric field magnitude

S9
enhancement excited at 543 nm. Light is incident from below as in the experiments. (c)
Another view of the same particle, this time showing the outer metallic surface. The
scale bar is saturated to an electric field magnitude enhancement of 31 but there are
areas on the particle that are enhanced further. Panels (b) and (c) share the same scale
bar.
Supplementary Figure 6. Further laser power and temperature experiments supporting
those shown in the main text. (a) A SIF trace as the laser power was ramped from 1%
power to 4% power over several seconds. The signal strength clearly increases as does
(b) the calculated instantaneous SIF rate. A doubling of the laser intensity from 2% to
4% increased the SIF rate from ~200 Hz to ~2000 Hz. The SIF rate is plotted with a
moving average filter to smooth out large fluctuations. (c) Traces of the instantaneous
SIF rate for two particles at two different temperatures again slightly smoothed. (d)
Histogram of the particle SIF rate data shown in (c). The higher temperature particle
had a higher sustained SIF rate with 6078 captured events. The lower temperature
particle had intermittent bursts of activity with only 638 captured events over the same
amount of time.
Supplementary Note 4. Molecular Dynamics Simulations. Time-scales for
adsorption, molecular migration and cluster reorganization were followed using the ab
initio Molecular Dynamics algorithms in Siesta 4.018. Simulations used a 1-fs time step,
and the temperature was ramped up and held at 300 or 400 K using a Nose thermostat.
Forces on atoms were calculated using SCF calculations with the Perdew-Burke-

S10
Erzenhof functional, with Tkachenko’s scaling implementation of Grimme’s D2
correction for metal-molecule interactions.19, 20, 21 A DZP basis set was used, with norm-
conserving Troullier-Martins pseudopotentials for core electrons, and a pseudo-atomic-
orbital (PAO) energy shift of 1 mRy, effectively extending the PAO radius of the C
orbital with the smallest ζ to 6 Å.22, 23, 24
0 fs
900 fs
1500 fs
g(r)
(a)
(b)
(c)
Supplementary Figure 7. Structural snapshots and pair distribution functions of a 216-
atom defect nanoparticle and adsorbate molecule. Snapshot times are indicated in the
column headings. In (a), the pair distribution functions for one adatom are compared at
300 K (red) and 400 K (black), with averaging done over 3000 snapshots. The slightly
broader and up-shifted black curve is indicative of the higher mobility of all atoms in
the nanoparticle at higher temperature. In (b), the pair distribution function of the
adatom is shown early in the simulation (black) and after 1200 fs (red), with averaging
done over 100 snapshots close to the indicated time. Again, broader peaks at later time
are indicative of atom mobility as the simulation heated up. The higher red peak at low
separation is indicative of reorganization of the adatom at 1200 fs. The meaning of the
lines in the (c) row g(r) plot is the same as for (b) – in this case, the adsorbate has bound
to a different adatom.

S11
The simulation results confirmed the mobile characteristic of the surface atoms in a
cluster. Although the time scale of the simulations was faster than the experimental
data, the goal was to confirm that the dynamic cluster reconstruction of surface atoms
was temperature dependent, and affected by organic (thiol) adsorbates. Those
adsorbates improved adatom stabilization. A better picture of the dynamical behavior
can be found in a video, a frame of which is shown in Supplementary Fig. 8.
Supplementary Figure 8. Frame from animation. Notice that the adatoms and atoms in
a cavity defect are distinctively coloured in the movie to emphasize their dynamic
behavior. (See video)

S12
Supplementary References
1. Huff J. The Airyscan detector from ZEISS: confocal imaging with improved signal-to-
noise ratio and super-resolution. Nature Methods 12, 1205 (2015).
2. Huff J, Bergter A, Birkenbeil J, Kleppe I, Engelmann R, Krzic U. The new 2D
Superresolution mode for ZEISS Airyscan. Nature Methods 14, 1223 (2017).
3. Olson AP, Ertsgaard CT, Elliott SN, Lindquist NC. Super-Resolution Chemical Imaging
with Plasmonic Substrates. ACS Photonics 3, 329-336 (2016).
4. Olson AP, Spies KB, Browning AC, Soneral PAG, Lindquist NC. Chemically imaging
bacteria with super-resolution SERS on ultra-thin silver substrates. Scientific Reports 7,
9135 (2017).
5. Willets KA. Super-resolution imaging of SERS hot spots. Chem Soc Rev 43, 3854-3864
(2014).
6. Raillon C, Granjon P, Graf M, Steinbock LJ, Radenovic A. Fast and automatic processing
of multi-level events in nanopore translocation experiments. Nanoscale 4, 4916-4924
(2012).
7. Brito-Silva AM, Sobral-Filho RG, Barbosa-Silva R, de Araújo CB, Galembeck A, Brolo AG.
Improved Synthesis of Gold and Silver Nanoshells. Langmuir 29, 4366-4372 (2013).
8. Sobral-Filho RG, Brito-Silva AM, Isabelle M, Jirasek A, Lum JJ, Brolo AG. Plasmonic
labeling of subcellular compartments in cancer cells: multiplexing with fine-tuned gold
and silver nanoshells. Chemical Science 8, 3038-3046 (2017).
9. Chen H-Y, Lin M-H, Wang C-Y, Chang Y-M, Gwo S. Large-Scale Hot Spot Engineering for
Quantitative SERS at the Single-Molecule Scale. Journal of the American Chemical
Society 137, 13698-13705 (2015).
10. Schmidt MK, Esteban R, González-Tudela A, Giedke G, Aizpurua J. Quantum
Mechanical Description of Raman Scattering from Molecules in Plasmonic Cavities. ACS
Nano 10, 6291-6298 (2016).
11. Carnegie C, et al. Room-Temperature Optical Picocavities below 1 nm3 Accessing
Single-Atom Geometries. The Journal of Physical Chemistry Letters 9, 7146-7151
(2018).

S13
12. Li J-M, et al. Multiplexed SERS detection of DNA targets in a sandwich-hybridization
assay using SERS-encoded core–shell nanospheres. Journal of Materials Chemistry 22,
12100 (2012).
13. Andreev YM, et al. A reusable biosensor chip for SERS-fluorescence dual mode
immunoassay. 9543, 954317 (2015).
14. Dieringer JA, Lettan RB, Scheidt KA, Van Duyne RP. A Frequency Domain Existence
Proof of Single-Molecule Surface-Enhanced Raman Spectroscopy. Journal of the
American Chemical Society 129, 16249-16256 (2007).
15. Ru ECL, Etchegoin PG. Single-Molecule Surface-Enhanced Raman Spectroscopy. Annual
Review of Physical Chemistry 63, 65-87 (2012).
16. Zrimsek AB, et al. Single-Molecule Chemistry with Surface- and Tip-Enhanced Raman
Spectroscopy. Chemical Reviews 117, 7583-7613 (2017).
17. Weber V, Feis A, Gellini C, Pilot R, Salvi PR, Signorini R. Far- and near-field properties of
gold nanoshells studied by photoacoustic and surface-enhanced Raman
spectroscopies. Physical Chemistry Chemical Physics 17, 21190-21197 (2015).
18. Soler JM, et al. The SIESTA method forab initioorder-Nmaterials simulation. Journal of
Physics: Condensed Matter 14, 2745-2779 (2002).
19. Perdew JP, Burke K, Ernzerhof M. Generalized Gradient Approximation Made Simple.
Physical Review Letters 77, 3865-3868 (1996).
20. Grimme S. Semiempirical GGA-type density functional constructed with a long-range
dispersion correction. Journal of Computational Chemistry 27, 1787-1799 (2006).
21. Liu W, et al. Structure and energetics of benzene adsorbed on transition-metal
surfaces: density-functional theory with van der Waals interactions including collective
substrate response. New Journal of Physics 15, 053046 (2013).
22. Ting EC, Popa T, Paci I. Surface-site reactivity in small-molecule adsorption: A
theoretical study of thiol binding on multi-coordinated gold clusters. Beilstein journal
of nanotechnology 7, 53-61 (2016).
23. Popa T, Paci I. Structure and Chirality in Sulfur-Containing Amino Acids Adsorbed on
Au(111) Surfaces. The Journal of Physical Chemistry C 119, 9829-9838 (2015).

S14
24. Chapman CRL, Ting ECM, Kereszti A, Paci I. Self-Assembly of Cysteine Dimers at the
Gold Surface: A Computational Study of Competing Interactions. The Journal of
Physical Chemistry C 117, 19426-19435 (2013).