Quantitative super-resolution imaging uncovers reactivity ... - Peng Chen

LETTERS PUBLISHED ONLINE: 19 FEBRUARY 2012 | DOI: 10.1038/NNANO.2012.18

Quantitative super-resolution imaging uncovers reactivity patterns on single nanocatalysts Xiaochun Zhou, Nesha May Andoy, Guokun Liu, Eric Choudhary, Kyu-Sung Han, Hao Shen and Peng Chen* Metal nanoparticles are used as catalysts in a variety of important chemical reactions1,2, and can have a range of different shapes3–8, with facets and sites that differ in catalytic reactivity1,2,9. To develop better catalysts it is necessary to determine where catalysis occurs on such nanoparticles and what structures are more reactive. Surface science experiments or theory can be used to predict the reactivity of surfaces with a known structure1,2,10, and the reactivity of nanocatalysts can often be rationalized from a knowledge of their well-defined surface facets3–5. Here, we show that a knowledge of the surface facets of a gold nanorod catalyst is insufficient to predict its reactivity, and we must also consider defects on the surface of the nanorod. We use super-resolution fluorescence microscopy to quantify the catalysis of the nanorods at a temporal resolution of a single catalytic reaction and a spatial resolution of ∼40 nm. We find that within the same surface facets on the sides of a single nanorod, the reactivity is not constant and exhibits a gradient from the centre of the nanorod towards its two ends. Furthermore, the ratio of the reactivity at the ends of the nanorod to the reactivity at the sides varies significantly from nanorod to nanorod, even though they all have the same surface facets. We chose gold nanorods (diameter, 21 nm; length, 150–700 nm) as nanocatalysts because of their highly anisotropic structure and faceted surfaces (a consequence of their nanocrystal nature; Supplementary Fig. S2)6–8. We further coated these nanorods with a 80-nm-thick mesoporous silica shell (hereafter Au@mSiO2 nanorods; Fig. 1b and Supplementary Figs S3,S4)11. This shell allows us to use calcination to remove the capping ligands on the nanorod surface to prepare for catalysis, while preventing aggregation of the nanorods and stabilizing their morphology. Meanwhile, the reactants are still able to access the gold surface for catalysis through the mesopores of the shell. This silica shell also mimics an oxide support, as is often used in heterogeneous catalysis1. Analysis of transmission electron microscopy (TEM) results showed that, of the many possible structural models, the gold nanorod and shell structure can both be modelled sufficiently well as a cylinder with two hemispherical caps (Fig. 1c, Supplementary Figs S5–S10 and Scheme S1). We used a super-resolution imaging approach to study the catalysis of single Au@mSiO2 nanorods. Study at the single nanocatalyst level is necessary, as individual nanocatalysts can differ greatly9,12–18. Our super-resolution imaging of catalytic reactions is based on wide-field single-molecule microscopy of a fluorogenic catalytic reaction: the gold-nanorod-catalysed oxidative deacetylation of the non-fluorescent Amplex Red (S) to the fluorescent resorufin (P) by H2O2 in aqueous solution (Fig. 1a and Supplementary Figs S11–S13). To image this reaction on single Au@mSiO2 nanorods, we dispersed the nanorods on a quartz slide in a microfluidic reactor cell and applied the reactant solution. These nanorods

scatter the laser light and are also emissive19,20, making them easily identifiable under an optical microscope for one-to-one correlation with their scanning electron microscopy (SEM) images (Fig. 2a; Supplementary Sections 10 and 13). Each reaction catalysed by a Au@mSiO2 nanorod generates a P molecule, the laser-induced fluorescence of which was imaged by an electron-multiplying charge-coupled device camera at a frame rate of 25 ms. The fluorescence intensity versus time trajectory from a single nanorod shows stochastic intensity bursts on top of the constant nanorod emission signal (Fig. 1d), where each burst marks the catalytic generation of a fluorescent P molecule. The position of each P molecule can be localized with nanometre accuracy by Gaussian-fitting its fluorescence image (Fig. 1e,f and Supplementary Fig. S16)21. By tracking the P position frame by frame in a movie22 and analysing the width of its image point spread function (PSF), we found that P molecules temporarily adsorb to the silica shell (for 100 ms, on average) and do not diffuse appreciably within the shell after they are generated from the gold surface and before they desorb and disappear into the surround solution (Supplementary Fig. S20). By localizing the position of every P molecule, we mapped all catalytic reactions on a single Au@mSiO2 nanorod (Fig. 2b). Both this map and the two-dimensional histogram (Fig. 2c) of the P positions clearly resolve the rod shape with 40 nm resolution (Supplementary Fig. S19), comparable to related super-resolution optical microscopy techniques23–25. The wide-field imaging format of our approach further offers the facility to study hundreds of nanorods in parallel, giving a high data throughput. Similar imaging strategies have also been used to differentiate catalysis on different facets of metal hydroxide and oxide microcrystals12,14, resolve catalytic domains in porous oxide micro-needles/crystals26, and probe the dimensions of reactive sites on carbon nanotubes27. We next mapped the structural contour of a Au@mSiO2 nanorod from its SEM image onto the two-dimensional histogram of its P positions and dissected the nanorod into 90 nm segments (Fig. 2c and Supplementary Figs S21–S23) (the gold surface area in each segment was determined using the model in Fig. 1c, yet our conclusions are independent of the model; Supplementary Section 7). For each segment, we determined its specific turnover rate v by counting the number of P molecules per unit time and per gold surface area. As expected, v is dependent on the concentration of the reactant Amplex Red ([S]); for all segments it shows saturation kinetics with increasing [S] while the other reactant H2O2 is kept at excess (Fig. 2d). This saturation kinetics follows the Langmuir–Hinshelwood kinetics for the case where the two reactants adsorb non-competitively while one reactant is in large excess (Supplementary Section 16)28: v=

kK[S] 1 + K[S]

(1)

Department of Chemistry and Chemical Biology, Cornell University, Ithaca, New York 14853, USA. * e-mail: [email protected] NATURE NANOTECHNOLOGY | VOL 7 | APRIL 2012 | www.nature.com/naturenanotechnology

© 2012 Macmillan Publishers Limited. All rights reserved.

237

LETTERS

NATURE NANOTECHNOLOGY

a

DOI: 10.1038/NNANO.2012.18

c

b

R O

OH

O

O

O

R

2r

N

N C

H3C

O

r

O

Flow

1.6r

L

200 nm

Laser

L'

e

10k

8k + 6k

4k 2k 0

1,000 nm 100

104 Time (s)

± 6 nm

f 6k Intensity (a.u.)

Fluorescence intensity (a.u.)

d

3

108

6 x (pixe l)

3

9

y

9 6 el) ix p (

Figure 1 | Single-molecule super-resolution fluorescence microscopy of single Au@mSiO2 nanorod catalysis at single-turnover resolution. a, Experimental scheme using total internal reflection fluorescence microscopy and a microfluidic reactor to image the fluorogenic deacetylation reaction of Amplex Red to resorufin catalysed by individual Au@mSiO2 nanorods. b, TEM image of a Au@mSiO2 nanorod. c, Structural model of Au@mSiO2 nanorods (Supplementary Section 6). Both the gold nanorod and the silica shell are modelled by a cylinder with hemispherical caps. The end cap of the nanorod is not concentric with the shell cap; it protrudes into the shell cap by a length of 1.6r (r ¼ 10.7+1.6 nm). d, Fluorescence intensity versus time trajectory for a single Au@mSiO2 nanorod under catalysis with 0.02 mM Amplex Red and 60 mM H2O2 in 7 mM pH 7.3 phosphate buffer. e, Wide-field fluorescence image of a single P molecule during one burst (circled in d). f, Three-dimensional representation of the image in e. The fluorescence intensity of a single P molecule spreads over a few pixels (each pixel, 267 nm) as a PSF. The centre position of this PSF can be localized with nanometre accuracy and is marked as a red cross in e. The emission signal of the nanorod itself has been subtracted from e and f (see details in Supplementary Section 11, Fig. S16).

a

c

40 2

3

4

5

6 R 20

No. of molecules

1

R 0 500 nm

200 nm

d

b

1 2 3 4 5 6

Specific turnover rate (×10−5 s−1 nm−2)

4

2

500 nm 0 0.0

0.4 [S] (µM)

0.8

Figure 2 | Quantitative reaction kinetics of a single Au@mSiO2 nanorod at subparticle resolution. a, SEM of a Au@mSiO2 nanorod. b, Positions of product (P) molecules detected on the nanorod in a. Each dot is one P molecule. The reactions were performed with 0.02–0.8 mM Amplex Red and 60 mM H2O2 in 7 mM pH 7.3 phosphate buffer. c, Two-dimensional histogram of b in 20 × 20 nm2 bins. The orientation of the image has been rotated so that the nanorod lies horizontally. The red line is the outer structural contour from its SEM image in a, and encircles the silica shell. Vertical lines dissect the nanorod into six segments; the two end segments include just the hemispherical caps of the shell, each of which contains the gold nanorod hemispherical cap and part of its cylinder, as depicted in Fig. 1c. For clarity, b only includes the first 2,800 of 8,700 P molecules presented in c. d, [S] dependence of specific turnover rates of the segments from c. All error bars indicate s.e.m. unless specified otherwise. 238

NATURE NANOTECHNOLOGY | VOL 7 | APRIL 2012 | www.nature.com/naturenanotechnology


LETTERS

a

b k (×10−5 s−1 nm−2)

DOI: 10.1038/NNANO.2012.18

k (×10−5 s−1 nm−2)


4

2

c

6

No. of nanorods 13 96 76 88 17 11

4

2

0 −300

−150 0 150 300 Distance from centre (nm)

40

40

K (µM−1)

K (µM−1)

60

20 0 −300

20

0 −300


193±24 283±45 366±27 425±23 486±18 548±22

8

−150 0 150 300 Distance from centre (nm) k (×10−5 s−1 nm−2)

−300

L±s.d. (nm)

4

0 −300


−150 0 150 Distance from centre (nm)

300


300

e

4 3 2 1

6

kend kcentre knanorod

4

200 400 600 Nanorod length, L (nm)

800

40

20

2

0 0

K (µM−1)

kend / kcentre

5

k (×10−5 s−1 nm−2)

60

d

200 400 600 Nanorod length, L (nm)

0 −300

Figure 3 | Spatial reactivity patterns of single Au@mSiO2 nanorods. a, Specific catalytic rate constant k and reactant adsorption equilibrium constant K of each segment from Fig. 2d. End segments are highlighted in red and centre segment in blue. X error bars indicate the segment lengths. b, As in a, but for another nanorod. c, As in a and b, but with the data averaged over many nanorods in six length groups as shown in the scheme at the top. X error bars are omitted for clarity. d, Correlation of the rate constant ratio kend/kcentre of the end and centre segments with nanorod length for 301 nanorods (grey dots). For each nanorod, kend is the average of the two end segments. The nanorods are then grouped every 50 nm by their lengths and averaged within each group to obtain the dependence of kend/kcentre on nanorod length (hollow squares). The red line is a visual guide. e, Dependence of kend , kcentre and knanorod on nanorod length. knanorod is obtained by analysing a nanorod as a whole (Supplementary Section 20, Fig. S29) and it follows the same trend as kend. Each point is an average of many nanorods.

Here, k is the specific catalytic rate constant, and v ¼ k when [S] 1. K is the adsorption equilibrium constant of S. Fitting Fig. 2d with equation (1) gives k and K for each segment, directly quantifying the catalytic property of a single nanorod at subparticle resolution. As the reaction kinetics was titrated across a range of reactant concentrations over a few hours, k and K here represent the timeaveraged properties of each nanorod segment, and are therefore not affected by possible surface-restructuring-coupled temporal activity fluctuations, which we have observed previously on small pseudo-spherical gold nanoparticles13. We then compared the reactivities of the end and middle segments of a single Au@mSiO2 nanorod (the segments comprising different gold surface facets). In the nanorod for which results are shown in Fig. 3a, the two end segments have larger values of k (that is, higher catalytic reactivity) than all its middle segments. In all segments, a larger k tends to be associated with a smaller K, with a cross-correlation coefficient of rk,K ≈ –0.3 (Supplementary Fig. S27). When averaged for more than 300 nanorods, this relative reactivity between the end and middle segments is largely maintained (Fig. 3c). Catalysis imaging with such nanometre resolution has therefore revealed that for these Au@mSiO2 nanorods their ends generally have a higher catalytic reactivity but weaker reactant adsorption than their sides. Surprisingly, the relative reactivity of the ends and sides of a gold nanorod varies among individual nanorods, even though they share the same types of facets on their ends and sides6–8. Figure 3b shows

the results for another Au@mSiO2 nanorod for which the middle segments have higher reactivity and correspondingly weaker reactant adsorption than the end segments. This variant of reactivity pattern is exhibited by a small population of the nanorods (15%) and is hidden in ensemble-averaged results (Fig. 3c). To quantify the relative reactivity between the ends and sides of every nanorod, we computed kend/kcentre , the rate constant ratio between the end and centre segments (Fig. 3d). Most of the nanorods have kend/kcentre . 1, but a small population have kend/kcentre ≤ 1. Generally, kend/kcentre decreases with increasing nanorod length, and when the length is greater than 600 nm, kend/kcentre , 1. This trend is associated with the opposite behaviours of kend and kcentre: with increasing nanorod length, kend decreases and kcentre increases slightly (Fig. 3e). The behaviour of kcentre is particularly interesting. It is completely hidden when a nanorod is analysed as a single entity rather than in segments, as kend dominates (Fig. 3e), and it is only discoverable by spatially resolving the reactivity of single nanorods at subparticle resolution. We can understand the relative reactivity of the ends and sides of these gold nanorods if we assume the catalytic sites are low-coordination metal sites (for example, corner, edge and defect sites), which are often more reactive because of their large coordinative unsaturation1,2. The ends of gold nanorods generally have a higher percentage of corner and edge sites, and consequently higher catalytic reactivity. However, the sides of these gold nanorods have many defects, visible under high-resolution TEM29, and occasionally



239

LETTERS No. of molecules

a

60

c 4

40

kc,L (×10−5 s−1 nm−2)


20 0 200 nm

DOI: 10.1038/NNANO.2012.18

2

0 0

2 4 βL × L (×10−5 s−1 nm−2)

d 6 βL (×10−8 s−1 nm−3)

4

2

1.2 4

0.8

kc,L (×10−5 s−1 nm−2)

Specific turnover rate (×10−5 s−1 nm−2)

b

2 −300


300

150

300 450 Nanorod length, L (nm)

Figure 4 | Reactivity gradient along the nanorod sides. a, Two-dimensional histogram of 9,530 product positions for the nanorod in Fig. 3b in 20 × 20 nm2 bins. Vertical lines separate out the end segments. b, Dependence of specific turnover rate v on location along the length of the nanorod in a at [S] ¼ 0.5 and 0.8 mM, which are close to saturating reactant concentrations where v ≈ k (equation (1)). The nanorod is sliced into 20 nm segments. X error bars indicate segment lengths. c, Correlation between kc,L and bL × L. Each point is from one nanorod. The red line is a linear fit with a slope of 1 and it intercepts the y-axis at (2.2+0.3) × 1026 s21 nm22. d, Dependence of bL and kc,L on nanorod length. Each point is an average of many nanorods. X error bars are s.d.

some nanorods can have many defects and thus higher reactivity on their sides. When compared among individual nanorods, longer nanorods grow faster than shorter ones during their one-pot seeded synthesis, and they should therefore have more surface defects on their sides30, thus explaining their larger values of kcentre. The growth rate of gold nanorods during synthesis is not constant and has been observed to decrease linearly with increasing length31. A gradient of defect density should therefore develop along their sides30, with the highest at the centre and becoming lower towards the two ends as they grow outwards from seeds. Consequently, a reactivity gradient should exist along the sides (excluding the two ends). Excitingly, this prediction is exactly what we have observed: the specific turnover rate of a single nanorod decreases linearly along its sides from the centre toward the two ends (Fig. 4a,b; Fig. 3a,b, top). A similar reactivity gradient is also visible in the results averaged over many nanorods (Fig. 3c, middle). We formulated a model to describe the location-dependent reactivity along the nanorod sides, assuming that the catalytic reactivity, kL(x), at a distance x from the centre of a nanorod of length L is linearly proportional to the local surface defect density, which in turn is linearly proportional to the nanorod growth rate at that location (Supplementary Section 22): kL (x) = −2bL x + kc,L

(2)

Parameter bL is the reactivity gradient from the centre towards the two ends; this also reflects the underlying gradient of defect density and that of the growth rate. kc,L is the reactivity at the nanorod centre. This model also dictates kc,L ¼ bLL þ k0 for each nanorod; here, k0 represents the reactivity of the nanorod sides close to the ends (that is, x ≈ L/2), where the growth rate approaches zero (that is, with perfect side facets). Using equation (2) to fit the data as in Fig. 4b, we obtained bL and kc,L for every nanorod. When 240

the results from many nanorods are compiled, a linear correlation between kc,L and bLL is clear, with a slope of 1 (Fig. 4c), thus validating our model. The y-intercept of this linear correlation gives k0 ≈ 2.2 × 1026 s21 nm22, the reactivity of perfect side facets. This k0 is an order of magnitude smaller than kc,L (1025 s21 nm22), indicating that the reactivity of the nanorod sides is dominated by surface defects rather than by the surfacefacet sites. The nanorod length dependencies of kc,L and bL provide further new insights into the growth of these gold nanorods during synthesis (Fig. 4d). For longer nanorods, kc,L is larger, consistent with kcentre (Fig. 3e), reflecting the higher defect density at the centre and thus faster initial growth rate from the seed. In contrast, bL is smaller for longer nanorods, reflecting a shallower gradient of defect density along the sides of longer nanorods and therefore a slower decay of their growth rates during synthesis. Both the faster initial growth rate and the slower decay of growth rate contribute to the longer nanorods eventually being longer. It is surprising that catalytic reactivity varies along the sides of any gold nanorod, considering that the same set of side facets span its length. This discovery indicates that identifying the surface facets of nanocatalysts is insufficient for correlating with and predicting their reactivity. Surface defects, which are abundant on nanocatalysts and have a spatial distribution that is strongly affected by the nanocatalyst growth pattern and synthesis procedure, must also be considered, and they can play dominant roles in determining surface reactivity. Single-molecule super-resolution catalysis imaging provides unique insights into the reactivity of nanocatalysts, and when performed in parallel with electron or X-ray microscopy for characterizing the structure and chemical nature of nanocatalysts32,33, it will offer even more insights into catalysis at the nanoscale. Received 7 November 2011; accepted 19 January 2012; published online 19 February 2012




DOI: 10.1038/NNANO.2012.18

References 1. Somorjai, G. A. & Li, Y. Introduction to Surface Chemistry and Catalysis 2nd edn (Wiley, 2010). 2. Ertl, G. Reactions at Solid Surfaces (Wiley, 2009). 3. Burda, C., Chen, X., Narayanan, R. & El-Sayed, M. A. Chemistry and properties of nanocrystals of different shapes. Chem. Rev. 105, 1025–1102 (2005). 4. Xia, Y., Xiong, Y., Lim, B. & Skrabalak, S. E. Shape-controlled synthesis of metal nanocrystals: simple chemistry meets complex physics? Angew. Chem. Int. Ed. 48, 60–103 (2009). 5. Lee, H. et al. Morphological control of catalytically active platinum nanocrystals. Angew. Chem. Int. Ed. 45, 7824–7828 (2006). 6. Murphy, C. J. et al. Gold nanorod crystal growth: from seed-mediated synthesis to nanoscale sculpting. Curr. Opin.Colloid Interface Sci. 16, 128–134 (2011). 7. Carbo´-Argibay, E. et al. The crystalline structure of gold nanorods revisited: evidence for higher-index lateral facets. Angew. Chem. Int. Ed. 49, 9397–9400 (2010). 8. Katz-Boon, H. et al. Three-dimensional morphology and crystallography of gold nanorods. Nano Lett. 11, 273–278 (2011). 9. Weckhuysen, B. M. Chemical imaging of spatial heterogeneities in catalytic solids at different length and time scales. Angew. Chem. Int. Ed. 48, 4910–4943 (2009). 10. Nilsson, A., Pettersson, L. G. M. & Norskov, J. K. Chemical Bonding at Surfaces and Interfaces (Elsevier, 2008). 11. Botella, P., Corma, A. & Navarro, M. T. Single gold nanoparticles encapsulated in monodispersed regular spheres of mesostructured silica produced by pseudomorphic transformation. Chem. Mater. 19, 1979–1983 (2007). 12. Roeffaers, M. B. et al. Spatially resolved observation of crystal-face-dependent catalysis by single turnover counting. Nature 439, 572–575 (2006). 13. Xu, W., Kong, J. S., Yeh, Y-T. E. & Chen, P. Single-molecule nanocatalysis reveals heterogeneous reaction pathways and catalytic dynamics. Nature Mater. 7, 992–996 (2008). 14. Tachikawa, T., Yamashita, S. & Majima, T. Evidence for crystal-face-dependent TiO2 photocatalysis from single-molecule imaging and kinetic analysis. J. Am. Chem. Soc. 133, 7197–7204 (2011). 15. Novo, C., Funston, A. M. & Mulvaney, P. Direct observation of chemical reactions on single gold nanocrystals using surface plasmon spectroscopy. Nature Nanotech. 3, 598–602 (2008). 16. Tang, M. L., Liu, N., Dionne, J. A. & Alivisatos, A. P. Observations of shapedependent hydrogen uptake trajectories from single nanocrystals. J. Am. Chem. Soc. 133, 13220–13223 (2011). 17. Meier, J., Friedrich, K. A. & Stimming, U. Novel method for the investigation of single nanoparticle reactivity. Discuss. Faraday Soc. 121, 365–372 (2002). 18. Tel-Vered, R. & Bard, A. J. Generation and detection of single metal nanoparticles using scanning electrochemical microscopy techniques. J. Phys. Chem. B 110, 25279–25287 (2006). 19. Eustis, S. & El-Sayed, M. Aspect ratio dependence of the enhanced fluorescence intensity of gold nanorods: experimental and simulation study. J. Phys. Chem. B 109, 16350–16356 (2005). 20. Geddes, C. D., Parfenov, A., Gryczynski, I. & Lakowicz, J. R. Luminescent blinking of gold nanoparticles. Chem. Phys. Lett. 380, 269–272 (2003).

LETTERS

21. Yildiz, A. et al. Myosin V walks hand-over-hand: single fluorophore imaging with 1.5-nm localization. Science 300, 2061–2065 (2003). 22. Michaelis, J. & Braüchle, C. Reporters in the nanoworld: diffusion of single molecules in mesoporous materials. Chem. Soc. Rev. 39, 4731–4740 (2010). 23. Betzig, E. et al. Imaging intracellular fluorescent proteins at nanometer resolution. Science 313, 1642–1645 (2006). 24. Rust, M. J., Bates, M. & Zhuang, X. Sub-diffraction-limit imaging by stochastic optical reconstruction microscopy (STORM). Nature Methods 3, 793–796 (2006). 25. Hess, S. T., Girirajan, T. P. K. & Mason, M. D. Ultra-high resolution imaging by fluorescence photoactivation localization microscopy. Biophys. J. 91, 4258–4272 (2006). 26. Roeffaers, M. B. J. et al. Super-resolution reactivity mapping of nanostructured catalyst particles. Angew. Chem. Int. Ed. 48, 9285–9289 (2009). 27. Xu, W. et al. Single-molecule electrocatalysis by single-walled carbon nanotubes. Nano Lett. 9, 3968–3973 (2009). 28. Satterfield, C. N. Heterogeneous Catalysis in Practice (McGraw-Hill, 1980). 29. Wang, Z. L., Gao, R. P., Nikoobakht, B. & El-Sayed, M. A. Surface reconstruction of the unstable {110} surface in gold nanorods. J. Phys. Chem. B 104, 5417–5420 (2000). 30. Martin, J. J. & Armington, A. F. Effect of growth rate on quartz defects. J. Crys. Growth 62, 203–206 (1983). 31. Gulati, A., Liao, H. & Hafner, J. H. Monitoring gold nanorod synthesis by localized surface plasmon resonance. J. Phys. Chem. B 110, 22323–22327 (2006). 32. Zheng, H. et al. Observation of single colloidal platinum nanocrystal growth trajectories. Science 324, 1309–1312 (2009). 33. De Smit, E. et al. Nanoscale chemical imaging of a working catalyst by scanning transmission X-ray microscopy. Nature 456, 222–225 (2008).

Acknowledgements The authors acknowledge the Army Research Office (W911NF0910232), National Science Foundation (CBET-0851257, to P.C.; DGE0903653, to E.C.), the US Department of Energy (DE-FG02-10ER16199) and the Sloan Research Fellowship for funding, E. Zubarev for gifts of gold nanorod samples and J. Sambur for comments. Part of the work was carried out at the Cornell Center for Materials Research (DMR-0520404) and the Cornell NanoScale Facility (ECS-0335765).

Author contributions P.C. conceived the experiments. X.Z. performed the experiments. N.M.A., G.L., E.C., K-S.H. and H.S. contributed to the experiments. X.Z. and P.C. analysed the data and wrote the paper.

Additional information The authors declare no competing financial interests. Supplementary information accompanies this paper at www.nature.com/naturenanotechnology. Reprints and permission information is available online at http://www.nature.com/reprints. Correspondence and requests for materials should be addressed to P.C.



241