Response to “Stripy Nanoparticles Revisited”


  • Miao Yu,

    1. Institute of Materials, École Polytechnique Fédérale de Lausanne, EPFL-STI-IMX-SuNMIL Station 12, Lausanne, CH-1015, Switzerland
    Search for more papers by this author
  • Francesco Stellacci

    Corresponding author
    1. Institute of Materials, École Polytechnique Fédérale de Lausanne, EPFL-STI-IMX-SuNMIL Station 12, Lausanne, CH-1015, Switzerland
    • Institute of Materials, École Polytechnique Fédérale de Lausanne, EPFL-STI-IMX-SuNMIL Station 12, Lausanne, CH-1015, Switzerland.

    Search for more papers by this author

  • The copyright line for this article was changed on 15 Aug 2017 after original online publication.


original image

In a paper entitled “Stripy Nanoparticles Revisited”, Lévy and co-workers contest the interpretation of scanning tunneling microscopy images of monolayer protected gold nanoparticles that our group has presented. We show that the two arguments they use are based on flawed assumptions and contradict each other. We also show new evidence for the existence of stripe-like domains on mixed monolayer-coated gold nanoparticles.

This Correspondence is written in response to the paper entitled “Stripy Nanoparticles Revisited”, published in this same issue.1 In 2004 we found, using scanning tunneling microscopy (STM), that self-assembled monolayers composed of a binary mixture of immiscible ligand molecules spontaneously develop stripe-like domains when assembled onto nanoparticles.2 The STM evidence produced was backed up by:

  • Trasmission electron microscopy (TEM) and X-ray diffraction (XRD) data2

  • Fourier transform infrared (FTIR) spectroscopy studies3

  • Atomic force microscopy (AFM) images in water4

  • computational studies5, 6

In the latter case, a remarkable agreement both in terms of domain spacing5 and in terms of the existence of a size regime where these stripes occur6 was found. It was also established that these particles have a non-monotonic dependence of their saturation concentration (solubility)7 and their interfacial energy4 on their ligand shell composition, in good agreement with molecular dynamics simulations. They exhibit protein-resistance properties,2, 8 can penetrate cell membranes,8 and posses two polar point defects that can be selectively functionalized with a place-exchange reaction9 that has been modeled so as to calculate the free energy of the reaction.10

In ref. 1, Lévy and co-workers present a series of arguments questioning the interpretation of our data as well as some of our results. It appears that some parts of our work might be misinterpreted and hence we are grateful for this opportunity to clarify these aspects. Our experiments are no doubt challenging. Through the years, we have dedicated a good amount of time to improving our experimental approaches and to derive a more reliable picture of our system, both through experiments and through simulations. We appreciate every effort to stimulate a constructive scientific debate and hope that this exchange will contribute to a better understanding of self-assembled monolayers (SAMs) on nanoparticles as well as their investigation with scanning probe microscopy.

The central argument presented in ref. 1 has to do with the interpretation of our STM images of mixed ligand nanoparticles, which Lévy and co-workers claim to be incorrect. We will present here a rebuttal to all of the points raised in ref. 1. In most cases we will start by correcting the selected presentation of our results given in ref. 1, to properly represent our body of work. We will also present new data to testify to our continuing research toward a deeper and more accurate interpretation of our STM data. In ref. 1, there exists the suggestion that our work on cell membrane penetration by striped nanoparticles cannot be reproduced; it should be stated up front that all of the data presented in our original publication8 have been reproduced in identical systems by our group,11 by our collaborators,12, 13 and in similar systems by groups not related to us.14

Summary of Previous Results

The main scientific argument developed in ref. 1 revolves around the interpretation of the STM images used to characterize our nanoparticles. We agree that this point merits an open discussion in the literature and we take this opportunity to deepen our discussion of the topic. Since our first publication,2 we have dedicated considerable attention to this point and are continuing to do so, as shown by the new results presented here. The authors of ref. 1 acknowledge our body of work on these particles, including four papers that deal almost entirely with STM images.2, 15–17 However, they choose to comment only on our first publication, ignoring answers available in the later ones. Science is a process, and it would be fair to assume that later papers offer a better understanding of a system and represent more accurately the position of a group of scientists.

For the sake of clarity, we will first briefly restate the relevant results and conclusions that we have published over the last eight years and then proceed to answer the different points raised in ref. 1.

In 2004, we presented a series of STM images whose interpretation was corroborated by XRD and TEM data.2 In these publications, we suggested that mixtures of dislike molecules spontaneously form striped domains on nanoparticles. In 2007, in a theory paper supported by experimental evidence, coarse grain and molecular dynamics simulations of our particles were presented. The simulations showed the presence of stripe-like domains with local order.5 Since then we have used in all of our publications a definition of our particles as containing “stripe-like” domains. We have since found a good number of structure–property relationships that link this particular configuration of ligands with macroscopic properties. None of these properties require a ‘perfect’ stripe alignment but only a local one. Representing our work as ‘perfect’ stripes on nanoparticles is a misrepresentation, and we hope that this communication will help to clarify this point. The simulation results shown in ref. 5 and the STM images presented in Figure 1 of ref. 15 (our second publication on this topic) show that our particles are made of stripe-like domains that are locally ordered. These are the representative images that we show in all of our presentations and proposals. Even in the case of our paper on the formation of chains of ‘striped’ nanoparticles via the functionalization of their polar defects,9 no ‘perfect’ alignment is required (as the authors of ref. 1 imply we state). For example, Figure 3d in ref. 5 shows that these poles also exist in the case of an imperfect ‘stripe-like’ arrangement.

Figure 1.

a) Molecular resolution STM topography image of hexanethiol-coated nanoparticles. b) Current image. c) Zoom of the delimited region of (a) showing headgroup arrangement. d,e) Height profiles along the dark lines in (a) together with Gaussian fits (dashed red lines) used to extract headgroup spacings (Bias voltage = 500 mV, Set current = 600 pA). The scan lines (1) and (2) (shown in (d) and (e), respectively) are taken on the middle and edge of a nanoparticle, respectively, yet they show identical spacing between the headgroups.

Response to STM Image Interpretation

The central claim of ref. 1 is based on two arguments: (1) that a simplified representation of striped particles would lead to a decreasing stripe width at the edges of our particles; (2) that Fourier transform (FT) analyses of our STM images evidence a periodicity in our images that is inconsistent with the presence of striped domains on nanoparticles. The former point is affected by a fundamental misunderstanding of the working principle of a scanning probe microscope, and can be disproved experimentally, while the latter is affected by a fundamental mistake in the interpretation of Fourier transforms of digitalized images, that can be readily identified through critical image analysis.

Image Formation

The first argument in ref. 1 deals with what we believe is a nontrivial problem. Imaging curved surfaces in scanning probe microscopy is complex, as arguably the microscopes have been engineered to image mostly-flat samples. Figure 1 in ref. 1 does not capture the reality of STM image acquisition, as it suggests (without showing it) that the tip follows a horizontal, sample-independent trajectory onto which sample-derived features (tunneling currents) are projected. Hence, the central argument in Lévy's paper is based on two assumptions: (1) an STM tip moves horizontally on a sample, and (2) tunneling currents flow perfectly vertically from the tip into the substrate holding the sample. Both assumptions are invalid. In reality, the tip follows the contour of the sample.23 The correct projection of the sample features being imaged with STM is onto the true tip trajectory, not onto an imaginary flat line. If we assume a tip trajectory that maintains a constant distance from the particle's center of mass (hence making a semicircular trajectory), then images of the idealized particle shown in ref. 1 would be projected onto such a semicircle and consequently should show stripes with a spacing of ∼1 nm. It should be noted that, were the tip to really move horizontally over the sample, there would be no feedback needed, no feedback loop-artifact possible, and the whole interpretation of the images presented in ref. 1 (feedback loop artifacts) would be in contradiction with the initial argument. The second assumption in ref. 1, that is, directionality of the tunneling current, is also invalid as, while very little is known about the exact path the current will take from the STM tip to the substrate, there is much evidence suggesting that the current will tunnel into the gold cores of the nanoparticles first.

Despite this, ref. 1 raises a valid point about imaging curved surfaces, given that the true tip and the true current trajectories are hard to determine. We have studied this problem in our group by analyzing images of homoligand nanoparticles. For these particles, our images show locally ordered patterns (of dots) that are simply an extension of the 2D crystals that the same molecules form on flat surfaces.18, 19 Hence, the interpretation of images of these particles is more straightforward than for mixed-ligand particles, providing an ideal reference system. We note that the arguments presented in ref. 1 should also apply to the intermolecular distance between adjacent ligand headgroups on homoligand nanoparticles; thus, according to the reasoning of ref. 1, such a distance should significantly decrease at the edges of particles (keeping the same proportionality, the spacing should decrease from 5 Å to 3 Å). As shown in Figure 6 of ref. 15 as well as in new data shown here in Figure 1, single molecules can be clearly identified both in the center and at the edges of nanoparticles, yet no distortions in the intermolecular spacing at the particle's edges is observed. These experimental observations disprove the main point in ref. 1.

The representation of our particles given in ref. 1 also assumes that all of the ligands are very rigid, something in contradiction with our physical insight into the flexible nature of the ligand molecules that form the stripes.5 Importantly, it should be noted that Lévy and co-workers chose not to discuss the fact that, in most of the published images of particles with stripe-like domains, stripes are indeed clear only in the center of the nanoparticles with the edges showing unclear features; this can be observed for example in Figure 5a of ref. 17. These images—which, as discussed in ref. 17, are the majority of the images we have—are in fact not in contradiction with the argument presented in ref. 1.

Image Fourier Transform

The second argument from ref. 1 is based on FT analysis carried out by the authors on two of our published STM images. Unfortunately, their arguments appear to be based on an erroneous interpretation of the FT results, suggesting that their conclusions are derived from an image-analysis artifact. The argument in ref. 1 is that an FT analysis of a single STM image from our work shows two lateral parallel bands. This, according to the authors of ref. 1, means that our images “have a defined wavelength, λx, along the fast scanning axis but no defined λy wavelength and therefore no overall wavelength λ”. As shown in Figure S1 in the Supporting Information, a series of aligned bands (with a very well-defined wavelength λ) shows two parallel bands in the FT, questioning this interpretation of the FT images.

Furthermore, the authors of ref. 1 perform an FT analysis on an image that they artificially construct from a series of cartoons of idealized striped particles. They find that an FT analysis of their image yields circular features. They claim that the circularity of their features confirms their argument. They seem to suggest that the circularity in their images is due to a ±5° rotation in the alignment of their ‘idealized’ particles. As shown in Figure S2 of our Supporting Information, an FT of the same image with all of the ‘idealized’ particles aligned keeps showing the same circularity, disproving this point.

These observations raise questions about the interpretation of the FT images. More questions can be raised by noticing that in the FT analysis of the STM image presented in ref. 1, a periodicity of 1.3 nm is found. This is in direct contradiction not only with our own measurements of average stripe width (based on 60 measurements) but also with those presented by Lévy and co-workers, in Table S1 of in ref. 1 (average width 1.08 ± 0.15 nm). Similar questions could be asked about the periodicity measured by the FT analysis of their cartoons, given the absence of an overall periodicity in those images.

Hereafter, we show that these features have little to do with the periodicity in the images but mostly depend on the shape of the objects present in the images. It is well known that FTs of images containing large features (compared to the size of the image) will tend to produce patterns reflecting the shape of the feature (a circle for a circle, rectangles and ovals turned 90° for rectangles and ovals, respectively). Hence, a circular shape like the one used to represent striped particles in ref. 1 leads to a circular band in the FT image. This can be proven by performing transformations on the image that preserve the periodicity the authors of ref. 1 claim to be studying. As shown in Figure 2, the FT of the cartoon used by the authors to re-create ‘idealized’ striped nanoparticles yields a circular pattern (central images), as shown in ref. 1. If both ends of the nanoparticles in that image are clipped so as to make the object appear more rectangular, or if the image is stretched parallel to the direction of the stripes, the FT images result in rectangular and oval patterns, respectively. Given that in both cases the periodicity is not altered, the arguments used by the authors of ref. 1 have to be incorrect.

Figure 2.

Images of idealized particles (top row) with their FT analysis (middle row). The bottom row show the same images shown in the middle row with the shape highlighted in the top row images simply turned 90° as one would expect going from direct to reciprocal space. In the middle column is the cartoon and the FT image that Lévy and co-workers use to substantiate their claims, left and right are two variations on the image that do not affect the ‘periodicity’ in the image but lead to completely different FT patters. It is easy to interpret these images, as each image is dominated by the FT of the shape of the particles that composes it (highlighted by the blue shapes). A more rectangular feature on the left, a circular on in the middle and an oval one on the right. All images analyzed had the same number of pixels. The blue features were added to the images after FT analysis.

To further prove this point, one can analyze two more types of images. It is possible to take the ‘idealized’ nanoparticle images from ref. 1—with a ±5° misalignment—and paste them in an overlapped manner (closer to what is observed in the STM). In such a case (as shown in Figure S3), the FT shows indeed two parallel bands (because there is no longer a prevalent circular shape in the image). Alternatively, it is possible to take an STM image of one of our particles and paste it in a pattern similar to that shown in ref. 1. As expected in this case, the result is an oval shape in the FT that is the reciprocal image of the shape of the particles (see Figure S4). Hence the interpretation of the FT images in ref. 1 appears to be based on an image-analysis artifact not directly related to the periodicity of our images.

In ref. 1, the authors claim to see a phase shift in images of our nanoparticles, describing this as a ‘hallmark’ of our images. We should first state that the images shown for example in references 16 and 17 do not present this feature, and also that this image feature can be simply interpreted as defect in the ‘stripe-like’ domains present on our particles, perfectly consistent with our argument; hence, one cannot use these features to discriminate between different image interpretations. Ref. 1 also claims that, in ref. 2 the STM images of our particles do not match the size of our particles reported in the same paper. To make this argument, the authors use the average size of our particles without considering their polydispersity (also reported in ref. 2). Obviously in ref. 2 we did not image with STM particles of exactly the average size seen under TEM, and the images are still consistent with the size distribution given. Moreover, in ref. 16 we used a monodisperse sample and showed rigorously that the STM size images do match the sizes of the particles shown under TEM.

STM Image Analysis

The analysis presented in ref. 1 does not explain any of the arguments used to substantiate our image representation. We have based our approach on a series of qualitative and quantitative analyses behind whose validity we stand. Through the years we have shown that there are significant differences between images of homoligand nanoparticles and those of particles coated with binary mixtures of ligands,15 as can be seen by comparing Figure 1 to Figures 3 and 4. We analyzed images of locally aligned stripe-like domains using many approaches. The main quantitative approach has been to vary the tip speed and to show that the average stripe width is almost invariant as a function of tip speed.17 We performed a full regression analysis to statistically validate and support our conclusions.17 Ref. 1 suggests that the images we generated are due to a feedback loop artifact such as the accidental scatter of our tip on a nanoparticle.24 In such a case, the resulting width of the image stripes would depend linearly on tip speed (and be zero at zero speed), as shown in ref. 17. This is not the case for images of our particles, as proven by a thorough statistical analysis based on 12 image sets and ∼1500 data sets.17 This was further confirmed by comparing multiple images of the same particles with and without stripes. We have shown that the subset of particles showing stripes stayed the same as successive images at varying tip speeds were taken,16, 17 ruling out any accidental event.

Figure 3.

STM images obtained in vacuum on an Omicron Vakuumphysik ‘micro-STM’ of striped nanoparticles at three different tip speeds (a–e) 300 nm/s, with average stripe width 0.98 nm; (f–j) 800 nm/s, with average stripe width 0.93 nm; (k–o) 1300 nm/s, with average stripe width 1.0 nm. Images in the middle row are isolated particles enlargements to better show stripe-like domains, the bottom row show the same images with lines to guide the eye on the striped domains.

Figure 4.

STM images obtained in vacuum on an Omicron Vakuumphysik ‘micro-STM’ of striped nanoparticles at two different scan angles. Panels (b) and (e) are enlargements of the blue boxes shown in panels (a) and (d), respectively. Panels (c) and (f) are identical to (b) and (e), respectively; the blue and green circles highlight the same particles with stripe-like domains, rotated according to the image rotation.

Finally, nanoparticles of the same composition show the same average stripe width over time, while particles of different composition show a different stripe width (see Figure 6 in ref. 15, and Figure 18 in ref. 17). Here in Figure 3 and Figure 4, we also show that striped nanoparticle images qualitatively and quantitatively comparable to the ones we have presented through the years (imaged in air with a Veeco Multimode IIIa microscope) can also be obtained in vacuum on an Omicron Vakuumphysik ‘micro-STM’, exhibiting a stripe width consistent with similar particles presented four years ago.16 Figure 3 and Figure 4 also show that the stripes of octanethiol:methylbenzylthiol 2:1 nanoparticles are invariant with tip speed and rotate with scan direction.

We believe that these results cannot be explained in the framework of ref. 1’s interpretation. Given the significant amount of work that we have done to address the interpretation of our images, it would seem fair to expect an alternative interpretation to explain all of our findings and not a selected subset.

Ref. 1 does not disclose the fact that simulations have been performed on mixed-ligand coated nanoparticles5 and have consistently shown the formation of stripe-like domains. These simulations have captured all of the features that we observe on nanoparticles. For example, recently we have proven correct the prediction that small particles are indeed Janus.20 These simulations do not have a faceted core but a spherical one. This is mainly because it is known that at room temperature the gold–sulfur interface is highly mobile and simulating it in a frozen configuration is at least as far from reality as simulating a sphere. Also, as discussed in ref. 9 and 15 for small particles, molecules tend to have a global alignment that goes beyond the faceted nature of the particles.

Nanoparticle Properties

Particle Solubility

In ref. 1, comments on our 2009 publications7 about the solubility of striped nanoparticles state that the saturation concentration that we report is too low if compared to their results. They use as reference a single experiment on homoligand nanoparticles. The lack of details provided to describe the particles they used and the experiment performed (and the fact that it was not performed under the conditions described in our paper) does not allow us to comment. We can only say that, through the years, we have noticed that the more the particles are cleaned of unbound molecules in their ligand shell the less soluble they are. Ref. 1 claims that there are contradictions between our statements on particle solubility in references 2 and 7 and particle interdigitation in references 2, 7, and 16. Yet ref. 2 describes octanethiol, mercaptoproprionic acid-coated particles, ref. 7 describes octanethiol, mercaptoproprionate-coated particles, and ref. 16 describes dodecane thiol, methylbenzyl thiol-coated particles. Why would or should all of these particles behave the same way in terms of solubility and interdigitation, given that the molecules that coat them are different?

Particles and Cells

In 2008, we produced a study that shows that striped nanoparticles coated with a 2:1 mixture of mercaptoundecane sulfonic (MUS) acid and 1-octanethiol (OT) (hereafter MUS:OT 2:1) are able to penetrate cell memebranes in an energy-independent way without overt poration.8 In our paper, we determine cytosolic presence of the nanoparticles by showing a diffuse cytosolic background in confocal fluorescence images of cells as well as by free nanoparticle images in TEM studies. Ref. 1 (i) fails to acknowledge the existence of relevant TEM images and of their statistical analysis, and (ii) claims that we are affected by quenching problems in our fluorescence microscopy studies. This cannot be relevant since our study compares nanoparticles (homoligand all-MUS, non-striped MUS/3,7-dimethyl 1-octanethiol 2:1, and striped MUS:OT 2:1 and 1:2) that differ only in the molecular arrangement of the ligand shell, apart from the homoligand all-MUS used as a negative control. These nanoparticles have statistically the same size distribution and thickness of the ligand shell, and hence cannot be differentially affected by quenching problems, as quenching would equally affect every particle. Moreover, ref. 1 claims that our choice of cell lines (dendridic mouse cell (DC2.4) and human fibroblasts) as well as our imaging technique are not proper for testing cell penetration by nanoparticles. In other words, in order to reproduce our experiments, Lévy and co-workers have changed most of the experimental parameters we used in our initial work. Nonetheless, we carefully tested the claim in ref. 1 that Hela cells are more suitable for cell penetration experiments and recently published work showing the same results we obtained with dendritic cells using Hela cells.11

The authors of ref. 1 also claim that photothermal heterodyne imaging (PHI) is a better imaging tool than fluorescence microscopy for our purposes. We also find that indeed PHI is a truly useful technique.21, 22 PHI work carried out in his lab has investigated the interaction of striped nanoparticles with DC2.4 cells. The results obtained are highlighted in a recently published paper,12 where we confirm the cytosolic presence of striped nanoparticles and show that PHI can detect orders of magnitude lower concentrations of nanoparticles in cells as compared to fluorescence microscopy (something that helped us cut down the incubation time to a single hour, avoiding most endocytotic pathways). Importantly, using the free movement of striped nanoparticles in the cytosol, we show that the combination of these particles and PHI can be used to calculate local cytosol viscosity. Summarizing, we have shown that an energy-independent penetration mechanism exists for MUS OT 2:1 nanoparticles, independent of (i) cell lines and (ii) imaging technique. These results confirm the reproducibility of our earlier findings. It should be stressed that four laboratories outside ours have used our particles and reproduced our data.


In conclusion, the alternative interpretation to our scanning tunneling microcopy images proposed in ref. 1 does not account for all of the arguments presented in our body of work and focuses only on our first publication. Importantly, the authors of ref. 1 disregard large parts of our work, concealing most of the relevant results. Their arguments are based on the wrong understanding of image acquisition in scanning probe microscopy and a particular interpretation of Fourier transform analysis of our images that we prove to be wrong. We show definite evidence that their analysis is an artifact of the image definition. The paper tries to claim that some of our presented data are not reproducible, yet the only experiments presented to support this claim are carried out under conditions significantly different to those presented in our papers. By contrast, work carried out in the laboratories of collaborators was able to successfully reproduce our results.

Supporting Information

Supporting Information is available from the Wiley Online Library or from the author.


We are grateful to Dr. Kislon Voitchovsky for very helpful discussions and critical reading of the paper, and to Dr. Cedric Dubois for the images of the homoligand nanoparticles.