The company Aurasense (now Exicure) was established in 2009. Founders are Shad Thaxton and Chad Mirkin. Its first employee was David Giljohann: Principal Scientist (2009-11), then COO 2012-13, now CEO.
According to this 2009 New York Times article, Thaxton and Mirkin founded Aurasense to commercialise a technology based on “tiny particles that mimic those good carriers of cholesterol before it can grow into dangerous deposits of plaque”
Thaxton, Mirkin & Giljohann are coauthors of a 2012 ACS Nano article: “Tailoring of Biomimetic High-Density Lipoprotein Nanostructures Changes Cholesterol Binding and Efflux”. Their affiliations are listed as Northwestern University. The article does not include any competing interest statement. There is no mention of the company Aurasense. Last sentence of article:
Thaxton published also in 2012 another study in PNAS where the conflict of interest was disclosed: “Conflict of interest statement: C.S.T. is cofounder of AuraSense, LLC, which holds licenseto synthetic HDL nanoparticles from Northwestern University”.
It seems most people (in my twitter followers anyway) agree that failing to declare relevant affiliation to a company and failing to declare competing interests is a breach of publishing ethics. But is there any remedy?
Note: this is a short case study focusing on one specific article. There is a lot more to consider about Aurasense/Exicure/Mirkin articles and competing interests (watch this space but don’t be too impatient).
Apparently* that is the title of a seminar I will be giving on Thursday in Topics In Bioengineering, a student-led seminar series within the Harvard John A. Paulson School of Engineering and Applied Sciences. I am grateful to the students and postdocs who invited me. The seminar will be online and you can register here.
* My original title was a bit more provocative :).
PS: credit for the portrait above goes to my daugther Laura.
… applications of nanoscience are revolutionizing the fields of biology and medicine. Nano-sized artificial molecular motors and machines are created using biological molecules, while nanomedicine is utilized to target cancer cells, deliver drugs and fight antibiotic resistant bacteria
The recent coronavirus crisis shows how the field of nanobiosensors may be key to finding innovative solutions for a quick response, and also to cope with future infectious outbreaks. And yet the discussion about the ethics of nanotechnology is ongoing: great opportunities come with huge responsibilities.
Humans are now learning to build machines on the nanometer scale, imitating the elegance and economy of nature. Nanotechnology can provide new formulations and routes for drug delivery, enormously broadening their therapeutic potential. Nanotechnology will allow earlier detection of cancer and other diseases. Nanoparticles are showing promise for the delivery of drugs to specific tissues (e.g., a tumor) where they are needed.
Mother Nature has produced some of the worst threats to humans — HIV, TB, and the Ebola virus, to name a few. Bionanodevices will revolutionize medical diagnostics, making sophisticated blood/urine/saliva tests inexpensive and routine operations at the doctor’s office. … when radically new technologies are developed, social and ethical issues can arise.
The latter is composed of sentences extracted from the National Nanotechnology Initiative: Leading to the Next Industrial Revolution, a report prepared by the US Interagency Working Group on Nanoscience, Engineering and Technology of the National Science and Technology Council’s Committee on Technology… and addressed by Bill Clinton’s White House to members of Congress on February 7, 2000.
This article is a collaboration between a scientist and a philosopher. It tackles a very important issue : how do wrong beliefs sustain themselves in spite of lack of supporting evidence? Specifically, the article considers the belief in nanoparticles diffusing through biological membranes, i.e. one that is the topic of many computational studies as well as being regularly cited as a source of concern (e.g. toxicity) and potential (e.g. drug delivery).
This blog article has benefitted from discussions with philosopher Yasemin J Erden. The analysis presented here was originally part of a Perspective article co-written with her and focused on the question of how wrong beliefs (such as diffusion of nanoparticles through membranes) persist for a long time. We have decided to separate the detailed critique of Liu et al’s article (below) from the general argument about persisting wrong beliefs. The Perspective article is currently submitted to a scientific journal and already available as a preprint.
I have discussed previously the article by Liu et al (here). However, I have since given a lot more thought to two points which had been troubling me: 1) given that there is no convincing experimental evidence that diffusion of nanoparticles across lipid membranes exists, how do the authors make the case for their theoretical study? 2) given that they include a short experimental section: how do they compare their theory with the experimental results?
HOW LIU ET AL MAKE THE CASE TO STUDY DIFFUSION OF NANOPARTICLES THROUGH MEMBRANES
They start by noting that some small organic compounds can diffuse through lipid membranes but that this is not the case for polar macromolecules. Then they make a general case for the importance of nanomaterials in biological applications. In table 1 below, I reproduce short statements from the Liu et al’s introduction which are supported by citations. I contrast what is likely to be understood by the readers with what is actually shown in the articles cited. This analysis reveals significant hype, but also the interesting fact that the claim for applications is made by citing, not articles reporting the existence of applications in biology and medicine, but instead chemistry articles reporting on the development of nanoparticles and making promises of future applications.
Table 1 also shows that, even though diffusion through membranes is the phenomenon they are modelling, none of the experimental articles cited in their introduction claim that nanoparticles diffuse through membranes. Instead, most cited articles report (correctly) entry of nanoparticles into cells by endocytosis. The erroneous claim that nanoparticles with one dimension smaller than 10-15 nm can enter by “diffusing in the hydrophobic region of the membrane and then on the other side” is made without reference. The introduction then moves to the review of theoretical and computational articles studying diffusion of nanoparticles through membranes. Thus, the case for studying the non-existing phenomenon of nanoparticles diffusion through membranes is a succession of four claims:
Some small molecules diffuse through membranes.
Nanoparticles have all sorts of important applications in biology and medicine (mostly, promises of).
Small nanoparticles can diffuse through membranes (they don’t).
All those previous theoretical efforts at studying diffusion of nanoparticles through membranes have shortcomings so more research is necessary.
Table 1: How Liu et al justify their study of the nanoparticles-diffuse-across-membranes (non-existing) phenomenon. References noted in brackets, e.g. [2-4], refer to Liu et al own bibliography.
Quotes from Liu et al introduction.
This may give the impression that:
Relevance of the cited experimental work to the question of diffusion of nanoparticles across membrane:
Tunable […] properties of […] nanomaterials enable engineering for a number of applications, such as drug delivery,[2−4] …
Nanomaterials are enabling drug delivery, or may do so soon.
Nanomaterials enable engineering for applications, but the cited articles are chemistry research articles making promises for the future. Ref 2 is so far from drug delivery that it does not even have cell biology experiments. Ref 3-4 are also mostly materials synthesis and characterization (some basic cell biology, no preclinical work).
Ref 2: None. Ref 3: None. The article report that the particles enter by endocytosis. They are supposed to escape later. No evidence, quantification or mechanism provided. Ref 4: None. The article reports that the particles enter by endocytosis.
Tunable […] properties of […] nanomaterials enable engineering for a number of applications, such as […], controlled release,[3,5]…
Nanomaterials are enabling controlled release, or may do so soon.
Ref 3, see above. Ref 5 is a highly-cited 2008 JACS communication. Mostly materials synthesis and characterisation with one figure showing cell viability and toxicity. 12 years later, the medical application of these materials (PEGylated Nanographene Oxide) does not seem any closer.
Ref 3: None. See above. Ref 5: The article reports that the particles enter by endocytosis.
Tunable […] properties of […] nanomaterials enable engineering for a number of applications, such as […], deep tissue imaging, and sensing of cellular behavior.[6−10]
Biologists and medical doctors are using those nanomaterials for deep tissue imaging, and sensing of cellular behaviour, or may do soon.
Ref 6-10 are chemistry papers largely describing the synthesis of new materials which could, according to their authors, be useful for deep tissue imaging etc. Some of these are 5+ years old but no biologist currently use these materials for their imaging or sensing needs.
Ref 6: No comment on mechanisms of uptake. Ref 7: None. See also PubPeer comment.9 Ref 8: None. The article reports receptor-mediated endocytosis. Ref 9: None. Ref 10: None.
Nonetheless, over the years, a number of systems have been reported to give cytoplasmic access to biomacromolecules, most notably cell-penetrating peptides, supercharged proteins,[18,19] and bacterial toxins and different types of NPs.[20−22]
Those strategies are an effective strategy for cytoplasmic access.
In fact, effective delivery to the cytosol remains a highly challenging task as some of these papers make clear. In the rare cases where cytosolic delivery is quantified, authors find that the proportion of nanomaterials reaching the cytosol is very small.10 That is also our experience, e.g. with TAT.11
Ref 17: None. Entry into cells is by endocytosis. Endosomal escape is supposed to involve specific membrane disruption mechanisms. Ref 18: None. Entry into cells is by endocytosis. Endosomal escape is supposed to involve membrane fusion between the lipid-coated “super-charged protein” and the target cell. Ref 19: None. Entry into cells is by endocytosis. Endosomal escape is via a protein pore. Ref 20. None. It is a 20 years old review of synthetic DNA Delivery Systems. Ref 21: None. It is a review of extracellular vesicles for drug delivery. It includes very little on mechanisms of uptake and no suggestion it could be by diffusion through the membrane. Ref 22: None. It is a review of nanoparticulate drug delivery system. In the section on accessing intracellular targets, there is no mention of any potential role of diffusion through membranes.
For particles with the smallest dimension larger than the membrane thickness, approximately above 10−15 nm, the permeation is generally controlled by membrane deformation  and endocytosis .
General rules about what controls permeation for “particles with the smallest dimension above 10-15 nm” have been obtained.
Multiple levels of confusion. Ref 23 is about graphene sheets; their smallest dimension is well below 10-15 nm (and there is no reason to extrapolate any general rule from that particular study). Ref 24 is about endocytosis but the smallest dimension of the smallest particle tested is 100 nm so it cannot tell anything about a supposed threshold around 10-15 nm.
Ref 23: None. The experimental section of the article uses graphene oxide sheets, which are 200 to 700 nm in lateral size. There is no experimental evidence that such objects could diffuse through membranes. Ref 24: None. Entry into cells is indeed by endocytosis.
Smaller nanoparticles can instead cross the membrane by passive transport, that is, by displacing, sometimes irreversibly, the lipids or by diffusing in the hydrophobic region of the membrane and then on the other side.
Particles with the smallest dimension smaller than 10-15 nm can cross membranes by passive transport.
No reference provided for this essential claim. There is no serious evidence for its validity – and there is plenty of conflicting evidence. In fact, if true, most proteins would diffuse through membranes, which is obviously not the case.
THE MODEL AND THE COMPARISON WITH EXPERIMENTS
Liu et al start building their model from an observation. But not an observation from physical reality. They observe that in a molecular dynamics time series of a nanoparticle close to a lipid membrane, nanoparticle penetration into the membrane is happening at the same time as a low-density fluctuation of the lipids. From that observation, they build an ad hoc phenomenological theoretical model composed of four mechanistic steps. Those four steps are not illogical, but they are also not the only possible ad hoc description available (e.g. a fifth step could be added accounting for the rotation of the nanoparticle in the correct orientation that matches the low density area). How those steps are transcribed into mathematical equations and then into predictions also involves several somewhat arbitrary decisions (e.g. a density threshold defining what is a ‘low density area’ and a decorrelation threshold defining a decorrelation time). This results in underdetermination, i.e. even if the model were in agreement with experimental results, it would not give much confidence that it really describe the physical reality.
In contrast to many theoretical articles, Liu et al do include experimental results. Liu et al’s model predicts, as the title of the article indicates, the “time of entry of nanoparticles in lipid membranes”. But what does ‘time of entry’ actually mean? How is the model tested experimentally?
Permeability, not time of entry, is usually measured to characterise transport through a membrane. The permeability relates the flux of a substance (permeant) through a membrane to the difference in the permeant’s concentration on either sides of the membranes1:
The permeability depends on the permeant and membrane properties. The difficulty with time of entry is that whilst it may seem to have an intuitive meaning, it is not a well-defined physical parameter: it does not just depend on the permeant and membrane properties. Instead, the time a nanoparticle takes to enter a membrane (or more precisely the statistical distribution of the times of entry) will depend on the starting position of the nanoparticle, and in particular its distance from the membrane. In Liu et al’s molecular dynamics simulations, the nanoparticles are confined within 2 nm of the membrane by ‘a harmonic potential that pushes the particles toward the membrane when its distance from the membrane interface exceeded 2 nm’. Altering or removing this confinement would dramatically alter the time taken by a nanoparticle to enter the membrane. By reading very carefully Liu et al, one can conclude that the quantitative definition of what they call time of entry Tentry is in fact as follows: molecules or nanoparticles confined within 2 nm of the membrane will penetrate that membrane with an average rate of 1/Tentry .
To compare with other models or with experiments, one needs to relate this rate with permeability. Here, we face two problems: 1) the fact that Liu et al’s model only considers the first half of the process, i.e. the entry of nanoparticles into membranes, whilst permeability relates to the passage of permeants through the membrane, i.e. the entry and the escape; 2) the need for a quantitative relationship between the rate of passage and the permeability. Thankfully, both problems can be solved as we will show below.
To address the first problem, one can easily extend Liu et al’s model by applying their approach to the escape. Liu et al’s model is shown in Figure 1 (top). The nanoparticles they studied are highly lipophilic. They report Pt, the probability of the NP to move from the hydrophilic phase to the hydrophobic phase, as being equal to 100% in all cases (table 1 and 2 in Liu et al). In such a scenario, if a particle encounters an LDA, they will always enter as their affinity for the lipid phase is infinite. But, with Pt equal to 100%, particles would never escape from the membrane (Figure 1, bottom): they would just accumulate in the lipid phase as the probability of escape from the lipid phase would be 0. In a private communication, the authors have provided a finite value for the partition coefficient Kd (that they deem appropriate for the particles used in their experimental validation). This value can be used to calculate Tescape, the time of escape using exactly the same reasoning as Liu et al (Table 2). One can observe that for such lipophilic particles, the time of escape is three orders of magnitude larger than the time of entry (therefore a temporary accumulation in the membrane should be observed which does not seem to be the case from the images shown in the article).
Finally, to compare between the model, the extended model, and the experiments, one needs to convert the rate of entry, or rate of passage, into a permeability. If one considers a short time interval delta-t, the amount of nanoparticles crossing an area of the membrane is:
The amount of nanoparticle crossing an area A of the membrane in a short time interval delta-t can also be obtained by calculating the number of particles at a distance smaller than h =2 nm, i.e. A*h*Cin and multiplying that number by the rate of entry:
By comparing equations (2) and (3) above, one can deduce that there is a simple relationship between the permeability and the time of entry:
Using equation 4, one can now compare the results of the experiments of Liu et al and the predictions from the models (Table 3). The experiments were done for three membrane compositions and one type of nanoparticles. The permeability was measured by quantifying the leakage of the nanoparticle from a spherical vesicle. First, one can remark that there is a very large uncertainty over the permeabilities measured by Liu et al for the first two membrane compositions, to the extent that they are not significantly different from each other and not significantly different from zero either. This large uncertainty is due mostly to the fact that over the course of the experiment (60 minutes), only a small percentage (less than 20%) leakage is observed for these two membrane compositions, and, combined with the data variability, it is not possible to measure an exponential decay with any confidence. Second, one can remark that the entry-only model predicts values which are 6 orders of magnitude larger than the observed permeability (the predicted values are larger than what has been observed for water – clearly unrealistic for nanoparticles). Third, the extended model predicts values that are still three orders of magnitude too large. The conclusion therefore should be unambiguous: the model’s predictions do not match with physical reality. This is however not the conclusion reached by the author.
Table 3: Comparison between the models and the experiments.
Permeability measured by Liu et al (Fig 9b of Liu et al), nm/s
Permeability deduced from the entry-only model (nm/s)
Permeability deduced from the extended model (nm/s)
0.2 ± 0.7
1.0 ± 1.8
3.9 ± 0.7
Liu et al do not discuss the fact that their model only describes entry into the membrane and not the escape. They do not discuss the very large uncertainty of two of their measured experimental permeabilities. They do not deduce an absolute relation between rate of entry and permeability. Instead, they postulate the following equation which has the same form as equation (4):
They then calculate c from a fit of the experimental data (Liu et al, private communication). Here we see the effects of underdetermination in practice. Liu et al bundle what they have not modelled, about which there is uncertainty or a lack of knowledge, into a constant. This in turn means that the constant is therefore unavoidably vague. And its value is determined by a fit of the data that are supposed to test the model. Unsurprisingly they obtain ‘a very close match’ and they conclude that the experiments validate their approach.
He has kindly responded by email and has allowed me to reproduce it here. As usual, the comments box welcomes further discussion.
(((me))):- But, to the question, “where have all the nanoscientists gone?”, my answer would be markedly different. They have gone home to save lives, including those of “our clinical and healthcare colleagues, who work on the frontline of the pandemic”. Right and proper.
Kostas:- If the ‘cancer model’ of early detection, monitoring and targeting (not only in treatment and confinement but ALSO targeted protection) was taking place early on (as some countries attempted to do) the biomedical scientists SHOULD have been mobilised to creatively contribute against this biological outbreak immediately, instead of being locked in at home asked to binge-watch Netflix.
(((me))):- It may be hard to hear that we, “highly trained etc” are non essential whilst people working in our local supermarket are essential. But it is nevertheless a fact.
Kostas:- It is not about being ‘highly trained’, ‘more valuable’, or ‘highly educated’ at all. It is about mobilising your resources to find solutions. If politicians were thinking clearly, free from political manipulations and weightings of voter swings, and the system was properly prepared, we should not have construction workers and nail parlours considered more ‘essential’ than biomedical scientists at a time of biological crisis. That is wrong. I do NOT have anything against any of the above professions at all and as you know I am not snobbish or socially exclusive. But biomedical scientists should be essential in biological crisis. Same as physicists in a radiological crisis.
(((me))): Re the cancer analogy section, on one level, I agree: the 3 principles are key, but they are what everyone (e.g. WHO) is saying : the analogy does not add much.
Kostas:- This is not correct. No country has shown either the willingness and decision or has the capacity to apply all three principles as early in the outbreak as possible. My hypothesis is because everyone thinks it is a ‘biological tsunami’ that will go away. The essence of what I tried to express in the Nature Nanotechnology article is that the situation should not be considered a ‘tsunami’, but a chronic condition. And we have learnt how to deal with chronic conditions better through the years, as in the case of cancer. Yes, some countries and the WHO have applied some of the principles (e.g. test, test, test) but the disasters in most countries in Europe at least have happened and are happening because ‘targeted protection’ did not take place at all. Most elderly care homes were not targeted for protection. Most hospitals that suffered dramatic collapses were not protected. And we are still hearing about lack of protection for a lot of our healthcare, frontline staff.
(((me))): On another level, I draw different conclusions from looking at “cancernanotech” in the context of COVID. Cancer nanotech is an area where vast amount of money have been invested on the basis of hyped promises and flawed concepts. Not exactly a model to follow right now.
Kostas:- I know your views about cancer nanotech and I disagree, but I am happy to keep discussing with you and other critics or sceptics. This discussion I think is for another time though.
Video recording of a talk given by Raphaël Lévy (University of Liverpool, UK) at the European Association for Cancer Research meeting: Nanotechnology in Cancer: Engineering for Oncology Cambridge, UK : 12 – 14 September 2019
So I did. Not the details of the simulations (beyond my expertise) but the introduction and some key aspects of the work, including the choice of nanomaterials and the comparison with experimental results.
I have annotated the article using hypothes.is. You can read those annotations here. There are several things in the introduction which are pretty common in the bionano litterature but nevertheless irritating.
So, for example, if you read the title and you are not an expert in this field, you will be thinking that nanoparticles in general enter lipid membranes. But do they? The authors (nearly) clarify in the introduction that in fact they don’t, e.g. “… proteins or nanoparticles (NPs) […] hydrophilicity and large size hamper direct diffusion through the membrane lipid bilayer” and ” In general, cells do not permit access of polar macromolecules to their cytosol, and phospholipid membranes constitute an effective barrier“. But then, they also state the contrary, i.e. that nanoparticle do penetrate cell membranes (science fiction?), e.g. “Smaller nanoparticles can instead cross the membrane by passive transport, that is, by displacing, sometimes irreversibly, the lipids or by diffusing in the hydrophobic region of the membrane and then on the other side“. That extraordinary statement is not supported by references. Which one is true? How do we reconcile these contradictory statements?
The introduction then moves away from the experimental world towards the world of simulation. That is fine (and largely beyond my expertise), but it means the rationale for studying nanoparticles penetration in membranes rests on the juxtaposition of two contradictory statements, one of which is not backed by experimental results.
So the introduction (and title) is about nanoparticles (in general) going through membranes, but the nanoparticles modeled (see below, reproduced from their Figure 8) are three tiny (even for nanoparticles) objects. Their sizes are certainly not typical of contrast agents or drug delivery vehicles discussed in the introduction.
And those nanoparticles have another feature really not typical of nanoparticles for biomedical applications: they are extremely hydrophobic. Their partition coefficient between lipids and water is 100% (table 1 in the paper). This is very far from your typical nanoparticle. And it raises another question related to the experiments presented in the last figure of the article.
UPDATE 24/09 (following Alan’s comment):
Fig 9 (reproduced below) shows the “leaking” of GQDs from giant vesicles (~50-70 micrometers). These are spinning disk confocal images taken with a 60X objectives. So we should see a section through the vesicle (I think it is more likely to be a total intensity projection but they don’t say).
If those “particles” love the membranes so much, they should simply accumulate in membranes, not diffuse through. Why don’t we see any accumulation in the membranes in Figure 9?
I can’t work out exactly the details of the comparison between the experimental data and the simulations so I will get in touch with the authors.
I am in charge of a module entitled “Advanced Skills for Biochemistry“. Our third year Biochemistry (Honours) students take this course. One of their tasks is to prepare and present a poster on a hot topic or technique. I have therefore asked the world (via Twitter) and my colleagues at the Institute of Integrative Biology to come up with suggestions of topics for these posters, as well as references that students could use as a starting point.
Evolutionary covariance for protein structure prediction; suggested by Dan Rigden via email: “The topic of evolutionary covariance, with myriad uses but particularly for protein structure prediction, goes from strength to strength. Unfortunately, Google decided not to make the code available or (I think) to publish anything in a journal [there’s a bit of a separate lesson to the students there]. However, they can read about it here and here. This paper, out this week, is the most similar approach I’m aware of and works extremely well. It has a server (that the students could try…) and the code is available.”