Interview: Armin Knoll
Rocking Brownian motors go viral
"If you are really able to manipulate the energy landscape you can manipulate the behaviour of the particle"
Armin Knoll from IBM Zürich is working on rocking Brownian motors — nanofluidic cells with nanoscale ratchet-like surface structures. These devices can sort nanoparticles with down to 1 nm difference in radius, and are very promising for diagnostics and protein purification. Dr Knoll told The Lithographer about these nanofluidic devices and their future applications.
The Lithographer (TL): Hi Armin, thanks for finding time for an interview. First of all: what are Brownian motors, and why are they called like that?
Armin Knoll (AK): Let's begin with a Brownian motor definition. It was discovered in the 90's that biology uses so-called molecular motors to transport molecules in cells, for example to do cell division, to actuate muscles, and so on. When people studied the basic principles of molecular motors they also began trying to build artificial machines based on the same principles to do similar things — transport particles or other cargo along a guiding track. This is when the term Brownian motor was coined by Peter Hänggi, a Swiss physicist. Brownian motors are called so after Lord Brown who discovered Brownian motion in 1827 — the thermal motion of particles in liquids. Temperature activates the movement of all molecules in the liquid, so what you actually see when these particles are moving around is the result of water molecules randomly bouncing off them. The energy input that makes them jiggle around is what you want to exploit in a Brownian motor, i.e. a device that uses Brownian energy to transport the particles inside the nanofluidic cell. On top of that, we have to do two things: create an energy landscape inside the Brownian motor and bring the system out of equilibrium. The Brownian motion enhances the properties of the motor.
TL: Why are they rocking?
AK: For a Brownian motor in a nanofluidic cell, we need to create an asymmetric energy landscape for the particles, sawtooth-like, for example. One way to make such a system is to switch the energy landscape on and off. When it's switched on, the particles will be drawn to the lowest point of the energy landscape, and when it is off, they will diffuse in all directions. When the potential is applied again, some particles will have moved closer towards the next minimum, so — because the energy landscape is asymmetric — in average you get transport of the particles. This implementation is called flashing Brownian motors. Our implementation of a Brownian motor is different. We don't switch the landscape on and off but rather keep it static. Instead, we apply a zero-mean external force. It is applied in forward direction with the same magnitude and duration as in the backwards direction, so the force cancels to zero for the particles. It is the asymmetric potential that rectifies the motion of the particles, so they only go in the forward direction. And that is why they are called rocking: we rock the potential back and forth. The particles feel the applied force as a potential with a linear slope, which means that you tilt the potential to one side and then to the other side, and so on. That's where the name comes from.
If you are really able to manipulate the energy landscape you can manipulate the behaviour of the particle.
TL: What can such devices be used for?
AK: First of all, Brownian motors transport the particles through liquids, and that's what they are used for in biology, too. The only prerequisite in our case is that the particles need to be charged. And if that is fulfilled the Brownian motor works with DNA, with proteins, gold nanoparticles, viruses — all this stuff can be transported.
TL: So the shape of the particles doesn't matter?
AK: The size of the object needs to be smaller than the length of our sawtooth unit. Around half the size is a good number. Transporting large DNA molecules would be a problem, you would have to make really large ratchets. Small DNA molecules can be transported directly, but if DNA spans several teeth of the motor, it would get stuck.
TL: How did the idea of particle transport in liquid lead to sorting?
AK: In 2010, we already had this nanofluidic project running, and since we have the capability to write very high-precision patterns, we were looking for applications. A paper from Madhavi Krishnan appeared more or less at the same time, where she showed that the geometrical recess of a surface can produce an energy landscape, at least a potential minimum. So, we thought it would be a nice fit to combine single-digit nanometer precision and the ability to generate potential energy landscapes. Energy always controls the behaviour of particles, and if you are really able to manipulate the energy landscape you can manipulate the behaviour of the particle.
A nice thing about lithography is that you really control all the energy barriers because you write them. When you know the energy barriers, you know the timing in the system, you can tune it from the top down. For Brownian motors, in our implementation at least, the particle transport strongly depends on the size of the particle because a bigger particle feels an amplified energy landscape. So, a larger particle rests closer to the interfaces of the nanofluidic channel, and therefore, the interaction energy amplifies exponentially. The energy landscape needs to be stronger than the thermal fluctuations, which is at least 3–4 kBT. But you don't want it to be at 20-30 kBT, because then you don't have sufficient force to tilt the energy landscape anymore. At around 5-10 kBT, the transport works nicely. And now, if there are different-sized particles, the smaller ones might not see enough of the energy landscape, so they will not be transported, and the ones that are too big will get stuck. But there is a size range in which the particles will be nicely transported. When we built our motor it was clear that this would happen. Then, we designed a slightly more complicated device optimized to achieve very high efficiency in separating particles. [1]
A nice thing about lithography is that you really control all the energy barriers because you write them. When you know the energy barriers, you know the timing in the system, you can tune it from the top down. For Brownian motors, in our implementation at least, the particle transport strongly depends on the size of the particle because a bigger particle feels an amplified energy landscape. So, a larger particle rests closer to the interfaces of the nanofluidic channel, and therefore, the interaction energy amplifies exponentially. The energy landscape needs to be stronger than the thermal fluctuations, which is at least 3–4 kBT. But you don't want it to be at 20-30 kBT, because then you don't have sufficient force to tilt the energy landscape anymore. At around 5-10 kBT, the transport works nicely. And now, if there are different-sized particles, the smaller ones might not see enough of the energy landscape, so they will not be transported, and the ones that are too big will get stuck. But there is a size range in which the particles will be nicely transported. When we built our motor it was clear that this would happen. Then, we designed a slightly more complicated device optimized to achieve very high efficiency in separating particles. [1]
a nanofluidic Brownian motor setup: a: Schematic cross-section and top view of the nanofluidic slit. B: The top view (not to scale) depicts the electrodes (yellow, spaced by ~1.2 micron) and the pillar in the center (width ~100 micron). C: The ratchet topography in polyphthalaldehyde and the 60-nm gold nanoparticle, drawn to scale. The particle experiences a ratchet-shaped energy landscape due to the ion-cloud interactions (orange) between like-charged surfaces. The gap distance is 150 ± 1 nm. [1]
TL: In one of your papers, you say that the theoretical limit of the size separation of the particles is 1 nm, in radius. Do you think you can achieve it?
AK: Actually, if you look at it theoretically, it can be even finer. But due to fabrication constraints set by the tools we have at our disposal you cannot be more precise than about 1 nm, so this is in the end where this limitation comes from. But also the physics "stops" at around this point. Even if you had a perfect device, sorting particles by 1 Angstrom would be very tough.
TL: Have you actually built such a system yet?
AK: We are currently working on that. We have a few demonstrations now. They don't reach 1 nm yet but they are on the order of a few nanometers, and we should get down to 2 nm at least.
TL: How do you engineer the energy landscape?
AK: A basic motor geometry is very trivial, you just need a saw-tooth topography with linear profiles. We pattern them on the bottom of the nanofluidic cell, and the topography exponentially translates into the energy landscape. Actually, the resulting energy landscape is exponential, but this effect is not very strong. Ideally, the sawtooth slopes should be logarithmic to create linearly rising energy landscapes, but it doesn't make a big difference.
TL: And what are the dimensions?
AK: The dimensions depend on the concentration of salts in your solution. The more salt you have the lower will be the range of the interaction between particles and walls. We typically work with salt concentrations in the millimolar range, which gives you a size range of about 10 nm, and the tooth height should be on the order of a few times more, so we usually work with 20–30 nm high "teeth". It is important to explore the entire energy range. If a particle is too far from the surface it no longer "feels" its effect, so the energy will be flat. If the particle is close to the surface, however, then you see the right energy landscape, about 10 kBT or so. So, we have to make sure we don't recess the topography too much.
TL: How sensitive is the energy landscape to the precise topography?
AK: During the motor operation 30 nm high topography features result in 10 kBT energy landscape. Therefore, 1 nm of topography corresponds to approximately 0.3 kBT. So, an error of 1 nm in your device produces 0.3 kBT error in the energy landscape, which is acceptable because it's less than kBT.
TL: 1-nm accuracy in the structure height is not trivial to achieve. How do you do it?
AK: We use thermal scanning probe lithography (t-SPL), or NanoFrazor technology, and polyphthalaldehyde, the typical polymer resist for t-SPL, to write these landscapes. It is a well-established process for us. There are some tricks we have to use, of course. For example, when you put a piece of silicon with the polymer on top in water, the polymer floats off. We control the silicon-polymer interface by adding a cross-linked polymer in between. The cross-linked polymer attaches very firmly to the substrate, and the interface between our resist and the cross-linked polymer is very stable because both are hydrophobic. We can also use other methods to stabilize the stack, such as removing the oxide by HF treatment from the Si interface, or silanisation of the surface. But, typically, we work with crosslinking.
TL: You also made a few videos of particles being sorted. It is fascinating that one actually see it! How do you detect what happens in the cell?
AK: It was quite a tricky setup to build. Luckily, we could use the knowledge of other people to make it work. The two slides forming the nanofluidic cell are contacted by electrodes to apply the potential. We shine a laser onto the system and record the interference of the light scattered by the particles and the substrate. In this way, we can detect very small particles with high frame rates of up to 1000 frames per second. Currently, the optical setup is quite large and complex but it could be reduced to a size of a small box with very simple parts.
On-demand shuttling of 60 nm Au particles between reservoirs. A sequence of electric field directions (bottom left) leads to the transfer of the particles into the bottom left reservoir. [1]
Slow-motion (factor 3) movie for sorting of nominally 60 and 100 nm-diameter particles. An electric field in y-direction of 4 V @ 30 Hz is applied after ~5 s. [1]
TL: What are the current bottlenecks and the goals of this project?
AK: We demonstrated that we can separate particles with a device just a few tens of microns in size, achieving a complete separation of populations within seconds. Then, we extrapolated which kind of devices we could make, which would be relevant for real-world applications. One could build devices that continuously separate the particles, for example. There would be input and output channels, and the output channels would deliver different particle sizes. Unfortunately, the throughput of such a device would not be very high. In order to get quantitative amounts of liquids, a lot of devices need to be stacked together, which is very complicated to manufacture. A simple device, on the other hand, can already handle the detection of nanoparticles in liquids. Our target at the moment is to detect small viruses in water. We think that we can build a device that can detect attomolar concentrations of viruses within 1 hour. And that is interesting for surveillance of water samples: to check if they contain viruses. Since we can transport the viruses, we can also trap them locally so they cannot escape. We simply write a deep pocket and then they are trapped there. Then, it is easy to detect these viruses even without special markers on them. This way, we can realize markerless virus detection. And this would be a device that cannot be made/built with any other technology at the moment.
TL: How expensive would it be?
AK: If we make the devices as we do them now it would be quite expensive. But, in principle, the system is based just on a topography on a polymer surface so you could use nanoimprint lithography to replicate the devices. We will even try injection moulding. That would be really cheap. If we could make the devices by injection molding it would open a large use space. But we have to find out how to do it first.
TL: Are you developing a high-throughput manufacturing method for such devices?
AK: To replicate and scale up the devices? Of course. Thermal probe lithography can address small areas, and in order to manufacture for the big market we should find a way to replicate the nanostructures. This is a general problem – not just of our devices but of the nanolithography techniques that operate on small areas in general. Parallelized t-SPL wouldn't solve the problem with the throughput either. Engineering-wise, implementing a large number of cantilevers is a big effort. This approach will never reach the prices and costs of the cheap techniques: if injection molding works, you can never beat this price, there is no chance.
TL: What are the other challenges of the project?
AK: We would like to achieve particle sorting in bodily fluids. It would be extremely helpful for the doctors and the patients if our devices could detect viruses in bodily fluids like blood or sweat within a short time frame at attomolar concentrations. Then, it would be possible to immediately determine if there is a viral infection. However, the salt concentration is too high, so we would have to dilute the blood before we can test it. The other complication is that bodily fluids are very complex systems. They contain a wide range of different molecules and particles, and we are not sure how the motor will function in such conditions. We will have to find out how good our motors are at rejecting these other things. In principle, as I said, the transport only works for certain object sizes (within a certain object size range), so it should be more or less robust. But if things start to attach and coat the surface, blocking becomes a problem. For this kind of applications there is still a long way to go.
TL: A recent paper from your group shows reversal of the particle current direction [2]. Could you please tell a little bit about that?
AK: The reversal of the transport direction in a motor is something that has been known for a long time, but nobody has done it for overdamped systems. Particles in fluids are heavily overdamped; if you would apply a force to a particle it would stop within less than a nanometer or so. It was not clear if the particle reversal would happen in such a system. It was predicted theoretically, and we thought we'd check our system, and indeed this current reversal happened at higher frequency. It has to do with how fast particles move in a potential. If you apply the force for too short a duration, so they cannot reach the next element, then they never jump forward but there is still a low probability for them to jump over the potential barrier in the backward direction. At some point, this probability is higher, and you get transport in the other direction.
TL: Would it also be of interest for pharma industry to separate very small molecules and proteins, since it is not possible to separate nanoparticles scale by size?
AK: After my presentations people often approach me with questions about separating proteins, for example. We plan to go in this direction soon. But at the moment, it is not completely clear how it is going to work. We would also need to work at reduced salt concentrations, which means the proteins would not be in their native environment. But, in some cases, it is not a problem, so people are interested in this approach. They want to purify and crystallize the proteins in order to study their structure. It would be a very nice use case for us if we were able to get quantitative information on the number of proteins separated from the residues and to extract them into a separate container.
TL: The reversed particle current must be slower.
AK: Indeed, it is a factor of 30 slower than the current in forward direction. But, in principle, it could also be interesting for separation and sorting because it is also frequency and size dependent.
TL: Do you have other applications in mind?
AK: At the moment, we are focused on controlling the nanoparticles in nanofluidic cells because we think we have established something new, and we want to see what kind of applications we can get out of that. But we want to go beyond use for mere transport since we can also trap the particles, transport them to a compartment or specific locations [3], and increase their concentration for analysis or detection.
For virus detection, size separation is interesting because each virus family has its distinct and rather monodisperse size. So, knowing the size already gives you a hint about what kind of virus is there, and then you can still do DNA or RNA identification to really find out which specific virus from this family you have. But you would know if you have a family that might be dangerous. At the moment, the detection of viruses is based mostly on PCR-amplified DNA and RNA of viruses, so you always need to know what kind of virus you are dealing with (Polymerase chain reaction, or PCR, is used to copy a specific DNA segment billions of times in vitro. PCR is used to diagnose deseases, identify pathogens or match criminals to crime scenes.) Otherwise, you have to do metagenomics — sequence all the DNA in the sample and then analyse it using bioinformatics tools. Currently, there's no method to operate directly.
So, here everything comes together: transport, size sorting and accumulation in a separate zone for detection. We are trying to establish a toolbox that we can operate with different nanofluidic systems and complement it with other things, for example, electric gates for steering transport, switching it on and off, and things like that. This is the main thrust of our work.
For virus detection, size separation is interesting because each virus family has its distinct and rather monodisperse size. So, knowing the size already gives you a hint about what kind of virus is there, and then you can still do DNA or RNA identification to really find out which specific virus from this family you have. But you would know if you have a family that might be dangerous. At the moment, the detection of viruses is based mostly on PCR-amplified DNA and RNA of viruses, so you always need to know what kind of virus you are dealing with (Polymerase chain reaction, or PCR, is used to copy a specific DNA segment billions of times in vitro. PCR is used to diagnose deseases, identify pathogens or match criminals to crime scenes.) Otherwise, you have to do metagenomics — sequence all the DNA in the sample and then analyse it using bioinformatics tools. Currently, there's no method to operate directly.
So, here everything comes together: transport, size sorting and accumulation in a separate zone for detection. We are trying to establish a toolbox that we can operate with different nanofluidic systems and complement it with other things, for example, electric gates for steering transport, switching it on and off, and things like that. This is the main thrust of our work.
We can separate particles with a device just a few tens of microns in size, achieving a complete separation of populations within seconds.
TL: What is the outlook for this project?
AK: We want to explore the limits in terms of particle size for interesting molecules like proteins. We also want to demonstrate that such devices can work with very small concentrations and still be able to detect the particles. This small world seems to be largely unexplored. The field of metagenomics appeared 10–20 years ago. For example, in one study, people analysed 200 litres of sea water. They filtered and then sequenced it. And they found more than 5000 different viruses. More than 99% of them were not known at the time. There seems to be so much going on at this scale!
References:
[1] Skaug et al., Science 359, 1505–1508 (2018)
[2] C. Schwemmer et.al., Phys. Rev. Lett. 121, 104102 (2018)
[3] S. Fringes et.al., Nano Letters (2019)
[1] Skaug et al., Science 359, 1505–1508 (2018)
[2] C. Schwemmer et.al., Phys. Rev. Lett. 121, 104102 (2018)
[3] S. Fringes et.al., Nano Letters (2019)
You can access the NanoFrazor Explore fact sheet or send a request to the team of Heidelberg Instruments Nano here.