Skip navigation

Stanford University

Stanford Microfluidics Laboratory

Microsecond Mixing for Investigation of Protein Folding
-in collaboration with
Lawrence Livermore National Laboratory

Important structural events in protein folding are known to occur on sub-millisecond time scales, but studies of fast protein folding kinetics have been limited by mixing rates of protein and denaturant solutions. Simple proteins often have folding rates on the order of 10 microseconds, whereas fast mixing experiments have yet to probe these short time scales.

Project Description
We have developed a diffusion-based microfluidic mixer to investigate protein folding on a microsecond time scale. Our mixer uses hydrodynamic focusing of pressure driven flow to create a sub-micron wide stream of protein and denaturant solutions. The denaturant diffuses out of this thin sheet faster than the protein and the proteins see a local drop in denaturant concentration causing them to fold.  The mixer geometry and flow conditions are optimized with numerical simulations, and we use micro-PIV to visualize velocity fields in the mixer. 

Figure 1. Schematic of mixer and SEM of silicon device. A) Red is denatured protein solution, blue is buffer. A solution of denatured proteins is injected from the north channel and hydrodynamically focused into a thin sheet with buffer from the side channels. The denaturant diffuses out of the focused stream inducing the proteins to fold. The folding kinetics are then optically detected downstream.

Mixer optimizations

Our mixers use over eight orders of magnitude less labeled protein sample mass flow than a previously reported ultra-fast protein folding mixer, with flow rates of 3 nl/s and protein concentrations of tens of nanomolar.  We have demonstrated the fastest mixers of any kind, and can mix two streams in a few microseconds.  Our work includes the design and optimization of mixers using models of convective diffusion phenomena; and a characterization of the mixer performance using micro-particle image velocimetry, dye quenching, and Förster resonance energy transfer (FRET) measurements of single-stranded DNA. 




Figure 2. Simulations of flow field in mixer including velocity field and convective diffusion of denaturant molecules from the center stream to the outer, sheath streams (on the left).  These simulations have been validated using micro-PIV experiments (Hertzog, 2004).  On the right is an oblique view of streamlines at relatively high Reynolds numbers (Re = 30), where flow becomes strongly three-dimensional and centripetal flow effects lead to dispersion and harder-to-control mixing.


Figure 3. An example mPIV velocity field obtained in the actual mixer and used to validate the model predictions and optimization process. (Image inverted for clarity). Maximum measured velocity here was 1.8 m/s. 


Our silicon mixer design is shown below.


Figure 4. Image of silicon microfluidic mixer.  (a) CCD reflected light image of mixing region showing  inlet channels with integrated filters, nozzle region at intersection, and exponential mixer exit region (south channel).  The exponential mixer allows observation of both microsecond and millisecond dynamics. Dots along the channel are distance markers.  (b) Schematic (mask) of the entire chip.  The inlet (center and side) and exit channels were sized (length and width) to suppress instabilities of focused stream from pressure controller fluctuations.  (c) SEM image of filter posts in inlet channels, just upstream of nozzle-region.  The filters consist of rows of posts spaced 1 to 2 mm apart to avoid clogging.  (d) SEM images of nozzles and mixing region.  Nozzle widths are ~1 to 3 mm wide and all channels are 10 mm deep.


Two counter-intuitive results

We have performed optimizations of mixer geometry and flow conditions (particular center-to-side flow rate ratio).  Our optimization calculations resulted in two important counter-intuitive results for the optimization of such mixers.  These are summarized below.



Figure 5. Simulation of convective diffusion (and subsequent mixing rate) of protein denaturant from center stream (a-c).  Model shows that the thinnest stream (c) is not the fastest mixing condition.  On the right (d) is a quantitative (experimental) image of species concentration in the optimized mixer. 


In the right-most flow field above, a thin center stream results in low advection velocities and mixing time is dominated by two-dimensional convective-diffusive transport in the “arrow-head” shaped focusing region.  On the right-most flow field, the transport in the focusing region is dominated by advection and mixing time is limited by diffusion in the spanwise direction (normal to streamlines) from the relatively thick center stream.  The fastest mixing time (for these conditions, less than 8 us) are for the center, intermediate case. 


A second counter-intuitive result is that, for near-optimal conditions, denaturants with higher diffusivities actually mix more slowly than denaturants with lower diffusivities.  That is, mixing time increases with diffusivity for near optimal conditions.  This second counter-intuitive result is due to the non-linear effects associated with convective diffusion in the initial focusing region.  Low diffusivity denaturants enable optimal mixer conditions with thin, low center stream flow rates which have spanwise-diffusion-dominated transport and minimize tmix.  High diffusivity denaturants require relatively thick, high center streams flow rates which exhibit substantial two-dimensional diffusion transport in the initial focusing region.


Experimental Mixing

We use techniques such as dye quenching and collapse of single stranded DNA to measure the mixing times of our mixers.  Figure 6a (below-left) shows results of a dye quenching experiment with two different mixer times.  We have also demonstrated feasibility of protein folding by measuring collapse and folding rates of a well studied benchmark protein (below-right).

Figure 6. (a) Results from fluorescence quenching experiments showing concentration of potassium iodide versus time for our original mixer design and a shape-optimized design.  Mixing is complete in approximately 7 ms for the original design and 4 ms for our newest optimized design. (b)  Experiment measuring folding rates of ACBP.  Signal indicates proximity of two fluorophores (FRET) attached to opposite ends of the protein.  A large signal indicates the protein is in a folded state.  Our measured kinetic rates at 90 ms and 8 ms agree well with published results.  By increasing the detection channel width downstream of the focusing region we are able to measure kinetics from under 10 ms to as long as 100 ms.

We have optimized, designed, and characterized a microsecond microfluidic mixer using hydrodynamic focusing, and have demonstrated as short as 4 ms mixing times with fluorescence quenching and collapse of single-stranded DNA.  Our mixer has been used to measure the folding rates of the protein acyl-CoA-binding protein (ACBP) and chymotrypsin inhibitor-2 (CI2), and is currently being used to investigate cytochrome-c with UV and three-photon detection techniques.

Our fastest mixers to date are able to mix in less than 5 us.


Hertzog, D.E., Santiago, J.G., Bakajin, O., "Microsecond Microfluidic MIxing for Investigation of Protein Folding Kinetics," Proceedings of the Seventh International Conference on Micro Total Analysis Systems, Squaw Valley, CA, USA. October 5-9, 2003


Hertzog, D.E., Michalet, X., Jager, M., Kong, X., Santiago , J.G., Weiss, S., Bakajin, O., Femtomole Mixer for Microsecond Kinetic Studies of Protein Folding.  Analytical Chemistry, vol 76, no. 24, 7169-7178. December 15, 2004 .