Fast microfluidic mixers are a valuable tool for studying solution-phase chemical reaction kinetics and molecular processes with spectroscopy. However, microfluidic mixers that are compatible with infrared vibrational spectroscopy have seen only limited development due to the poor infrared transparency of the current microfabrication material. We describe the design, fabrication, and characterization of CaF2-based continuous flow turbulent mixers, which are capable of measuring kinetics in the millisecond time window with infrared spectroscopy, when integrated into an infrared microscope. Kinetics measurements demonstrate the ability to resolve relaxation processes with 1 millisecond time resolution, and straightforward improvements are described that should result in sub-100 µs time-resolution.

Most solution-phase chemical reactions and biophysical processes are diffusion-controlled bimolecular reactions1–3 occurring when reactants are mixed under nonequilibrium conditions. Rapid mixing of two or more reactants is, therefore, the most versatile and common method of initiating chemical reactions and biophysical processes, such as protein folding and binding.4–6 By mixing reactants faster than the reaction rate, the equilibration of the system can be investigated using spectroscopic probes. For reactions in water at room temperature, the diffusion-limited rate constants, k, are 108–109M−1 s−1 for small molecules, as expected from the Smoluchowski expression, k = 4πRD, where R is the hydrodynamic radius and D is the sum of diffusion coefficients for the reactants (10−5–10−6 cm2/s). For biomolecular processes, such as those involving DNA oligos and proteins, this corresponds to the relaxation times, τ ∼ 10–100 µs in 1 mM solution. Therefore, to study such reactions, devices that can mix solutions into a homogeneous state within tens of microseconds are needed.

All mixing, whether in turbulent or laminar flow, is ultimately rate limited by molecular diffusion across boundary layers. To set the scale, the diffusion length scale for mixing in water, =4Dt1/2, is ∼30 µm for a 50 µs period. Therefore, designing a kinetic instrument capable of mixing in tens of microseconds necessitates miniaturization, typically mixing interfaces to the micrometer scale. This has resulted in great interest in microfluidic mixing devices, which generate chaotic advection and/or turbulence. These conditions in which the fluid motion becomes irregular cause quantities, such as pressure and velocity, to vary randomly in both space and time. Turbulent microfluidic mixer designs, such as zigzag channels,7 intersecting channels,8 convergent-divergent channels,9 3-D structures,10 embedded barriers,11 staggered herringbone structures,12 and tesla structures,13 have been proposed in the literature.14 A rapid mixing instrument should be able to both initiate the reaction by ultra-fast mixing of two species and monitor the formation and decay of all reaction products and reactants, as well as possible intermediates. Rapid mixing devices developed for use with fluorescence,15,16 UV–Vis,17,18 circular dichroism,17 and nuclear magnetic resonance (NMR)19,20 have been used to derive kinetic and mechanistic information on protein folding,21,22 enzyme catalysis,23,24 and polymerization.25 Infrared (IR) spectroscopy is sensitive to intermolecular interactions and changes in chemical structure without the use of exogenous labels, making it an excellent spectroscopic probe of molecular processes. So far, microfluidic mixers are not well developed for use with IR spectroscopy primarily due to poor mid-IR transparency of the current microfabrication material. Regardless, there has been a significant effort to develop IR compatible micro-mixers. Most offer modest time resolution of tens to hundreds of milliseconds, limiting their utility to slower biophysical processes rather than fast chemical reaction kinetics.26–32 

The challenge with incorporating IR spectroscopy into microfluidics due to poor transparency can be overcome by opting to fabricate microchannel in substrates that transmit in the mid-infrared vibrational fingerprint region (ν = 4000–400 cm−1, λ = 2.5–25 µm). CaF2 appears to be the most attractive choice because it is transparent throughout visible and up to 10 µm in the mid-IR allowing simultaneous use for visible and mid-IR measurements, in addition to having mechanical durability, chemical compatibility, and relatively low cost. Silicon is another material that is widely used because lithography is optimized for use with Si as a substrate;33,34 however, its high refractive index and lack of transparency to visible light makes it susceptible to problems with reflections and etaloning, and integration into microscopes. ZeSe, although transparent to redder visible light and longer IR wavelengths than CaF2, is less commonly used because it is significantly more expensive.35 

Although significant advances have been made to deal with IR compatibility, challenges, such as device durability and the ability to prototype sophisticated mixer designs, remain unaddressed. Early work from Lendl and co-workers paved a path for time-resolved IR spectroscopic measurement in IR compatible micromixers.36,37 They leveraged the advances in photolithography to fabricate diffusion-based mixers on Si substrates and bonded with an optically transparent CaF2 substrate to allow IR spectroscopy and microscopy measurements across mid-IR regions.38,39 Dyer and co-workers developed a new way to incorporate mixer designs that have microfluidic channels cut through a polymer spacer and sandwiched between two IR transparent windows.40 Other attempts of fabricating microchannels into CaF2 were based on wet etching with strong acids;41,42 however, CaF2 resistance to acids limits the design of the microchannel to a simple linear geometry. Although microfluidic devices constructed from polydimethylsiloxane (PDMS) are readily fabricated with soft lithography and are suitable for rapid prototyping, PDMS is not an optimal material due to strong absorption bands in the mid-IR. Krummel and co-workers reported a PDMS-based microfluidic device that could be used for IR spectroscopy by reducing the polymer thickness to hundreds of micrometers.43,44 These devices were used with transmission IR microscopy to study the enzyme kinetics of glucose oxidase. However, reducing the thickness of the polymer compromises the durability and mechanical strength of the device, which becomes crucial for mixers that require high flow rates and pressure drops for sufficient mixing.

Here, we describe the design, microfabrication, and characterization of a versatile and robust CaF2 based continuous flow device. The goal of this technology is to enable high structural and kinetic resolution, facilitating investigations of chemical reaction kinetics and biomolecular processes occurring in tens or hundreds of microseconds. The microfluidic channels are fabricated directly on a CaF2 substrate using photolithography, bonded to a drilled CaF2 window, and housed in a 3D-printed compression flow cell. We characterized the mixer performance using fluorescence microscopy and performed a pH dependent kinetic study of the reduction of ferricyanide to ferrocyanide by L-ascorbic acid (AA) using IR micro-spectroscopy. In practice, we were able to resolve kinetics as fast as 500 µs, which is limited primarily by the IR microscope’s 320 µm spatial resolution.

We have developed a method to fabricate microchannels directly onto the CaF2 window using photolithography, followed by bonding to a second drilled CaF2 window with heat and pressure. The schematic of the microfluidic mixing flow cell is shown in Fig. 1. It consists of one drilled 35 mm CaF2 window and one 40 mm CaF2 window with the microchannel fabricated on sandwiched layers of photoresist. The mixer is compressed in a 3D-printed hard-plastic housing, which directs the flow of fluid into and out of the drilled holes of the 35 mm CaF2 window. Five screws in the top flange of the housing are used to make an O-ring seal between the drilled CaF2 inlet and outlet ports and the bottom flange of the housing. The inlet ports of the bottom flange are connected by tubing to a high-pressure syringe pump that drives the reactants through the microchannels. In continuous flow mixers, the time-resolution of the measurement is governed by the time needed for the fluid to flow across the focused IR beam. This requires that the full device be integrated into an IR microscope with a working distance of less than 20 mm. A rapid prototyping of low-profile housing that incorporates fluid delivery and returns within the bottom flange is made possible by 3D printing. The total thickness of the current version’s flow cell, upon complete assembly, is 15 mm. This can be further reduced to fit the need for ultra-low working distance microscopes by choosing thinner plastic housing and CaF2 windows.

FIG. 1.

Design of the rapid mixing compression cell. (a) 3D printed top compression flange, (b) 1 mm thick, 40 mm ∅ CaF2 window onto which the mixer is fabricated, (c) 33 µm thick AZ 40XT photoresist film, (d) 1 mm thick, 35 mm ∅ CaF2 window with drilled inlet and outlet ports, and (e) 3D-printed bottom flange containing internal tubing connections, and 10 mm ∅ aperture for transmission measurement.

FIG. 1.

Design of the rapid mixing compression cell. (a) 3D printed top compression flange, (b) 1 mm thick, 40 mm ∅ CaF2 window onto which the mixer is fabricated, (c) 33 µm thick AZ 40XT photoresist film, (d) 1 mm thick, 35 mm ∅ CaF2 window with drilled inlet and outlet ports, and (e) 3D-printed bottom flange containing internal tubing connections, and 10 mm ∅ aperture for transmission measurement.

Close modal

The design of the microfluidic mixer is shown in Fig. 2. The mixer is fabricated onto a circular, 40 mm diameter window, incorporating three equally spaced inlet/outlet ports on a 15 mm radius, which are aligned to the drilled holes of the compression cell’s bottom window. Two inlet ports lead to the microfluidic mixing region positioned near the top of the accessible imaging region that exits into an observation channel and flows to the outlet port [Fig. 2(a)]. In addition to the mixer, the mixing region incorporates a 350 µm diameter sampling region into the input channels that transmit the entire focused IR beam and is used to acquire reference IR spectra of the input solutions and an air background [Fig. 2(b)]. Since minimizing mixing time is critical to our device design, we selected a continuous-flow chaotic zig-zag mixer geometry that has been previously reported to mix in sub-microsecond timescales.45 Computational fluid design simulations determined the turning angle of the zigzag to be the key parameter that affects the performance of the mixer with sharper turns yielding a higher degree of mixing.46 We modified the mixer geometries into “serpentine” and “inch-worm” [Fig. 2(c)], which has a turning angle of 23° to enhance mixing. The fabrication of a mixer with such sharp turns was made possible by direct-write lithography, which provides superior optical resolution compared to traditional mask-based lithography techniques. Both mixers consist of a “Y” junction made up of a 20 µm wide, 33 µm deep channel, followed by either a 210 µm long serpentine or a 130 µm long inchworm, which flares into a 350 µm wide observation channel to accommodate measurement with a commercial IR microscope with the beam diameter of 320 µm (1/e2) at the sample focal plane.

FIG. 2.

Design of microfluidic mixer. (a) Overview of the mixer lithographic fabrication onto 40 mm ∅ CaF2 window (red circle – 25 mm ∅ photoresist film), showing the two inlet ports, the mixing region, the observation channel, and the outlet port. The region accessible to IR imaging (10 mm ∅) with the full assembly is shown by the dashed green line. (b) Design of mixing region. Three reference regions are included to measure the IR absorption spectrum of the two input channels and to acquire an air background. The two inlets merge into a 20 µm channel at the beginning of the mixing region and flare out after the mixing region into a 350 µm observation channel. (c) Illustration of the serpentine and inchworm mixer designs. (d) Illustration of three observation channel designs: linear channel (7 mm long path), alternating channel (50 mm long path), and spiral channel (40 mm long path).

FIG. 2.

Design of microfluidic mixer. (a) Overview of the mixer lithographic fabrication onto 40 mm ∅ CaF2 window (red circle – 25 mm ∅ photoresist film), showing the two inlet ports, the mixing region, the observation channel, and the outlet port. The region accessible to IR imaging (10 mm ∅) with the full assembly is shown by the dashed green line. (b) Design of mixing region. Three reference regions are included to measure the IR absorption spectrum of the two input channels and to acquire an air background. The two inlets merge into a 20 µm channel at the beginning of the mixing region and flare out after the mixing region into a 350 µm observation channel. (c) Illustration of the serpentine and inchworm mixer designs. (d) Illustration of three observation channel designs: linear channel (7 mm long path), alternating channel (50 mm long path), and spiral channel (40 mm long path).

Close modal

To extend the time window for observations, the straight observation channel (7 mm path length) was extended to create a 40 mm long observation channel. To constrain a long observation channel within our 10 mm diameter imaging window, we employ two designs; a long serpentine with curved bends (“alternating”) and a circulating “spiral” channel, as shown in Fig. 2(d). At a flow rate of 700 µl/min, this extends the observation time window from 5 ms in the straight channel to 35 ms.

The positive photoresist AZ 40XT was selected for its high reproducibility with laser-based exposure and high solubility in organic solvents that allows the CaF2 window to be reused. This photoresist consists of phenolic novolak resin with diazonaphthoquinone sulphonate (DNQ) as a photo initiator. Exposure of DNQ to UV light triggers its conversion into an indene carboxylic acid (ICA) that increases the solubility of the phenolic resin by several orders of magnitude. The exposed area can then be removed in a suitable alkaline developer. IR absorption pathlengths vary by the type of applications, from less than 50 µm in water samples due to strong background absorption to greater than 100 µm for solutions in organic solvents. We targeted a pathlength of ∼30 to 40 µm as a compromise between the ease of depositing the photoresist reliably and the typical 50 µm pathlength used for the spectroscopy of aqueous solutions.

The fabrication of the microchannel began with the cleaning of the CaF2 window with acetone and methanol to remove organic impurities followed by drying in a 110 °C vacuum oven. The clean window was placed on a spinner chuck of PWM32 Controller Spinner (Headway Research Inc., USA) for spin coating photoresist. About 3 ml of the photoresist was poured onto the substrate directly from an amber storage jar since its high viscosity precluded pipetting. The spin coating follows a series of steps. To allow the photoresist to spread evenly across the entire surface of the substrate, the spin speed was started at 500 rpm and held constant for 10 s. The speed was then increased to 2000 rpm with an acceleration of 1000 rpm/s and held for 30 s. This process resulted in an average photoresist film thickness of 34 µm with a variation of ±0.9 µm among various devices. The film height uniformity for any one device is ±0.2 µm across the observation region, as measured by profilometry.

After spin coating, the residual solvent in the resist film is removed by a soft bake, placing the substrate on a 125 °C hotplate for a total time of seven minutes. To minimize fracture of the CaF2 substrate due to thermal shock and bubble formation from uneven heating of the resist film, the soft bake was performed on a hot plate with lift pins for contact or proximity baking. The height of pins was consecutively held at 0.25, 0.1, and 0.05 in. at two min intervals and finally in contact for the last 1 min. After soft baking, the photoresist film was exposed by direct writing of the structure with a Heidelberg Mask Less Aligner (MLA)-150 using a 375 nm laser. Two exposure files were prepared in AutoCAD software. The first exposure file consisted of a microchannel design while the second one consisted of two concentric circles––25 mm and 40 mm in diameter. The second exposure removes the “edge bead”––elevation around the substrate edge, resulting from the spin coating of high viscosity photoresist so that a uniform layer of 25 mm ∅ photoresist film is deposited on the CaF2 substrate. The key factors for determining the optimal time and dose for exposure are the reflectivity of the substrate, the type of photoreaction occurring in the photoresist, and the thickness of the photoresist film. The reflectivity of CaF2 is ∼10% as compared to 30% for Si wafers and, therefore, requires a larger exposure dose. For AZ 40 XT, the photoreaction is a one-photon process, and the required dose was optimized experimentally. A series of exposures with doses ranging from 100–800 mJ/cm2 with a step of 50 mJ/cm2 was performed to fabricate a 100 × 100 µm2 square. Doses larger than 500 mJ/cm2 yielded fabrication of high-aspect ratio microstructure. For positive resists, the optimal light dose for most lithographic processes is close to the value at which the development rate begins to saturate. To avoid possible thermal or mechanical damage to the resists at a high light intensity, 500 mJ/cm2 was chosen as an optimal dose. The total time needed for the exposure was 7 min, controlled by the speed of the high-precision motorized stage that held the substrate.

The photoactivation is followed by a post-exposure bake, placing the substrate on a 110 °C hotplate for 50 s. Since the nitrogen gas is formed as a by-product, this step is monitored carefully to avoid bubble formation within the film. Finally, the exposed photoresist was washed away by submerging in a fresh AZ MIF 300 alkaline developer solution every minute for 3 min with light agitation. Figure 3(a) shows a 3D image of a microchannel, demonstrating the sharp features and film thickness uniformity of a “serpentine” mixer acquired using a confocal laser scanning microscope (LEXT OLS5000, Olympus, USA).

FIG. 3.

Depth characterization of microchannel before and after bonding. (a) 3D image from a laser confocal microscope of the microchannel before bonding shows the high-resolution lithography of the “serpentine” mixer. (b) Image of sandwiched CaF2 assembly after bonding with heat and pressure. (c) Comparison of depth profile before and after bonding using laser confocal profilometry on the same microwell, images shown on the bottom. The compression due to bonding is less than 2 µm. (d) Image of the fully assembled flow cell.

FIG. 3.

Depth characterization of microchannel before and after bonding. (a) 3D image from a laser confocal microscope of the microchannel before bonding shows the high-resolution lithography of the “serpentine” mixer. (b) Image of sandwiched CaF2 assembly after bonding with heat and pressure. (c) Comparison of depth profile before and after bonding using laser confocal profilometry on the same microwell, images shown on the bottom. The compression due to bonding is less than 2 µm. (d) Image of the fully assembled flow cell.

Close modal

The bonding step seals the microchannels by compressing the photoresist between the 40 mm CaF2 substrate and the 35 mm CaF2 drilled window. Novolak based photoresists, like AZ 40XT, have been previously reported to exhibit adhesion properties on CaF2 without the use of any adhesion promoter.47 This is crucial because the bonding should be strong enough to sustain the high pressure generated in the microfluidic channels. Prior to bonding, the holes of drilled CaF2 window (1.5 mm ∅) were aligned to the channel inlets and outlet holes. To simplify this alignment, the microchannel port in the photoresist film was set to 2.2 mm in diameter. The sandwiched CaF2 was then bonded thermally using a heat press (Dulytek, USA) containing two 2.5 × 6 in.2 temperature-regulated aluminum plates. The top plate is fixed while the bottom plate rests on the ram of a hydraulic bottle jack, where a pressure gauge was added in house to monitor the pressure during bonding. To prevent breaking or fracture of brittle CaF2 windows during the bonding, we cushioned the CaF2 windows between 100 µm thick Teflon sheets and placed a 2 mm thick metal shim shock on either side of the aluminum plates. To minimize the risk of breaking windows during bonding, our cell design is readily adaptable to thicker CaF2 windows, as long as one accounts for the working distance of the IR microscope.

Optimal bonding was achieved at a temperature of 65 °C, pressing the CaF2 assembly at a pressure of 400 psi and holding at that pressure for 5 s. If the sandwiched CaF2 is placed at the center of the plate and pressed, air bubbles are randomly trapped in the microchannel. This is problematic because once the fluid flows, it spreads throughout the bubble, which disintegrates the assembly. However, if the windows are placed such that only one-half of the cell is sealed, then the trapped air travels to the non-bonded side. We can then use multiple cycles of rotation and compression to completely remove air bubbles. Figure 3(b) shows the image of the CaF2 cell assembly after bonding with no trapped air between the photoresist film and drilled window.

To characterize the compression of photoresist film during bonding, we performed microchannel depth measurements before and after bonding using confocal laser scanning microscopy. The depth of a 310 µm ∅ dummy well was measured to compress from a thickness of 34.7–32.9 µm after bonding [Fig. 3(c)], indicating a compression due to bonding of ∼5%.

Following bonding, the window assembly is ready to be integrated into the 3D-printed flow cell. The complete cell assembly after compression of the CaF2 cell to the housing is shown in Fig. 3(d). The entrance and exit ports of the bottom flange of the flow cell are connected to 1/16 in. OD Teflon tubing (1622L, IDEX, USA) using NanoTight PEEK fittings for 10–32 tapped holes (F-333 NX, IDEX, USA). A small layer of epoxy glue was applied onto the threads of the fittings, which performed better in sealing compared to Teflon tape. The two inlet tubes were connected to 8 ml stainless-steel syringes (702267, Harvard Apparatus, USA) using Swagelok fittings. Prior to establishing this connection, the syringes were filled with sample solutions and filtered using a 0.2 µm syringe filter (UX-15945-42, Cole-Parmer, USA). Since the dust can easily clog the mixer, we opted to filter the sample solutions and assemble the flowcell inside a clean room taking extra effort in cleaning the syringes before each experiment.

The filled syringes were loaded into a syringe pump with two independently controlled pumping channels (Pump 33 DDS, Harvard Apparatus, USA). This pump was controlled using a touch screen interface, which allowed users to set the flow rate, the inner diameter of the syringe, and measurement mode. For all measurements presented here, we utilized a “twin” measurement mode in which both pumping channels are identical in flow rate and direction. The syringe pump advances the syringe plunger at a set linear rate, which, by multiplying by the cross-sectional area of the syringe, provides the volumetric flowrate. The flowrate set by the user is the rate at which fluid is exiting from each syringe, and, therefore, the effective flowrate in the microchannel after mixing is the sum of the individual flowrates.

For microfluidic applications requiring high flowrate, the pump’s linear force is an important specification that should be sufficiently large enough to overcome the back pressure generated in the microchannel. The pressure drop ΔP, in a rectangular microchannel with length L, width w, depth h, for a fluid with viscosity η flowing at a flowrate Q can be estimated using Poiseuille formula for frictional pressure drop, ΔP = 12ηLQ/(h3w). For our case, we estimate the pressure drop to be ∼250 psi at a flowrate of 350 µl/min, requiring syringes designed for high pressure applications and syringe pumps with a linear force of at least 50 lbs.

To prepare the flow cell for optical measurement, the inlet Teflon tubes were first primed with respective reactant solutions to minimize air bubbles in the microchannel. Then, the CaF2 window assembly was compressed in the cell housing, aligning, and sealing the inlet/outlet ports to the bottom flange O-rings. The assembled flow cell was mounted onto the motorized microscope stage of an infrared microscope using a 3D-printed holder. IR absorption measurements were performed in transmission measurement mode using a commercial FTIR instrument (Vertex 70, Bruker Optics Inc., USA) connected to an IR microscope (Hyperion 2000, Bruker Optics Inc., USA) with 15× IR objective, using the video assisted measurement wizard in the Bruker OPUS software. The microscope is equipped with a single channel MCT infrared detector, a visible camera, and a computer controlled x–y stage with an adjustment accuracy of 0.1 µm and repeatability of 1 µm or better. A visible image of the microchannel is collected by the camera and used to identify a series of predefined positions for measurements of IR spectra along the observation channel, an air background, and input solution reference spectra. Pinhole transmission measurements indicate that the IR focal spot in the measurement channel is 320 µm in diameter (1/e2 intensity), and thus, our first measurement is made Δx = 150 µm from the exit of the mixer to avoid clipping the IR beam by the photoresist film. The remaining 30–35 measurement points were spread across the spiral and/or alternating observation regions.

Generally, in selecting the total number of IR measurement points, one needs to take into account a series of tradeoffs in flowrates and total sample volume, the number of time points to properly sample relaxation processes, and the desired spectral resolution and baseline noise of the IR spectra. A total volume of 8 ml permits 22 min of measurement time at a flowrate of 350 µl/min. To adequately get to the baseline of a relaxation process, the measured time points should span three times the 1/e decay time. For most kinetic studies, measurement can be performed with a modest spectral resolution (4–16 cm−1), but this could heavily depend on the type of system being investigated and the linewidth of IR modes of interest. The IR brightness of the sample relative to the solvent background determines the minimum average needed to collect spectra with a good signal to noise. We found an average of four spectra with a spectral resolution of 2 cm−1 to be adequate for collecting high quality spectra of our model system. With these measurement parameters, the collection time for each spectrum is ∼10 s; therefore, we chose between 30 and 35 measurement points for each experiment. Once the measurement points were set, the flowrate was set to 350 µl/min in the pump, and the mixer was monitored using the visible camera for around 30 s. This was to confirm that no dust particles stuck in the mixer affected the mixer’s performance. Then, the reference spectra were collected at each inlet and confirmed to be the desired sample. Finally, IR spectra were collected at each point in the observation region. Once the last measurement was taken, the flow was stopped, the mixer was examined for any damage, and the flow cell was inspected for any leakage.

Time-resolved spectroscopy in continuous-flow mixers is made possible by mapping the position of a measurement to the time after mixing by using the steady-state velocity of the mixed fluid flow along the observation channel. Although the fluid’s velocity field can have complex spatial variations, in calculating the time-delays for a given measurement position, we assume that the fluid is homogeneously mixed upon exiting the mixer and that the velocity in the channel at a particular point along the direction of fluid flow, ux, is governed only by the cross-sectional area through the device σx. The height of our channel h is constant, but the width of the channel through the mixer and the exit into the observation channel w(x) varies with x so that σx=hw(x). The time-delay after mixing at which an observation is made, τx, can then be determined from V(x), and the volume of fluid in the device between the exit of the mixer (x0) to that point (x),

Vx=x0xσ(x)dx,
(1)
τx=V(x)/Q.
(2)

Here, Q is the volumetric flow rate. For a linear channel of fixed height and width, the velocity u = Q/wh, and the time-delay increases linearly with distance as τx=ΔxwhQ1=Δx/u, where Δx = xx0. For the dimensions of our observation channel, w = 350 µm and h = 30 µm, and a flow rate of Q = 700 µl/min, v = 1.12 µm/μs, or equivalently an increase of time delay along the channel of Δτx = 0.9 µs/μm.

In practice, a given volume within the observation channel can be obtained by calculating the surface area of the observation channel on the substrate up to the observation point and then multiplying it by the height of the channel. This provides a way of accounting for variations in the width and curvature of the observation channel. The exceptional fidelity of the photolithography allows us to estimate the surface area from AutoCAD design using a measure area tool. The first and shortest time delay measured at Δx = 150 µm corresponds to a volume of 1.3 µl and τ = 110 µs at a flow rate of 700 µl/min. Separately, we define mixing time, τmix, as the time spent by the reactants in the mixer and estimate it with a similar strategy. The volumes of the “inchworm” and “serpentine” mixers are 0.85 and 1.4 nl, respectively. This corresponds to 7 and 12 µs for the time needed to traverse the “inchworm” and “serpentine” mixers, respectively.

For the purpose of quantifying the very fastest kinetic relaxation processes, the observed change of absorbance in the transmitted beam as a function of position along the observation channel will also depend on the variation in τ(x) across the spot size of the measurement beam, Δr. The influence of Δr on the observed relaxation kinetics can be calculated from the convolution of the beam intensity profile with the relaxation kinetics in terms of τ. The minimum time-resolution with which kinetics can be quantified is governed by the spread of time-delays across the microscope’s imaging beam within the channel, Δτ, which we define through the variance of time-delays measured at full-width half-maximum in the beam intensity profile, Δr, which in our case is ∼200 µm. For a Gaussian intensity distribution to the beam in a straight channel Δτ=hwΔr/(8ln2Q), in principle, Δτ ≈ 90 µs at a flow rate of 700 µl/min for our device design. This variance is negligible for all but the closest measurements to the mixer exit. We see that decreasing channel width, decreasing spot size, or increasing flow rate each decreases Δτ in a linear manner.

The spatial uniformity of mixing was initially characterized by fluorescence imaging of the observation channel upon mixing a rhodamine 6G solution with water. Images were acquired using an inverted fluorescence microscope (Nikon Ti-eclipse, 10× objective, 532 nm excitation) with a CCD camera (iXon X3, Andor), and the details of the filter cubes used are provided in the supplementary material. The two stainless-steel syringes were filled with a 50 µM dye solution and water and loaded into the syringe pump. Figures 4(a) and 4(c) show images of the fluorescence distribution in the observation channel at various flowrates for the “inchworm” and “serpentine” type mixer, respectively. Figures 4(b) and 4(d) show the corresponding transverse fluorescence count profile 150 µm downstream from the mixer exit. At a flowrate of 100 µl/min, the flow is laminar in both mixers with little diffusion across the interface separating dye and water solutions. At flowrates between 300 and 500 µl/min, swirls from the folding of flow between solutions are observed. A spatially uniform fluorescence distribution is observed at flowrates of greater than 600 and 700 µl/min for the “serpentine” and “inchworm,” respectively, as verified by plateauing of the transverse fluorescence count profiles at these flowrates.

FIG. 4.

Fluorescence images of the observation channel following the mixing of 50 µM rhodamine 6G solution and water in (a) serpentine and (c) inchworm mixer at flowrates of 100, 300, 500/600, and 700 µl/min. Fluorescence distribution across the channel 150 µm from the exit of the mixer, dashed magenta line, is shown in (b) and (d), suggesting homogeneous mixing at flowrate >300 µl/min. (e) Fluorescence images of the observation channel following the quenching reaction of 1 µM fluorescein by 0.5M potassium iodide in an “inchworm” mixer. (f) Fluorescence distribution suggests a 70% drop in fluorescein fluorescence upon complete mixing.

FIG. 4.

Fluorescence images of the observation channel following the mixing of 50 µM rhodamine 6G solution and water in (a) serpentine and (c) inchworm mixer at flowrates of 100, 300, 500/600, and 700 µl/min. Fluorescence distribution across the channel 150 µm from the exit of the mixer, dashed magenta line, is shown in (b) and (d), suggesting homogeneous mixing at flowrate >300 µl/min. (e) Fluorescence images of the observation channel following the quenching reaction of 1 µM fluorescein by 0.5M potassium iodide in an “inchworm” mixer. (f) Fluorescence distribution suggests a 70% drop in fluorescein fluorescence upon complete mixing.

Close modal

To further test that homogeneous mixing has reached the molecular scale, we imaged the quenching of fluorescein fluorescence by 0.5M potassium iodide (KI) in the “inchworm” mixer. The results were similar in behavior to that observed by fluorescence dilution [Figs. 4(e) and 4(f)]. Laminar flow with persistent fluorescent streams is observed in the observation channel below 500 µl/min, but at a flowrate of 700 µl/min, the peak fluorescence and integrated area under the fluorescence profiles are reduced by ∼70%, as expected for completion of the quenching reaction.33,48 At the highest flow rates, a small excess of fluorescein fluorescence is observed within 10 µm of the right wall in the observation channel, suggesting that the “inchworm” design does underperform the “serpentine” in terms of transverse mixing homogeneity. However, quenching uniformity following the inchworm mixer is otherwise constant across the measured regions of the observation channel.

To confirm that these fluorescence imaging observations are consistent with infrared spectroscopy measurements, we monitored pH-jump experiments of adenosine monophosphate (AMP) with the “serpentine” mixer. AMP can be used as a ratio metric pH indicator near its 3.8 pKa using the strong and clearly resolved absorbance bands at 1666 and 1624 cm−1 that arise from the protonated and deprotonated states of the adenosine ring, respectively.49,50 Since the protonation/deprotonation reactions in water are expected to occur within a few microseconds,40,51–54 a homogeneously mixed solution should be fully equilibrated at the final pH upon exiting the mixing channel. Because the AMP vibrational transitions overlap with the H2O bending vibration, we performed experiments in D2O. The pH of the solution was measured using a standard glass electrode and is reported as pH*, which can be converted to pD using pD = pH* + 0.41.55 We monitored the mixing of a 16 mM, pH* 3 AMP solution with 50 mM, pH* 6.9 sodium phosphate buffer as a function of flowrates. The large IR spot size in the microscope limits the ability to fully assess the spatial variation of pH within the observation channel; however, we can test for spectral variations between the left and right sides of the channel. Figure 5 shows peak normalized FTIR spectra with the measurement focus centered on either side of the observations channel near the exit of a “serpentine” mixer and compares these spectra with those of steady-state AMP solutions at the initial (dashed magenta) and final pH (dashed red). Consistent with the fluorescence characterization, the transverse spatial variation of the spectra varies from reproducing the individual input solutions at 100 µl/min to variations consistent with inhomogeneous mixing for 300 µl/min flowrates, to spatially uniform and consistent with the final pH for a flowrate of 600 µl/min.

FIG. 5.

Peak normalized FTIR spectra at flowrates of 100, 300, and 600 µl/min of AMP pH jump experiment monitored at the exit of the “serpentine” mixer. The blue and green dots in the inset show the measurement location, the relative size of the IR beam, and the steady state spectra, and the initial and final pH are shown in dashed magenta and red lines, respectively. The rise in 1624 cm−1 and drop in 1666 cm−1 mode with increasing flowrate, which becomes uniform across both measurements at 300 µl/min, suggest homogeneous mixing, consistent with fluorescence imaging.

FIG. 5.

Peak normalized FTIR spectra at flowrates of 100, 300, and 600 µl/min of AMP pH jump experiment monitored at the exit of the “serpentine” mixer. The blue and green dots in the inset show the measurement location, the relative size of the IR beam, and the steady state spectra, and the initial and final pH are shown in dashed magenta and red lines, respectively. The rise in 1624 cm−1 and drop in 1666 cm−1 mode with increasing flowrate, which becomes uniform across both measurements at 300 µl/min, suggest homogeneous mixing, consistent with fluorescence imaging.

Close modal

The performance of the mixer for fast chemical kinetics measurements was tested using the reduction of potassium ferricyanide by ascorbic acid, a well-studied test reaction for rapid mixing.56–58 While this reaction is typically studied with UV/visible absorbance, the C≡N stretch vibrations of ferricyanide and ferrocyanide ions are distinct and, therefore, are suitable to investigate with IR spectroscopy.59–61 The net reaction scheme is given by

AH2+2FeCN63A+2FeCN64+2H+,

where AH2 represents the ascorbic diacid and A is dehydroascorbic acid; however, the underlying reaction mechanism is multistep, involving the ascorbic acid/base equilibria and an ascorbate free radical intermediate,

AH2KA1AHKA2A2(fast),AH+FeCN63k2(slow)AH+FeCN64,AH+FeCN63k3(fast)A+FeCN64+H+,A2+FeCN63k4A+FeCN64.

As a result, the reaction is pH sensitive, allowing one to use pH to smoothly vary kinetics across the micro-to-millisecond timescale. For the empirical rate law,

12dFeCN63dt=kobsFeCN63AH2,

the observed kinetic relaxation rate constant kobs can be computed using the following formula:56 

kobs=2k2+k4KA2[H+]1+[H+]KA1+KA2[H+].
(3)

Here, KA1 (9.16 × 10−5M) and KA2 (4.57 × 10−12M) are the acid dissociation constants for ascorbic acid, and the rate constants k2 and k4 have been previously found to be 4.58 × 102 and 6.02 × 106M−1s−1, respectively.56 

Potassium ferricyanide K3[Fe(CN)6], ferrocyanide K4[Fe(CN)6], L-ascorbic acid, anhydrous potassium phosphate monobasic, dibasic, and tribasic salts were purchased from Fisher Scientific and used as received. Phosphate buffer solutions were prepared at a 0.5M concentration, with pH varying from 4 to 12. Both 0.1M L-ascorbic acid and 25 mM potassium ferricyanide were prepared in the buffer solution day of the experiment to avoid unwanted reactions with oxygen and moisture in the atmosphere. The pH of each reagent and the final mixture were measured using a standard glass electrode pH meter. The measured pH of two initial and final solutions are summarized in Table I, which is consistent with theoretically calculated pH.

TABLE I.

Summary of pH-dependent reduction kinetics of ascorbic acid (AA). Measured pH of the initial input AA and ferricyanide solutions and the equilibrated final mixed solution are shown together with the pH dependent observed relaxation rate constants and relaxation times, τobs=kobs1.

Solution pHRelaxation kinetics
Final mixedInitial AAInitial ferricyanidekobs (s−1)τobs (ms)
11.3 11.1 11.8 ⋯ ⋯ 
9.4 8.3 11.4 2650 0.4 
8.3 7.9 10.7 680 1.5 
7.9 7.7 8.9 500 2.0 
7.7 7.5 8.3 240 4.2 
7.3 7.1 7.7 190 5.2 
6.6 6.5 7.0 130 7.9 
4.2 4.0 5.5 50 20 
Solution pHRelaxation kinetics
Final mixedInitial AAInitial ferricyanidekobs (s−1)τobs (ms)
11.3 11.1 11.8 ⋯ ⋯ 
9.4 8.3 11.4 2650 0.4 
8.3 7.9 10.7 680 1.5 
7.9 7.7 8.9 500 2.0 
7.7 7.5 8.3 240 4.2 
7.3 7.1 7.7 190 5.2 
6.6 6.5 7.0 130 7.9 
4.2 4.0 5.5 50 20 

To illustrate a measurement, the raw transient FTIR absorption spectra between 2000 and 2150 cm−1 as a function of mixing time delays from 0.1 (blue) to 35 ms (red) are shown in Fig. 6(a) for pH 6.6 together with the reference spectra at the inlets of ferricyanide and ascorbic acids. A smoothly sloping baseline in the spectra is dominated by the bend–libration combination band of H2O, and the C≡N stretch vibrations of the ferro- and ferricyanide ions are observed as sharper peaks on top of this background at 2037 and 2115 cm−1, respectively. This solvent background is corrected by subtracting the spectra of the ascorbic acid solution collected at one of the reference channels. Any residual baseline offsets, which we attribute to clipping of the beam by the flowcell bottom flange at long delays, are corrected by subtracting a linear offset to match at a frequency where there is no molecular absorption (2085 cm−1 for this case). Figure 6(b) shows the resulting processed absorption spectra. The characterization of reproducibility of data acquisition is shown in the supplementary material, Fig. S1. Peak height variations are due to the 3× difference in molar extinction coefficient of the ferrocyanide mode (4.24 × 103M−1 cm−1) relative to the ferricyanide mode (1.18 × 103M−1 cm−1). By measuring the integrated IR peak area, we determined the ratio of transition dipole strength of ferrocyanide and ferricyanide mode to be 7:1, which is consistent with values reported in the literature.60–62 

FIG. 6.

Potassium ferricyanide reduction kinetics. (a) Raw transient FTIR absorption spectra in the C≡N stretching region for pH 6.6, and (b) the same spectra following background subtraction and baseline correction. The ferricyanide mode is centered at 2115 cm−1, with a full width at half maximum (FWHM) bandwidth of 9 cm−1. The ferrocyanide mode is centered at 2037 cm−1, with a FWHM bandwidth of 17 cm−1. (c) Example of time-dependent peak absorbances for the ferricyanide (red) and ferrocyanide (blue) vibrations for pH 7.7. (d) and (e) pH-dependent reduction kinetics showing expected exponential growth and decay. (f) Experimentally determined relaxation times as a function of pH (points) and the predicted behavior based on Eq. (3) (line). The pH reported in the figure is the pH of the final mixture.

FIG. 6.

Potassium ferricyanide reduction kinetics. (a) Raw transient FTIR absorption spectra in the C≡N stretching region for pH 6.6, and (b) the same spectra following background subtraction and baseline correction. The ferricyanide mode is centered at 2115 cm−1, with a full width at half maximum (FWHM) bandwidth of 9 cm−1. The ferrocyanide mode is centered at 2037 cm−1, with a FWHM bandwidth of 17 cm−1. (c) Example of time-dependent peak absorbances for the ferricyanide (red) and ferrocyanide (blue) vibrations for pH 7.7. (d) and (e) pH-dependent reduction kinetics showing expected exponential growth and decay. (f) Experimentally determined relaxation times as a function of pH (points) and the predicted behavior based on Eq. (3) (line). The pH reported in the figure is the pH of the final mixture.

Close modal

The time-dependence of the spectra follows the decay of the ferricyanide mode at 2115 cm−1 and the growth of the ferrocyanide mode at 2037 cm−1. At the earliest mixing delay, the ferricyanide absorbance is almost halved relative to the inlet due to dilution upon mixing, and some ferrocyanide is already present, indicating the extent of reduction up to the first detection point. The time-dependence of the peak absorption for the two modes is shown in Fig. 6(c) along with fits, illustrating that the relaxation process for both peaks follows the expected exponential growth and decay kinetics with the same time constant. Figure 6(c) also illustrates that the two earliest time points (τ = 0.12, 0.44 ms) deviate from the monotonic exponential behavior of the remaining points. This behavior is consistently observed in our data, indicating that unresolved challenges remain to reliably access sub-millisecond behavior.

To extract kinetic parameters, the ferrocyanide absorption growth and ferricyanide absorption decay were fitted simultaneously to Eqs. (4) and (5), respectively,

At=A1ekobst+A0,
(4)
at=a0ekobst+a.
(5)

For comparison among datasets, data were normalized with the fit parameters using the following scheme:

Fe2+t=AtA0AA0,
(6)
Fe3+t=ataa0a.
(7)

The resulting pH dependent kinetics traces of the peak absorption by the ferrocyanide and ferricyanide modes are shown in Figs. 6(d) and 6(e), respectively. To exclude the measurements for τ < 0.5 ms from our interpretation, these data have been normalized to their upper baseline in the case of ferrocyanide growth kinetics and normalized by the τ = 0 value of the fit in the case of ferricyanide relaxation kinetics. The kinetic traces of ferrocyanide mode at pH 9.4 have already grown within the first few measurement points, indicating that the reaction rate is too fast to be measured with the current time resolution of the device. Below pH of 4.2, the reduction rates are in hundreds of milliseconds, beyond the time window allowed by the current design. The values of kobs and corresponding relaxation time obtained from fitting for different pH conditions are summarized in Table I. Between pH of 6.6–8.30, the reaction rates speed up, which is consistent with an observation reported in the literature.56,63 To verify our experimental result, the pH dependence of the second-order rate constant was plotted as a function of pH, as shown in Fig. 6(f). The predicted behavior based on Eq. (3) agrees well with the observed rate constants below pH 9. Above pH 9, the curve leaves the experimental points, and this suggests that the instrument time-resolution is limited to resolve kinetics faster than 500 µs.

Our fabrication methods provide a robust platform for the rapid prototyping of mixer designs on CaF2 substrates for time-resolved IR spectroscopy. Fabrication using photoresists is direct, enabling high aspect ratio structures and precise control of microchannel depth compared to other fabrication materials, such as paraffin and PDMS. However, there are several avenues for improving and extending this technology. Overall, we achieved a success rate of greater than 90% in the photolithography step and roughly 50% during the bonding while making devices with AZ 40XT photoresist. This photoresist has poor chemical compatibility in organic solvents and at temperatures above 80 °C limiting its application to the investigation of reactions in water. Additionally, we also noticed the micromixer disintegrated while flowing fluids at a flow rate greater than 600 µl/min if exposed to ambient light for greater than 48 hr. As a result, most of our devices are good for one time use, but CaF2 windows are reusable. The bonded CaF2 windows were taken apart by sonication in acetone, and individual windows were cleaned with methanol. Repeated ultrasonic cleaning in organic solvents ultimately degraded the surface quality, and compression during bonding introduced minor cracks, which limited the window reusability to less than 50 cycles.

Before considering AZ 40 XT, we performed preliminary work with SU-8 photoresists. SU-8, an epoxy based photoresist, was attractive due to its low cost, chemical compatibility, and availability in a wide range of film thicknesses ranging from 4 to 120 µm.64 The lithography of SU-8 with a laser-based source commonly led to over-polymerization when channel dimensions were less than 25 µm, likely due to local heating, Fig. S2. As another option, we tried traditional mask based exposure with SU-8, but this has its own set of challenges, such as inefficient prototyping from the lead time to order masks, mask-substrate misalignment, and decreased optical resolution. Furthermore, SU-8 is difficult to remove from the CaF2 window once polymerized. AZ 125 nXT could possibly resolve the over polymerization issue because its resin cross-linking reaction occurs at room temperature and can be developed right after exposure.

In terms of the mixer designs, we believe that both mixers are capable of homogeneously mixing the reagents, but “serpentine” performed better and at lower flowrates than “inchworm.” Despite this, the “inchworm” was our preferred mixer because the “serpentine” mixture was susceptible to wall collapse during a prolonged flow of reactants at a flowrate of greater than 600 µl/min, Fig. S3. In the future, we hope to make changes in the mixer design, observation channels, and design of the flow cell to improve various aspects of the device, such as mixing time/efficiency, time resolution, and time window extension.

In order to shorten mixing time and improve mixing efficiency, we plan to reduce the mixer channel width from 20 to 10 µm, as this results in a larger pressure drop that stretches and folds the fluids more effectively. The time-resolution for kinetics measurements can be improved through tighter focusing of the IR beam and a narrowing of the observation channel, ideally closer to the spot size of the focused beam. Using high numerical aperture optics, an IR beam from a laser source can be tightly focused to a spot size of ∼10 µm. The time resolution of 10 µs can be achieved with a near-diffraction-limited 10 µm IR beam focused on a 20 µm wide observation channel at a flowrate of 700 µl/min, as calculated from Δτ=τmix+hwΔr/(8ln2Q). To extend the time window of observation to hundreds of milliseconds, several steps can be taken. For the linear observation channels, the width of the channel can be gradually flared out so that the time increases nonlinearly with the distance from the mixer. Larger CaF2 substrates enable flowcell bottom flange design with a wider aperture for measurement, facilitating much longer observation channels.

The microfluidic technology presented here allows rapid prototyping of continuous flow mixer designs on CaF2 substrates to be used for kinetic measurements on microsecond to millisecond timescales with IR spectroscopy. The mixer can also potentially be used with other spectroscopic probes compatible with optical microscopies, such as UV/Vis absorption, fluorescence, and Raman. Although we have demonstrated the acquisition of mixing kinetics down to a time-scale of 1 ms, the current design is primarily limited by the IR focusing, and several strategies for further improvement are available to lower the time-scale for kinetics measurements to less than 50 µs.

See the supplementary material for characterizing the reproducibility of kinetics measurements, fabrication with SU-8 photoresist, failure of a serpentine mixer, and additional instrumentation details.

The authors thank Amber Krummel for numerous discussions and advice on IR compatible microfluidic technologies. This work was supported by grants from the U.S. Department of Energy (Grant No. DE-SC0014305) and the National Science Foundation (Grant No. CHE-1856684). This work made use of the Pritzker Nanofabrication Facility of the Institute for Molecular Engineering at the University of Chicago, which receives support from Soft and Hybrid Nanotechnology Experimental (SHyNE) Resource (Grant No. NSF ECCS-2025633), a node of the National Science Foundation’s National Nanotechnology Coordinated Infrastructure. This work also made use of the shared facilities at the University of Chicago Materials Research Science and Engineering Center, which receives support from the National Science Foundation under Award No. DMR-2011854.

The authors have no conflicts to disclose.

Ram Itani: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Methodology (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Max Moncada Cohen: Conceptualization (supporting); Data curation (supporting); Investigation (supporting); Methodology (equal). Andrei Tokmakoff: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (lead); Investigation (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

1.
O. G.
Berg
and
P. H.
von Hippel
,
Annu. Rev. Biophys. Biophys. Chem.
14
(
1
),
131
158
(
1985
).
2.
D.
Shoup
,
G.
Lipari
, and
A.
Szabo
,
Biophys. J.
36
(
3
),
697
714
(
1981
).
3.
A.
JohnáElliot
,
J. Chem. Soc., Faraday Trans.
86
(
9
),
1539
1547
(
1990
).
4.
R. L.
Berger
,
B.
Balko
, and
H. F.
Chapman
,
Rev. Sci. Instrum.
39
(
4
),
493
498
(
1968
).
5.
P.
Regenfuss
,
R. M.
Clegg
,
M. J.
Fulwyler
,
F. J.
Barrantes
, and
T. M.
Jovin
,
Rev. Sci. Instrum.
56
(
2
),
283
290
(
1985
).
6.
O.
Bilsel
,
C.
Kayatekin
,
L. A.
Wallace
, and
C. R.
Matthews
,
Rev. Sci. Instrum.
76
(
1
),
014302
(
2005
).
7.
V.
Mengeaud
,
J.
Josserand
, and
H. H.
Girault
,
Anal. Chem.
74
(
16
),
4279
4286
(
2002
).
8.
J.
Li
,
G.
Xia
, and
Y.
Li
,
J. Chem. Technol. Biotechnol.
88
(
9
),
1757
1765
(
2013
).
9.
A.
Afzal
and
K.-Y.
Kim
,
Chem. Eng. J.
203
,
182
192
(
2012
).
10.
T. W.
Lim
,
Y.
Son
,
Y. J.
Jeong
,
D.-Y.
Yang
,
H.-J.
Kong
,
K.-S.
Lee
, and
D.-P.
Kim
,
Lab Chip
11
(
1
),
100
103
(
2011
).
11.
S. H.
Wong
,
P.
Bryant
,
M.
Ward
, and
C.
Wharton
,
Sens. Actuators, B
95
(
1–3
),
414
424
(
2003
).
12.
A. D.
Stroock
,
S. K. W.
Dertinger
,
A.
Ajdari
,
I.
Mezić
,
H. A.
Stone
, and
G. M.
Whitesides
,
Science
295
(
5555
),
647
651
(
2002
).
13.
A.-S.
Yang
,
F.-C.
Chuang
,
C.-K.
Chen
,
M.-H.
Lee
,
S.-W.
Chen
,
T.-L.
Su
, and
Y.-C.
Yang
,
Chem. Eng. J.
263
,
444
451
(
2015
).
14.
C.-Y.
Lee
,
W.-T.
Wang
,
C.-C.
Liu
, and
L.-M.
Fu
,
Chem. Eng. J.
288
,
146
160
(
2016
).
15.
M.
Arai
and
K.
Kuwajima
,
Folding Des.
1
(
4
),
275
287
(
1996
).
16.
H.
Roder
,
K.
Maki
,
H.
Cheng
, and
M. C.
Ramachandra Shastry
,
Methods
34
(
1
),
15
27
(
2004
).
17.
J.
Luchins
and
S.
Beychok
,
Science
199
(
4327
),
425
426
(
1978
).
18.
J. S.
Olson
,
Methods in Enzymology
(
Academic Press
,
1981
), Vol. 76, pp.
631
651
.
19.
D. B.
Green
,
J.
Lane
, and
R. M.
Wing
,
Appl. Spectrosc.
41
(
5
),
847
851
(
1987
).
20.
B.-F.
Shi
,
Y.-H.
Zhang
,
J. K.
Lam
,
D.-H.
Wang
, and
J.-Q.
Yu
,
J. Am. Chem. Soc.
132
(
2
),
460
461
(
2010
).
21.
P. S.
Kim
and
R. L.
Baldwin
,
Annu. Rev. Biochem.
51
(
1
),
459
489
(
1982
).
22.
C. R.
Matthews
,
Annu. Rev. Biochem.
62
,
653
683
(
1993
).
23.
S. J.
George
,
G. A.
Ashby
,
C. W.
Wharton
, and
R. N. F.
Thorneley
,
J. Am. Chem. Soc.
119
(
27
),
6450
6451
(
1997
).
24.
S. J.
George
,
S.
Kurkin
,
R. N. F.
Thorneley
, and
S. P. J.
Albracht
,
Biochemistry
43
(
21
),
6808
6819
(
2004
).
25.
B.
Liu
,
H.
Matsuoka
, and
M.
Terano
,
Macromol. Rapid Commun.
22
(
1
),
1
24
(
2001
).
26.
E.
Kauffmann
,
N. C.
Darnton
,
R. H.
Austin
,
C.
Batt
, and
K.
Gerwert
,
Proc. Natl. Acad. Sci. U. S. A.
98
(
12
),
6646
6649
(
2001
).
27.
P.
Hinsmann
,
M.
Haberkorn
,
J.
Frank
,
P.
Svasek
,
M.
Harasek
, and
B.
Lendl
,
Appl. Spectrosc.
55
(
3
),
241
251
(
2001
).
28.
J.
Tang
and
F.
Gai
,
Appl. Spectrosc.
60
(
12
),
1477
1481
(
2006
).
29.
H.
Fabian
and
D.
Naumann
,
Methods
34
(
1
),
28
40
(
2004
).
30.
A. J.
White
,
K.
Drabble
, and
C. W.
Wharton
,
Biochem. J.
306
(
3
),
843
849
(
1995
).
31.
A.
Perro
,
G.
Lebourdon
,
S.
Henry
,
S.
Lecomte
,
L.
Servant
, and
S.
Marre
,
React. Chem. Eng.
1
(
6
),
577
594
(
2016
).
32.
T. M.
Floyd
,
M. A.
Schmidt
, and
K. F.
Jensen
,
Ind. Eng. Chem. Res.
44
(
8
),
2351
2358
(
2005
).
33.
T.
Egawa
,
J. L.
Durand
,
E. Y.
Hayden
,
D. L.
Rousseau
, and
S.-R.
Yeh
,
Anal. Chem.
81
(
4
),
1622
1627
(
2009
).
34.
R.
Herzig-Marx
,
K. T.
Queeney
,
R. J.
Jackman
,
M. A.
Schmidt
, and
K. F.
Jensen
,
Anal. Chem.
76
(
21
),
6476
6483
(
2004
).
35.
K. L. A.
Chan
,
S.
Gulati
,
J. B.
Edel
,
A. J.
de Mello
, and
S. G.
Kazarian
,
Lab Chip
9
(
20
),
2909
2913
(
2009
).
36.
P.
Hinsmann
,
J.
Frank
,
P.
Svasek
,
M.
Harasek
, and
B.
Lendl
,
Lab Chip
1
(
1
),
16
21
(
2001
).
37.
P.
Svasek
,
E.
Svasek
,
B.
Lendl
, and
M.
Vellekoop
,
Sens. Actuators, A
115
(
2–3
),
591
599
(
2004
).
38.
N.
Kaun
,
M. J.
Vellekoop
, and
B.
Lendl
,
Appl. Spectrosc.
60
(
11
),
1273
1278
(
2006
).
39.
W.
Buchegger
,
C.
Wagner
,
B.
Lendl
,
M.
Kraft
, and
M. J.
Vellekoop
,
Microfluid. Nanofluid.
10
(
4
),
889
897
(
2011
).
40.
D. P.
Kise
,
D.
Magana
,
M. J.
Reddish
, and
R. B.
Dyer
,
Lab Chip
14
(
3
),
584
591
(
2014
).
41.
T.
Pan
,
R. T.
Kelly
,
M. C.
Asplund
, and
A. T.
Woolley
,
J. Chromatogr. A
1027
(
1–2
),
231
235
(
2004
).
42.
B.
Lehmkuhl
,
S. D.
Noblitt
,
A. T.
Krummel
, and
C. S.
Henry
,
Lab Chip
15
(
22
),
4364
4368
(
2015
).
43.
M. V.
Barich
and
A. T.
Krummel
,
Anal. Chem.
85
(
21
),
10000
10003
(
2013
).
44.
M.
Srisa-Art
,
S. D.
Noblitt
,
A. T.
Krummel
, and
C. S.
Henry
,
Anal. Chim. Acta
1021
,
95
102
(
2018
).
45.
Y.
Li
,
C.
Liu
,
X.
Feng
,
Y.
Xu
, and
B.-F.
Liu
,
Anal. Chem.
86
(
9
),
4333
4339
(
2014
).
46.
Y.
Li
,
D.
Zhang
,
X.
Feng
,
Y.
Xu
, and
B.-F.
Liu
,
Talanta
88
,
175
180
(
2012
).
47.
G.
Birarda
,
G.
Grenci
,
L.
Businaro
,
B.
Marmiroli
,
S.
Pacor
,
F.
Piccirilli
, and
L.
Vaccari
,
Vib. Spectrosc.
53
(
1
),
6
11
(
2010
).
48.
D. E.
Hertzog
,
X.
Michalet
,
M.
Jäger
,
X.
Kong
,
J. G.
Santiago
,
S.
Weiss
, and
O.
Bakajin
,
Anal. Chem.
76
(
24
),
7169
7178
(
2004
).
49.
M.
Tsuboi
,
Y.
Kyogoku
, and
T.
Shimanouchi
,
Biochim. Biophys. Acta
55
(
1
),
1
12
(
1962
).
50.
K.
Nakanishi
,
A.
Hashimoto
,
T.
Pan
,
M.
Kanou
, and
T.
Kameoka
,
Appl. Spectrosc.
57
(
12
),
1510
1516
(
2003
).
51.
J. C.
Flanagan
and
C. R.
Baiz
,
Opt. Lett.
44
(
20
),
4937
4940
(
2019
).
52.
K. K.
Smith
,
K. J.
Kaufmann
,
D.
Huppert
, and
M.
Gutman
,
Chem. Phys. Lett.
64
(
3
),
522
527
(
1979
).
53.
T. P.
Causgrove
and
R. B.
Dyer
,
Chem. Phys.
323
(
1
),
2
10
(
2006
).
54.
C. S.
Peng
,
C. R.
Baiz
, and
A.
Tokmakoff
,
Proc. Natl. Acad. Sci. U. S. A.
110
(
23
),
9243
9248
(
2013
).
55.
P. K.
Glasoe
and
F. A.
Long
,
J. Phys. Chem.
64
(
1
),
188
190
(
1960
).
56.
B. i.
Tonomura
,
H.
Nakatani
,
M.
Ohnishi
,
J.
Yamaguchi-Ito
, and
K.
Hiromi
,
Anal. Biochem.
84
(
2
),
370
383
(
1978
).
57.
K. W.
Watkins
and
J. A.
Olson
,
J. Chem. Educ.
57
(
2
),
158
(
1980
).
58.
L. J. A.
Martins
and
J. B.
da Costa
,
J. Chem. Educ.
65
(
2
),
176
(
1988
).
59.
G. M.
Sando
,
Q.
Zhong
, and
J. C.
Owrutsky
,
J. Chem. Phys.
121
(
5
),
2158
2168
(
2004
).
60.
J. J.
Alexander
and
H. B.
Gray
,
J. Am. Chem. Soc.
90
(
16
),
4260
4271
(
1968
).
61.
D. M.
Drew
,
Anal. Chem.
45
(
14
),
2423
2424
(
1973
).
62.
P.
Yu
,
F.
Yang
,
J.
Zhao
, and
J.
Wang
,
J. Phys. Chem. B
118
(
11
),
3104
3114
(
2014
).
63.
U. S.
Mehrotra
,
M.C.
Agrawal
, and
S.P.
Mushran
,
J. Phys. Chem.
73
(
6
),
1996
1999
(
1966
).
64.
A.
del Campo
and
C.
Greiner
,
J. Micromech. Microeng.
17
(
6
),
R81
(
2007
).

Supplementary Material