Hossein
Davoodi
a,
Nurdiana
Nordin
ab,
Lorenzo
Bordonali
a,
Jan G.
Korvink
a,
Neil
MacKinnon
*a and
Vlad
Badilita
*a
aInstitute of Microstructure Technology (IMT), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz Platz 1, 76344 Eggenstein-Leopoldshafen, Germany. E-mail: neil.mackinnon@kit.edu; vlad.badilita@kit.edu
bDepartment of Chemistry, Faculty of Science, University of Malaya, 50603 Kuala Lumpur, Malaysia
First published on 15th July 2020
Combining microfluidic devices with nuclear magnetic resonance (NMR) has the potential of unlocking their vast sample handling and processing operation space for use with the powerful analytics provided by NMR. One particularly challenging class of integrated functional elements from the perspective of NMR are conductive structures. Metallic electrodes could be used for electrochemical sample interaction for example, yet they can cause severe NMR spectral and SNR degradation. These issues are more entangled at the micro-scale since the distorted volume occupies a higher ratio of the sample volume. In this study, a combination of simulation and experimental validation was used to identify an electrode geometry that, in terms of NMR spectral parameters, performs as well as for the case when no electrodes are present. By placing the metal tracks in the side-walls of a microfluidic channel, we found that NMR RF excitation performance was actually enhanced, without compromising B0 homogeneity. Monitoring in situ deposition of chitosan in the microfluidic platform is presented as a proof-of-concept demonstration of NMR characterisation of an electrochemical process.
All these beautiful characteristics make NMR an extremely interesting method with which to investigate the bio-chemical processes and phenomena usually studied within LOC devices. However, scientists relying on NMR to characterise processes in an LOC device must be aware of the sword of Damocles potentially endangering their results: NMR has a notoriously low intrinsic mass sensitivity. As an order of magnitude, consider protons in a 9.4 T magnetic field at room temperature, where only 10 out of one million spins contribute to the NMR signal, so that concentrations below 1 mM are hard to observe.
One proven way to address the issue of poor mass sensitivity in small samples is to resort to miniaturised NMR detectors.1 An additional advantage is that many micro-MR topologies can be readily integrated with microfluidics. The straightforward geometry and fabrication process of planar coils have been employed to demonstrate multinuclear single and double resonance NMR in a microfluidic detection volume of 25 nl.2 Boero et al.3 successfully demonstrated NMR experiments on sub-nL samples, i.e. ova of microorganisms, by combining highly sensitive single-chip CMOS detectors with a high spatial resolution 3D printed microfluidic channel. Microcoil phased arrays combine the advantage of increased signal-to-noise ratio with an enlarged field of view,4 offering a 2.5 D sensitive region that may include microfluidic networks or sample chambers. The stripline detector topologies employed by Kentgens et al.5,6 and Utz et al.7,8 have proven to be easily amenable to microfluidic integration, at the same time providing excellent B1 and B0 homogeneity. Solenoidal microcoils9 must be integrated with three-dimensional microfluidic networks10 for sample delivery, which makes the fabrication process rather challenging. Another NMR detector which naturally lends itself to facile microfluidic integration is the Helmholtz coil geometry.11,12 Microfabrication technologies in general, and microfluidics in particular, offer several other valuable additions to NMR characterisation. Bordonali et al. and Eills et al. have independently reported microfluidic solutions which provide bubble-free hydrogen gas contact with the liquid sample, thus enabling the use of signal enhancement by means of parahydrogen induced hyperpolarisation at the microscale.13,14 Similarly, an earlier paper reports on a 3D-printed system to optimise the dissolution of hyperpolarised gaseous species for microscale NMR, and the functionality is demonstrated using 129Xe.15
Planar microfabrication including thin layer deposition of metals, UV-photolithography and etching, allow for facile integration of electrodes in the microfluidic system, thus enabling an entirely different class of experiments to be coupled with NMR, within a microfluidic environment, for small amounts of samples. In a recent paper, spatial and temporal control over multilayer assembly and composition of chitosan hydrogel, by means of electrodeposition in a microfluidic environment, was demonstrated.16
There are several application scenarios that would benefit from the introduction of metallic electrodes in, or in the vicinity, of the NMR detection region. These scenarios include NMR hyphenated with electrokinetic separation methods (e.g., capillary electrophoresis), dielectrophoresis, and a wide range of other electrochemical experiments. The presence of a metallic electrode structure impacts the NMR measurements in two ways: 1) it affects the B0 homogeneity due to a magnetic susceptibility jump at the leading edges of the tracks, with direct implications on the spectral resolution, and 2) it affects the B1 strength and homogeneity due to radiofrequency shielding effects, with direct implications on the sensitivity and uniformity of the NMR measurement. Metal track susceptibility is a rather old issue which has been reported during the very early EC-NMR attempts.17 Potential solutions to the B0 homogeneity problem, which stem from the Gauss law, have been identified in the re-design of metal electrodes with a high degree of symmetry,18,19 or rotation of the sample in a similar way as is done in solid-state NMR.18–21 The second challenge in many ways is an orthogonal problem, with a different physical origin, namely stemming from Ampere's law, and therefore requiring a different strategy. However, the solution for both problems is, to some extent, geometrical, so that neither the B0 nor the B1 field should intercept the largest projected surface of the metal electrodes. When the B1 field lines penetrate through a metal layer, the amplitude of the RF pulse is attenuated due to the skin effect. The characteristic quantity governing this effect is the skin depth, i.e., the material thickness which attenuates the RF amplitude by 1/e. The skin depth depends on the conductivity σ of the electrode material, as well as on the frequency ω of the electromagnetic field (assuming non-magnetic electrode materials). Therefore, higher B0 fields and thus higher Larmor frequencies admit thinner metals as EC electrodes, which is an additional limitation for EC applications requiring high applied currents. Addressing this issue, some authors19,22 have employed electrodes with nanometer thickness, i.e., less than 1% of the skin depth. Chen et al. provided an analytical analysis of the field penetration inside a metallic cylinder.23 According to their study, metallic layers thinner than 0.1% of the skin depth are transparent to the external RF field. In microsystems such as LOC devices, B0 and B1 perturbations are further exacerbated by the inherently smaller available real-estate.24
In this paper, acknowledging the important potential of NMR to perform in situ characterisation and monitoring of electrochemical reactions, we address the issues that arise by introducing metallic electrodes in the proximity of the NMR detection region, in particular for the case when LOC devices, miniaturised NMR detectors, and small (sub-μL) sample quantities are involved. We conjecture that this work is relevant for the μ-TAS community, because the theoretical considerations derived and experimental characterisations presented herewith are also applicable to other NMR-hyphenated scenarios where metallic structures must be similarly introduced, i.e., in electrophoresis and dielectrophoresis, and digital microfluidics based on electrowetting, in which individual droplets are manipulated on a surface of an array of electrodes by means of individually applied voltages.25–27
As a case study, we consider a microfluidic channel oriented along the static magnetic field B0 and various electrode topologies within this channel. The miniaturised NMR detector is a Helmholtz coil introduced previously by Spengler et al.12 In this way, we gain complete freedom to design a suitable microfluidic insert and the distribution of electrodes, the only geometrical precondition being that the insert fits inside the Helmholtz structure, as shown in Fig. 1a.
Various detector geometries were considered theoretically and a selection of these variants have been fabricated and tested. The study focuses on the effect of metal electrodes on the static field B0, RF field B1, and the additional noise captured in the detector. As a test platform to explore the practical effects of electrode integration with NMR-compatible microfluidics, we monitor in situ electrodeposition of chitosan inside the microfluidic channel. Chitosan (CS) is a biocompatible polyaminosaccharide with the ability to undergo reversible gelation in response to a change in solution pH.28 Chitosan plays the role of forming an interface between the technical device (here, the NMR detector) and some biological sample/process.29–31 The electrodeposition process and the change in solution pH are monitored by NMR spectroscopy.
The electrode-free reference insert is a simple microfluidic channel defined in SU-8 photoresist. The channel height was 90 μm and was sandwiched between two glass wafers, each having a thickness of 210 μm. Starting from this basic microfluidic channel structure, various electrode configurations were chosen with respect to the direction of the B1 field generated by the Helmholtz pair. From a fabrication perspective, planar disk electrodes in the zx-plane were the easiest to introduce in the microfluidic channel. A working electrode was positioned within the field of view (FOV) of the Helmholtz detector, and the counter electrode was placed 5 mm away within the microfluidic channel. The thickness of the working electrode determined the current induced inside the electrode, and hence also the field perturbation and the dissipated power. For the purpose of this study, three inserts with planar electrode thicknesses of 340 nm (∼10% × δAu), 34 nm (∼1% × δAu), and 3 nm (∼0.1% × δAu) were considered.
Patterning the electrodes on the sidewalls of the microfluidic channel was a second configuration that was explored. It was chosen because it minimises the total electrode surface being penetrated by the B1 RF field. To this end, two high aspect-ratio metallic walls were designed as the active electrodes in the sidewalls of the microfluidic channel, i.e., in the yz-plane. The width of the electrodes was set to 30 μm to ensure their mechanical stability, while minimising the cross-section to the B1 wave front. As shown in Fig. 1b, two different situations were considered: placing the sidewall electrodes 2 mm apart (outside the FOV of the Helmholtz coil), or placing inside the FOV of the detector. In the latter case, the optimum distance between the electrodes, w.r.t. RF homogeneity, mass sensitivity, and spectral resolution at the sample region, was found to be 1.1 mm.
The third option was to place a structured planar electrode in the FOV of the Helmholtz detector. The three different structures studied were mesh, meander, and comb patterns. The thickness of the electrodes was 34 nm, with a filling factor of 50%, and a linewidth of 50 μm. These structures offered more resistance to the eddy current and hence less B1 perturbation, albeit with smaller electrode surface area. Further analysis had been performed through COMSOL simulations.
In terms of its relative magnetic susceptibility to water, copper is a better choice than gold. Nevertheless, gold was chosen as electrode material for the following reasons: i) the value of the standard electrode potential (SEP) of gold is higher than that of copper, i.e., 1.69 V for gold and 0.339 V for copper;34,35 ii) copper oxidation limits the lifetime of the electrodes especially for the thin films; iii) in case of chitosan deposition, copper ion migration results in metal particles trapped in the gel.36
Of the eight different configurations, four variants were fabricated and experimentally investigated: planar disk electrodes with 34 nm and 3 nm thickness, and both versions of the sidewall electrodes.
Chitosan was coupled to PEG featuring diacid functionality. For modification of chitosan, a 1:1 CS:PEG molar ratio was chosen in order to maximise the PEG degree of substitution to chitosan amines.16 PEG was activated with EDC and NHS in 20 m of deionized water for 30 min at 24 °C. The PEG:EDC:NHS ratios was 1:0.5:0.5. This solution was then added to a 1% w/v chitosan solution containing the appropriate quantity of amine reaction centres. The reaction was allowed to proceed for 3 h at 24 °C. NaOD was added dropwise to the modified polymer solution to form a hydrogel before purification by extensive dialysis. After the dialysis process, DCl was added to redissolve the hydrogel and the pH was adjusted to pH 6. The solution was then filtered through a 5 μm syringe filter before use.
A nutation experiment was performed in order to determine the B1 field homogeneity. The nutation spectrum consists of 300 single scans using 0.8 W applied power with an increment of 1 μs. A relaxation delay of 10 s was set between two consecutive scans. Each data point in the nutation spectrum, represented by a single NMR peak, was integrated and plotted as part of a continuous curve to facilitate comparison with the simulated experiments.
Taking the gold skin depth (δgold ∼ 3.37 μm at 500 MHz) into account, eight different electrode topologies (previously introduced in sections 2.3) were simulated, critically evaluated, and compared to the situation when no electrode is inserted in the NMR sensitive volume. Four such topologies were fabricated and experimentally tested. All results are presented in a condensed manner in Table 1.
Fig. 2 (a) Simulated (dash) and measured (solid) nutation curves for different inserts: electrode-free, planar disk electrodes (340 nm-, 34 nm-, and 3 nm-thickness), structured planar (34 nm thickness) electrodes (mesh, meander, and comb), sidewall electrodes (wide channel (2 mm), narrow channel (1.1 mm)). All nutation signal amplitudes normalised to their 90° intensity. The differences between simulated and measurement signals of the 34 nm-thick planar electrode are discussed in detail in section S4.† (b) Simulated (dash) and measured (solid) 1H NMR spectra of the TSP signal in different configurations. The peak heights are normalised to unity for easier comparison. (c) Measured 1H NMR spectra of the sucrose sample in five different configurations. |
From the point of view of B1 performance, a first conclusion offered by the comparative presentation in Table 1 and Fig. 2a was that almost all structures perform in a similar manner: their simulated B1 field homogeneity was at least as good as for the electrode-free case, with the factor A450/A90 between 68% and 75%. The only exception was the 340 nm-thick planar electrode that degraded the B1 field homogeneity to A450/A90 = 25%, i.e., almost one third compared to the electrode-free case. Three structures (34 nm-thick disk, the mesh, and the narrow channel sidewall electrode) showed slightly improved field homogeneity.
The eddy currents induced in the disk metallic structures depleted the B1 field from the centre and concentrated it around the edges of the electrodes. This field distortion increased with the thickness of the metal layer. For a 34 nm-thick layer, this effect led to a flattened field distribution, which enhanced the B1 field homogeneity. The mesh electrode provided multiple closed loops for the eddy current paths, by spreading the field over the entire detection zone, and further enhancing the B1 homogeneity to A450/A90 = 71%. The inter-electrode distance for the narrow channel sidewall configuration was optimised for B1 field homogeneity, reaching a value of A450/A90 = 75%, achieved by both constraining the sample inside the certain region which had the best field homogeneity, and compressing the field inside the sample volume using the sidewall electrodes.
As expected from the analytical expressions given by Chen et al.,23 the thickest planar disk electrode showed the worst performance in terms of mass sensitivity – Srel = 86.4%. For all other structures except for the narrow channel sidewall configuration, the penalty to be paid in comparison with the electrode-free configuration, in terms of mass sensitivity, was <5%. The narrow channel sidewall electrode configuration clearly outperformed all other designs. The sensitivity enhancement for the narrow channel sidewall structure could also be observed through the nutation frequency (Fig. 2a). The slightly higher (1.04×) nutation frequency of the narrow channel sidewall electrode configuration, compared to the electrode-free insert, suggests that the presence of the vertical electrode is constructive and enhances the overall sensitivity.
The planar electrode configurations generate a material interface in the immediate vicinity of the FOV, which degrades the spectral resolution. As discussed in section S4,† this degradation was found to be more severe with increase of the thickness of the metal layer (FWHM = 16.59 ppb for 340 nm and FWHM = 9.05 ppb for 34 nm). However, for a thickness of 3 nm, the spectral resolution returned to a value comparable to the electrode-free configuration, and similar to other structures which had been evaluated. It is interesting to note that all patterned planar electrodes improved the spectral resolution in comparison to a disk electrode with the same thickness, the FWHM value being comparable to the electrode-free case, as shown in Table 1. The spectra collected from the wide and narrow channel sidewall configurations have FWHM of 4.18 ppb and 4.57 ppb, which were similar to the FWHM of the spectrum from the electrode-free insert (4.28 ppb). This was achieved by extending the electrodes along B0 to shift the material interface far from the FOV.
Three additional structures were selected for fabrication and experimental characterisation. These geometries had similar simulated performance and were, importantly, compatible with the electrochemistry experiment designed in section 3.4: the wide channel sidewall electrode structure and, for the ease of fabrication, the 34 nm and 3 nm disk electrode structures. These four structures together with the electrode-free configuration were investigated experimentally and are discussed in the next section.
The absolute values for the A450/A90 factor, extracted from the nutation curves in Fig. 2a and presented in Table 1, were found to be smaller compared to the simulated values. This is attributed to the fabrication tolerances for the Helmholtz detector, e.g., Helmholtz coil size, slight misalignment of the windings, the distance between two windings, and the effect of the extra tracks which have not been considered in the simulations, as well as the alignment of the insert with respect to the coil.
In contrast with the simulated results, the experiment showed an inferior performance of the structures with planar electrodes in terms of B1 field uniformity when compared to the electrode-free configuration. We attribute this to severe B0 distortions (discussed in section 3.2.3 and S4†) which cannot be fully compensated by shimming. This broadens the linewidth and hence reduces the height of the peak, the effect being more pronounced for higher flip angles due to the additional signal phase deformation.
The two sidewall electrode configurations confirmed experimentally the simulated behaviour. The wide channel sidewall has the same B1 field homogeneity as the electrode-free configuration (A450/A90 = 56%), whereas the narrow channel sidewall electrode exhibits a slightly higher field homogeneity (A450/A90 = 59%).
B 1 maps, collected from MRI experiments, at a flip angle corresponding to 19.2 μW for all five configurations are presented in Fig. 3a. Further B1 profiles at various excitation powers are depicted in Fig. S3.† In order to highlight the field homogeneity, the maps were scaled to unity. Fig. 3b depicts the profile of the B1 field along the x-axis at the middle of the detection zone. For the narrow channel insert, the measurements confirmed that the field homogeneity was enhanced and that a steep drop of the field happened as the sample volume was restricted by the metallic sidewall electrodes.
Fig. 3 (a) Normalised distribution of the B1 field (%) at the sample volume excited with 19.2 μW power for different types of inserts. B1 distributions using different excitation powers can be found in Fig. S3.† (b) Normalised profile of B1 along the x-axis at excitation power of 19.2 μW. (c) B0 field map at the detection zone of the coil. |
In order to investigate the B0 distribution more precisely and avoid shimming influence on the results, the B0 maps of different configurations were collected and are plotted in Fig. 2b. The standard deviation calculated from the B0 map was used as a second figure of merit for B0 distribution analysis. These values are presented in Table 1. In all configurations, the detection volume was surrounded by a rim whose voxels were severely distorted due to the partial volume effect (the sample does not fully occupy a voxel) and low B1 field. To calculate the standard deviation of the colour maps, a rim of one pixel was excluded to minimise the noise contribution. At the bottom and top of the detection zone, similar field distortions were observed in all the B0 field maps. This distortion originated from the coil windings.
The B0 pattern at the sensitive zone of the coil showed that the planar 34 nm electrode distorted the static field especially at the top and bottom edges of the electrode, which perfectly aligned the material interface intersections with B0. The ultra-thin electrodes (3 nm thickness) introduced less perturbations; however, the overall pattern appeared similar as expected. These distortions are likely introduced because of the chromium seed layer (see section S3† for further details). Conversely, the narrow channel sidewall electrodes had almost no effect on the overall field pattern, except for the left and right edges where the 1H NMR signal was excluded due to the presence of the electrodes. These results correlated with the measured spectral FWHM results.
Since the narrow channel sidewall electrode configuration proves to be the best compromise in terms of B1 field homogeneity and mass sensitivity, as well as in terms of B0 field distortion, i.e., spectral resolution, this structure was further used for the experiment of NMR in situ monitoring of chitosan electrodeposition.
Using this electrode configuration, the compatibility of an electrochemical experiment with NMR spectroscopy within a LOC environment was demonstrated using chitosan electrodeposition. This experiment is attractive first because of its relative simplicity, requiring only a voltage applied between 2 electrodes to initiate water hydrolysis, resulting in a local pH gradient as required for chitosan hydrogel deposition. Second, the NMR spectrum of chitosan in solution is significantly different compared to the gel state, and thus the deposition process, as a function of time, can be monitored. Third, chitosan can be chemically modified so that NMR signals from covalently attached molecules can also be monitored as a function of gel formation. Finally, the chitosan hydrogel can be controlled in terms of chemical composition and architecture.16
To perform the in situ experiment, a chitosan (CS) solution was injected into the narrow-channel sidewall insert (1% CS, 50 mM TSP, pH 5.5). Two versions of chitosan were investigated: the native biopolymer, and one chemically modified with polyethyleneglycol (PEG). Prior to applying a current, a 1H NMR spectrum of the solution was measured and NMR calibrations were performed. After the reference spectrum was measured, the current was applied over a working electrode area of 0.42 mm2. The current flowed for the entire experiment, and after intervals of 5 min a new 1H NMR spectrum was acquired. This cycle was repeated so that a total of 5 spectra were measured, the results are summarised in Fig. 4.
The 1H NMR signals from chitosan in solution are clearly visible in the chemical shift range 3 to 4 ppm (pink region in Fig. 4a). As expected, these signals begin broadening as a function of deposition time. The broadening results from constrained molecular motion as would be expected in the hydrogel state. A similar result was observed for chitosan when modified with PEG (Fig. 4c and d). In contrast, the PEG signal (∼3.8 ppm and ∼3.3 ppm, blue region) remained relatively sharp as the hydrogel was formed, suggesting this highly hydrated polymer maintained a degree of molecular motion. This was an interesting result, since bi-functional PEG can be used to attach interesting molecules to chitosan, which would then be potentially decoupled from the deleterious line-broadening effects when deposited as the hydrogel. This feature is currently being explored in continuing work in our group.
We conjuncture that the results of this work will be highly relevant for other applications and processes that involve the presence of metallic structures, when these are combined with NMR monitoring. Electrokinetic separation methods, where analytes move through electrolytes under the influence of an applied electric field, are widely used in μ-TAS environments and require the presence of metal electrodes within the device. Dielectrophoresis is another phenomenon extensively used in lab-on-a-chip system to immobilise dielectric particles in non-uniform electric fields. More recently, droplet microfluidics which involves manipulation of droplets of analyte using a planar array of electrodes has been combined with in situ NMR monitoring facing similar challenges.
The methodology presented here can also be extended to other electrode materials compatible with microfabrication and relevant to various electrochemical studies. For example, electrode materials including Ag/AgCl, platinum, and carbon are rather common in 3-electrode electrochemical setups, and indium-tin-oxide (ITO) can be used as transparent electrode, thus enabling direct optical observation at the surface of the electrode where deposition is taking place, or under the electrode in the process chamber.
All similar applications of interest to the lab-on-a-chip community will benefit from the present study, opening new avenues by hyphenating a very chemically specific characterisation method such as NMR with electrochemistry.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/d0lc00364f |
This journal is © The Royal Society of Chemistry 2020 |