3.1 System setup and fabrication of nanofilter membrane device
The schematics and the actual setup of the direct electrophoretic nucleic acid preparation using the nanofilter membrane device were displayed in Fig. 1. The principle of the electrophoretic preparation was to simply drive the negatively charged nucleic acid in the reservoir chamber (sample) to the collection chamber (buffer), where the positive voltage was applied via external DC power source (Fig. 1a). To construct the electrical preparation system, the membrane and the chambers were assembled with elastomer gaskets on each side of the chambers to prevent leakage of the liquids (Fig. 1b). The sample and the collection buffer (1× TE buffer in this work) were injected into each chamber, and a pair of electrodes was loaded. After setup, the nanopores acted as the separation layer between the chambers and the only transport path for the ions, the fluid, and the biomolecules.
The nanofilter membrane should prevent the mixing of the buffers in the two chambers while facilitating the transport of the nucleic acid molecules. In the molecular transport perspective, conventional molecular filter membranes such as track-etched polycarbonate membrane were unfavorable to use in this work because their 10 µm-order thicknesses resulted in low molecular fluxes [33]. In contrast, the membranes of sub-micron thicknesses were advantageous in promoting the molecular transport to the opposite chamber, while the thinness may sacrifice the mechanical vulnerability of the membranes.
Consequently, to create a thin and robust nanoporous membrane, silicon nitride was selected as the membrane material and thus semiconductor fabrication technique as the device fabrication method. From its excellent mechanical robustness and chemical stability, SiNx thin film has been widely used in solid-state nanopore fabrication [34, 35]. The semiconductor fabrication technology, lithography in particular, has a strong advantage in forming a highly aligned structure with designed feature size [36, 37]. This allowed the fabrication of a packed, orderly network of identical pores.
Figure 2 illustrated the fabrication procedure of the nanofilter membrane device with the images of the completed devices and the porous membrane [32, 34, 35, 38, 39]. Figure 2a presented the nanofilter membrane fabrication sequence, explained in detail in the experimental section. Nanoimprint technique is an efficient lithographic method that features a stamping process of a reusable polymeric mold onto nanoimprint resist, reducing time and cost for the fabrication of nanostructures [32]. As the result, Fig. 2b, c clearly displayed that the nanofilter membrane consisted of well-aligned and uniform nanopores of 200-nm diameter was formed at the center of a 1 cm × 1 cm chip. The nanopore density calculated from Fig. 2c was 7.22 × 108/cm2.
In determining the dimensions of the pore, the membrane, and the device, the reproducibility in fabrication, as well as the separation principle, were considered. Since the operation principle was mainly based on the electrical charge and the size of the particles, nanopores should be small enough to sieve relatively large-sized impurities and large enough to allow the flux of small particles including nucleic acids [33]. To stably produce a well-arranged array of small nanopores of nanometer-level size, 200 nm was selected as the optimum pore diameter. Nanoporous membranes with smaller pore sizes have been reported in a number of articles [40,41,42]; however, the distinct feature of the nanofilter membrane in this work was the ordered network of uniform pores spanning in a whole membrane to increase the channel area for the biomolecular transport. The area (700 μm to 1 mm in width) and the thickness (~ 100 nm) of the membrane were determined upon its mechanical stability during the electrical preparation experiments. During the device fabrication, the thickness of the membrane was indirectly monitored by measuring the SiNx thickness of the partially etched backside of the device using ellipsometry. Finally, the chip size was designed for the convenient handling of the device.
3.2 Direct and electrophoretic miRNA preparation: a proof-of-concept
First, to validate that the nucleic acid can electrically transport and be collected across the nanofilter membrane, hsa-mir-93-5p (miR93-5p) mimic was electrophoretically drawn from the pure miRNA mimic solution with a known concentration of 100 pg/µl. In this experiment, 1–10 V was applied to the collection chamber for 30 min to drag the negatively charged miRNA in the reservoir chamber across the membrane. Voltage application time was first set as above referring to the previous reports on lab-on-a-chip nucleic acid extraction protocols [13,14,15,16,17,18]. Longer operation times would allow more miRNA to transport to the collection chamber, but the buffers started to dry significantly after 30 min, limiting the processing time up to half an hour. Closed packaging would enable long-term and stable runs, yet still shorter times would be preferable in terms of efficiency as long as the collected miRNA is analyzable in downstream applications. The input sample volume (150 µl in the reservoir chamber) was attributed to the commercial columnar-based methods, a reference to assess the performance and efficiency of the direct electrophoretic preparation system. Following the input volume, the output volume was set to 75 μl for stability that the collection buffer stayed intact for 30 min.
The collected miRNA was amplified using qRT-PCR to compare the miRNA collection efficiency under each applied voltage condition. Threshold cycle (Ct) values were obtained in all voltages, indicating that the miRNA has transported across the nanofilter membrane under the bias voltage. When the same setup was left for 30 min without bias voltage to assess the diffusive contribution to the transport, the miRNA concentration in the collection chamber was less than 5% of those collected under the electrically biased conditions. This result suggested that the electric field generated by the applied voltage was indeed the major driving force for the miRNA movement.
To compare the molecular transport by the applied voltage and optimize the voltage condition, the collected quantities of miRNA relative to the input were calculated from the Ct values and the buffer volumes. Figure 3 presented the transported % of miRNA under 1, 2, 5, 10 V. Two regions are identifiable in the graph; 1 V and 2 V with lower percentages but clustered data points, and 5 V and 10 V with a couple of points above 5% but scattered distributions. The data fluctuation at 5 V and 10 V may be attributed to uneven evaporation and migration of the fluid under these conditions, which will be discussed below with graphics. Considering performance reproducibility, the lower voltages, especially 2 V, were the favorable collection voltage in this work. Additionally, the molecular stability of the collected miRNA was confirmed in gel electrophoresis, where the band positions obtained from the collected miRNAs corresponded to that from the reference miRNA mimic. In conclusion, from the preliminary experiments using a known miRNA mimic solution, the idea of direct electrophoretic nucleic acid preparation was practically validated in consecutive qRT-PCR. Applying 2 V for 30 min was suitable in terms of quantitative consistency over multiple experiments. Under this condition, the average collection rate of miRNA from the pure miRNA solution was 12.2 pg/min.
3.3 System stability in the electrophoretic preparation setup
As well as the consistency in data, electrical and electrochemical stability of the system during the experiment should be considered for optimizing the collection voltage. Ionic current during 30 min and change in the chamber and buffer pH after 30-min experiments were presented by the applied voltage (Fig. 4). In Fig. 4a, the ionic current stayed stable after a capacitive delay at 2 V, but the current curve started to decrease after 1200 s (5 V) and ~ 200 s (10 V) at the higher voltages.
The reducing ionic current would have resulted from the decrease in the collection buffer volumes under higher voltages; the buffer partly evaporated and migrated to the reservoir chamber. Figure 4b clearly showed that the collection buffer was significantly depleted after 30-min application of 10 V. Under 5 V, though hardly noticeable in the figure, about 20–30% of the collection buffer in average moved to the opposite chamber during 30 min while it was steady under 1–2 V. The direction of the fluid movement was the same as the electroosmotic flow direction expected from the movement of the positive counterion gathered near the negatively charged SiNx surface at pH 8.0 [43]. By principle, electroosmotic flow inside the nanopores existed also under lower voltages, but the changes in the buffer volumes were minimal in these conditions presumably due to the lower electric field.
In addition to the change in the collection buffer quantity, air bubbles could be found in the chambers at 5 V and 10 V. As Pt electrodes were used, the evolved bubbles would be hydrogen and oxygen generated from water electrolysis reaction. The gas evolutions were indirectly identifiable through pH alterations of the buffers after 30 min (Fig. 4c). The direction of pH change corresponded to the water electrolysis reactions in a basic solution (pH of TE buffer: 8.0) shown below:
$$\left( {{\text{cathode}}} \right){\text{ 4H}}_{{2}} {\text{O }} + {\text{ 4e}}^{ - } \to {\text{2H}}_{{2}} + {\text{ 4OH}}^{ - } \left( {\text{pH increase}} \right),$$
$$\left( {{\text{anode}}} \right){\text{ 4OH}}^{ - } \to {\text{O}}_{{2}} + {\text{ H}}_{{2}} {\text{O }} + {\text{ 4e}}^{ - } \left( {\text{pH decrease}} \right).$$
The standard potential (1 M, 25 °C, 1 atm) of this reaction is 1.23 V. Under 2 V, although the reaction would also have been present, there was a minimal effect in disrupting the buffering capability according to Fig. 4c. The migration of fluid, the air bubble generations, and the pH change in solutions after 30-min runs were observed in all 5 and 10-V runs. Therefore, the electrically induced fluid migration and water electrolysis reaction were active under the high voltages, and the instability of the system would have adversely affected the reproducibility of the transport yield. To summarize, considering the quantitative and qualitative discussions above—though miRNA was reported to be stable under a wide range of pH [44]—2 V was set as the optimum voltage condition for the electrophoretic miRNA collection. Henceforth, all experiments were conducted under 2 V for 30 min in the sections below.
3.4 Yield and quality of the electrically collected miRNA
Even though miRNA successfully and stably transported across the nanofilter membrane as above, the electrically collected amount of miRNA was as small as a few percent of the input quantity (Fig. 3). To analyze the origin of the migration yield or the electrophoretic driving force, the chamber system was constructed as a simple circuit model in Fig. 5a. The circuit consisted of a power source (V), resistances of the fluidic chambers (Rres, Rcol) and the nanopores (Rpore), and the resistive and capacitive components in the electric double layer (EDL) at the electrode-solution interface (REDL, CEDL) [45]. Here, the capacitance of the silicon chip was neglected from its small value (in pF order) [46, 47] and for simplicity. From the nonzero saturation current at 2 V in Fig. 4a, the platinum electrodes were assumed a realistic electrode, having the resistive term REDL [48]. In principle, the electrical driving force was proportional to the electric field thus the voltage drops in each part of the system: more fundamentally, the relative magnitude of each resistance.
To quantitatively identify each component in the circuit, the ionic current curve was recalled from Fig. 4a. At 2 V, the time-dependent ionic current can be mathematically interpreted as in the equation below:
$$I\left(t\right)=\left\{I\left(0\right)-I\left(\infty \right)\right\}\cdot {\mathrm{e}}^{-\left(\frac{1}{{R}_{EDL}{C}_{EDL}}+\frac{1}{\left({R}_{res}+{R}_{col}+{R}_{pore}\right){C}_{EDL}}\right)\cdot t}+I\left(\infty \right),$$
where I(t) is the ionic current at 2 V as a function of time t, I(0) is the initial ionic current at t = 0, and I(∞) is the saturated ionic current [46]. By curve fitting and calculating the chamber and the nanopore resistance from the known dimensions, total EDL resistance was 1.13 × 105 Ω, Rres + Rcol was 1.23 × 104 Ω, and Rpore was 39.8 Ω.
Calculating from the resistances, nearly 90% of the total voltage drop occurred at the electrode-buffer interfaces. Pt electrodes were ideal in the electrochemical aspect—the byproducts of the electrochemical reaction were only hydrogen and oxygen—but the electrophoretic driving force inside the chambers was substantially reduced by the Pt electrode-solution interface. The resistance of the parallel-aligned nanopore was minimal in the electrical setup. Nevertheless, due to the ~ 100 nm thickness of the membrane, the electric field inside the nanopores was as high as 6 × 103 V/m from a simple calculation of voltage drop/pore length. This electric field strength was higher than that in a typical gel electrophoresis setup (~ 100 V/10 cm). Therefore, despite the large parasitic resistance at the EDL of the electrode, the electrophoretic driving force inside the nanopores was effective in transporting miRNA across the membrane.
On the other hand, from their millimeter-scale length, the electric field inside the chambers was insufficient to drive the charged molecules to the high-field nanopore region. This was a limitation in the electrical setup in this work, where the chamber volume and dimensions have been designed with a view to comparison with the conventional miRNA extraction methods. Now that the electrophoretic preparation strategy was proven valid in this work, a microfluidic structure would be an efficient improvement of the fluidic chambers to exert adequate electric field throughout the whole system and increase the migration efficiency.
To confirm the feasibility of the direct electrophoretic preparation from clinical samples, miRNA was collected from human blood serum using the electrophoretic nanofilter system and the conventional columnar method as a reference. As previously mentioned, the input sample and the output buffer volumes were fixed to 150 µl and 75 µl respectively in all experiments. In the electrophoretic trials, 2-V, 30-min condition was employed and TE buffer was used as the collection buffer.
From the sequential gel electrophoresis result in Fig. 5b, the preparations of miRNA from a clinical sample were successful in both protocols. Therefore, the electrophoretic preparation system was also capable of collecting miRNA from the serum and compatible with qRT-PCR. Further, to identify the chemical components in the collection buffer after the experiment, spectrophotometry analysis was performed for both samples (Fig. 5c). The absorbance spectrum of the directly collected miRNA exhibited a peak at 280 nm wavelength, representing the presence of proteins in the solution (Fig. 5c, upper panel). A260/280 purity of the electrophoretically prepared solution was 0.594, also corresponding to the value of protein [49]. In contrast, the absorbance spectrum of the chemically extracted miRNA solution showed no sign of protein as they were excluded and washed out during the extraction process. A260/280 purity of the conventionally extracted solution was 1.72. Instead, a peak at 270 nm was observed in the spectrum (Fig. 5c, lower panel), representing phenol residue from the lysis reagent used in the extraction protocol. Phenol is a PCR inhibitor known to degrade the DNA polymerase [50]. A 260-nm peak, indicating the presence of nucleic acids, was absent in both spectra. This observation was reasonable though, from that the typical miRNA concentration in blood serum was lower than the detection limit of the spectrometer used in this study [26, 51].
Hence, protein co-migration along with miRNA apparently existed in the electrical setup, where the nanofilter membrane was incapable of completely blocking the negatively charged proteins in the blood sera moving in the same direction as miRNA. By principle, the electrophoretic transport mechanism applied to all charged particles in the chambers, including nucleic acids, albumins, and globulins, where the last two are negatively charged proteins abundantly existing in blood serum [52]. Therefore, a fundamental solution to eliminate the proteins from transporting with nucleic acids would be separation by size. However, the differences in the diameters of nucleic acid strands (single-strand: ~ 1 nm, double-strand: ~ 2 nm) and the proteins (albumins and globulins in blood serum: 4–5 nm to < 10 nm) are only a few nanometers [53], thus the window for the separation is limited. Another limitation to the separation by size is difficulties in fabrication of nanoporous structure with uniform few-nm sized nanopores. Nevertheless, despite the imperfect separation, from Fig. 5b and the results in the section below, the electrophoretically collected miRNA was still able to be analyzed by the consecutive qPCR.
In summary, the electrophoretic miRNA collection protocol showed limitations of low miRNA transport yields and imperfect separation of the nucleic acids from proteins in clinical samples. The low yield and specificity in the electrophoretic transport were native issues in the system, where the electrochemically favorable Pt electrodes acted as a large parasitic resistance and the electrophoretic migration heavily relied on the electrical nature of the biomolecules. Nevertheless, the validity of the direct electrophoretic miRNA preparation method was still proven from the gel image of the collected miRNA from clinical serum samples.
3.5 Direct electrophoretic miRNA preparation from human blood sera
As the final demonstration of the direct miRNA preparation from clinical samples, the same procedure was conducted with the serum donated by hepatocellular carcinoma (HCC) patients and healthy individuals. In liquid biopsy studies, miR93-5p concentrations in the tissue and the blood were reported to be upregulated in diverse cancer patients including HCC compared to the control healthy group [30, 54,55,56,57]. Similar to the previous section, miRNA was prepared from each serum of 150 µl chemically (using the column-based kit) and physically (using the electrophoretic preparation system and nanofilter membrane) to 75 µl of the clean solution to proceed to qRT-PCR. The amplification results were presented in Fig. 6. As previously mentioned, the conventional method could only be performed once due to the limited amount of the sample provided, while the electrophoretic experiments were repeated 3 times. Interestingly, the Ct value from the electrophoretic preparation was comparable to those obtained from the conventional extractions of the same sample. The average Ct values from the HCC patients were 32.38 (kit extraction) and 32.69 (electrophoretic preparation) when the values from the healthy controls were 33.83 (kit) and 34.06 (electrical). Therefore, the degradation of the qPCR efficiency was small on average, or it was rather insignificant considering that the physical protocol also allowed some protein molecules to move along with miRNA. The average miR93-5p levels in the two groups were close to each other, which was an unexpected trend. One recognizable reason would be the small sample size (n = 5 for each group). In conclusion, the direct electrophoretic system was capable of collecting miRNA from human blood sera in both experiments, and the collected gene was identifiable and analyzable using qRT-PCR.
To summarize, the simple electrical method was successful in collecting miRNA from the clinical serum samples and was compatible with the conventional downstream application of genetic analysis, despite the impurities transported with miRNA that may have negatively affected PCR amplification.