We demonstrate a microfluidic platform with an integrated surface electrode network that combines resistive pulse sensing (RPS) with code division multiple access (CDMA), to multiplex detection and sizing of particles in multiple microfluidic channels.
Microfluidic processing of biological samples typically involves differential manipulations of suspended particles under various force fields in order to spatially fractionate the sample based on a biological property of interest. For the resultant spatial distribution to be used as the assay readout, microfluidic devices are often subjected to microscopic analysis requiring complex instrumentation with higher cost and reduced portability. To address this limitation, we have developed an integrated electronic sensing technology for multiplexed detection of particles at different locations on a microfluidic chip. Our technology, called Microfluidic CODES, combines Resistive Pulse Sensing with Code Division Multiple Access to compress 2D spatial information into a 1D electrical signal. In this paper, we present a practical demonstration of the Microfluidic CODES technology to detect and size cultured cancer cells distributed over multiple microfluidic channels. As validated by the high-speed microscopy, our technology can accurately analyze dense cell populations all electronically without the need for an external instrument. As such, the Microfluidic CODES can potentially enable low-cost integrated lab-on-a-chip devices that are well suited for the point-of-care testing of biological samples.
Accurate detection and analysis of biological particles such as cells, bacteria or viruses suspended in liquid is of great interest for a range of applications1,2,3. Well-matched in size, microfluidic devices offer unique advantages for this purpose such as high-sensitivity, gentle sample manipulation and well-controlled microenvironment4,5,6,7. In addition, microfluidic devices can be designed to employ a combination of fluid dynamics and force fields to passively fractionate a heterogeneous population of biological particles based on various properties8,9,10,11,12. In those devices, the resultant particle distribution can be used as readout but spatial information is typically accessible only through microscopy, limiting the practical utility of the microfluidic device by tying it to a lab infrastructure. Therefore, an integrated sensor that can readily report particles' spatiotemporal mapping, as they are manipulated on a microfluidic device, can potentially enable low-cost, integrated lab-on-a-chip devices that are particularly attractive for the testing of samples in mobile, resource-limited settings.
Thin film electrodes have been used as integrated sensors in microfluidic devices for various applications13,14. Resistive Pulse Sensing (RPS) is particularly attractive for integrated sensing of small particles in microfluidic channels as it offers a robust, sensitive, and high-throughput detection mechanism directly from electrical measurements15. In RPS, the impedance modulation between a pair of electrodes, immersed in an electrolyte, is used as a means to detect a particle. When the particle passes through an aperture, sized on the order of the particle, the number and amplitude of transient pulses in the electrical current are used to count and size particles, respectively. Moreover, the sensor geometry can be designed with a photolithographic resolution to shape resistive pulse waveforms in order to enhance sensitivity16,17,18,19 or to estimate vertical position of particles in microfluidic channels20.
We have recently introduced a scalable and simple multiplexed resistive pulse sensing technology called Microfluidic Coded Orthogonal Detection by Electrical Sensing (Microfluidic CODES)21. Microfluidic CODES relies on an interconnected network of resistive pulse sensors, each consisting of an array of electrodes micromachined to modulate conduction in a unique, distinguishable manner, so as to enable multiplexing. We have specifically designed each sensor to produce orthogonal electrical signals similar to the digital codes used in code division multiple access22 (CDMA) telecommunication networks, so that individual resistive pulse sensor signal can be uniquely recovered from a single output waveform, even if signals from different sensors interfere. In this way, our technology compresses 2D spatial information of particles into a 1D electrical signal, permitting monitoring of particles at different locations on a microfluidic chip, while keeping both device- and system-level complexity to a minimum.
In this paper, we present a detailed protocol for experimental and computational methods necessary to use the Microfluidic CODES technology, as well as representative results from its use in analysis of simulated biological samples. Using the results from a prototype device with four multiplexed sensors as an example to explain the technique, we provide protocols on (1) the microfabrication process to create microfluidic devices with the Microfluidic CODES technology, (2) the description of the experimental setup including the electronic, optical, and fluidic hardware, (3) the computer algorithm for decoding interfering signals from different sensors, and (4) the results from detection and analysis of cancer cells in microfluidic channels. We believe that using the detailed protocol described here, other researchers can apply our technology for their research.
1. Design of Coding Electrodes
Note: Figure 1a shows the 3-D structure of the micropatterned electrodes.
2. Microfabrication of Surface Electrodes
Note: Figure 2b shows the fabrication process of surface electrodes.
3. Fabrication of the SU-8 Mold for Microfluidic Channels
Note: Figure 2a shows the fabrication process of the mold for microfluidic channels.
4. Assembly of the Microfluidic CODES Device
5. Preparation of the Simulated Biological Sample
6. Running the Microfluidic CODES Device
Note: Figure 3 shows the experimental setup.
7. Processing of Sensor Signals
A Microfluidic CODES device consisting of four sensors distributed over four microfluidic channels is shown in Figure 1b. In this system, the cross-section of each microfluidic channel was designed to be close to the size of a cell so that (1) multiple cells cannot pass over the electrodes in parallel and (2) cells remain close to the electrodes increasing the sensitivity. Each sensor is designed to generate a unique 7-bit digital code. The device was then tested using a cell suspension. Recorded electrical signals corresponding to four individual sensors are shown with associated ideal digital codes in Figure 4. Recorded signals closely match with the ideal square pulses, while small deviations do exist. Such deviations result from a combination of several factors including the non-uniform electric field between coplanar electrodes, coupling between different electrode pairs, spherical shape of cells, as well as the constant flow speed of cells in microfluidic channels. We created a template library based on the recoded sensor signals. By correlating the recorded signals with all of the templates in the library, we determined a template that produced the maximum auto-correlation peak (Figure 4). As the digital codes for microfluidic channels are designed to be orthogonal to each other, a dominant auto-correlation peak could robustly be identified in this process. Using this approach, we could computationally determine the microfluidic channel the cell passed through, the duration of the sensor signal, and therefore the flow speed of the cell.
The Microfluidic CODES technology can resolve situations when multiple cells simultaneously interact with coding electrodes. When such overlaps occur, the signals from individual sensors interfere and the resultant waveform cannot readily be associated with any single template corresponding to a specific sensor. Accurately decoding such overlapping signals is particularly important for reliable processing high-density samples, where interferences are more likely to occur. To resolve overlapping events, we developed an iterative algorithm based on a successive interference cancellation (SIC) scheme24,25, which is typically used for multi-user detection in CDMA communication networks. Figure 5 demonstrates how the SIC algorithm is implemented in resolving a waveform that resulted from four overlapping cells in four different microfluidic channels. In each iteration, we first determined the dominant auto-correlation peak (Figure 5a, 2nd column), corresponding to the strongest interfering signal, by correlating the input waveform (Figure 5a, 1st column) with the template library. Based on the selected template and the resultant auto-correlation amplitude, we then estimated the strongest interfering signal (Figure 5a, 3rd column) and subtracted it from the input waveform. The remaining waveform was passed to the next iteration as the input. This process continued until the correlation of the residual signal with the template library did not produce a clear auto-correlation peak (Figure 5a, 5th row, 2nd plot). Following the termination of the interference cancellation process, we reconstructed an estimate of the waveform by combining all the estimated signals from each iteration (Figure 6a). Using an optimization process based on a least squares approximation to minimize the mean square error between the original waveform and the reconstructed signal, we updated our estimates for the amplitude, duration, and relative timing of individual sensor code signals (Figure 6b). We also estimated the size of the cells detected based on the amplitude of the estimated individual sensor signals. To achieve this, we calibrated the electrical signal amplitudes with optically measured cell sizes using linear regression (Figure 6b). A comparison of our results from the Microfluidic CODES with the information obtained from the simultaneously recorded high-speed microscope images shows that the cell size and speed can be accurately measured, which validates our results (Table 1). Figure 6c shows the simultaneously recorded high-speed microscopy image used for validating the decoding result.
To demonstrate the reproducibility of our results and also the performance of the Microfluidic CODES technology for a high-throughput sample processing, we analyzed electrical signals corresponding to >1,000 cells. The signals were automatically decoded in MATLAB by running the algorithm explained above and the accuracy of our results was evaluated by directly comparing our results with optical data from simultaneously recorded high-speed video. Our analysis indicates that electrical signals from 96.15% of cells (973/1,012) were accurately decoded. Success rate for decoding non-overlapping and overlapping cell signals is 98.71% (688/697) and 90.48% (285/315), respectively.
Figure 1. Design of the four-channel Microfluidic CODES device. (a) Electrodes in each microfluidic channel are micropatterned to generate a unique digital code. The impedance modulation due to sequential interactions of flowing cells with electrode pairs leads to electrical pulses. (b) A microscope image of the Microfluidic CODES device. During the fabrication process, glass substrate with coding surface electrodes is aligned with PDMS microfluidic channels under a microscope. (c) A close-up image of coded surface electrodes producing 7-bit Gold sequences: "1010110", "0111111", "0100010", "0011000". Please click here to view a larger version of this figure.
Figure 2. Microfabrication process. (a) The PDMS microfluidic channels are fabricated using soft lithography27. (b) The surface electrodes are fabricated using a lift-off process. (c) A cross-sectional schematic of the final device. PDMS microfluidic channels are aligned and bonded to the glass substrate with surface electrodes. Please click here to view a larger version of this figure.
Figure 3. Experimental setup. Using a syringe pump, the cell suspension is run through the Microfluidic CODES device at a constant flow rate. A 400 kHz AC signal is applied to the reference electrode using a function generator. Current signals from positive and negative sensing electrodes are first converted into voltage signals using two transimpedance amplifiers and subtracted from each other using a differential amplifier. The differential bipolar signal is extracted by a lock-in amplifier and then sampled into a computer for signal processing and decoding. High-speed optical microscopy is used to optically record operation of the device for validation and characterization purposes. Please click here to view a larger version of this figure.
Figure 4. Recorded electrical signals from individual sensors and their correlations. Recorded signals and their correlation with each other are given for four code-multiplexed resistive pulse sensors. Sensor 1 (a), sensor 2 (b), sensor 3 (c) and sensor 4 (d) were designed to produce 7-bit digital waveforms "1010110", "0111111", "0100010", and "0011000", respectively. For each sensor, the top figure shows that normalized signal recorded from each sensor matches closely with the ideal square pulse sequence that the sensor was designed to produce. For each sensor, the lower panel shows recorded sensor signal's autocorrelation and cross-correlation with signals corresponding to three other code-multiplexed sensors in the network. In all cases, an autocorrelation peak can robustly be identified because the digital codes from individual sensors are designed to be orthogonal to each other. Please click here to view a larger version of this figure.
Figure 5. Decoding an overlapping waveform with successive interference cancelation. In each iteration, the input waveform (1st column) is correlated with the preassembled template library to identify the specific template that results in the maximum correlation amplitude (2nd column). Using this specific template, the strongest interfering signal is estimated based on the amplitude and timing information from the correlation peak (3rd column). The estimated signal is then subtracted from the original waveform, effectively canceling the strongest interference due to the largest cell. The process is iterated until no correlation peak can be determined (i.e., correlation coefficient < 0.5) in the residual signal. Please click here to view a larger version of this figure.
Figure 6. Decoding result analysis. (a) Estimated signals are refined based on an optimization algorithm that aims to obtain the best fit between the reconstructed and the original recorded waveform using the least-squares approximation. (b) At the end of the optimization process, the timing and amplitude of calibrated signals accurately reflect the cell parameters measured by high-speed microscopy. (c) Simultaneously recorded high-speed microscopy image validates our results from electrical measurements. Please click here to view a larger version of this figure.
Measurement type | rch1 (μm) | rch2 (μm) | rch3 (μm) | rch4 (μm) | ∆t1 (ms) | ∆t2 (ms) | ∆t3 (ms) |
Electrical | 8.010 | 6.490 | 5.300 | 6.550 | 0.465 | 1.705 | 0.744 |
Optical | 8.320 | 6.770 | 5.680 | 7.040 | 0.375 | 1.625 | 0.750 |
Table 1. Comparison of electrically and optically measured cell parameters of Figure 6b. To validate our estimations, we optically measured the cell sizes from the high-speed microscopy image. Relative timing between different cells is optically measured from the number of frames between the cells in the high-speed video recorded at 8,000 frames per second.
Multiple resistive pulse sensors have previously been incorporated into microfluidic chips28,29,30,31,32. In these systems, resistive pulse sensors were either not multiplexed28,29,30,31 or they required individual sensors to be driven at different frequencies32. In both cases, dedicated external connections were needed for each resistive pulse sensor on the chip and therefore a large number of sensors could not be integrated without greater hardware complexity. The important advantage of Microfluidic CODES is that it allows simultaneous reading of multiple resistive pulse sensors from a single output in a simple device. We achieve this by utilizing multiplexing techniques commonly used in telecommunications to design micromachined resistive pulse sensors integrated into microfluidic devices. In essence, our technology relies on code-multiplexing a network of on-chip Coulter counters by designing each to produce a distinguishable signal when a particle is detected. Each micromachined sensor in the network consists of multiple coplanar surface electrodes ordered in differing configurations such that the sequential interaction of flowing particles with these electrodes produces orthogonal impedance modulations waveforms. To accommodate asynchronous particle-sensor interaction, we specifically designed each sensor to produce Gold codes33, pseudo-orthogonal digital codes that are typically used in the uplink of the CDMA telecommunication networks. Gold codes maintain a certain level orthogonality even when they are misaligned with random phase differences34.
Microfluidic CODES is easily scalable. Although we presented results from a prototype Microfluidic CODES device with four sensors in this paper, more sensors can be incorporated in the device when designed to produce output signals distinguishable from the rest. One way to expand the sensor network is to design sensors based on larger orthogonal code sets with longer digital codes. Longer orthogonal codes with more bits provide higher processing gain in decoding and can be distinguished from each other when there is interference. On the other hand, longer Gold codes in the device also means larger sensing volume, which increases the expected number of interfering sensors. Likewise, increasing the number of sensors for a given sample density will lead to more particles overlapping due to an increase in the overall sensing volume. As such, the density of the particles in the sample is a critical parameter that needs to be considered in using the Microfluidic CODES technology. The maximum particle density that can be resolved (in analogy with the channel capacity of a CDMA telecommunications network) depends on several factors such as the individual sensor signals and their relation, the decoding scheme, layout of the microfluidic device, and the electronic noise level. Depending on the application, the sample can be diluted to reach a particle density that produces an acceptable error rate.
From signal processing perspective, decoding of time waveforms from a Microfluidic CODES device is not computationally intensive using current systems as evidenced by the fact that cell phone communications on a CDMA network can be demultiplexed in real-time. Furthermore, the physical events to be decoded in microfluidic devices happen much slower than bit transmission rate in cell phone communications allowing us to use more advanced and time-consuming algorithms such as SIC and an optimization processes, which we use to iteratively resolve overlapping signals from sensors.
Taken together, Microfluidic CODES is a versatile, scalable electronic sensing technology that can be readily integrated into various microfluidic devices to realize quantitative assays by tracking particles as they are processed on the chip. The technology is very easy to implement, because (1) it is very simple from a hardware perspective (2) it is directly compatible with the soft lithography (3) it provides a direct electronic read-out without any active on-chip component, and (4) it relies on simple computational algorithms for signal processing and data interpretation.
The authors have nothing to disclose.
This work was supported by National Science Foundation Award No. ECCS 1610995. The authors would like to thank the Institute of Electronics and Nanotechnology and the Parker H. Petit Institute for Bioengineering and Bioscience staff for their support in using shared facilities. The authors also would like to thank Chia-Heng Chu for his help in preparing the manuscript.
98% Sulfuric Acid | BDH Chemicals | BDH3074-3.8LP | |
30% Hydrogen Peroxide | BDH Chemicals | BDH7690-3 | |
Trichlorosilane | Aldrich Chemistry | 235725-100G | |
NR9-1500PY Negative Photoresist | Furuttex | ||
Resist Developer RD6 | Furuttex | ||
Acetone | BDH Chemicals | BDH1101-4LP | |
SU-8 2015 Negative Photoresist | Microchem | SU8-2015 | |
SU-8 Developer | Microchem | Y010200 | |
Polydimethylsiloxane (PDMS) | Dow Corning | 3097358-1004 | Sylgard 184 Silicone Elastomer Kit |
Isopropyl Alcohol | BDH Chemicals | BDH1133-4LP | |
RPMI 1640 | Corning Cellgro | 10-040-CV | |
Fetal Bovine Serum (FBS) | Seradigm | 1500-050 | |
Penicillin-Streptomycin | Amresco | K952-100ML | |
Phosphate-Buffered Saline (PBS) | Corning Cellgro | 21-040-CM | |
PHD 22/2000 Syringe Pump | Harvard Apparatus | 70-2001 | |
HF2LI Lock-in Amplifier | Zurich Instrument | ||
HF2TA Current Amplifier | Zurich Instrument | ||
Eclipse Ti-U Microscope | Nikon Corporation | ||
DS-Fi2 High-Definition Color Camera | Nikon Corporation | ||
v7.3 High-speed Camera | Phantom | ||
PCIe-6361 Data Acquisition Board | National Instruments | 781050-01 | |
BNC-2120 Shielded Connector Block | National Instruments | 777960-01 | |
PX-250 Plasma Treatment System | Nordson MARCH |