Introduction

The outbreak of coronavirus disease 2019 (COVID-19), which resulted from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection at the end of 2019, posed significant challenges to medical systems worldwide. Societal vulnerability in the face of a pandemic was highlighted by this event. In most countries, access to large-scale detection of COVID-19 and subsequent treatment remains limited. Countries that developed such capabilities experienced a massive shift in resource allocation and significant disruptions to daily life1. The pandemic has reinforced the importance of preventive measures, specifically the need for continuous pathogen surveillance and prompt quarantine measures, to curb the transmission of infectious diseases.

Numerous methods are available for virus detection2, and the most commonly used methods are antibody detection3 and nucleic acid amplification tests (NAATs)4. Once invaded by SARS-CoV-2, the human body produces IgM or IgG antibodies for defense. Methods for antibody detection measure the levels of these antibodies in blood samples, providing an indirect indication of the presence of SARS-CoV-2. This method of detection, which does not require cell culture, can be completed within minutes. However, the production of detectable antibody levels may not occur in the early stages of infection, complicating the confirmation of SARS-CoV-2 infection. In contrast, NAATs can label and amplify viral sequences with fluorescent or other markers in vitro, facilitating the detection of the virus regardless of the infection stage. Thus, NAATs prove to be relatively suitable for detecting SARS-CoV-2, especially in early infection stages.

Viral diagnostics typically involve the extraction of ribonucleic acid (RNA) and its subsequent detection. Initially, RNA is isolated and purified from a nasal or throat swab. This process is followed by either reverse transcription-quantitative polymerase chain reaction (RT–qPCR) or reverse transcription loop-mediated isothermal amplification (RT–LAMP), techniques that amplify RNA to a fluorescently detectable level. In recent years, developed instruments, both commercial and prototype instruments, have combined RNA extraction and detection via either RT‒PCR or RT‒LAMP to obtain a sample-to-answer diagnosis5. Such instruments reduce the potential for cross-contamination during sample transfer between instruments and enhance diagnostic efficiency6. The market has offered quite a few sample-to-answer systems for COVID-19 diagnostics, including the Vivalytic Analyser (Bosch GmbH, Germany)7, Accula™ (Mesa Biotech, Inc., San Diego, CA, U.S.A.)8, and Cobas® Liat® (Roche Diagnostics International AG, Basel, Switzerland) instrument9. These systems perform the required tests as well as scan test cartridge identification codes, linking test results with patient data to expedite the diagnostic process10,11,12. The Convergys® RT‒PCR COVID‒19 testing platform can be used to conduct a fluorescence test within 90 min13. A system utilizing RT-LAMP has been shown to be capable of detecting as few as 10 to 100 RNA copies within one hour14, automating the process through the use of magnetic beads for nucleic acid extraction. However, the size and complexity of these instruments, along with their cost, make them impractical for point-of-care testing (POCT) and challenging to use in outdoor conditions. Furthermore, the demanding operation and long detection times of POCT systems have hindered their broader adoption.

Microfluidic techniques are well suited for POCT applications due to their simple structure, compact size, and affordability. Integrated off-center microfluidic platforms can store and release reagents, whereas most centrifugal systems depend on valves15,16, resulting in complexity and higher manufacturing costs17,18,19,20,21,22,23. Rapid paper-based diagnostic systems exist for DNA capture through electrostatic adsorption onto filter paper fibers, but these often exhibit less than 80% efficiency24,25,26. Microarray-based DNA detection systems capable of simultaneously diagnosing multiple viruses are also available. However, their long production cycles and high costs make them less ideal for POC diagnostics27,28,29,30,31. Recent developments in sample-to-answer systems for COVID-19 diagnosis include rapid testing, high-throughput methods, and automated detection32,33,34. However, these systems have not become commonplace due to their cost, size, or design deficiencies. Thus, simple, rapid and low-cost diagnosis remains a significant challenge for the application of microfluidic sample-to-answer systems.

Here, we developed a rapid, versatile, and low-cost sample-to-answer system for SARS-CoV-2 diagnostics. The system consists of RNA extraction using magnetic beads, reverse transcription, and subsequent amplification and detection. Compared with commercial sample-to-answer systems for COVID-19 diagnostics approved by the Food and Drug Administration (FDA), the proposed system can implement both PCR and LAMP methods based on diagnostic and reagent requirements (Fig. 1). The major advantage of LAMP is its ability to function with minimal power consumption. Designing hardware specifically for LAMP requires thermal insulation of the heated sections. Running PCR on such a system would result in a markedly slower process. Consequently, using a real-time PCR system to run LAMP has limited practicality, as it utilizes the advantages of the simple system of LAMP compared to that of PCR. However, demonstrating the system’s ability to perform both PCR and LAMP analyses is crucial, highlighting its versatility for POCT. The distinctive features of LAMP include rapid amplification speed relative to traditional PCR and straightforward end-point detection methods such as turbidity35 or colorimetry36. In contrast, PCR offers the flexibility to design multiplex assays within a single reaction.

Fig. 1: Workflow of a microfluidic diagnostic system for RNA detection from sample collection to result display.
figure 1

a Sample collection: A sample is collected under an outdoor tent. b Reagent setup: Test tubes filled with various reagents are used for RNA detection. c RNA preparation protocol: A flow diagram details the steps of the RNA preparation process using magnetic beads. d Sample extraction and subsequent amplification on a microfluidic chip: An illustration of the microfluidic chip shows the sample and reagents loaded into the corresponding port. e Photograph of the POCT microfluidic system: A vertical view of the microfluidic diagnostic device, highlighting its overall features with a chip in place for analysis

Full size image

Additionally, our system can complete the entire assay within 28 min, including sample loading (5 min), RNA extraction (3 min), and RT-LAMP (20 min), making it competitive with most FDA-approved devices (Table 1). The poly(methyl methacrylate) (PMMA)-based sample-to-answer chip allows for mass production without complex treatment procedures, significantly reducing the cost of a single test to ≈$9.5, which is much less than that of commercial alternatives. Further cost reductions are achievable through mass production techniques such as roll-to-roll hot embossing37 or injection molding38. A key advantage of our compact, portable system is its ability to provide immediate diagnoses at the point of interest, eliminating the need to transport samples to a centralized facility.

Table 1 Comparison of sample-to-answer systems
Full size table

Materials and methods

System setup

The system comprised a sample-to-answer chip, a temperature control module, and an optical detection module (Fig. 2). A single-chip controller with built-in firmware controls all system operations, including magnet motion, temperature control, and fluorescence detection. Users can program all steps using four buttons at the POCT unit, and a rotating wheel allows for manual control of magnet locations. For added convenience, the system included a universal serial bus (USB) interface to facilitate communication with a personal computer (PC). A block diagram is available in Supplementary Section A.

Fig. 2: Components of the microfluidic diagnostic system.
figure 2

a System overview: The microfluidic diagnostic device featuring a stepper motor assembly and user interface with operational buttons and display. b Cross-sectional view of the temperature system: Illustration depicting the temperature system containing a thermoelectric cooler (TEC), Pt100 temperature sensor and heat sink with an electric fan. c Cross-sectional view of the optical system (left) and optical pathway schematic (right): diagram showing the LED light source passing through an excitation filter to the sample, with the emitted light collected by a photodiode (PD) after passing through an emission filter. d Software control user interface

Full size image

Sample-to-answer chip

Chip design and fabrication

We used a JAVA-based script in Nanolithography Toolbox software39 with integrated microfluidics features40 to design the chip, generating a file in the graphic design system II (GDS II) format. The design included features tailored for microfluidics, such as a chamber with tangential connections and a rounded path for fluid delivery, ensuring minimal dead volume22. The PCR/LAMP chamber, the system’s main feature, had a nominal volume of 6 μl and an assumed depth of 100 μm. Additionally, the system included two smaller chambers: one for capturing magnetic particles and the other for purifying these particles with mineral oil to remove debris produced by cell/virus lysis (Fig. 3). We transformed the GDS II file into a data exchange format and then into a product model data format, commonly used for computer numerical control (CNC) machining. A ≈ 3 mm-thick PMMA chip with an area of 25 × 75 mm², was vertically milled to form the microfluidic channels. The vertical and horizontal channels had widths of ≈955 µm and ≈1350 µm, respectively. This CNC-fabricated chip was then bonded to a ≈ 1 mm thick flat piece of PMMA through chemically assisted thermal bonding following the protocol described in 2007 41.

Fig. 3: Microfluidic chip for RNA detection and schematic overview.
figure 3

a Actual chip: A photograph of the microfluidic chip filled with fluorescein illuminated with blue light, thus increasing the contrast of all microfluidics channels. b Schematic diagram: The diagram illustrates the microfluidic pathway within the chip, detailing the flow (yellow dashed line) of RNA samples through various processing stages, including extraction, oil, and PCR/LAMP chambers

Full size image

Sample in-chip workflow

We used an adsorption technique with magnetic nanoparticles for RNA extraction, which involved several key steps: virus lysis, RNA/DNA adsorption, debris washing, and RNA/DNA elution. Off-chip, we spiked synthetic viral RNA into lysis buffer containing magnetic nanoparticles42. In this step, RNA was released from the lysed cells and adsorbed onto the magnetic nanoparticle surface. The RNA-magnetic nanoparticle (RNA-MNP) complex was then sequentially pumped into the chip, where a permanent magnet mounted on an arm attached to a stepper motor in the system trapped and moved the RNA-MNPs. In the second step, we washed the RNA-MNP complex in the oil chamber using M5904 mineral oil (Sigma‒Aldrich, MO, U.S.A.) to remove PCR/LAMP inhibitors attached to the magnetic nanoparticles. Subsequently, we moved the complex to the final position—the PCR/LAMP chamber—where the RNA template was mixed with the master mix. We also used the same oil to block all the microfluidic chip input ports to prevent sample evaporation and removal of the chamber during amplification. Isothermal reverse transcription (RT) of RNA and complementary DNA amplification were performed by RT-LAMP, and we detected the resulting amplicons using a fluorescence-based method.

Temperature control

We heated/cooled the microfluidic chip using a thermoelectric cooler (TEC) powered by an H-bridge with pulse-width modulation (PWM) that used electrical current pulses at a nominal frequency of 100 kHz. We controlled the dissipated power through the duty cycle modulation of the PWM. The direction of the electrical current determines whether the TEC is cooling or heating. We equipped the top plate of the TEC with a Pt100-type temperature sensor integrated into a closed feedback loop system for proportional integrative derivative (PID) control.

Optical detection

We developed and tested a single-point fluorescent detection system using a light-emitting diode (LED) with a nominal wavelength of 465 nm and a luminous intensity of 4.7 cd (Fig. 2). This system was powered by electrical current pulses set to an amplitude of 200 mA, created by 5 V pulses and a series resistor with a nominal value of 10 Ω. The pulse frequency was set to 998 Hz, with a duty cycle of 10%. The optical system included a fluorescein isothiocyanate (FITC) filter set comprising three filters. The low-pass filter, with a nominal cutoff wavelength of 490 nm, prevented the passage of light with a longer wavelength. Subsequently, a longpass filter with a cutoff wavelength of 500 nm served as a dichroic mirror, reflecting the light. We then focused the light into the PCR/LAMP chamber of the microfluidic chip through a lens with a 6.35-mm diameter, a 1.8-mm focal length, a 1.8-mm working distance, and a 0.7-mm numerical aperture.

The light from the excited fluorescence passed through a dichroic mirror and was then filtered through a bandpass filter with a nominal bandwidth of (525 ± 25) nm before being detected by a photodiode (PD) detector. We first converted the generated photocurrent into voltage using a transimpedance amplifier with a conversion factor of 3.3 × 108. A lock-in amplifier then processes the output signal. We used a computer-aided design system to design the optical system hardware. Both the system casing and optical housing were fabricated using a CNC system and subsequently assembled, as detailed in Supplementary Section B.

Our control software enabled the adjustment of the lock-in amplifier’s parameters, such as background removal (compensation) and phase shift, to optimize the system’s gain. We determined that the optimal phase shift for the system was 73°, with the background removal setting adjusted to 920 (Supplementary Section B).

Software control

Upon connecting the system to a PC through a USB interface and initiating the software, several windows became available. These included the Welcome window, Manual Control, Protocol Control, Fluorescence Data, Temperature Chart, and Stepper Motor Control windows, as detailed in Supplementary Section C.

When the program was initiated, users first encountered the Welcome window, which led them to the Manual Control page. Here, the software established a connection with the hardware using the USB communication interface. Once the connection was successful, indicated by the connection icon turning green, the program displayed the last PID values from the sample-to-answer system. Users had the discretion to modify the PID constants if needed. The program also allowed manual setting of the TEC and the TEC support temperatures, along with activation of the cooling fan.

The Protocol Control window, designed to be self-explanatory, allowed users to select data for the RT‒PCR protocol. Users can hasten the system by incorporating preannealing and preelongation steps. Users also had the option to use the RT-LAMP protocol, keeping the temperature constant during both the denaturation and annealing/elongation phases. Once a protocol was defined, it could be uploaded to the sample-to-answer system. Alternatively, the protocol can be stored on a PC and transferred to or from the sample-to-answer system as needed. The PC screen showed the selected temperature, actual TEC temperature, temperature difference, step name, remaining time for the step, and number of cycles. An icon was also presented to commence and stop the protocol.

In the Fluorescence Data window, users can fine-tune the lock-in amplifier parameters to compensate for background noise and adjust the phase shift for optimal gain. The system also displayed the actual fluorescence analog-to-digital conversion value at the lock-in amplifier output and recorded it in a file.

The Temperature Chart window allowed users to control the temperature recording frequency and adjust the sampling rate between 0.1 s and 1.0 s. Users can read the temperatures of both the TEC plate, where the microfluidic chip was situated, and the substrate. Since increasing the temperature of the bottom part of the TEC enhances the heating rate and reduces the cooling rate, users can optimize the system by using substrate temperature control for the quickest heating/cooling rate. The data can be stored in a text file on the PC.

The Stepper Control window enables users to manually maneuver the magnet using the large knob in the sample-to-answer system or to control it semiautomatically for hands-free detection of RNA. This window displays the stepper motor location and status and allows users to set the stepper motor parameters. When initiating the motor, we set the arm to move from a step position of 1000 to 9500; subsequently, we used the wheel to control the motor position. Alternatively, the motor can be controlled by software as well to achieve fully automated operation.

Experimental protocol

Sample Preparation and Nucleic Acid Purification

For all the experiments, we utilized the quantitative synthetic N gene of SARS-CoV-2 RNA supplied by Genewiz, Inc. (Suzhou, China), with a stock solution concentration of 3.46 × 1012 copies·µl−1. We extracted viral RNA using SeraSil-Mag 400 (Cytiva, MA, U.S.A.) superparamagnetic coated silica beads. We spiked 100 µl of RNA into 200 µl of RNeasy Lysis Buffer (buffer RLT) from the RNeasy Mini Kit (Qiagen, catalog number: 74004, Hilden, Germany) and added 10 µl of magnetic nanoparticles at a concentration of 20 mg·µl−1 to the water. After manually mixing the lysate, we pumped the sample into the microfluidic chip. We then washed the RNA-MNP complex with 10 µl of M5904 light mineral oil suitable for PCR (Sigma‒Aldrich, MO, U.S.A.). In the final step, we mixed the washed complex with RT‒qPCR/RT‒LAMP reagents.

RT‒qPCR of artificial RNA

We used the Evo M-MLV Reverse Transcription Kit II (with gDNA removal reagent for qPCR) made by Accurate Biotechnology (Changsha, China). We employed both a commercial Applied Biosystems StepOne qPCR instrument (ABI) and a microfluidic chip sample to test the ability of the system to target RNA from the N gene of SARS-CoV-2. An ≈ 100 μl reaction mixture was prepared consisting of 50 μl of 2× RT‒PCR buffer, 4 μl of forward and reverse primers (0.1 mM each) (Table 2), 2 μl of Pro TaqHS polymerase (5 U·µl−1), 2 μl of M-MLV reverse transcriptase (5 U·µl−1), 4 μl of EvaGreen 20× (25 μM) in deionized H2O (DI-H2O), 10 μl of the RNA sample, and 24 μl of DI-H2O.The RT‒PCR protocol was as follows:

  • RT step at 42 °C for 5 min

  • Hot start at 95 °C for 30 s

  • PCR cycling (40 cycles)

    1. ○

      Denaturation at 95 °C for 15 s

    2. ○

      Annealing and extension at 62 °C for 75 s

  • Melting curve analysis (MCA).

After completing the PCR, we conducted an MCA to determine the melting temperature (TM) of the amplicon, thereby validating the specificity of the PCR amplification. Additionally, we extracted the threshold cycle (CT) value from the PCR amplification data.

Table 2 RT‒PCR primer sequences
Full size table

RT-LAMP reaction with a fluorescence-based readout

We conducted parallel RT-LAMP tests using an RT-LAMP kit from New England BioLabs, Ltd. (Beijing, China), along with six primers targeting the N-A region. These tests were performed on a commercial qPCR instrument maintained at a constant temperature of 65 °C for 60 min, during which we recorded the fluorescence amplitude (F) every 30 s. The reaction mixture volume was set at 25 µl, comprising 12.5 µl of RT-LAMP master mix, 0.5 µl of 50× fluorescent dye, 2.5 µl of RT-LAMP primer mix (10×), 4.5 µl of DI-H2O, and 5 µl of synthesized target RNA solution. We achieved a range of RNA concentrations by diluting the RNA sample in a 10-fold serial manner, from 108 copies·µl−1 to 104 copies·µl−1. Each diluted RNA solution was subjected to three separate reaction tests to generate accurate standard curve. We always ran a negative control (NTC) once we had prepared the samples, as we benchmarked them using a commercial StepOne qPCR system made by Applied Biosystems, Inc. (ABI) for both RT‒PCR and RT‒LAMP. Since the NTC results in the ABI qPCR system were negative, we have proven that the samples were not contaminated; thus, we did not analyze the NTC using the chips.

We prepared the RT-LAMP solution for microfluidic chip testing in a manner similar to the previous method, with slight modifications. These included the addition of 5 µl of bovine serum albumin (BSA) solution at a concentration of ≈ 20 mg·ml−1 and ≈ 10 µl of synthesized target RNA solution. The RNA concentrations ranged from 109 to 105 copies·µl−1 and were adjusted in 100-fold increments. The use of relatively high RNA concentrations aimed to address concerns about exceeding the system limit of quantification (LoQ). This approach ensures that CT values derived from high concentrations of DNA or RNA are sufficiently reliable for plotting standard curves 43.

The six RT-LAMP primers used were designed using software (https://www.neb.com/neb-primer-de-sign-tools) (Table 3). All primers were synthesized by Sangon Biotech Co., Ltd. (Shanghai, China).

Table 3 RT-LAMP primer sequences
Full size table

Results

Heating/cooling rate

We initiated testing of the system by optimizing the PID values through an empirical tuning method. The protocol involved two steps: denaturation and annealing, set at 93 °C and 60 °C for 10 s each. Initially, with the integral (I) and derivative (D) parameters set to zero, we gradually increased the proportional (P) gain until the system exhibited instability and oscillation. After observing this difference, we reduced the P value and progressively increased the I and D values. This adjustment continued until the system displayed satisfactory behavior, which we defined by achieving the highest possible heating and cooling rates of 5.5 K·s−1 and −6.7 K·s−1, respectively, as shown in Fig. 4a.

Fig. 4: Temperature profiling and distribution in the PCR amplification process.
figure 4

a PCR protocol: Graphical representation of the PCR protocol indicating rapid heating (blue line) and cooling (red line) cycles at rates of ≈ 5.5 K·s−1 and ≈ −6.7 K·s−1, respectively, which are essential for efficient DNA amplification. b Infrared radiation intensity: An infrared image capturing the distribution of radiation intensity emitted from the sample area used to infer the temperature distribution across the microfluidic chip. c Melting curve analysis (MCA) and temperature mapping: 3D map derived from the MCA showing the extracted apparent melting temperature (TM*) values across the sample area with a mean ± standard deviation of (79.02 ± 0.77) °C (mean value ± standard deviation), demonstrating uniformity in the thermal profile essential for precise nucleic acid amplification

Full size image

Chip temperature uniformity/temperature calibration

The temperature of devices is often determined by noncontact methods such as infrared (IR) imaging via the use of thermal cameras that operate between wavelengths of 8 µm and 14 µm44. However, IR systems do not measure the temperature itself but rather the amplitude of emitted radiation influenced by surface properties such as emissivity (ε), in accordance with the Stefan–Boltzmann law. A significant challenge arises even when the exact value of ε is known. This is because the measured radiation comes from the surface of an ≈ 3 mm thick PMMA device, which may have a different temperature than the solution in the chamber where PCR/LAMP processes take place.

A previous proposal introduced a noncontact, photobleaching-free MCA-based measurement method45, which has since become routine for measuring temperature inside microfluidic systems, either measuring heat transfer46 or calibrating temperature sensors in PCR systems47 and microcalorimeters48. This MCA-based method, which focuses on the melting curve analysis of double-stranded DNA (dsDNA), provides a single temperature measurement. Its precision remains unaffected by surface properties or photobleaching, unlike commonly used IR44 or fluorescein-based methods 49.

The tested solution consisted of PCR master mix after completing PCR. This mixture included DNA with a known TM value of 79.7 °C, as measured by a commercial qPCR system using EvaGreen intercalating dye (Supplementary Sections D, E). We filled the chip with this solution and placed it in the heated area of the system. The PCR/LAMP chamber within the chip was situated beneath a fluorescent imaging system. This system comprised an LED with a nominal wavelength of 470 nm, an FITC filter set, and a commercial macrolens with a 180 mm focal length set to an aperture of eight. The setup also included a single-lens reflex camera with a 45-megapixel imaging chip. We captured fluorescence images from the chip at temperatures ranging from 70 °C to 84 °C (Supplementary Section F) and processed the data as described previously47. The F value was measured for random pixels and their adjacent pixels, averaging the F values for eight adjacent pixels. We then binned these measurements into 8 × 8 pixels and plotted apparent TM* values based on location (Fig. 4b). This process resulted in a 3D map (Fig. 4c) showing that the chip’s temperature was uniform at (79.02 ± 0.77) °C (mean ± standard deviation). The minor discrepancy of ≈ 0.68 °C between the mean TM* values and the TM from the commercial PCR machine indicates good agreement between the two temperature sensors. The standard deviation, although small, is thought to be primarily due to image noise rather than actual temperature differences as shown in the MCA image 47.

Sample extraction test on a chip (with external PCR amplification)

We initially conducted tests to verify the nucleic acid extraction efficiency of our chip. Five different DNA concentrations ranging from 297 to 2.97 × 106 copies·µl−1 were prepared and divided into two groups, one for the microfluidic chip sample extraction experiment and the other for reference.

We sequentially introduced the reagents into the respective chambers into the chip 10 μl of oil, 10 μl of elution buffer, 40 μl of a mixture consisting of 10 μl of DNA, 20 μl of magnetic beads as previously described42, and 10 μl of lysate at a flow rate of 100 μl·min−1. We stored the DNA samples after sample extraction in a 4 °C refrigerator for subsequent external qPCR analysis via commercial qPCR (ABI) according to the following protocol: 0.3 μl of Takara Taq (5 U·μl−1), 1 µl of 10 × PCR buffer (Mg2+), 0.8 μl of dNTP mixture (2.5 mM), 0.4 μl of forward primer (10 μM), 0.4 μl of reverse primer 10 μM, 0.5 μl of EvaGreen (20×), 1 μl of DNA template, and 5.6 μl of DI-H2O. The other group of DNA samples without an in-chip extraction step was prepared with the same protocol and the same instrument (Fig. 5a).

Fig. 5: PCR amplification efficiency and standard curve analysis.
figure 5

a Amplification curves generated using a commercial qPCR system showing amplification curves of DNA samples after extraction via a microfluidic chip benchmarked against original DNA samples. b Standard PCR curves extracted from the PCR amplification curves shown in (a) with an extracted Ct shift value of ≈ 4.9, indicating a sample extraction efficiency of ≈ 3.4%

Full size image

Both standard curves had nearly identical slopes (Fig. 5b), with values of (−3.47 ± 0.08) cycle·dec−1 and (−3.63 ± 0.16) cycle·dec−1 for PCR of the reference sample solution and PCR of the sample after extraction in the microfluidic chip, respectively. The CT value slope obtained from DNA extracted on the chip was greater by ≈ 4.9 cycles in comparison with that of the original samples, indicating ≈ 96.6% DNA loss during extraction. We tested the system with a minimum copy number of 297 copies, as the system was still in its dynamic range even with this sample loss.

RT‒qPCR results

We prepared an ≈100 µL RT‒PCR master mix with an RNA sample concentration of 1.314 × 105 copies·µL−1 following the protocol and temperature guidelines outlined in the previous section. The master mix was divided into two portions. The first portion, consisting of approximately ≈10 µl each, was allocated for three parallel experiments using a commercial qPCR system. The second portion, approximately ≈ 70 µl in volume, was designated for experiments conducted within the microfluidic chips.

Using the commercial qPCR system, we obtained three sets of results. We plotted these results to calculate the mean F value along with the corresponding error bar (Fig. 6a). This resulted in a CT value of (20.65 ± 0.58) cycles (mean value ± standard deviation). Following RT‒PCR, we conducted an MCA at a temperature ranging from 65 °C to 92 °C. This analysis yielded raw MCA data, from which we derived a TM value of (83.46 ± 0.02) °C (mean value ± fitting error) based on the derivative of the curve (Fig. 6b, c).

Fig. 6: Comparative analysis of RT‒PCR results and melting curves from commercial qPCR machine and microfluidic chip.
figure 6

a Commercial qPCR raw data: Amplification plots of N gene RNA sequences performed in triplicate on a standard qPCR machine. b MCA on a commercial machine: Post-RT‒PCR MCA showing the specific melting temperatures of the amplified products. c Derivative of melting curves: Graphical derivative representation of the melting curves highlighting the peaks corresponding to the specific melting points. d Microfluidic chip raw data: Amplification plots from RT‒PCR performed on a microfluidic chip. e Normalized chip data: RT‒PCR data from the microfluidic chip after normalization and Boltzmann fitting, revealing the amplification efficiency. f Chip MCA results: Melting curve analysis derived from the microfluidic chip’s RT‒PCR raw data indicating the melting temperature of the nucleic acids postamplification, revealing the amplification specificity

Full size image

We subsequently conducted the assay using a microfluidic chip with a chamber volume of ≈6 µl. Therefore, the ≈70 µl volume of the master mix was sufficient for seven experiments. The procedure began by filling almost the entire chip, except for the reaction chamber, with ≈ 40 µl of M5904 mineral oil to prevent evaporation of the RT‒PCR master mix. After introducing ≈ 10 µl of the RT‒PCR master mix into the reaction chamber, we placed the chip into the sample-to-answer system. This system runs a temperature protocol identical to the commercial protocol, employing a continuous method for monitoring both fluorescence and temperature50. We recorded temperature and fluorescence data at approximately 0.5 s intervals.

We processed the raw data (Fig. 6d) using our MATLAB code. After normalization and Boltzmann curve fitting (described in the next section), we obtained a CT value of (18.6 ± 0.01) cycles and a TM value of (82.68 ± 0.05) °C, both (of which are mean values ± fitting errors) (Fig. 6e, f). The results indicated that the RT‒PCR system performed similarly to the commercial qPCR system when the temperature difference was less than 1 °C.

RT-LAMP results

Off-chip RT-LAMP results

We conducted RT-LAMP tests with a commercial qPCR system using five different RNA template concentrations, including an NTC. The concentration of each sample was tested in triplicate. For the RT-LAMP process, we maintained the sample at ≈65 °C for 60 min while recording the fluorescence amplitude every ≈30 s. Instead of using cycle numbers, we expressed the collected data as a function of time (t). For each RNA concentration, we calculated the mean F value along with its standard deviation and plotted these mean values with error bars (Fig. 7a). As previously described, we normalized the data by fitting the mean value of F using the Boltzmann function 51:

$$F={F}_{2}+\frac{{F}_{1}-{F}_{2}}{1+{e}^{\frac{t-{t}_{0}}{{F}_{0}}}}$$
(1)

where F1 is the minimum value of F, F2 is the maximum value of F, and F0 defines the fluorescence slope at inflection point t0 (Fig. 7b). Subsequently, we subtracted the value of F1 from the fitted curve by Eq. (1) and then divided it by (F1–F2) to normalize it to the minimum value of zero. We defined the CT value as the cycle number having a 10% F value increase over the baseline. From Eq. (1), we derived CT as follows:

$${C}_{{\rm{T}}}={t}_{0}-{F}_{0}\cdot \mathrm{ln}\left(\frac{{F}_{1}-{F}_{2}}{F-{F}_{2}}+1\right)$$
(2)

by substituting (F1–F2)/10 + F1 for F. Then, we plotted (Fig. 7c) the RT-LAMP standard curve (blue) and obtained a slope of (–1.913 ± 0.053) min·dec−1 (mean value ± fitting error).

Fig. 7: RT-LAMP amplification efficiency and standard curve evaluation.
figure 7

a RT-LAMP amplification curves: Real-time amplification data from RT-LAMP reactions performed on a commercial qPCR machine with five different RNA concentrations tested in triplicate, including no-template controls (NTCs). b Normalized RT-LAMP data: The raw RT-LAMP amplification data were analyzed on chips and normalized using Boltzmann function curve fitting to demonstrate the amplification progress and efficiency. c Standard RT-LAMP curves: Comparative standard curves showing the amplification efficiency of RT-LAMP assays. The left slope represents the standard curve obtained from the commercial qPCR machine with a slope of (–1.913 ± 0.053) min·dec−1, and the right slope shows the standard curve derived from the microfluidic chip data with a slope of (–1.282 ± 0.077) min·dec−1, both indicating the mean value and fitting error

Full size image

RT-LAMP results inside the chip

We initiated the RT-LAMP process by injecting ≈40 µl of M5904 mineral oil into the oil channel through the oil injection port. This step was essential for preventing the RT-LAMP solution from evaporating. Then, we introduced ≈10 µl of the RT-LAMP reaction mixture into the amplification chamber of the chip. The chip was placed in the sample-to-answer system, after which the temperature was maintained at ≈65 °C for a minimum duration of ≈30 min. This process ensured the effective performance of RT-LAMP inside the chip while continuously monitoring the excitation fluorescence signal through the optical components of the system.

The distribution of on-chip RT-LAMP results involved dividing the strains into three groups, each characterized by different RNA concentrations. We ran each sample within the same RNA content group three times, consuming a total of nine chips for these tests. The raw data of these tests are presented in supplementary Section G. Individual normalization of the captured curves from the raw data of the chip amplification was performed (Fig. 7b), leading to the creation of the RT-LAMP standard curve (red curve) (Fig. 7c). This curve displayed a slope of (–1.282 ± 0.077) min·dec−1 (mean value ± fitting error), similar to the benchmark data obtained from the commercial qPCR system, which had a slope of (–1.913 ± 0.053) min·dec−1 (mean value ± fitting error). The square of the Pearson correlation coefficient for the fitting curve in the chip was 0.996, whereas it was 0.998 in the commercial qPCR system, indicating good standard curve linearity. These results demonstrated the effective performance of RT-LAMP on the chip, as evidenced by the establishment of the RT-LAMP standard curve.

Integrated system

The operation of the integrated system involved a sequence of steps. We began by introducing ≈1 µl of magnetic bead solution into ≈10 µL of RNA solution. Uniform mixing was achieved by pipetting the mixture for ≈30 s. This prepared solution was then injected into the sample-injection aperture. Pumping continued until the mixture reached the periphery of the oil chamber.

In the next phase, we pipetted ≈10 µL of the RT-LAMP master mix into the chip’s reaction chamber via the sample input port to fill the chamber. Subsequently, we pipetted ≈2 µl of M5904 mineral oil to fill the oil chamber and the channel. After completing these sample injection stages, we positioned the chip on the sample-to-answer system with the arm having a permanent magnet above the sample extraction chamber. Subsequently, we pipetted the sample to a volume of ≈5 µl in ≈5 min, after which all the other fluid inputs/outputs were blocked except for that for the raw sample waste (Fig. 3b). The magnet captured the particles with bound RNA and moved via the oil chamber, effectively removing all the debris from the solution, leaving only magnetic particles with bound RNA. Then, we moved the magnetic beads into the reaction chamber containing RT-LAMP master mix and initiated the RT-LAMP protocol to facilitate the amplification procedure. This entire sample pretreatment required ≈8 min, including ≈5 min for sample loading and ≈3 min for nucleic acid extraction.

We tested RNA sample concentrations of 105, 107, and 109 copies·µl−1, and each concentration was tested in triplicate to establish the standard curve of the integrated system (Fig. 8). The chips were disposable, and nine chips were used for nine experiments. We extracted CT values via the same procedure as mentioned in the previous section, and the integrated system demonstrated a LAMP amplification slope of (–2.788 ± 0.098) min·dec−1 (mean value ± fitting error), as illustrated in Fig. 8d. The fitting error of the curve was acceptable even though the middle point (concentration of 107 copies·µl−1) was off the fitting point, but the curve itself contained a test error bar. These results indicate that the system is capable of performing RT-LAMP following the extraction function of the integrated system and enables relative quantification with a standard curve comparable to that of a commercial qPCR system.

Fig. 8: Standard curve analysis of the integrated RT-LAMP system.
figure 8

Normalized Boltzmann function curve derived from raw RT-LAMP data from the chip without sample extraction and with the function of the integrated system with RNA contents of a 109 copies·µl−1, b 107 copies·µl−1, and c 105 copies·µl−1; each experiment was conducted in triplicate, and the results are shown as different color curves. d The standard curve (black) shows a slope of (–2.788 ± 0.098) min·dec−1 (mean value ± fitting error), which is indicative of the system’s amplification capability after sample extraction. This curve is compared to the red curve with a slope of (–1.282 ± 0.077) min·dec−1 from the chip without the sample extraction process, emphasizing the influence of the integrated extraction on the amplification efficiency

Full size image

Discussion

Our study introduces a rapid and versatile sample-to-answer system designed for automated nucleic acid extraction and analysis. This platform uses both RT‒qPCR and RT‒LAMP methods and offers a simplified workflow suitable for real-time molecular diagnostics 52.

An important feature of this system is the controlled heating and cooling rate53. This aspect is crucial for PCR-based techniques that depend on specific temperatures for enzymatic reactions54. In this study, we optimized only the PID constants to achieve the fastest possible heating and cooling rates that the system hardware allows, at 5.5 K·s−1 and −6.7 K·s−1, respectively. However, this system optimization was not accompanied by PCR protocol optimization. We employed a conservative PCR protocol with rather long denaturation and annealing/extension steps to demonstrate that the system can also be applied to perform RT‒PCR analysis, the gold standard in DNA/RNA diagnostics. Modifications to the microfluidic chip to improve heat transfer are essential for running a reasonably fast RT‒PCR protocol.

Precise temperature management during PCR-based amplification can significantly influence diagnostic sensitivity and specificity55. Here, we implemented a noncontact, photobleaching-free MCA-based measurement technique45,47,48. This technique excels at providing accurate and consistent temperature readings56 and is a significant advantage over traditional methods that often encounter issues related to surface properties and potential photobleaching effects.

A critical step in molecular diagnostics is nucleic acid extraction57. Our system performs on-chip extraction, but some loss of RNA was observed, indicating the need for further optimization. The extraction process is reasonably good at detecting the presence of the target nucleic acids even at lower concentration sensitivities. We tested the system with a minimum copy number of 297 copies, as the system was still in its dynamic range even with this sample loss. This limit is sufficient for SARS-CoV-2 screening, as the method was conducted in China58 in 2023 even when the pooling technique was used59 to further dilute the sample by combining 20 samples from different individuals. This indicates that in such a case, our system limit of extraction being better than 297 would be sufficient for detecting the presence of SARS-CoV-2 RNA60. However, additional diluted samples should be tested because certain viruses, such as human immunodeficiency virus, have a low viral load, and monitoring the number of RNA copies indicates HIV treatment efficacy 61.

The results from our on-chip system closely align with those from commercial RT‒qPCR and RT‒LAMP platforms, confirming the robustness of our approach. The on-chip RT-LAMP method, known for its speed and high amplification efficiency, shows strong potential for field applications 36,62.

Conclusion

Our study introduces a rapid, versatile, and low-cost sample-to-answer system for SARS-CoV-2 diagnostics. It integrates the extraction and purification of nucleic acids, followed by amplification using either RT‒qPCR or RT‒LAMP. This versatility, coupled with the use of conventional chemistry, makes the system competitive with most FDA-approved devices, completing analyses within 28 min. This timeframe includes ≈5 min for sample loading, ≈3 min for target extraction and ≈20 min for RT-LAMP. However, the RT‒PCR protocol is significantly slower because the heat transfer of the microfluidic chip is not optimized. We demonstrated the RT-PCR capability of the chip, but it is not the system’s primary focus. The total consumption of either the RT-LAMP or RT‒PCR master mix was only ≈10 µl. Currently, the major factor influencing detection limits is a high level of NA loss during the extraction step, which was calculated to be 96.6%. Even with such low extraction efficiency, the system is still sufficiently sensitive for COVID-19 diagnosis due to the high viral load of SARS-CoV-2 but is not adequate for identifying HIV and similar viruses with low viral loads.

The PMMA-based sample-to-answer chip developed in this work enables mass production without complex treatment procedures, leading to a lower cost of ≈$9.5 per test compared to commercial options. This cost can be further reduced by replacing CNC-based chip manufacturing techniques with more economical methods, such as roll-to-roll hot embossing or injection molding. This system requires further refinement, especially for minimizing RNA loss, and it has robust and high performance. Its portability, ease of use, and automation capabilities make the platform suitable for on-site diagnosis by minimally trained personnel, eliminating the need for centralized facilities. This study lays the groundwork for the development of more sophisticated, reliable, and user-friendly microfluidic systems for molecular diagnostic testing.