Transfer of blocker-based qPCR reactions for DNA methylation analysis into a microfluidic LoC system using thermal modeling

Abstract

Changes in the DNA methylation landscape are associated with many diseases like cancer. Therefore, DNA methylation analysis is of great interest for molecular diagnostics and can be applied, e.g., for minimally invasive diagnostics in liquid biopsy samples like blood plasma. Sensitive detection of local de novo methylation, which occurs in various cancer types, can be achieved with quantitative HeavyMethyl-PCR using oligonucleotides that block the amplification of unmethylated DNA. A transfer of these quantitative PCRs (qPCRs) into point-of-care (PoC) devices like microfluidic Lab-on-Chip (LoC) cartridges can be challenging as LoC systems show significantly different thermal properties than qPCR cyclers. We demonstrate how an adequate thermal model of the specific LoC system can help us to identify a suitable thermal profile, even for complex HeavyMethyl qPCRs, with reduced experimental effort. Using a simulation-based approach, we demonstrate a proof-of-principle for the successful LoC transfer of colorectal SEPT9/ACTB-qPCR from Epi Procolon® colorectal carcinoma test, by avoidance of oligonucleotide interactions.

I. INTRODUCTION

Methylation of the 5’-cytosine base within CpG dinucleotides in DNA is one of the best understood epigenetic marks and is involved in many biological processes.^1–3 DNA methylation has regulatory functions in gene expression, as, for example, hypermethylation of promotor regions can lead to their transcriptional silencing.⁴ This inactivating character of DNA methylation can also be seen with X-chromosome inactivation and parental imprinting.^5,6 The dynamic interplay of maintenance methylation, de novo methylation, and demethylation is essential for embryogenesis and cell differentiation.^7,8 Therefore, it becomes obvious that aberrant DNA methylation landscapes are associated with many diseases among others cancer.^2,4,7,9 The hypermethylation of CpG islands in promoter regions is a frequently observed phenomenon in cancer cells, which can lead to the inactivation of tumor suppressor genes (TSPs) and, therefore, contribute to tumorigenesis.^10,11 These hypermethylated promoter regions can serve as robust DNA methylation markers, due to the covalent bond between the methyl group and the cytosine base. DNA methylation markers can be detected with highly sensitive PCR-based methods after a previous bisulfite conversion reaction, desulfonation, and purification of the bisulfite-converted DNA.^5,12,13 In comparison to genetic biomarkers, DNA methylation can enable more sensitive detection in early tumor stages, as the number of recurrent mutations is lower in comparison to hypermethylation of several CpGs in a promoter region at the same time.^14,15 Because changes in the DNA methylation landscape occur in early stages of tumorigenesis^2,7,11,16 and can also be found on cfDNA in body fluids like blood and urine, these markers are promising targets for early minimally invasive diagnostics.¹⁷ One of these markers that is frequently silenced in colon cancer and that is used as an early cancer detection marker in the FDA approved Epi Procolon® cancer test (Epigenomics AG) is the v2 region of the SEPT9 promoter. Using an additional oligonucleotide blocker, quantitative HeavyMethyl-PCR (HM-qPCR) technology, can repress the amplification of unmethylated sequences.^18–22 The SEPT9 test has been extensively characterized and validated as reported in several publications.^23–25

High mortality of cancer is related to a lack of early diagnostics before metastasis of primary tumors.^5,26 An early diagnosis increases therapy success and enables monitoring of tumor progression that can lead to better survival chances.^5,14,17,26 Standard colon cancer screening programs are relatively poorly accepted as they are highly invasive and show poor patient compliance.^18,27,28 For this reason, effective minimally invasive blood-based colorectal carcinoma (CRC) tests are expected to increase participation in screening programs.¹²

Nevertheless, DNA methylation analysis is a laborious and time-consuming procedure. Therefore, automated execution of such an analysis in microfluidic LoC systems will bring epigenetic testing a major step closer to its broad application in Point-of-Care (PoC) diagnostics, making it economically lucrative. Several LoC systems for DNA methylation detection have been described in the literature^29–35 and have been extensively reviewed.^36–40 Although all of the known approaches have substantially contributed to DNA methylation analysis within LoC systems, many hurdles still have to be overcome before DNA methylation analysis finds its way into routine PoC diagnostics. None of the above-cited systems applies a blocker-based qPCR reaction. This would be essential, when sensitive detection in early stage liquid biopsy patient samples, with high background of unmethylated sequences, is needed.

It is almost impossible to transfer a qPCR reaction that was optimized for performance in a qPCR cycler 1:1 to a LoC system, due to different heating and cooling rates. These differences become even more critical in complex qPCR systems like HeavyMethyl (HM) qPCRs. Therefore, substantial experimental effort is required to transfer these qPCR reactions to LoC systems. As shown by Devos et al.,²⁰ the oligonucleotides for detection of methylated SEPT9 show overlaps, making a LoC transfer challenging due to occurring oligonucleotide interactions caused by aberrant thermal profiles. In the presented work, a blocker-based qPCR has been transferred to the Vivalytic system from Bosch Healthcare Solutions GmbH (Fig. 1). Within this pneumatically actuated LoC system, diverse and complex protocols can be performed in a fully automated and reproducible manner.^41,42 In the Vivalytic system, PCR reactions are not performed by heating and cooling of one compartment. Rather, the fluid plug is shuttled between three reaction chambers, each of which is heated by one or two separate heaters, providing the temperature levels needed for denaturation, annealing, and elongation.

FIG. 1.

Overview of the microfluidic Vivalytic cartridge. Left: overview of the compartments and microfluidic system. The cartridge consists of several chambers for pumping and specific reactions like PCRs or purification processes connected through microfluidic channels. Pneumatically driven valves help us to move fluids precisely and reproducibly from one compartment to another. Up to nine reagents can be stored in reagent chambers. Additional chambers for sample input, mixing, waste storage, as well as a reaction chamber are available. Chambers close to the PCR chambers are suited for storage of surface dried oligonucleotides and lyophilized mastermix reagents. Right: overview of the different layers of the Vivalytic cartridge. Long term storage of reagents is possible within sealed reagent bars. PC: polycarbonate, TPU: thermoplastic polyurethane. Modified with permission from Kärcher et al., $\mu$TAS Conf. J. 25, 41 (2021). Copyright 2021 Chemical and Biological Microsystems Society.

Thermal models of the specific LoC system can suggest a suitable actuation profile for the desired qPCR reaction and thus reduce experimental iterations to a large extent. This is especially beneficial when using complex single-use microfluidic prototype systems for development but also helps to avoid unnecessary waste of expensive reagents. By computing the temporal evolution of fluid plug temperature as a function of heater temperatures as well as fluid plug position and by comparing the fluid temperature with the target qPCR protocol, it is possible to iteratively adjust the control parameters for the Vivalytic analyzer until the computed thermal profile fulfills the requirements.

II. EXPERIMENTAL

Sample material was generated by cultivating the CRC cell line HCT116 and shearing the extracted gDNA to around 200 bp with the bioruptor pico (Diagenode). For bisulfite conversion of 1000 ng sheared HCT116 DNA, the protocol and reagents from Premium Bisulfite Kit (Diagenode) were used. Subsequent purification of bisulfite-converted DNA was performed using the buffers from the same kit. For qPCR reactions, qPCR ProbesMaster Lyophilisate Bead (Jena Bioscience) was used. Regarding mastermix constitution and temperature profiles for qPCR performed with the Vivalytic One Analyser (on-chip) and with the 7500 Real Time PCR cycler from Applied Biosystems (off-chip), see Tables I and II in the supplementary material. Oligonucleotides were purchased from Biomers as published by Devos et al.²⁰ in order to simulate the Epi Procolon® test with its SEPT9/ACTB duplex qPCR assay for CRC detection. Sample quantification has been performed using the Qubit 4© fluorometer (Invitrogen). DNA quality control and determination of size distribution was performed with Fragment Analyser (Agilent).

For performance of LoC experiments, Vivalytic cartridges were used. 200  $\mu$l of qPCR reaction mix (see Table III in the supplementary material) containing 5 ng/ $\mu$l bisulfite-converted DNA in elution buffer or H₂O, was pipetted into the sample chamber (Fig. 1). Elution buffer for purging of microfluidic system was also stored in a reagent chamber (Fig. 1). The system was also purged with PCR reaction mix from the sample chamber and after filling of PCR chamber with PCR reaction mix, PCR was initiated.

Microfluidic cartridges were processed by the Vivalytic Analyser and automated pictures were taken during the annealing step of every qPCR cycle.

After the manual selection of image sections, fluorescence intensities were determined by a proprietary image analysis software. The data were exported to an excel file for further analysis. For the definition of the threshold, tenfold standard deviation of the first 20 cycles was added to the average value of the first 20 cycles. Afterward, the cycle in which the threshold was first exceeded was extracted. Due to deviations in the manual selection of image sections for image analysis and bubble formation during the qPCR reaction, the curves had to be normalized by subtracting each value of a data row from the value of the first cycle in order to shift the curves to a starting value of 0, enabling the comparison of different curves. In addition, curves were subjected to exponential smoothing using the following formula:

$$x_i=\alpha y_i+(1-\alpha )x_{i-1},$$ (1)

where $x_i$ denotes the smoothed value at time $t_i$, $y_i$ the measured value at $t_i$, and $\alpha$ the smoothing factor, respectively.

III. RESULTS

A. Identification of oligonucleotide interactions in on-chip experiments

As shown in Fig. 1S in the supplementary material, oligonucleotide interactions occurred in the LoC system when using all of the seven oligonucleotides for the SEPT9/ACTB duplex qPCR. Therefore, qPCR reactions were performed in which the SEPT9 oligonucleotides were omitted one by one, showing that with primers and the blocker as well as with the primers and the probe, amplicons for SEPT9 and ACTB were successfully amplified. This demonstrated that only the complete mix of all types of oligonucleotides was susceptible to oligonucleotide interactions. As all components are needed for adequate qPCR performance showing high sensitivities and specificities, further effort was undertaken to find a thermal profile that allows for the amplification of SEPT9 and ACTB without the formation of oligonucleotide interactions when using the SEPT9 blocker.

B. Model-based estimation of thermal protocols

In order to identify a suitable thermal protocol with little experimental iteration, a thermal network model of the Vivalytic cartridge has been constructed and implemented in the software package Mathematica.⁴³ In principle, the physical processes associated with a shuttle qPCR would justify a sophisticated scale-resolved simulation model with coupled fluid flow and heat transfer (such as in Ref. 44). However, since the focus of the present work is on the variation of thermal protocols, the computational cost of applying a detailed model has been considered too high. Though this limitation could be overcome by reduced-order methods (see, for instance, Ref. 45), the effort to arrive at the simple, discrete model described below is significantly less than following the path of reduced-order modeling applied to a scale-resolved simulation model.

The model used in the present study is instead based on a manually defined discretization of the cartridge material (see Fig. 2), each discrete cell of which is represented by a time-dependent temperature value. Using the discretization, the thermal network can be represented by a graph data structure. Therein, the nodes (vertices) of the graph represent thermal capacities of the form $C_i={(c_p\varrho V)}_i$ for each node $i$, where $c_p$ indicates the material’s thermal capacity, $\varrho$ the material’s density, and $V$ the volume of the discretization cell. The edges of the graph represent thermal resistances $R_{ij}$ (or, alternatively, thermal conductances ${\mathrm{\Lambda }}_{ij}=R_{ij}^{-1}$) between two temperature states $i,j$ of the form $R_{ij}=l_{ij}/({\lambda }_{ij}{A}_{ij})$, where (neglecting the indices) $l$ denotes the distance between the two states, ${\lambda }_{ij}$ the (effective) thermal conductance of the material in between the two states, and $A_{ij}$ the cell-to-cell interface area between the two cells $i$ and $j$. Balancing thermal energy ( $C_i{T}_i$) for all states then leads to a system of ordinary first-order differential equations, which reads

$$\begin{array}{rl}& \left[\begin{array}{ccc}{C}_1& \cdots & 0\\ ⋮& \ddots & ⋮\\ 0& \cdots & C_n\end{array}\right]\left[\begin{array}{c}\dot{T_1}\\ ⋮\\ \dot{T_n}\end{array}\right]\\ & =[\begin{array}{ccc}-({\mathrm{\Lambda }}_{12}+\cdots +{\mathrm{\Lambda }}_{1m})& {\mathrm{\Lambda }}_{12}& \cdots \\ {\mathrm{\Lambda }}_{12}& -({\mathrm{\Lambda }}_{21}+\cdots +{\mathrm{\Lambda }}_{2m})& \cdots \\ ⋮& ⋮& \ddots \\ {\mathrm{\Lambda }}_{n1}& {\mathrm{\Lambda }}_{n2}& \cdots \end{array}\\ & \begin{array}{ccc}\cdots & \cdots & {\mathrm{\Lambda }}_{1m}\\ \cdots & \cdots & {\mathrm{\Lambda }}_{2m}\\ \cdots & \cdots & ⋮\\ -({\mathrm{\Lambda }}_{n1}+\cdots +{\mathrm{\Lambda }}_{nm})& \cdots & {\mathrm{\Lambda }}_{nm}\end{array}]\left[\begin{array}{c}{T}_1\\ ⋮\\ T_m\end{array}\right].\end{array}$$ (2)

Equation (2) contains $n$ differential equations for each of the temperature states, and a total of $m$ temperature variables of which the last $m-n$ denotes the heater (input) temperatures. Additionally, the relation $R_{ij}=R_{ji}$ holds because of heat flux symmetry. Heat flux between cartridge material and ambient air is included with the model in the following way. For each discretization cell $i$ in contact with ambient air (state 0), an effective thermal resistance $R_{i0}={\left(\alpha A_{i0}\right)}^{-1}$ between the constant ambient air temperature ( $T_0=40{}^{°}\mathrm{C}$) and the temperature $T_i$ is computed using a constant heat transfer coefficient of $\alpha =6.2\mathrm{W}/({\mathrm{m}}^{2}K)$ (estimated based on correlations for free convection at a horizontal flat plate, cf. Ref. 46).

FIG. 2.

Scheme of the discretization of the Vivalytic cartridge for the purpose of creating a network model. Left: in-plane discretization. The red dots indicate the location of the temperature state while the blue lines mark cell boundaries. Each state is referred to by a unique identifier (number or letter). Right: out-of-plane discretization vertical to the one on the left. The large number of states (identifiers stated in boxes) as compared to the in-plane discretization ensures a sufficiently accurate approximation of the temperature field in the region of interest, i.e., around the chambers. The heater elements are depicted in red. Note that the ideal heater target temperature (e.g., Target 1, VI) differs from the actual heater temperature (e.g., 1, VI) by a first-order temporal response with an empirical time constant.

In principle, Eq. (2) can now be solved after specifying initial conditions for the temperature states and time-dependent heater temperatures. However, the above system does not yet account for the fluid plug shuttling. There are several options for including the plug motion. For the present work, the fluid temperature has been included as a temperature state with its thermal relations, i.e., those values $R_{ij}$ involving to the fluid, changing when the fluid plug is shifted between chambers, thus changing parts of the right-hand side of Eq. (2) dynamically and making some of the coefficients $R_{ij}$ effectively time-dependent. The simulation setup is based on the actual control parameters used in Vivalytic actuation profiles, which are then transferred to the computational form suitable for solving Eq. (2) numerically using Mathematica’s NDSOLVE function. Figure 3 depicts a typical input–output pair of actuations and thermal responses for the thermal model.

FIG. 3.

Example of temperature prediction during a shuttle PCR by means of a thermal model. Specifying heater target temperatures (upper left) and fluid plug position (upper right) as a function of time allows for predicting the thermal response (bottom) of the cartridge and the fluid plug.

In order to validate the thermal model and assess its accuracy in predicting thermal profiles, a number of comparisons with thermal measurements were conducted. The simulation has been designed to use the same set of parameters as the experimental protocol used with the Vivalytic system. An example validation is depicted in Fig. 4 where in the right-hand side plot, the experimental settings were imitated while the left-hand side plot visualizes the effect of removing the undershoot.

FIG. 4.

Experimental validation of the thermal model focusing on undershoot effects. The two plots contain identical experimental data (symbols) recorded by means of thermocouples (cf. Fig. 5) but compare different simulation results (lines) against it. The curves on the left result from simulating constant heater settings, namely, 90  ${}^{°}$C at heaters 1 and 2 (i.e., at chamber PCR1), 65  ${}^{°}$C at heaters 3 and 4 (i.e., at chamber PCR2) and 60  ${}^{°}$C at heater 5 (i.e., at chamber PCR3), while the curves on the right correspond to a temporary temperature setting (undershoot) of 20  ${}^{°}$C for 7 s in the annealing step (heater 5). The fluid temperature is shown as a red solid line that is temporarily touching or at least approaching the thermocouple temperatures depending on which chamber the fluid plug is located in. The total residence times (including undershoots) are 15 s for PCR1, 30 s for PCR2, and 30 s for PCR3.

Within this work, the focus lies on the optimization of the annealing step, as this step is especially relevant for oligonucleotide hybridization.^47,48 For better performance, the annealing step was divided up into two phases. In the first one, a so-called undershoot was performed in order to obtain higher cooling rates by applying temperatures lower than the final annealing temperature. In the second phase, the final annealing temperature of 56  ${}^{°}$C was set [cf. Fig. 3(a) and Table IV in the supplementary material)].

Note that the input temperatures generally deviate from the temperature experienced by the fluid plug due to thermal conduction which is one of the reasons why a thermal model is useful. Thermal actuation profiles for qPCR reactions were deduced from the simulation results and verified by thermal measurements with thermocouples (Fig. 5) build into the PCR chambers of the Vivalytic cartridge. These so-called thermocartridges were connected to an National Instruments data acquisition card and processed by the Vivalytic Analyser running the different thermal profiles.

FIG. 5.

Thermal cartridge with thermocouples built into the PCR chambers for temperature measurement. These cartridges were used for the validation of temperature protocols generated by the simulation tool.

C. Selection of thermal programs

Several thermal programs (see Table V in the supplementary material) for qPCR performance were derived from simulation results and verified by thermal measurements using integrated thermocouples (Fig. 5). Figure 6 depicts measured time series of chamber temperatures resulting from different protocols. The recorded temperature curves were used to adapt a thermal cycler-based qPCR protocol for its application within the Vivalytic LoC platform. TP2 was selected due to its long holding times of temperatures between 60 and 56  ${}^{°}$C and a better cooling rate as with TP1 and TP5. A even higher cooling rate was achieved with TP4. TP7 and TP8 showed high cooling rates as well as long holding times. Therefore, TP2, TP4, TP7, and TP8 have been verified with the SEPT9/ACTB duplex qPCR within the Vivalyic cartridge (Fig. 7).

FIG. 6.

Temperature–time series of the generated thermal programs measured with thermocouples. (a) Denaturation, annealing, and elongation phases for the PCR chambers PCR21, PCR22, and PCR23, respectively. (b) Denaturation, annealing, and elongation phases from all thermal programs (TP1–TP8). (c) Annealing phase from all thermal programs for chamber PCR23.

FIG. 7.

Verification of selected thermal programs, generated with the help of a thermal simulation approach. For verification, SEPT9/ACTB-duplex-qPCR was performed in duplicates within the Vivalytic cartridge. 100 ng of sheared and bisulfite-converted HCT116 DNA was used as a template in each qPCR.

As shown in Fig. 7, ACTB was amplified successfully with every thermal profile that was applied. The DNA methylation marker SEPT9 was successfully amplified with the thermal profile TP4. Also, with TP8, in one of the replicates, a slight amplification could be observed as well.

IV. DISCUSSION

Transferring a PCR-based detection method, like a qPCR reaction, into a microfluidic lab-on-chip platform, is a challenging task, especially when multiple, sequentially overlaying oligonucleotides are applied.²⁰ Due to significantly different technical aspects like thermal conductivities, cooling rates, etc., a qPCR, that was initially developed for performance with a qPCR cycler, can normally not be performed within a lab-on-chip device, applying the same thermal program. In order to minimize experimental work as well as a waste of expensive reagents and lab-on-chip prototypes, we present a simulation-based approach in order to identify a suitable thermal program with minimal effort.

From Fig. 4, we can conclude that while the simulation model is fairly accurate, it especially underestimates the cooling rate in the annealing step and predicts an annealing temperature slightly below the experimental value. As a consequence, although the model could be used for a relative comparison of possible parameter settings throughout this study, it cannot be considered sufficiently accurate to replace thermal measurements. Nevertheless, the simulation-based estimation of candidate parameter sets has proven an initial guess accurate enough to reduce the required number of experimental iterations and finally led to the identification of a suitable thermal profile (TP4) for the performance of on-Chip HM-PCR. With further adjustment of the thermal model as well as the thermal profile, the amplification of the SEPT9 methylation marker could be further improved. Although Stark et al.³¹ have contributed substantially to LoC-based DNA methylation analysis,³¹ no HM-PCR applying an oligonucleotide blocker, was performed. Therefore, our approach has the potential to pave the way for more efficient LoC integration of complex molecular diagnostic assays.

V. CONCLUSIONS

Within this work, we presented an approach for LoC integration of a complex HM-PCR with strongly reduced experimental effort and waste of reagents and LoC prototypes. Based on a thermal model that characterizes the specific LoC system, possible thermal profiles for the specific qPCR can be interrogated by simulation and further deviated and optimized by thermal measurements using thermocouples and experimental experience. The occurring oligonucleotide interactions, preventing successful LoC integration of the SEPT9/ACTB duplex qPCR, could be eliminated by the fast identification of a suitable thermal profile using a simulation-based approach. A more sophisticated and, thus, more accurate model than the one presented above could, however, lead to a more precise initial guess of the thermal protocol and, therefore, reduce the number of required experimental iterations even more. Nevertheless, the successful application of the above simplified model demonstrates that, for qPCR protocols, the modeling effort can be kept low while still offering the advantage of a simulation-based approach. We suggest that the implementation and thermal characterization of a LoC system can accelerate and facilitate the development of automated molecular diagnostic tests performed in microfluidic LoC systems and could help us to pave the way for automated point-of-care diagnostics into routine diagnostics.

SUPPLEMENTARY MATERIAL

The supplementary material comprises tables of off- and on-chip protocols as well as details on experimental and simulation setups considered during the study.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts to disclose.

Author Contributions

Janik Kärcher: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Britta Schulze: Data curation (equal); Investigation (equal); Validation (equal); Visualization (equal). Aaron Dörr: Data curation (equal); Investigation (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal). Sascha Tierling: Supervision (equal). Jörn Walter: Conceptualization (equal); Supervision (equal).

DATA AVAILABILITY

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