Open Access
Add to BibSonomyAbstract
Fluorescence readout is an important technique for detecting biomolecules. In this paper, we present a multiplexed fluorescence readout method using time varied fluorescence signals. To generate the fluorescence signals, coded strands and a set of universal molecular beacons are introduced. Each coded strand represents the existence of an assigned target molecule. The coded strands have coded sequences to generate temporary fluorescence signals through binding to the molecular beacons. The signal generating processes are modeled based on the reaction kinetics between the coded strands and molecular beacons. The model is used to decode the detected fluorescence signals using maximum likelihood estimation. Multiplexed fluorescence readout was experimentally demonstrated with three molecular beacons. Numerical analysis showed that the readout accuracy was enhanced by the use of time-varied fluorescence signals.
© 2016 Optical Society of America
1. Introduction
High-throughput and cost-effective detection of DNA and RNA would have many applications in point-of-care testing, including clinical diagnostics and the monitoring of epidemics [1]. Fluorescence probes can report molecular signals as optical signals [2]. For multiplexed detection, typical fluorescence detection requires different fluorescence wavelengths to be assigned to individual target molecules. Thus, the maximum number of readouts in a single test tube is usually the number of fluorescence molecules that can be spectrally distinguished between. Owing to the bandwidth of fluorescence spectra, this number has been limited to six [3]. Optical parallelization provides massively large-scale simultaneous fluorescence readout. A good example is DNA microarray technology, which uses spatial dimensions to discriminate between target molecules and allows the detection of tens of thousands of targets simultaneously [4]. However, this method requires immobilization of specific DNA probes on a substrate to capture target molecules and may not be a cost-effective approach.
Another approach for detecting multiple targets is the use of wavelength and intensity dimensions, i.e., fluorescence color coding. The fluorescence color coding method allows the limitation of the number of detectable targets to be overcome [5]. In this technique, fluorescence color codes represented by combinations of fluorescence wavelengths and intensities are assigned to individual target molecules. Color coding can be carried out by tagging a probe with multiple fluorescence molecules, a nanostructure, or a bead [5–7]. This method has had great success in producing fluorescence color codes but requires nano- or micro-scale observation to readout fluorescence signals from each tag in a single sample solution. An alternative method of color coding is regulating the ratio of fluorescence probes [8–12], such as TaqMan probes and molecular beacons, to generate the coded color during the assay process. The molecules contained in a sample can be identified from the obtained fluorescence intensities. This approach offers the identification of multiple molecules in a single test tube with a simple fluorescence measurement. However, multiple detection with these color coding methods is limited because of degeneracy of the codes [10].
In this paper, we propose a multiplexed fluorescence readout method that uses the fluorescence signal time to detect multiple target molecules simultaneously in a single tube. Use of the time dimension increases the degree of multiplexing. Usually, bioassay kinetics are well analyzed for quantitative diagnosis and the reaction parameters are obtained in advance [13,14]. Moreover, there are several methods that can be used to change the dynamics of fluorescence signals with DNA chain reactions and enzymatic reactions [17]. Designed DNA reactions are used to control the amplification [18] and nonlinear response [19] of the fluorescence signals. The time responses created by these reactions could allow the multiplicity of the detectable target molecules to be increased. To demonstrate this approach, we introduce coded strands of DNA and universal fluorescence probes. The coded strands contained a series of code sequences that generated temporary fluorescence signals by binding to molecular beacons. Each prepared coded strand represented a state of the assigned target molecule. Furthermore, multiplexed fluorescence signals from each coded strand were modeled in advance and sets of target molecules were determined from the obtained fluorescence signals.
2. Model
Figure 1(a) details the concept of the proposed multiplexed detection with fluorescence color coding and time variation. Ntar of the target molecules are detected using Nλ of the fluorescence molecules. Each target molecule is converted into Nλ temporal fluorescence signals. The conversion reactions proceed independently for each target molecule. The intensity curve measured for each fluorescence is the sum of the fluorescence signals from Ntar target molecules. The kinetic model of the conversion process is obtained in advance using reaction equations and parameter acquisition. The molecular signals of the sample can then be estimated from the model and the obtained fluorescence signals.
Fig. 1 (a) Schematic of multiplexed fluorescence detection. (b) Implementation of fluorescence conversion of a target molecule with a coded strand and set of molecular beacons.
Figure 1(b) shows the fluorescence conversion process. In this study, we propose fluorescence conversion using a set of universal molecular beacons. Molecular beacons are prepared for individual fluorescence molecules. Each target molecule is converted into a coded strand, which can be carried out using enzymatic reactions such as strand displacement DNA synthesis [20]. Each coded strand has multiple sites, the sequence of which determines the binding of the assigned molecular beacons. The molecular beacons bind to the assigned sites and emit fluorescence. Thus, a combination of fluorescence signals can be generated following the composition of the sites in the coded strand. Importantly, the lengths of each site are adjusted to change the fluorescence time response. The kinetics of hybridization between the molecular beacons and the detected sequences depend on the length of their complementary regions [14]. The molecular signals are converted into time courses of fluorescence intensity according to the coded strand. Using the same sites in the design of all coded strands allows the molecular beacons to work universally. This means that the number of fluorescence probes needed for multiplex detection can be decreased.
In this study, a simple model is established to demonstrate the first example of multiplexed detection based on time responses. In the model, it is supposed that the concentration of the coded strands is binarized by amplifying the target molecules. Here, the received fluorescence signal y(t) = [y1(t) y2(t) · · · yNλ (t)]T represents the time variation of the i th fluorescence intensity (i = 1, 2, . . . , Nλ) at the t th timestep, t = 1, 2, . . . , T. The transmitted molecular signal x = [x1 x2 · · · xNtar]T represents the presence (xj = 1) or absence (xj = 0) of j th coded strand (j = 1, 2, . . . , Ntar), where Ntar is the total number of coded strands. Each coded strand is converted into the time responses of each fluorescence; the conversion matrix is expressed by H(t), the Nλ × Ntar matrix where entry hi,j (t) is the time response of the j th coded strand for the i th fluorescence. In this paper, each coded strand is assumed to be converted into fluorescence signals independently. Thus, the relation between the transmitted and the received molecular fluorescence signals is expressed as follows:
where z is the readout noise for the detection of the fluorescent signals. The readout noise z is assumed to have a Gaussian distribution with noise background offset b and variance σ2. Then, the likelihood of the model reconstructing the fluorescence signals L is expressed as follows: and the estimated molecular signal x̂ is obtained using maximum likelihood estimation as follows: The estimated molecular signal responding to the maximum of L is considered to be the transmitted molecular signal.The reaction between a molecular beacon B and a site in a coded strand S is assumed to be B + S ⇌ B · S, where · represents binding. Let us assume that the concentration of molecular beacons is much higher than that of the coded strands so that all coded strand sites bind with molecular beacons ([B] ≫ [S]). Additionally, to simplify the reaction model, it is approximated that the dissociation process may be ignored and the concentration of molecular beacons is constant during the reactions. The fluorescence is suppressed when B and S are separated and increased when B · S is formed. Under these assumptions, the time variation of the j th fluorescence signal from the i th coded strand is expressed as shown below:
where Ai,j, ki,j, and B0 are the fluorescence intensity at equilibrium, the reaction rate, and the initial molecular beacon concentration, respectively. Ai,j can be modulated by the number of sites and ki,j can be varied by the length of each site.3. Experiment
For the experimental demonstration, we prepared 10 coded strands and 3 molecular beacons. The sequences of the sites and fluorescent beacons are shown in Table 1. MB-A555, MB-A594, and MB-A647 represent the sequences of molecular beacons that were modified with Alexa 555, Alexa 594, and Alexa 647, respectively. The molecular beacons used in this study have previously been demonstrated in Refs. [15, 16]. xxA555, xxA594, and xxA647 (xx = 15, 16, 17) are the sites that binded MB-A555, MB-A594, and MB-A647, respectively. Each site was 15 to 17 bases long. The sequences of the coded strands are shown in Table 2. T20 was used as a spacer between the sites in the coded strands. Samples containing the molecular beacons were heated to 95 °C for 5 min and cooled rapidly to room temperature (22 °C) to form closed hairpin structures. The coded strands were then added to the molecular beacon solutions to obtain fluorescence signals. The final concentrations of each probe and each coded strand were 6.25 μM and 0.625 μM in 32-μL buffer (10 mM phosphate buffer and 250 mM NaCl), respectively. The composition of the coded strands were changed for each experiment.
Table 1. Sequences used in this study. *1, *2, *3, and *4 represent the positions modified with Alexa555, BHQ-2, Alexa594, and Alexa647, respectively.
Table 2. Sequences of coded strands.
Alexa 555 and Alexa 594 were excited with a 450-nm laser (PL450B, Osram Opto Semiconductors GmbH, Germany) and Alexa 647 was excited with a 660-nm laser (ML101J27, Mitsubishi, Japan, wavelength: 660 nm). The laser beams were combined with a beam splitter and passed into the sample to excite the molecular beacons simultaneously. The resulting fluorescence spectra were measured for 100 s at intervals of 5 s with a fiber-coupled spectrometer (BTC112E, B & W Tek, Inc., Newark, DE). The measured fluorescence spectra were decomposed to the spectral components of each fluorescence molecule using least square curve fitting [21]. The obtained peak fluorescence intensity values were plotted against time. The parameters for the prepared coded strands were obtained experimentally by fitting the measured fluorescence curves of each coded strand with Eq. (4) (shown in Table 3). All experiments were conducted at room temperature (22 °C).
Table 3. Measured parameters of fluorescence time variations for coded strands.
The coded strands were readout to confirm multiple detection. The fluorescence signals were measured 25 s after a coded strand was added to the sample and the time courses of each fluorescence intensity were obtained. Figure 2 shows (a) the measured fluorescence spectra at 125 sec and (b) the corresponding log-likelihoods. In the figure, the sample number corresponds to the added coded strand. Each coded strands produced a combination of individual fluorescence signals. The maximum log-likelihood was obtained in terms of contained coded strand number. All samples exhibited a maximum log-likelihood for the corresponding contained coded strand number. The results showed that a much larger number of targets was identified than the number of fluorescence signals used. Also, using the designed coded strands, the three universal molecular beacons allowed the generation of decodable molecular signals for multiple molecular detection. This feature lowers the number of fluorescence probes required for multiple detection.
To demonstrate multiplex detection, we prepared two samples containing different coded strands, [1 1 0 0 0 0 0 0 0 0]T for sample A and [0 0 0 0 0 0 1 0 1 0]T for sample B. The two samples had the same number of total sites corresponding to MB-A555 and MB-A594. Figure 3(a) and (b) show the time courses of the measured and estimated fluorescence. The estimated plots indicate the time courses of the molecular signal estimated based on Eq. (3) using solver CVX, a package for specifying and solving convex programs [22]. The estimated signals were [1 1 0 0 0 0 0 0 0 0]T (log-likelihood: −7.26 × 104) for sample A and [0 0 0 0 0 0 1 0 1 0]T (log-likelihood: −2.06 × 105) for sample B. This result shows that the multiplexed signals were transmitted correctly. To further confirm multiplexed detection, sample C [1 0 0 0 1 0 1 1 0 1]T was also tested. Figure 3(c) shows the time courses of the measured and estimated fluorescence intensities for input C. The molecular signal was estimated to be [1 0 0 0 1 0 1 1 0 1]T (log-likelihood: −6.10 × 105). The molecular signal was transmitted successfully with the multiplexed fluorescence signals. The differences between the measured and estimates signal intensities may have arisen from errors in pipetting and the fluorescence measurements.
Fig. 3 Estimated and measured fluorescence signals for (a) sample A, (b) sample B, and (c) sample C. “Estimated” refers to the intensities calculated taking the minimum likelihood.
As a performance evaluation, the accuracy readout rate Pa was determined by numerical simulation with noise-added data. Pa was defined as the rate of successful readout for 10,000 tests. The transmitted molecular signals were generated randomly. Figure 4(a) shows the dependence of Pa on the signal-to-noise ratio (SNR). As readout noise, Gaussian noise was added to the generated fluorescence signals. In this study, SNR is defined as:
where Imax is the sum fluorescence intensity when all coded strands were used. The fluorescence sampling number was set at 1, 2, 5, 10, 20, and 40 for 100 s. The higher the sampling number, the higher the accuracy readout rate.Fig. 4 Correct readout rate to (a) readout noise, (b) reaction speed error, and (c) signal intensity error.
Next, the tolerance for the signal distortion related to error in the reaction rate vector k was evaluated. Errors in k shift the time variation of the fluorescence signals. Here, the signal distortion ratio (SDR) is defined as:
where ‖ · ‖2 denotes the l2 norm and ysd(t) is the fluorescence signals generated using noise-added k. Errors would mainly be derived from the reaction conditions (e.g., temperature, concentration). Additionally, the tolerance for intensity modulation errors from changes in A was evaluated. These errors also derive from the reaction conditions, especially the concentration of coded strands. In this study, the modulation error ratio (MER) is defined as: where yme(t) is the fluorescence signals generated using the noise-added A. Figure 4(b) and (c) show Pa against SDR and MER. Higher SDR and MER (> 30 dB) are required for reliable readout (Pa > 0.9). An increase in the sampling number had little impact on SDR and MER compared with on the SNR results.4. Discussion
The experimental results show that multiplexed readout using time responses was achieved with the universal molecular beacons and coded strands. In this study, the kinetic parameters of the fluorescence signals were adjusted using the length of the complementary sequences. To estimate the multiplexing ability, supposing the number of available sites equals N and the available fluorescence molecules equals M, the possible number of fluorescence codes is N+M CN − 1. For example, if N = 5 and M = 4 are available, the possible number of multiplexed signals in a single test tube is 125. This multiplexing ability can not compete with that of other multiplexing enhancement methods using spatial dimensions such like DNA microarray technique (>10,000 target), which requires specific substrates and optical setup for readout. In contrast, our multiplexing approach can provide multiplexed detection with simple fluorescence measurement of one test tube. This feature offers prospective applications for on-site testing of microRNAs. The use of universal fluorescence probes to readout all coded strands is a cost effective method because labeled strands are usually more expensive than non-labeled ones. However, the coded strands must be generated and amplified enough to detect the fluorescence signals against the assigned target molecules. A promising approach would be to use the cascaded DNA reactions [17] and stem-loop real-time PCR [23].
The available bit depth for the measurement of fluorescence signals can be a limiting factor to the enhancement of the degree of multiplexing. The fluorescence intensity bandwidths assigned to individual coded strands become narrower as the multiplexed signal number increases. Fluorescence time variation can be used to decode multiplexed fluorescence signals and to improve accuracy rates, especially at low SNR as shown in Fig. 4(a). But, as shown in Fig. 4(b) and (c), errors in the fluorescence signals influence the accuracy and would degrade the performance. To solve this, the kinetic parameters should be optimized using appropriately designed sequences. The accuracy of the readout depends on the kinetics of the fluorescence conversion processes. The amplification curve should be modeled as a sigmoid curve during the real-time PCR process and the kinetic parameters should be analyzed in detail. This fluorescence conversion process would enhance the measurement performance compared with that achieved using the sum of the simple exponential curves. Moreover, other prior information regarding the target molecules (e.g., variability distributions of the kinetic parameters, triage based on the micro RNA biomarkers for cancer diagnosis [24]) could be used for likelihood modeling to enhance measurement performance. Another limiting factor is signal interference during the fluorescence conversion process. The conversion of each coded strand to fluorescence signals is supposed to be independent in this study. However, in practice, unspecified bindings occur and generate incorrect fluorescence signals that leads signal interferences. To reduce the signal interferences, the sequences of sites and probes should be optimized not to occur unspecified bindings.
5. Conclusion
We demonstrated that multiple molecules can be detected using time varied fluorescence signals. In experiments, three kinds of universal molecular beacons were used to detect combinations of ten different coded strands. The use of a universal fluorescence probe set offers a cost effective method for fluorescence readout. In traditional methods, the detectable range of fluorescence intensity is usually limited by the performance of the detection device. Thus, use of the time domain information provides an effective way to increase the number of simultaneously-detectable molecules.
Funding
Japan Society for the Promotion of Science KAKENHI Grant-in-Aid for Young Scientists (B) (16K16408).
References and links
1. M. R. Hartman, R. C. H. Ruiz, S. Hamada, C. Y. Xu, K. G. Yancey, Y. Yu, W. Han, and D. Luo, Nanoscale, “Point-of-care nucleic acid detection using nanotechnology,” Nanoscale 5, 10141–10154 (2013). [CrossRef] [PubMed]
2. Y. N. Teo and E. T. Kool, “DNA-Multichromophore Systems,” Chem. Rev. 1124221–4245 (2012). [CrossRef] [PubMed]
3. L.G. Lee, K.J. Livak, B. Mullah, R.J. Graham, R.S. Vinayak, and T.M. Woudenberg, “Seven-color, homogeneous detection of six PCR products,” BioTechniques 27, 342–349 (1999). [PubMed]
4. V. Boln-Canedo, N. Snchez-Maroo, A. Alonso-Betanzos, J.M. Bentez, and F. Herrera, “A review of microarray datasets and applied feature selection methods,” Information Sciences 282, 10141–10154 (2014).
5. M. Han, X. Gao, J. Z. Su, and S. Nie, “Quantum-dot-tagged microbeads for multiplexed optical coding of biomolecules,” Nat. Biotechnol. 19, 631–635 (2001). [CrossRef] [PubMed]
6. Y. Li, Y. T. H. Cu, and D. Luo, “Multiplexed detection of pathogen DNA with DNA-based fluorescence nanobar-codes,” Nat. Biotechnol. 23, 885–889 (2005). [CrossRef] [PubMed]
7. C. Lin, R. Jungmann, A. M. Leifer, C. Li, D. Levner, G. M. Church, W. M. Shih, and P. Yin, “Submicrometre geometrically encoded fluorescent barcodes self-assembled from DNA,” Nat. Chem. 4, 832–839 (2012). [CrossRef] [PubMed]
8. Q. Huang, Q. Hu, and Q. Li, “Identification of 8 Foodborne Pathogens by Multicolor Combinational Probe Coding Technology in a Single Real-Time PCR,” Clinical Chemistry 53, 1741–1748 (2007). [CrossRef] [PubMed]
9. Q. Huang, L. Zheng, Y. Zhu, J. Zhang, H. Wen, and J. Huang, “Multicolor Combinatorial Probe Coding for Real-Time PCR, ” PLoS ONE 6, e16033 (2011). [CrossRef] [PubMed]
10. A. Rajagopal, A. Scherer, A. Homyk, and E. Kartalov, “Supercolor Coding Methods for Large-Scale Multiplexing of Biochemical Assays,” Anal. Chem. 85, 7629–7636 (2013). [CrossRef] [PubMed]
11. V. Taly, D. Pekin, L. Benhaim, S. K. Kotsopoulos, D. L. Corre, X. Li, I. Atochin, D. R. Link, A. D. Griffiths, K. Pallier, H. Blons, O. Bouch, B. Landi, J. B. Hutchison, and P. Laurent-Puig, “Multiplex picodroplet digital PCR to detect KRAS mutations in circulating DNA from the plasma of colorectal cancer patients,” Clin. Chem. 59, 1722–1731 (2013). [CrossRef] [PubMed]
12. T. Nishimura, Y. Ogura, K. Yamada, Y Ohno, and J. Tanida, “Biomolecule-to-fluorescent-color encoder: modulation of fluorescence emission via DNA structural changes,” Biomed. Opt. Express 7, 2082–2090 (2014). [CrossRef]
13. R. G. Rutledge, “Sigmoidal curve-fitting redefines quantitative real-time PCR with the prospective of developing automated high-throughput applications,” Nucl. Acids Res. 32, e178 (2004). [CrossRef] [PubMed]
14. A. Tsourkas, M. A. Behlke, S. D. Rose, and G. Bao, “Hybridization kinetics and thermodynamics of molecular beacons,” Nucl. Acids Res. 31, 1313–1330 (2003). [CrossRef]
15. A. S. Piatek, S. Tyagi, A. C. Pol, A. Telentii, L. P. Miller, F. R. Kramer, and D. Alland, “Molecular beacon sequence analysis for detecting drug resistance in Mycobacterium tuberculosis, ” Nat. Biotechnol. 16, 359–363 (1998). [CrossRef] [PubMed]
16. M. B. Baker, G. Bao, and C. D. Searles, “In vitro quantification of specific microRNA using molecular beacons,” Nucl. Acids Res. 40, 1–12 (2012). [CrossRef]
17. K. Komiya and Y. Yamamura, “Cascading DNA Generation Reaction for Controlling DNA Nanomachines at a Physiological Temperature,” New Generat. Comput. 33, 213–229 (2015). [CrossRef]
18. R. M. Dirks and N. A. Pierce, “Triggered amplification by hybridization chain reaction,” Proc. Natl. Acad. Sci. USA 101, 15275–15278 (2004). [CrossRef] [PubMed]
19. M. Weitz, J. Kim, K. Kapsner, E. Winfree, E. Franco, and F. C. Simmel, “Diversity in the dynamical behaviour of a compartmentalized programmable biochemical oscillator,” Nat. Chem. 6, 295–302 (2014). [CrossRef] [PubMed]
20. C. Li, Z. Li, H. Jia, and J. Yana, “One–step ultrasensitive detection of microRNAs with loop-mediated isothermal amplification (LAMP),” Chem. Commun. 47, 2595–2597 (2011). [CrossRef]
21. K. Lao, N. L. Xu, V. Yeung, C. Chen, K. J. Livak, and N. A. Straus, “Nanometer-scale Fluorescence Resonance Optical Waveguides,” Nano Letters 4, 1035–1039 (2004). [CrossRef]
22. M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex programming,” http://cvxr.com/cvx (2013).
23. K. Lao, N. L. Xu, V. Yeung, C. Chen, K. J. Livak, and N. A. Straus, “Multiplexing RT-PCR for the detection of multiple miRNA species in small samples,” Biochemical and Biophysical Research Communications 343, 85–89 (2006). [CrossRef] [PubMed]
24. K. M. Foss, C. Sima, D. Ugolini, M. Neri, K. E. Allen, and G. J. Weiss, “miR-1254 and miR-574-5p: serum-based microRNA biomarkers for early-stage non-small cell lung cancer,” J. Thorac. Oncol. 6, 482–488 (2011). [CrossRef] [PubMed]
