Fluorescence imaging, due to its high sensitivity and selectivity, has become one of the most popular analytical tools for detecting biomolecular interactions [1–3]. Nowadays, fluorescence imaging is required in a number of fields, especially in biological, medical and food industries [4,5]. Conventional fluorescence scanners are filter-based, with multiple filters employed to distinguish different fluorescent labels [6]. There are some disadvantages for this filter-based mechanism. Firstly, when multiple fluorescent labels are needed, the chosen labels should have minimal overlap of their emission spectra, otherwise, it is difficult to identify various labels using only filters. Secondly, it poses challenges for filter-based scanners to determine whether the measured data have been corrupted by some extraneous emission sources. For instance, the contamination induced by fabrication chemistry as well as the emissions from the substrates [7] may affect the determination of gene or protein expressions [8]. To overcome these problems, we have previously developed a quasi-confocal, parallel scan hyperspectral fluorescence imaging system, which has been successfully applied to microarray analysis [9]. The hyperspectral fluorescence approach works by recording full spectra of all the points on the sample plane via a spectrometer [8–10]. Multivariate data exploitation is then employed for analysis of spectral information to accurately determine ratios of different emission sources [11,12]. As a result, the restrictions on the allowable fluorescent labels are reduced, and the detection accuracy is enhanced by removal of emissions of the contamination and substrates, exhibiting advantages in multi-label strategies.

Microfluidic systems are of great significance and have been widely employed for biochemistry applications [13–15]. A great many microfluidic systems only provide mono-channel detection with a single flow cell [16,17]. However, in some cases it is necessary to carry out multi-analyte sensing, while the mono-channel detection is difficult to do so. Moreover, a fixed point of the flow cell is conventionally monitored in a fluidic protocol. However, the biomolecular interactions vary among different points, and the information of a fixed point can hardly provide a comprehensive analysis of biological events. In order to carry out multi-analyte biosensing, microfluidic systems with multiple channels have been developed [18,19], offering the potential for high-throughput and multi-channel biomedical research.

In this work we present a line-monitoring, hyperspectral fluorescence setup which is optimized for applications in multi-channel microfluidic systems. The laser light is focused to a line to cover all the channels of the microfluidic system for fluorescence excitation, and thus multiple channels can be monitored simultaneously. Moreover, a small region of the fluidic channel is scanned and the final output is an average of signals of the scanned region. In this way we acquire information of an area instead of a fixed point, reducing measurement errors that might occur. The performance of this sensor is characterized, and its biomedical application has been demonstrated by a model experiment detecting the hybridization of short oligonucleotides.

In order to demonstrate the biomedical application of this line-monitoring, hyperspectral fluorescence biosensor, we designed a model experiment to detect short oligonucleotides for identification of bacteria using multi-channel microfluidic systems. Figure 4(a) shows the preparation of DNA probes. As can be seen, the 1st, 2nd and 5th channels immobilize Escherichia coli probes (5′ (SH)-ACG GTT ACC TTG TTA CGA CTT-3′), and the other two channels immobilize β-hemolytic streptococcus probes (5′ (SH)-ACC GTC ACT TGG TGG ATT-3′). Then five kinds of solutions that contain different fluorescence labeled complementary oligonucleotides are flowed through these fluidic channels at a flow-rate of 20 μL/min by means of a peristaltic pump. The sample solutions used corresponding to each channel are as follows: 1 μmol/L Cy5 labeled Escherichia coli oligonucleotides (channel 1), 0.5 μmol/L Cy5 labeled Escherichia coli oligonucleotides (channel 2), 2 μmol/L Dylight 680 labeled β-hemolytic streptococcus oligonucleotides (channel 3), 1 μmol/L Dylight 680 labeled β-hemolytic Streptococcus oligonucleotides (channel 4), and a mixture of 0.5 μmol/L Cy5 labeled Escherichia coli oligonucleotides and 2 μmol/L Dylight 680 labeled Escherichia coli oligonucleotides (channel 5). These synthesized oligonucleotides present in sample solutions are complementary to the substrate bound DNA probes. The excitation efficiency of Dylight 680 by a He-Ne laser is weaker than that of Cy5, and thus the applied concentration of Dylight 680 labeled sample solutions is higher than that of Cy5 labeled solutions. After the hybridization process, fluidic channels are rinsed thoroughly by deionized water to remove unbound oligonucleotides, and then ready for optical scan by this hyperspectral fluorescence setup.

Figure 4(b) shows the detection results with one spectral curve corresponding to each channel. Every spectral curve is acquired by two averaging processes. Firstly, a small region (∼1 mm) of these fluidic channels (∼0.8 mm in width per channel) is scanned, and all the scanning images are averaged. Secondly, all the rows corresponding to a channel are averaged. Finally we acquire a spectral curve representing each fluidic channel, by which a wealth of information can be obtained. The sample concentration and hybridization efficiency are demonstrated by the relative fluorescence intensity, and the biochemical interactions can be revealed via the spectral characterization due to the specificity of each fluorescent spectrum. The curves corresponding to the 1st and 2nd channels exhibit identical spectra due to the same fluorescent labels, except that the fluorescence intensities vary from each other because of different sample concentrations. The similar results are observed in the 3rd and 4th channels.

In the 5th channel, both Cy5 and Dylight 680 labeled complementary Escherichia coli oligonucleotides are present in the flowing sample solutions. Thus the acquired spectrum of channel 5 is different from absolute Cy5 or Dylight 680 spectrum. We apply multivariate data exploitation to the raw spectrum of channel 5 to determine ratios of each dye, as shown in Figure 5. The black curve represents the raw spectrum, and the red and blue curves represent the weighted Cy5 and Dylight 680 spectra, respectively, which are calculated by multivariate data analysis. The yellow curve is a sum of the weighted Cy5 and Dylight 680 spectra. As can be seen from Figure 5, the sum curve closely corresponds to the raw spectrum, revealing that the multivariate algorithm that we have developed can successfully model the hyperspectral stack using linear admixtures of these two dyes. Besides distinguishing different fluorescent tags, the significance of this experiment in channel 5 is that our system can also manage to identify and eliminate unwanted artifacts, e.g., the fluorescence emissions from the substrate or contamination, by spectrally resolving the overall raw spectra.

In this work we propose a line-monitoring, hyperspectral fluorescence biosensor designed and optimized for applications in multi-channel, microfluidic systems. By recording the fluorescence spectra instead of intensities, the hyperspectral technique enhances the detection accuracy by identification and elimination of some unwanted artifacts. Moreover, this sensor is able to detect multiple fluidic channels simultaneously, providing the potential for concurrent multi-analyte biosensing. The output of this sensor, behaving as a spectral curve, shows both relative intensity and the spectrum characterization of corresponding fluorescent labels, and thus the concentration of the sample solution as well as the information related to the biomolecules can be obtained. The sensitivity of this biosensor is better than 1 fluors/μm², and its homogeneity is proved good. On account of these features, we believe that this line-monitoring, hyperspectral fluorescence biosensor has the potential for providing high-throughput and high-accuracy analysis of biochemical interactions.