Computer vision vs. spectrofluorometer-assisted detection of common nitro-explosive components with bola-type PAH-based chemosensors

Computer vision (CV) algorithms are widely utilized in imaging processing for medical and personal electronics applications. In sensorics CV can provide a great potential to quantitate chemosensors' signals. Here we wish to describe a method for the CV-assisted spectrofluorometer-free detection of common nitro-explosive components, e.g. 2,4-dinitrotoluene (DNT) and 2,4,6-trinitrotoluene (TNT), by using polyaromatic hydrocarbon (PAH, PAH = 1-pyrenyl or 9-anthracenyl) – based bola-type chemosensors. The PAH components of these chemical bolas are able to form stable, bright emissive in a visual wavelength region excimers, which allows their use as extended matrices of the RGB colors after imaging and digital processing. In non-polar solvents, the excimers have poor chemosensing properties, while in aqueous solutions, due to the possible micellar formation, these excimers provide “turn-off” fluorescence detection of DNT and TNT in the sub-nanomolar concentrations. A combination of these PAH-based fluorescent chemosensors with the proposed CV-assisted algorithm offers a fast and convenient approach for on-site, real-time, multi-thread analyte detection without the use of fluorometers. Although we focus on the analysis of nitro-explosives, the presented method is a conceptual work describing a general use of CV for quantitative fluorescence detection of various analytes as a simpler alternative to spectrofluorometer-assisted methods.


Absorption and emission spectroscopies
Emission and excitation spectra were measured on the Horiba FluoroMax-4 spectrofluorometer. UV-Vis absorption spectra were recorded on the Shimadzu UV-1600 spectrophotometer.

Quenching experiments (spectrofluorometer setup)
Fluorescence titration experiments were carried out on the Horiba-Fluoromax-4 spectrofluorometer. All titration experiments were conducted in a standard 1 cm, 3 ml quartz cuvettes (Hellma Optik GmbH Jena, Germany). The fluorescence titration experiment started with the preparation of fluorophore solutions by dissolving sensors 1 or 2 compounds in DMSO:H2O [1:1 (v/v)]. The fluorescence quenching was measured by Single Point method with λEx = 353 nm, λEm = 507 nm for sensor 1 and λEx = 353 nm, λEm = 472 nm for sensor 2. 3 ml of the sensor solutions (1×10 -5 M for sensor 1 and 1×10 -6 M for sensor 2) were placed in quartz cells following by stepwise addition of 10-15 aliquots of the analyte (2,4,6-Trinitrotoluene (TNT)) dissolved in acetonitrile with concentration 1×10 -3 M for sensor 1 and 1×10 -4 M for sensor 2. Analysis of fluorescence emission intensity was performed on the base of a Stern-Volmer equation (1): where I0 is the intensity, or rate of fluorescence, without a quencher, I is the intensity, or rate of fluorescence, with a quencher, KSV is the Stern-Volmer quenching constant for complex formation, and [Q] is the concentration of the quencher. The quenching constant (KSV) was calculated as the slope of the graph intensity ((I0/I)-1) versus the concentration of the quencher ([Q]).

Computer vision-assisted quenching experiments (spectrofluorometer-free setup)
The Canon D3000 Kit camera was used for the images. The fluorescence quenching experiments were carried out in 10 ml glass vials (Medsteklo, Russia). The fluorescence quenching experiment started with the preparation of fluorophore solutions. For sensors 1 and 2 were used DMSO:H2O [1:1 (v/v)] solutions. 6 ml of the sensor solution (1×10 -5 M for sensor 1 and 1×10 -6 M for sensor 2) was placed in a glass vial following by the addition of 11 aliquots of TNT in acetonitrile (2×10 -3 for sensor 1 and 2×10 -4 M for sensor 2). The first aliquot volume was 3μl; the second was 7μl, and the following nine aliquots were 10μl each. Imaging of both vials at once has been done in a dark chamber under UV lamp (λ=365 nm) before and after the addition of each aliquot. The images are shown in Figure S3. Analysis of the fluorescence emission intensity was done by the image processing algorithm in Mathcad 15.0 packets (M020 [MC15_M020_20121127]) © Parametric Technology Corporation. The received values were plotted as the Stern-Volmer graph in a similar way described in the quenching experiments procedure. The detailed image-processing algorithm (computer vision) is shown on page S10 as the total listing of the Mathcad programming:

Dynamic Light Scattering experiments
Measurements were made on the Malvern Zetasizer Nano ZS (Malvern Panalytical, U.K.) equipped with a 4 mW He-Ne laser operating at a wavelength of 633 nm. The scattered light is detected at an angle of 173°, an optical arrangement known as non-invasive back scatter (NIBS) optic arrangement. Measurements were carried out in a quartz cell (volume is 1 ml) with round aperture at 25 °C. Data on viscosity and refractive index for solvent mixtures were taken from the literature 1 . The measurements were carried out in an automatic mode; the autocorrelation function was analyzed to assess the measurement quality.
The dynamic light scattering (DLS) experiments were carried out in different solvents: DMSO, DMSO:H2O [1:1 (v/v)] and cyclohexane. For measurements fresh solutions prepared within 1-2 hours were used. No particles were detected in pure DMSO and cyclohexane. Only fresh solutions of sensors 1-2 in aqueous DMSO showed presence of particles. However, in solutions of sensors 1-2 in a mixture of solvents, measured 24 hours after preparation, there were no particles, which may be due to their sedimentation. To study the interaction of chemosensors 1-2 with DNT, we used their solutions in a DMSO:H2O [1:1 (v/v)] at 5×10 -5 M concentration. To this solution of sensor a solution of DNT in acetonitrile (the concentration of DNT was 1×10 -2 M) or pure acetonitrile in an appropriate amount was added.

Calculation of the Limit of Detection (LOD)
The detection limit was determined based on the data of experiments on fluorescent quenching according to the method published previously 2 . A calibration curve was plotted between the fluorescence intensity and the analyte concentration to obtain a regression curve equation (see Fig. S4). The limit of detection was calculated using the equation (2): S3 = 3 ÷ (2) where σ is the standard deviation of the fluorophore intensity in the absence of an analyte obtained via STEYX function in Excel and k is the slope of the calibration curve.