The chemically responsive indicators used in prior colorimetric sensor arrays have been limited to soluble molecular dyes held in a semi-fluid polymer film and generally printed onto a porous membrane2–5, 20–24
. Relative to insoluble pigments, chemically responsive dyes in semi-fluid state are often unstable over long periods of storage. Non-porous pigments, on the other hand, generally do not provide sufficient contact between the analyte and the chromophores of the pigment, because the chromophores at the surface of the pigment are generally only a small fraction of the total number of chromophores.
There is, however, a general method to the incorporation of chemically responsive dyes into porous matrices, thus creating nanoporous pigments that can provide both stability and analyte access to the chromophores. Among various host materials, ormosils provide excellent matrices for a variety of organic and inorganic colorants28–31
. Furthermore, the final properties of the nanoporous pigments (e.g., hydrophobicity, porosity, and surface area) can be easily modified by controlling the physical and chemical parameters of the sol-gel process. For our printing of sensor arrays, we prepared sol-gel-colorant solutions by the simple hydrolysis of solutions containing commercially available silane precursors (e.g., tetraethoxysilane, methyltriethoxysilane, phenethyltrimethoxysilane and octyltriethoxysilane) with a variety of chemically responsive indicators.
This conversion of soluble dyes into nanoporous ormosil pigments significantly improves stability28–31
: we find that there is no change in the analyte response for arrays stored for more than three months27
. Importantly, our sol-gel formulation permits the simple printing onto ordinary polymer flat surfaces, which has allowed us an easy packaging of the sensor array into a self-sealing cartridge with minimal dead-space (Supplementary Fig. S1
) suitable for use with a handheld battery-operated scanner (Supplementary Fig. S2
). In addition, we find that the porous matrix improves the sensitivity of the sensor; compared to prior results with highly plasticized dye sensor arrays, we find an improvement in sensitivity of ~800% at low concentrations. We hypothesize that the nanoporous matrix may serve as an in situ
preconcentrator for analytes.
As an important application of this new optoelectronic nose, we have begun an examination of TICs, choosing 19 representative examples from lists generated by the International Task Force 25 and 40 reports32–33
. The IDLH concentrations of these TICs are given in Supplementary Table S2
. Detection and discrimination among the wide range of high priority TICs remains a major challenge34
and the subject of substantial recent research. For example, Hammond et al. recently reported35
on TIC identification using an array of ceramic metallic films; they were able to different-tiate among ten TICs with an error rate of ~10% using a linear discriminant analysis. Using metal oxide detectors combined with temperature programming, Meier et al. examined36
five toxic industrial chemicals and were able to reduce their error rate (both false negatives and positives) to 3%.
We have extensively tested our colorimetric sensor array against 19 TICs at their IDLH concentrations at 50% relative humidity (RH). The colorimetric sensor arrays were exposed for two minutes to a diluted gas mixture produced either from pre-mixed, certified gas cylinders or from saturated vapor, using digital mass flow controllers (configurations shown in Supplementary Fig. S3
). Importantly, gas stream concentrations were confirmed by in-line analysis by FT-IR in real time using a MKS multi-gas analyzer for most analytes or by Dräger detector tubes in the few cases where FT-IR cannot be used (e.g., homonuclear diatomics Cl2
The printed arrays were digitally imaged with an ordinary flatbed scanner before and after exposure to each analyte. Color difference maps for the arrays were generated by subtraction of the image before exposure from the image after exposure: such a difference map provides a molecular fingerprint that effectively identifies the analyte to which the array has been exposed. The difference maps permit facile detection and identification of the 19 representative TICs, as shown in . Even by eye, without statistical analysis, the array response to each TIC is represented by a unique pattern. For quantitative analysis of the difference maps, we can define a 108-dimensional vector (i.e., 36 changes in red, green and blue values for the after exposure image compared to the before exposure image for our 6×6 array of nanoporous pigments). In this fashion, it is not necessary to retain the full digital images: each analyte is represented digitally by the 108-dim. vector and may be compared by standard chemometric techniques. Simple comparisons of the Euclidean distance between an observed color difference vector and a pre-collected library are extremely rapid (sub-second); even a large library requires only minimal memory (e.g., 1000 entries needs 148 KB zip compressed). The raw digital color difference vectors from a total of 140 experimental trials are given in Supplementary Tables S3 and S4
Color change profiles of representative toxic industrial chemicals (TICs) at their IDLH (immediately dangerous to life or health) concentration after 2 min of exposure
For most analytes, the response of these colorimetric sensor arrays is based primarily on equilibrium interactions between the array pigments and the analyte, and consequently, each concentration of a TIC has a separate pattern which can be used to establish a limit of detection. As examples, color change profiles for three different analytes as a function of concentration can be seen in . Here the limits of detection (LODs) for all three examples are well below their respective PELs. While highly dependent on the analyte, our estimates for the LOD for TICs examined here are all well below their respective PELs.
The effect of concentration on array response to NH3, Cl2, and SO2
Most of the TICs can be identified from the array color change in a matter of seconds, and >90% of total response is observed in less than two minutes, as shown in for five representative TICs; all other TICs tested show very similar responses. For some aggressive analytes that undergo irreversible reactions with some pigments (e.g., bleaching), response time can be slightly longer, as demonstrated for Cl2 (), but even in these cases the color change pattern is distinctive and easily recognized.
Response time of the array
While the CSAs are meant to be disposable, they are still re-usable for most analytes. The CSA is best thought of as a “chemical fuse”: just as with an electric fuse, as long as the concentration of the odorant (i.e., “the electric current”) fluctuates within some range, the CSA (“fuse”) is unaffected. But, if the concentration increases to too high a value, the CSA will take too long to recover: i.e., the fuse is blown and the CSA should be replaced. As illustrated in , the CSA will reproducibly cycle between the IDLH and PEL concentrations of many toxic industrial chemicals (SO2 as an example). After switching from one concentration to the other, equilibrium response is achieved within two minutes. The reversibility of the array is dependent on the type of chemical interaction between the pigments and the analytes, and for irreversible reactions with highly aggressive analytes (e.g., those capable of bleaching or redox reactions), the array cannot be recycled as demonstrated in for Cl2. For other gases, notably arsine and phosphine, it is the reduction of metal salts generating metal nanoparticles that produces acidic byproducts that are detected by pH indicators incorporated into the nanoporous sol-gel spot. Finally, the response to phosgene is due to an alkylation reaction with 4-(4-nitrobenzyl)pyridine within the sol-gel matrix.
Reversibility of colorimetric array response
In real world situations, one is unlikely to have control over the humidity, which can be extremely problematic for prior electronic nose technologies. Because both our colorants and the sol-gel matrices are selected to be hydrophobic, we find that changes in humidity present no difficulty. As shown in Supplementary Fig. S4
, the color difference maps are unaffected by changes in RH from 10% to 90%.
The color change profiles are inherently digital data and are easily handled by routine chemometric analysis37–40
. The high dispersion of the colorimetric sensor array data requires a classification algorithm that uses the full dimensionality of the data. The simplest statistical approach is hierarchical cluster analysis (HCA), which is a classification scheme based on the Euclidean distance between data points in their full dimensionality. The advantage of HCA compared to other model-dependent statistical analysis (e.g. linear discriminant analysis) is that it makes no assumptions about the classification of results that one is trying to establish. Each experimental trial is defined as a 108-dimensional vector consisting of the changes in red, green, and blue values of each of the 36 nanoporous pigments in our array. Hierarchical clustering is based on the Euclidean distance in this 108-dimensional ΔRGB color space among these vectors, which generates a dedrogram, as shown in . Remarkably, in septuplicate trials, all 19 TICs and a control were accurately identified against one another with no errors or misclassifications out of 140 cases.
Hierarchical cluster analysis (HCA) for 19 TICs at IDLH concentrations and a control
The ability of our CSAs to discriminate so many analytes from one another is impressive and is due largely to the extremely high dimensionality of the array data. Principal component analysis (PCA) can be used to determine the number of meaningful independent dimensions probed by a cross-reactive array. The eigenvector of each principal component defines the linear combination of the response of each sensor parameter by the amount of variance in the data along each principal component. Based on the 140 trials on 20 analytes (i.e., 19 TICs plus background), the PCA of our colorimetric sensor array requires 9 dimensions for 90% of total variance (and 13 dimensions for 95%, as shown in Supplementary Fig. S5
). This extremely high dispersion reflects the very wide range of chemical-properties space being probed by our choice of chemoresponsive pigments. As a consequence, chemically diverse analytes are easily recognizable and even closely related mixtures can be distinguished2–4
. In contrast, data from most prior electronic nose technology are dominated by only two or three independent dimensions (one of which, analyte hydrophobicity, generally accounts for >90% of total variance); this is the inherent result of relying on van der Waals and other weak interactions (e.g., adsorption to metal oxide surfaces or absorption onto or into polymer films) for molecular recognition.