|Home | About | Journals | Submit | Contact Us | Français|
A partially overlapping population of random sequence 60mer DNA molecules consisting of many concatamers of varied lengths was spatially separated in one and two dimensions by electrophoresis in polyacrylamide and transferred to nitrocellulose membranes. The spatially separated library serves as a potential sensor interface on which many different molecular recognition events or target analyte-binding patterns may emerge, thereby theoretically representing a “universal sensor” surface. The separated DNA library has been referred to as a DNA combinatorial array recognition surface or “CARS.” After UV baking and various fluorescence staining or fluorescent probe interactions, the one-dimensional (1-D) and 2-D membrane-bound CARS were digitally photographed and subjected to image analysis with National Institutes of Health Image-Java software. Image analysis demonstrated relatively consistent and more similar spatial fluorescence patterns within CARS analyte treatment groups but noteworthy pattern differences before and after analyte addition and between different analyte treatments. Taken together, these data suggest a potential role for CARS as a novel, inexpensive, self-assembling universal molecular recognition surface that could be coupled to sophisticated Bayesian or other pattern recognition algorithms to classify analytes or make specific identifications, much like the senses of smell or taste.
Senses such as smell and taste represent “universal” biochemical sensor systems capable of detecting and at least partially identifying almost any target molecules that elicit a response. These senses are the results of pattern recognition events in the brain based on the interactions of sensory cells with analytes that they encounter in the environment. The brain rapidly examines various complex electrical signal patterns associated with the chemical interaction of an analyte and sensory cells to classify an interaction as tasting sweet, salty, or bitter. An animal's tongue is a natural biosensor array composed of spatially separated sensory cells that can detect various classes of molecules and report to the brain, which then categorizes the information.
Previous work by Bruno et al.1 represented an initial attempt to mimic gustatory and olfactory senses by ordering random sequence DNA molecules in a one-dimensional (1-D) array, primarily according to length by electrophoresis (Fig. 1). In theory, the random sequence DNA library actually constitutes a large collection of potential “aptamers” that can bind virtually any target molecule with varying affinities and in different spatial regions to produce characteristic optical patterns.1 It has become abundantly clear over the last two decades that DNA aptamers can bind a broad spectrum of different types of analytes from small molecules2–7 to whole cells.8–13 Bruno et al.1 named the array of partially hybridized or concatenated and electrophoresis-separated random sequence DNA chains “CARS”. The CARS concept is similar to an animal's tongue with regions that can detect various classes of molecules if appropriate transduction mechanisms and complex pattern recognition are used. Therein lies the real challenge for advancing CARS as a pragmatic “universal sensor” technology, i.e., the ability to detect binding events in a complex molecular “jungle” or “smear” of 1-D size-ordered (Fig. 1) or theoretical 2-D (size- and charge-ordered) random sequence DNA molecules.
Bruno et al.1 originally had marginal success with simply scanning the natural low-level autofluorescence of DNA14 in a polyacrylamide gel strip, in which a CARS DNA library had been run and formed a 1-D array. Some consistent differences were observed before and after various analyte binding in the gel strip, but the gel acted as a barrier to free diffusion of target molecules and therefore, did not permit rapid interactions or a practical format for portable field use.1 In addition, swelling or shrinkage of the gel over time affected reproducibility of fluorescence scans.1 It was thought that transfer of the 1-D CARS to a membrane might facilitate the speed and ease of detecting analyte interactions with CARS. The present paper focuses on this aspect of CARS development by demonstrating fairly reproducible differences in complex 1-D CARS patterns before and after interaction with different analytes by fluorescence image analysis. In addition, the first attempts at 2-D CARS detection and characterization are reported on nitrocellulose membranes.
To generate CARS immobilized on membranes, completely randomized 60-base DNA libraries (2–4 mg/mL) and all other oligonucleotides were obtained from Integrated DNA Technologies, Inc. (Coralville, IA, USA). Each position in the 60-base oligonucleotide is randomized for the four nucleotides (A, C, G, and T). This lyophilized DNA library was rehydrated in 1× aptamer-binding buffer (1×BB; 0.5 M NaCl, 10 mM Tris-HCl, and 1 mM MgCl2, pH 7.5–7.6) and mixed 1:1 with 5× nucleic acid loading buffer (Bio-Rad Laboratories, Hercules, CA, USA). The 1:1 diluted CARS random 60mer DNA (20 μL) was loaded into each well of 4–20% gradient Tris-boric acid-EDTA buffer (TBE)-polyacrylamide mini electrophoresis gels (Invitrogen, Carlsbad, CA, USA) and run for approximately 1 h at 200 V in cold TBE until the dye front reached the bottom of each gel, and gels were then rinsed briefly in 100 mL TBE and stained in 100 mL TBE plus 5 μL 1% ethidium bromide (EtBr) for 10 min, followed by a 10-min wash in 100 mL TBE. Gels were imaged on a UV transilluminator using a UVP, LLC (Upland, CA, USA), Doc-It® system with Life Sciences digital imaging software.
Gels were next overlaid with TBE-wetted nitrocellulose or nylon mini membranes (Bio-Rad Laboratories) and placed inside filter paper sandwiches, which were placed in cassettes and an electro-blotting tank (Bio-Rad Laboratories) with 1.5 liters cold TBE, and the DNA was transferred to membranes for 2 h at 32–38 V and 300 mA or 10 W in cold TBE. Membranes were removed, rinsed briefly in TBE, air-dried, and “baked” in a UV oven for 15 min, after which, they were re-wetted with TBE and imaged again using the Doc-It®. Membranes were often cut in half vertically between lanes to enable comparison of two different analyte-binding patterns from the same membrane, and membranes were equilibrated in 25 mL 1 × BB and exposed to various target analytes at final concentrations of approximately 50 μg/mL in 25 mL 1 × BB for 1 h with gentle mixing. All target analytes were obtained from Invitrogen or Sigma Chemical Co. (St. Louis, MO, USA). Thereafter, membranes were removed and washed in 50 mL 1 × BB plus 0.1% SDS detergent for 30 min, followed by two washes in 50 mL 1 × BB.
Membranes were imaged again on the Doc-It®. All images were subjected to image analysis using National Institutes of Health Image-Java (NIH Image-J) software, which was downloaded from the internet. The NIH Image-J profiling tool, surface plot, and 3-D plug-in features were used to generate the data in this paper within defined regions of interest (ROI), as indicated by boxes in each of the figures. All image analyses were performed on eight-bit color images without color channel (red green blue) splitting.
The facile Invitrogen ZOOM® immobilized pH gradient (IPG) Runner™ 2-D mini-polyacrylamide gel system was used throughout this project. Briefly, ZOOM® IPG [isoelectric focusing (IEF)] strips, having a gel side with a linear pH gradient from pH 3 (+) to 10 (−), were threaded into channels of a disposable plastic ZOOM® cassette along with 160 μL CARS 60mer DNA library (2–4 mg/mL) in 18 M Ohm deionized water with 1% ampholyte stabilizers (Invitrogen). After sealing the cassette with tape, the strips were equilibrated for 16 h at 4°C in a horizontal position. The tape was removed, and the cassette apertures were covered with filter paper wetted with 600 μL 18 M Ohm deionized water. The cassette was then placed in a ZOOM® IPG Runner™ IEF tank surrounded by 600 mL deionized water and subjected to a stepwise IEF protocol consisting of 200 V for 20 min, followed by 450 V for 15 min, 750 V for 15 min, and in some cases, 2000 V for 30 min at ≤2 mA or ≤2 W.
After IEF, the ZOOM® IPG strips were extracted from the cassette, trimmed on the ends up to the gel line, and placed horizontally in the long wells of 4–12% polyacrylamide mini Bis-Tris ZOOM® gels (Invitrogen). Strips were cemented in place with molten 4% agarose in TBE buffer, which was allowed to cool and harden. The adjacent well was loaded with 10 μl 0.5–12 Kb DNA ladder (Invitrogen, Cat. No. 15615-016), diluted 1:1 with 10 μl 5× nucleic acid loading buffer (Bio-Rad Laboratories) for second dimension electrophoresis in cold TBE buffer. The samples were then electrophoresed at 200 V for 3 h and stained with 5 μL 1% EtBr in 100 mL TBE for 10 min on an orbital shaker, followed by a 10-min wash in TBE and all subsequent transfer, processing, and imaging steps, as described previously for the 1-D CARS membranes.
The fact that 1-D CARS gel smears or “streaks” are fairly reproducible is illustrated by the EtBr-stained polyacrylamide gel image in the upper right-hand corner of Figure 1, which shows eight nearly identical lanes of smeared DNA, resulting from the partial hybridization and concatenation of the random sequence 60mer library. This image essentially represents the appearance and uniformity of 1-D CARS before analyte binding.
To provide an initial test for the CARS on a membrane surface, EtBr-stained 1-D CARS DNA was electro-transferred to nitrocellulose and nylon membranes and baked in a UV oven. The preliminary results (not shown) revealed that nitrocellulose gave a much better-defined 1-D CARS pattern than nylon, which appeared diffuse and lacked any discernable detail. Therefore, nitrocellulose was used in subsequent experiments.
Figure 2 illustrates results of an experiment in which two different proteins (bovine insulin and a HRP-labeled goat IgG antibody conjugate) interacted with several 1-D CARS to give perceptible differences in the visual appearance caused by exposure to these two different proteins. Figure 2 shows further enhancement of these differences by means of ROI (yellow-boxed region) profiling and a 3-D wire-frame representation of the fluorescence image. Averaging over the entire 1-D arrays or CARS lanes leads to a 2-D profile, which is “smoother” but still portrays differences in the interactions with these two different proteins. The primary difference between treatment groups in many of the analyzed images was fluorescence intensity. However, as the 2-D profile and the 3-D wire-frame representation demonstrate, clear textural differences could also be detected between the two treatments in Figure 2 and other figures. These textural differences appear to be more consistent or similar within rather than between the treatments and illustrate the sort of complex patterns that would be encountered by any algorithm designed to recognize a particular analyte's presence or “signature.”
Figure 3 illustrates results of a 1-D CARS fluorescent nucleic acid-binding experiment. In this case, two different DNA aptamers of known sequence2 were 5′-labeled with fluorescein and allowed to interact separately with several 1-D CARS lanes. After washing and image analysis, it appeared that location-dependent differences in aptamer binding or partial Watson-Crick hybridization in different regions of the lanes had occurred. This is revealed in visual assessment of the arrays and profiling of the ROIs across the various lanes, as shown in Figure 3 (insets). Again, the differences are much greater between the two treatments groups than within a given aptamer-treated group, as seen in the ROI profiles and 3-D wire-frame plot.
The next experiment compared the effects of binding another fluorophore (Alexa Fluor 647)-labeled DNA aptamer with that of a multi-colored and fluorescent set of molecular weight standards (Invitrogen, Cat. No. 57318) on a series of 1-D CARS lanes. These results are presented in Figure 4 and again demonstrate clear 2-D average profile changes between the two different treatment groups with much less difference among individual CARS lanes within either treatment group. Again, 3-D plots serve to enhance fluorescence intensity and textural differences or similarities, as illustrated in Figure 4. It is also interesting to note that the appearance and texture of the average 2-D profiles and 3-D plots in Figure 4 differ from those shown in Figures 2 and and3,3, thus further suggesting that unique, albeit complex, signature patterns exist for each type of analyte-CARS interaction.
In an attempt to increase detection of binding events further on 1-D CARS, a fluorescence polarization (FP) imaging approach was attempted, which was similar to that demonstrated by McCauley et al.15 for individual chip-immobilized fluorophore end-labeled aptamers. It was thought that perhaps sections of the 1-D array, which bound analytes, would slow the rotation of aptamers in that region and increase FP locally, thereby increasing light throughput at that point in the image, as light passed through a polarizing filter to a camera or photodetector. Unfortunately, as Figure 5 shows, it was not possible to perform the FP imaging experiment with this strategy, as when fluorescein is conjugated to the 5′ ends of DNA oligonucleotides in the random sequence 60mer library, it severely inhibits higher molecular weight DNA concatamer formation, thereby effectively preventing 1-D CARS formation.
Figure 6 represents the typical appearance of the EtBr-stained random sequence 60mer DNA library after IEF with stoppage at 750 V (upper) and 2000 V (lower) in 1% ampholytes (using a linear pH 3–10 IEF strip) and 2-D electrophoresis, with (left) and without (right) an orange emission filter and UV transillumination. Completion of the second electrophoresis dimension was expected to require only 45 min but oddly, required approximately 3 h for all attempts at a constant potential of 200 V in TBE, as assessed by migration of the bromphenol blue dye front to the bottom of the gel. As Figure 6 clearly illustrates, the application of 750 V moved much of the random DNA library as far as the mid-point of the IEF strip but yielded a more rectangular, albeit diffuse, distribution pattern (boxed in Fig. 6, upper panels). The rectangular pattern would likely be more useful as a constant and more predictable geometry for 2-D CARS-based analyte detection. The further application of 2000 V resulted in pushing the dispersed random sequence DNA further to the right (toward pH 3.0 or +pole) and arcing DNA near the top of the gel while producing an interesting aurora borealis pattern. These patterns were fairly reproducible in other 2-D CARS library electrophoresis experiments (data not shown). Therefore, IEF was capped at 750 V in subsequent attempts to produce a more rectangular area for detection of analytes on nitrocellulose membrane-bound 2-D CARS. It is difficult to state with certainty which bands of the DNA ladder represent which lengths of DNA because of the compression of the ladder, but it appears that the top band near the middle of the gel may represent migration of 12 Kb DNA, and the lowest band may represent migration of a 500-bp DNA molecule (Fig. 6).
The difficulty in producing highly reproducible 2-D CARS has impeded progress in this area. However, as Figure 7 illustrates, there is reason to believe that 2-D CARS on membranes may be possible if the density of random DNA concatamers can be increased, as it appears that some regions unstained previously (circled) emerge when a 5′-fluoresceinated aptamer probe is applied in the rectangular 2-D CARS region generated by stopping IEF after the 750-V step. Figure 7B demonstrates that much of the initially orange EtBr-stained DNA or intercalated EtBr itself can be washed away (when compared with Fig. 7A), but some DNA streaks still remain, and some new green fluorescent regions (circled in Fig. 7B) appear to emerge when a 5′-fluoresceinated DNA aptamer is applied, and the membrane is washed extensively.
The CARS concept is potentially quite powerful, versatile, facile, and inexpensive for detection and identification of virtually any analyte, thereby possibly making it a universal sensor interface technology. However, the system clearly still has several key technical difficulties to overcome to fulfill its potential. These hurdles include dealing with complex visual patterns, signal amplification, reproducibility within and between experiments, and even selection of the optimal mode of detection. Electrical detection, as used with some DNA hybridization arrays having discrete and insulated loci of different known sequences, is probably not a viable option for analyte detection at present, as the array is rather dense, complex, and contiguous (i.e., no discrete and insulated detection loci as with conventional biochips). Visual detection can provide very fine spatial resolution and detail versus electrical detection, and greater resolution equates to better discriminatory capability.
Rapid classification of complex visual or even “subvisual” (subconscious) patterns is actually an old and well-studied problem, especially in the area of automated cytopathology.16–20 In digital image analysis of stained cell images, Markovian and other texture analysis algorithms have long been used to develop Bayesian boundaries or rules for accurate classification of normal versus altered cells that appear identical to the naked eye.16–20 The area of 2-D and even 3-D visual pattern recognition is evolving toward faster classification algorithms21 and shows great potential for being applied to the CARS sensor concept once other technical problems are resolved.
In the present work, the issues of transfer to and immobilization of CARS onto membranes were largely resolved. It was determined that nitrocellulose was superior to the more sturdy nylon as a membrane material in terms of resolution and detail of the transferred CARS DNA. Although several methods of CARS fluorescent staining, processing, and visual detection were attempted, including FP imaging using 5′-fluorescein labeling15 of the CARS DNA library (which inhibited DNA concatamer formation), it was observed that simple EtBr staining of CARS in a polyacrylamide gel prior to membrane transfer yielded the best results. One can only speculate about the mechanism of visual detection of analyte-induced changes in the arrays at this point. However, it seems reasonable to postulate that some of the intercalated EtBr in double-stranded or stem regions of the array becomes displaced or somewhat quenched by analyte binding,22,23 leading to subtle changes in the visible smear patterns. The FP approach15 to CARS imaging has not been abandoned completely but will require a different strategy for fluorescently labeling CARS, which does not disintegrate the concatamers, as individual 5′ oligonucleotide labeling with fluorescein appears to have caused.
Finally, efforts were made in this project to develop and characterize the first ever 2-D random sequence DNA library gels and subsequent 2-D membrane-bound CARS. Little literature exists about 2-D electrophoresis of DNA libraries, and that which does exist focuses on separation of DNA replication intermediates and restriction endonuclease fragments derived from digestion of bacterial genomes.24–26 One of the primary findings in this study was that 2-D CARS electrophoresis (based on concatamer charge and length differences) is feasible. However, 2-D separation patterns appear to be largely influenced by IEF conditions and possibly other parameters such as temperature, ionic strength, and minor batch-to-batch variations in the random DNA (CARS library) synthesis. Still, it seems possible that a fairly rectangular and potentially reproducible 2-D CARS detection area can be developed for future use if chemical and physical parameters, including voltages and separation times, are tightly regulated. Even if 2-D CARS ultimately proves too complicated for commercialization, the 1-D CARS arrays are self-assembling and therefore, quite simple to mass produce. In addition, the 1-D CARS format from individual gel lanes appears to lend itself to a simple test strip embodiment, which could be coupled to an optical reader or scanner device.
Many in the molecular biology and biotechnology communities who read this article may dismiss the data as laden with artifacts or consider the CARS concept as being too simplistic for further consideration as a biosensor interface. However, the concept is built on solid tenets, such as the facts that random DNA sequence libraries contain numerous potential aptamers, or aptamer families that have the ability to bind similar or diverse analytes with different individual affinities. In addition, these potential aptamers can partially hybridize by Watson-Crick base pairing, concatenate into longer chains of DNA, and be separated by electrophoresis into size- and charge-ordered arrays. In essence then, the CARS concept satisfies the criteria for being a biochip that can bind analytes in specific locations in the array. CARS technology clearly has some major disadvantages, such as the fact that the DNA sequences within any specific binding region are not easily known. However, CARS has a distinct advantage in that it possesses a much greater diversity of “receptors” (aptamers) than any conventional biochip yet produced and therefore, may bind and detect a much larger diversity of ligands. As the CARS can take on so many different appearances as a result of interacting with different target molecules, one major key to CARS future success or abandonment as a technology will probably lie in computer-aided pattern recognition of the many complex patterns that can emerge.21 In conclusion, although many obstacles still exist for practical application of CARS technology, strides were made toward further proof-of-concept of this potentially “universal biosensor” interface technology.
This work was supported by a Small Business Innovative Research (SBIR) contract (contract no. FA8650-09-M-5422) from the U.S. Air Force.