Printed Colorimetric Arrays for the Identification and Quantification of

Subscriber access provided by Technical University of Munich University Library

Article

Printed Colorimetric Arrays for the Identification and Quantification of Acids and Bases Michael J Kangas, Rachel Lukowicz, Jordyn Atwater, Armando Pliego, Yasmine Al-Shdifat, Shana Havenridge, Raychelle Burks, Billy Garver, Miles Mayer, and Andrea E Holmes Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.8b02432 • Publication Date (Web): 20 Jul 2018 Downloaded from http://pubs.acs.org on July 21, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Printed Colorimetric Arrays for the Identification and Quantification of Acids and Bases Michael J. Kangas*, Rachel Lukowicz, Jordyn Atwater, Armando Pliego, Yasmine Al-Shdifat, Shana Havenridge, Raychelle Burks, Billy Garver, Miles Mayer, Andrea E. Holmes

Doane University, Department of Chemistry, Crete, Nebraska 68333

ABSTRACT: Solid supported colorimetric sensing arrays have the advantage of portability and ease of use when deployed in the field, such as crime scenes, disaster zones, or in war zones, but many sensor arrays require complex fabrication methods. Here, we report a practical method for the fabrication of 4x4 colorimetric sensor arrays using a commercially available inkjet printer, which are printed on nylon membranes. In order to test the efficacy of the printed arrays, they were exposed to 43 analytes at concentrations ranging from 0.001 M to 3.0 M for a total of 559 samples of inorganic and organic acids or bases including hydrochloric, acetic, phthalic, malonic, picric, and trifluoroacetic acid, ammonium hydroxide, sodium hydroxide, lysine, and water as the control. Colorimetric data from the imaged arrays was analyzed with linear discriminant analysis and k-nearest neighbors to determine the analyte and concentration with ~88-90% accuracy. Overall, the arrays have impressive analytical power to identify a variety of analytes at different concentrations, while being simple to fabricate.

Keywords: colorimetric arrays, linear discriminate analysis, k-nearest neighbors, ink-jet printing, nylon membrane, warfare, portability, in field testing, red green blue (RGB), analyte concentration, clustering, analyte identification.

INTRODUCTION Colorimetric sensor arrays have the ability to detect a variety of different analytes, in liquid and gas phases, including ions, acids and bases, metal nanoparticles, explosives, pesticides, warfare agents, drugs, various organic compounds, complex mixtures, such as coffee, beer, and soft drinks, and even biological molecules, such as steroids and proteins.1–11 Colorimetric arrays are typically composed of multiple colored dyes arranged in a two-dimensional grid, changing color upon interaction with specific analytes. The pattern of color changes can be used to analyze and identify the substance in question. The red, green, blue (RGB) color pattern recognition is based on the combined response from numerous sensors.5,12,13 Colorimetric arrays are available in liquid and solid forms. While colorimetric arrays in solution allow for bulk testing of analytes, they have the disadvantage of not being easily deployable in tactical situations, such as at a crime scene or in a war zone where field-ready tests kits are critical for presumptive testing.11 Thus, there is an important need to transform the liquid colorimetric arrays into a solid form for greater portability. This can be done by depositing selected sensors on a paper like substrate. Various printing methods exist, such as commercially available high-throughput micro-arrayers using piezoelectric print heads and non-contact printing from Arrayjet or Arrayit using high-throughput contact printing technology. While these printing methods have been employed successfully mostly for biological applications, such as antibody printing, they require expensive custom made printers and highly skilled operators. Printing chemical detection arrays with these types of high-end printers prohibits the quick and inexpensive printing of small arrays that can be custom made depending upon the application they are used for. Smaller research-based inkjet printing of arrays have been reported.10,14 We have developed a method that can be easily employed in any lab by formulating sensor solutions for use

ACS Paragon Plus Environment

Analytical Chemistry 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

with most desktop inkjet printers and printed on various surfaces, such a paper, nylon, polyvinyldifluoride (PVDF), glass membranes, Teflon, polyethersulfone (PES), etc. Figure 1 shows our overall array fabrication scheme.

Figure 1: The sensors are dissolved in a print solution and tested in a 96-well plate to determine sensor performance. Select sensors are printed and further tested against specific analytes. The sensors are scanned before and after analyte exposure, with RGB values collected from array images. Linear discriminant analysis (LDA) and k-nearest neighbors (KNN) are used to both in the classification, identification, and quantitation of analytes, as well as to probe the selectivity of each sensor.

MATERIALS AND METHODS Analyte Preparation All reagents were purchased from various chemical supply companies at technical grade or better and were used as received without further purification. Acetic acid (0.1 – 3.0 M), ammonia (0.1 – 3.0 M), picric acid (0.001 - 0.1 M), trifluoroacetic acid (0.001 - 3.0 M), hydrochloric acid (0.1 - 3.0 M), sodium hydroxide (0.1 - 3.3 M) solutions were prepared by diluting concentrated reagent solutions with milli-Q water (18 MΩ-cm). Solutions of malonic acid (0.001 – 3.0 M), phthalic acid (0.001 - 0.01 M), lysine (HCl salt, 0.1-2.0 M) were prepared by dissolving appropriate amount of the analytes in milli-Q water. Sensor formulation All reagents were purchased from various chemical supply companies at technical grade or better and were used as received without further purification. Saturated 1% weight-by-weight solutions of Congo red (CR), erythrosin B (EB), universal indicator (UI),12 and bromophenol blue (BB) were prepared by dissolving into aliquots of print solution. This print solution is a mixture of acetate buffer (0.1 M, pH 5), ethylene glycol, triethylene glycol monobutyl ether, and glycerol in a ratio of 14:1.6:1:3.2. The sensor solutions were sonicated for 1 hour in a bath sonicator (30 ºC), followed by 5 minutes mixing with a probe sonicator, and then vacuum filtered through Whatman #1 filter paper and Whatman nylon filter membranes (0.2 µm). Printing of colorimetric sensor arrays A 4x4 sensor array template was designed using Microsoft PowerPoint, which consisted of columns of black, cyan, yellow, and magenta squares (~5mm). A Canon MG5520 inkjet printer was used for all arrays fabricated in this study, but subsequent tests with Canon MG7720 and Canon TS6020 printers yielded similar quality arrays. Empty ink cartridges from InkOwl were filled with sensor solutions or print solution using a syringe with a syringe filter (0.8 µm) and placed in the printer. Both black cartridges were filled with the same solution. Before printing arrays, the print head was cleaned by printing with cartridges filled with the print solution. Arrays were then printed on plain copy paper using the high quality setting available through the PowerPoint print window until the deposition was sufficiently consistent. Consistency was judged by analyzing RGB values of printed arrays. A relative standard deviation of less than 5% for each color channel of each sensor was the “sufficiently consistent” benchmark. Once the sensor arrays were consistently printed, the print surface was switched to nylon membranes (Amershan Hybond - N+, GE Healthcare). Arrays were stored in the dark prior to use to prevent photo-bleaching. After the sensor arrays were printed, the print head was flushed by printing with cartridges filled containing print solution on to plain copy paper until no color was observed on the pages.


Page 2 of 12

Page 3 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Analyte Testing Control scans of the arrays were collected before application of the arrays. Arrays were then briefly dipped for about 1 second face down into a petri dish of the analyte solution. Initial attempts to apply the analytes by dipping the array on its edge or applying drops of the analyte resulted in smearing of the sensors. After removing the sensor arrays from the analyte, the arrays were allowed to dry for 1 minute before scanning (referred to as “wet scans”). These same arrays were then allowed to dry, and after 10 minutes were scanned again (referred to as “dry scans”). The data of the wet arrays will be the only data presented in this manuscript as there was no significant difference between wet and dry RGB data for the analyte set as judged by KNN and LDA. Image Collection and Processing All array images (24-bit color, 400 dpi) were collected using Canon MG5520 printer/scanner and are hereafter designated “scans”. Evaluation of the data by KNN and LDA, these two scanners were found to give equivalent results were used in the data collection. The images were analyzed with ImageJ15 and the extraction of mean RGB values was automated with a macro.16 No attempts were made to correct for image-to-image variation by subtracting a control row, as our previous work showed such a correction to be unnecessary.12 The RGB dataset is provided in the supplemental information as SI.1. Chemometric Data Analysis All statistical analysis was performed using the statistical programming language R (version 3.4.3).17 LDA was performed using the MASS library18 and the classification ability of the LDA model was evaluated using all but one cross validation. The concentrations of each analyte were treated as a separate class, and for classification, the prior probabilities of the classes were equal. KNN based on Euclidean distances was performed for k = 1, 3, 5, 7, 9, and 11 using the class package.18 In this case, the unknown was classified based on the most common class of the k-amount of its nearest neighbors. RESULTS Our previous work with bulk liquid arrays paired with nine different chemometric methods showed that LDA and KNN offered high classification accuracy and simplicity of execution for similar analytes as used here.12 Therefore, we chose LDA and KNN for studying the printed colorimetric arrays using analyte concentrations between 0.1 M and 3.0 M for hydrochloric, acetic, phthalic, malonic, picric, and trifluoroacetic acid, ammonium hydroxide, sodium hydroxide, lysine, and water as the control. Printed colorimetric arrays Figure 2 shows examples of the printed 4x 4 arrays that were exposed to 0.01 M acetic acid, 0.01 M trifluoroacetic acid (TFA), 0.1 M TFA, 0.1 M of lysine, 0.1 M of sodium hydroxide (NaOH), and 0.5 M of ammonium hydroxide (NH4OH). The printing of these arrays was described in section 3.3. None of the sensor squares leached or bled after exposure to the analyte solutions. When 0.01 M acetic acid and 0.01 M trifluoroacetic acid are compared, it is obvious that visual differences exist between the individual sensors even though the acids are at the same concentrations. A similar color pattern like 0.01 M TFA was also observed for 0.1 M samples of hydrochloric acid, indicating that acid strength determines sensor response, as expected for this array contains pH indicators. Differences in color changes are more subtle between 0.1 M NaOH and 0.5 M of NH4OH. A more dramatic change for acids is explained by the sensors used because EB has a color changing range of pH 2.2 - 3.6, CR between pH 3.0 - 5.0, and BB between pH 6 - 7.6.19 UI is composed as previously described,12 being a mix of methyl red, methyl orange, bromothymol blue, and phenolphthalein and changes above pH 7. Though the sensor array changes elicited by NaOH and NH4OH were not visually obvious, the changes in RGB values can still be discriminated via RGB data analysis.



Figure 2: Representative examples of 6 printed colorimetric arrays after they were exposed to 0.01M Acetic Acid, 0.01M Trifluoroacetic Acid, 0.1 M HCl, 0.1 M of lysine, 0.1 M of NaOH, and 0.5 M of Ammonium Hydroxide (NH4OH). Visual colorimetric differences are obvious between 0.01M Acetic Acid and 0.01 M Trifluoroacetic Acid where each sensor has a distinctly different color. Differences in color changes are more subtle between 0.1 M NaOH and 0.5 M of NH4OH and may not be visually obvious but the changes in RGB values can be discriminated by ImageJ. Each sensor square is 5x5mm.

Linear Discriminant Analysis LDA results are shown in Table 1. It demonstrates that out of 559 samples, 504 samples were correctly classified by identity and concentration. 55 samples were misclassified, giving an overall accuracy of 90.2%. The misclassified samples were predominantly NH4OH samples, showing a decreasing trend of accuracy from 76.9% to 0% as the concentration increases 0.1 M to 3.0 M. In comparison to sodium hydroxide, this is a significantly less accuracy than with sodium hydroxide at the same concentrations. This indicates that stronger bases such as NaOH are more easily discerned with LDA than weaker bases when colorimetric changes are not distinguishable among different concentrations of NH4OH samples. Seven out of the 55 misclassified samples were the wrong analyte, while 48 samples were classifying the sample correctly but not the concentration. Therefore, the accuracy of 98.7% is very high for analyte identification.


Page 4 of 12

Page 5 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table 1: LDA classification results of various acidic and basic analytes on printed arrays. Analyte

Correct

Incorrect

Correct %

Misclassified

Acetic Acid 0.1 M

13

0

100.0

-

Acetic Acid 0.5 M

11

2

84.6

2-1 M acetic

Acetic Acid 1 M

10

3

76.9

3- 0.5 M acetic

Acetic Acid 2 M

13

0

100.0

-

Acetic Acid 3 M

13

0

100.0

-

Ammonium Hydroxide 0.1 M

10

3

76.9

3- 0.5 M ammonia


3

10

23.1

3- 0.1 M ammonia, 7- 1 M ammonia

Ammonium Hydroxide 1 M

4

9

30.8

2-0.1 M ammonia, 7- 0.5 M ammonia


0

13

0.0

1-0.1 M ammonia, 5- 0.5 M ammonia, 7- 3 M ammonia


13

0

100.0

-

HCl 0.1 M

13

0

100.0

-

HCl 0.5 M

13

0

100.0

-

HCL 1 M

13

0

100.0

-

HCl 2 M

13

0

100.0

-

HCl 3 M

13

0

100.0

-

Lysine 0.1 M

9

4

69.2

3-0.5 M ammonia, 1-0.001 M phthalic

Lysine 0.5 M

12

1

92.3

1- 1 M lysine

Lysine 1 M

12

1

92.3

1- 2 M lysine

Lysine 2 M

12

1

92.3

1- 1 M lysine

Malonic 0.001 M

13

0

100.0

-

Malonic 0.01 M

13

0

100.0

-

Malonic 0.1 M

13

0

100.0

-

Malonic 0.5 M

13

0

100.0

-

Malonic 1 M

13

0

100.0

-

Malonic 2 M

13

0

100.0

-

Malonic 3 M

13

0

100.0

-

NaOH 0.1 M

13

0

100.0

-

NaOH 0.5 M

13

0

100.0

-

NaOH 1 M

11

2

84.6

2- 2 M NaOH

NaOH 2 M

11

2

84.6

1- 1 M NaOH, 1-3M NaOH

NaOH 3 M

12

1

92.3

1- 1 M NaOH

Acid

10

3

76.9

2- 0.001 M malonic, 1- 0.001 M TFA

Phthalic Acid 0.01 M

13

0

100.0

-

Picric Acid 0.001 M

13

0

100.0

-

Picric Acid 0.01 M

13

0

100.0

-

Phthalic 0.001 M



Page 6 of 12

TFA 0.001 M

13

0

100.0

-

TFA 0.01 M

13

0

100.0

-

TFA 0.1 M

13

0

100.0

-

TFA 0.5 M

13

0

100.0

-

TFA 1 M

13

0

100.0

-

TFA 2 M

13

0

100.0

-

TFA 3 M

13

0

100.0

-

Water

13

0

100.0

-

Overall

504

55

90.2

Figure 3 demonstrates the scores plot of LDA of all 559 samples. Some of the analytes in this plot were easily distinguishable because they formed distinct clusters that do not overlap with other analytes, such as 0.1 M picric acid, 3.0 M HCl, and 0.1 M Malonic Acid. However, due to the fact that many analytes and concentrations were included in the plot, it was visually difficult to recognize the individual clusters. Thus, it was necessary to separate the analytes into different LDA plots to demonstrate the actual clustering trends of the analytes and the concentrations. As each analyte data set is linearly independent, the LDA analysis was not affected by this separation.

Figure 3: LDA plot of Principal Component 1 and 2, demonstrating the clustering of all tested analytes between 0.1 M and 3.0 M. The plot shows distinct clustering of list all the samples that separated well. Some of the analytes did not form distinct separation and experienced some overlap.

This is seen in Figure 4. Figure 4 A shows how water and basic samples of sodium hydroxide, ammonium hydroxide, and lysine are located in the same quadrant of the LDA plot, but are distinctly separated from each other. It is also clearly shown how the different concentration color gradients are separated in lysine and sodium hydroxide with the lighter shades of color representing the lower concentrations and the darker shades of color representing the higher concentrations. The clustering separation and visualization of concentrations are even more apparent in Figure 4B, where the acidic samples malonic acid, HCl, and TFA clearly form individual clusters with obvious concentration trends from low to high depending on the shade of color. Figure 4C shows a clear separation of picric acid from the other two acids, acetic and phthalic acid. This was also seen in Figure 3. All plots in Figure 4 demonstrate that the analytes were clearly separated from water that served as the control in all studies.


Page 7 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4 A-C: LDA scores plots highlighting individual analytes in three different plots to visualize the individual analyte clustering.

K-Nearest Neighbor classification Table 2 shows the KNN classification results. K was varied and the Euclidean distance was calculated for the KNN analysis. Overall, KNN was used to correctly identify 88.0% of samples for K = 1, decreasing to 85.0.0% for K=11. This means that the best accuracy was calculated when the analyte was assigned to the class of its single nearest neighbor. The results indicate that KNN was a very robust method for sample classification for this sample size since the least accurate run with the largest number of neighbors (K = 11) was able to classify samples correctly 85.0% of the time. Overall KNN



Page 8 of 12

(k = 1) had 13 misclassifications on the analyte identity. However, given the very large sample size, it is apparent that these types of misclassifications do not occur very often, such as in only one sample out of a large sample pool like the one studied here.

Table 2: KNN classification results of various acidic and basic analytes on printed arrays. Analyte

Correct

Incorrect

Correct %

Misclassified

Acetic Acid 0.1 M

12

1

92.3

1– 2M lysine

Acetic Acid 0.5 M

13

0

100

-

Acetic Acid 1 M

11

2

84.6

1– 0.5 M acetic, 1- 0.001 M TFA

Acetic Acid 2 M

13

0

100

-

Acetic Acid 3 M

13

0

100

-


8

5

61.5

4- 1 M ammonia, 1- 2 M ammonia


4

9

30.8

8- 1 M ammonia, 1-2 M ammonia


1

12

7.7

4- 0.1 M ammonia, 7-0.5 M ammonia, 1-2 M ammonia


2

11

15.4

4- 0.5 M ammonia, 1-1 ammonia, 6-3 M ammonia


7

6

53.8

6- 2 M ammonia

HCl 0.1 M

13

0

100

-

HCl 0.5 M

12

1

92.3

1- 3 M HCl

HCL 1 M

13

0

100

-

HCl 2 M

13

0

100

-

HCl 3 M

13

0

100

-

Lysine 0.1 M

10

3

76.9

2- 0.5 M lysine, 1-0.001 M malonic

Lysine 0.5 M

11

2

84.6

2- 0.1 M lysine

Lysine 1 M

10

3

76.9

2- 0.5 M lysine, 1 water

Lysine 2 M

12

1

92.3

1 - 0.5 M acetic acid

Malonic 0.001 M

13

0

100

-

Malonic 0.01 M

12

1

92.3

1- water

Malonic 0.1 M

13

0

100

-

Malonic 0.5 M

12

1

92.3

1- 1 M malonic

Malonic 1 M

13

0

100

-

Malonic 2 M

12

1

92.3

1- 1 M malonic

Malonic 3 M

13

0

100

-

NaOH 0.1 M

13

0

100

-

NaOH 0.5 M

13

0

100

-

NaOH 1 M

12

1

92.3

1 - 0.1 M NaOH

NaOH 2 M

11

2

84.6

1 - 0.5 M ammonia, 1 - 0.1 M lysine

NaOH 3 M

13

0

100

-


10

3

76.9

2 - 0.001 M malonic, 1 water


13

0

100

-

Picric Acid 0.001 M

13

0

100

-

Picric Acid 0.01 M

13

0

100

-


M

Page 9 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


TFA 0.001 M

13

0

100

-

TFA 0.01 M

13

0

100

-

TFA 0.1 M

11

2

84.6

2-3 M malonic

TFA 0.5 M

13

0

100

-

TFA 1 M

13

0

100

-

TFA 2 M

13

0

100

-

TFA 3 M

13

0

100

-

Water

13

0

100

-

Overall

492

67

88.0

Increasing the k value has previously been shown to influence method performance and accuracy, but the relationship between k and performance can vary.20 In our study, increasing K = 1 to K = 11 resulted in lower classification accuracy (Table 3). This indicates that K = 1 was the optimal parameter. This coincides with our previous results that the increase of the K parameter does not increase the accuracy.12,21 Comparing the two classification methods, KNN (K = 1) is slightly inferior to LDA with 88.0% accuracy and 90.2% accuracy, respectively. While LDA utilizes input variables with optimal weights to separate the group means and KNN gives equal significance to all of the variables, both methods still give similar results. Alternative KNN algorithms do apply various transformations to the dataset to optimize the accuracy of KNN. However, these methods were not pursued in this study.22

Table 3: Effect of k on KNN performance K

Correct

Incorrect

Total

% Correct

1

492

67

559

88.0

3

494

65

559

88.4

5

493

66

559

88.2

7

492

67

559

88.0

9

481

78

559

86.0

11

475

84

559

85.0

DISCUSSION The printed colorimetric arrays demonstrated impressive analytical power to identify and quantify 559 analyte samples. These types of solid arrays have several advantages over liquid arrays. First, the newly discovered printing procedures allow for several hundred arrays to be easily printed in a short of amount of time, similar to the time it takes to print a simple 4x4 array graphic on an inkjet printer. The technology lends itself to printing smaller arrays than the ones presented here. This may be advantageous for smaller consumption of anlaytes. This allows for the quick accessibility of many assays in order to perform several replications of experiments to obtain meaningful experimental statistics. The newly developed printing formulation of these sensors opens up the possibility for any type of sensor that works in solution to also work on a surface. Furthermore, the ability of the printing formulations to attach non-covalently to the nylon surface allows for superior analyte testing without the sensors running off of the substrate. This also enables attachment of sensors without modifications of the surfaces for covalent bonding. The described method enables the sensors to be visualized with vibrant colors before and after analyte application and leads to RGB data that is well defined for chemometric analysis. The second advantage of the solid over of liquid arrays is that the user is not exposed to any liquid chemicals. Moreover, liquid chemicals may have safety requirements when being disposed. The printed arrays can simply be tested and thrown away without causing any toxic effects to the user or the environment.



LDA / KNN results Overall, both LDA and KNN were able to classify ~90% of the samples correctly. Similar, to previous studies,12,21 LDA performed slightly better than KNN with accuracies of 90.2% and 88.0%. The superior performance of LDA could be due to differences in the algorithm where LDA attempts to use the multivariate dataset to separate the group means, while KNN is a non-parametric method.23 However, LDA requires a larger data set which could be difficult for rare or expensive samples. Also previous results have shown that LDA struggles with outliers than other methods.12,21 Thus, the simplicity of KNN may make it attractive for some applications. Similar analysis was performed with arrays that were allowed to fully dry (~10 minutes) before scanning, based on the idea that arrays would be more uniform and would result in better classification. However, the classification accuracy between wet and dry arrays using LDA was very similar at 90.2% and 90.1%, respectively (data not shown). Similar results were also obtained using KNN with 88.0% correct with the wet samples performing slightly better at 90.3% correct for the dry samples (data not shown). For both data sets and chemometric methods, the majority of classification errors were the correct analyte, but the wrong concentration. Likewise, for both the dry and wet datasets the ammonium hydroxide samples had the lowest classification accuracy of the analytes. Based on the similarity of the classification results and for the efficiency of imaging right away, the wet samples were used for the final analysis. Comparison of results to liquid arrays The screening of the analytes in 96-well plates before printing provides a good preliminary indication which sensor would perform optimally in printed form. In comparison to our previous work in identifying similar analytes with an eight sensor array in 96-well plates, the accuracy of the printed arrays was slightly lower.21 This is probably due to the fact that the eight sensor array was decreased to a four sensor array containing one sensor that reacted particularly strong with ammonia.

Conclusion The portability and ease of analysis of colorimetric arrays make them ideal for deployment in a variety of areas. While the presented technology requires flatbed scanners, which is not particularly portable for field deployment, the arrays could be easily photographed with a cell phone and similar RGB analysis can be performed on the image. Arrays like these would be suitable for testing suspected explosives and warfare agents in combat situations. Additional applications for these versatile sensors include the analysis of street drugs by police officers or the detection of potentially dangerous compounds by first responders in emergency situations. The expedited process would provide these individuals with an effective method for testing threatening chemicals when their safety may rely on the speed of detection. Printed colorimetric arrays will likely continue to advance in the future, allowing for more rapid and accurate analysis in field and laboratory applications. Furthermore, continued improvements to these sensors will allow for more safe and effective testing in hazardous environments. Other improvements that are needed include the analysis of analyte mixtures. Finally, it is important to note that these printed colorimetric arrays may provide a low cost alternative to other available sensors, making them a useful primary or substitute detection method when portable means are necessary.

ASSOCIATED CONTENT Supporting Information The Supporting Information is available free of charge on the ACS Publications website. RGB Dataset in CSV format (docx)

AUTHOR INFORMATION Corresponding Author * [email protected]

Author Contributions The manuscript was written through contributions of all authors.


Page 10 of 12

Page 11 of 12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


ACKNOWLEDGMENT Funding was provided by NSF-IRES Grant Number 1459838, NSF-LSAMP, Grant Number HRD-1619654, 2016-2021; HRD1102461, 2011-2017, the National Institute for General Medical Science (NIGMS) (5P20GM103427), the Camille and Henry Dreyfus Foundation, and SBIR Phase II - Chemical Biological Radiological Nuclear and Explosives (CBRNE) Reconnaissance Sampling Kit (W911SR-16-C-0051)

REFERENCES (1) Askim, J. R.; Mahmoudi, M.; Suslick, K. S. Optical Sensor Arrays for Chemical Sensing: The Optoelectronic Nose. Chem. Soc. Rev. 2013, 42 (22), 8649–8682. (2) Bang, J. H.; Lim, S. H.; Park, E.; Suslick, K. S. Chemically Responsive Nanoporous Pigments: Colorimetric Sensor Arrays and the Identification of Aliphatic Amines. Langmuir 2008, 24 (22), 13168–13172. (3) Batres, G.; Jones, T.; Johnke, H.; Wilson, M.; Holmes, A. E.; Sikich, S. Reactive Arrays of Colorimetric Sensors for Metabolite and Steroid Identification. J. Sens. Technol. 2014, 4 (1), 1–6. (4) Capitán-Vallvey, L. F.; López-Ruiz, N.; Martínez-Olmos, A.; Erena, M. M.; Palma, A. J. Recent Developments in Computer Vision-Based Analytical Chemistry: A Tutorial Review. Anal. Chim. Acta 2015, 899, 23–56. (5) Chulvi, K.; Gaviña, P.; Costero, A. M.; Gil, S.; Parra, M.; Gotor, R.; Royo, S.; Martínez-Máñez, R.; Sancenóna, F.; Vivancos, J.-L. Discrimination of Nerve Gases Mimics and Other Organophosphorous Derivatives in Gas Phase Using a Colorimetric Probe Array. Chem. Commun. 2012, 48 (81), 10105–10107. (6) Johnke, H.; Batres, G.; Wilson, M.; Holmes, A. E.; Sikich, S. Detecting Concentration of Analytes with DETECHIP: A Molecular Sensing Array. J. Sens. Technol. 2013, 3 (3), 94–99. (7) Kitamura, M.; Shabbir, S. H.; Anslyn, E. V. Guidelines for Pattern Recognition Using Differential Receptors and Indicator Displacement Assays. J. Org. Chem. 2009, 74 (12), 4479–4489. (8) Lim, S. H.; Feng, L.; Kemling, J. W.; Musto, C. J.; Suslick, K. S. An Optoelectronic Nose for the Detection of Toxic Gases. Nat. Chem. 2009, 1 (7), 562–567. (9) Mahmoudi, M.; Lohse, S. E.; Murphy, C. J.; Suslick, K. S. Identification of Nanoparticles with a Colorimetric Sensor Array. ACS Sens. 2016, 1 (1), 17–21. (10) Soga, T.; Jimbo, Y.; Suzuki, K.; Citterio, D. Inkjet-Printed Paper-Based Colorimetric Sensor Array for the Discrimination of Volatile Primary Amines. Anal. Chem. 2013, 85 (19), 8973–8978. (11) Kangas, M. J.; Burks, R. M.; Atwater, J.; Lukowicz, R. M.; Williams, P.; Holmes, A. E. Colorimetric Sensor Arrays for the Detection and Identification of Chemical Weapons and Explosives. Crit. Rev. Anal. Chem. 2017, 47 (2), 138–153. (12) Kangas, M. J.; Burks, R. M.; Atwater, J.; Lukowicz, R. M.; Garver, B.; Holmes, A. E. Comparative Chemometric Analysis for Classification of Acids and Bases via a Colorimetric Sensor Array. J. Chemom. 2018, in press. (13) Askim, J. R.; Suslick, K. S. Hand-Held Reader for Colorimetric Sensor Arrays. Anal. Chem. 2015, 87 (15), 7810–7816. (14) Komuro, N.; Takaki, S.; Suzuki, K.; Citterio, D. Inkjet Printed (Bio)Chemical Sensing Devices. Anal. Bioanal. Chem. 2013, 405 (17), 5785–5805. (15) Schneider, C. A.; Rasband, W. S.; Eliceiri, K. W. NIH Image to ImageJ: 25 Years of Image Analysis. Nat. Methods 2012, 9 (7), 671–675. (16) Soldat, D. J.; Barak, P.; Lepore, B. J. Microscale Colorimetric Analysis Using a Desktop Scanner and Automated Digital Image Analysis Douglas. J. Chem. Educ. 2009, 86 (5), 617–620. (17) R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2017. (18) Venables, W. N.; Ripley, B. D. Modern Applied Statistics with S, Fourth.; Springer: New York, 2002. (19) Sabnis, R. W. Handbook of Acid-Base Indicators; CRC Press: Boca Raton, Fl, 2008. (20) Ma, C.-M.; Yang, W.-S.; Cheng, B.-W. How the Parameters of K-Nearest Neighbor Algorithm Impact on the Best Classification Accuracy: In Case of Parkinson Dataset. J. Appl. Sci. 2014, 14 (2), 171–176. (21) Kangas, M. J.; Wilson, C. L.; Burks, R. M.; Atwater, J.; Lukowicz, R. M.; Garver, B.; Mayer, M.; Havenridge, S.; Holmes, A. E. An Improved Comparison of Chemometric Analyses for the Identification of Acids and Bases With Colorimetric Sensor Arrays. Int. J. Chem. 2018, 10 (2), 36–55. (22) Nigsch, F.; Bender, A.; van Buuren, B.; Tissen, J.; Nigsch, E.; Mitchell, J. B. O. Melting Point Prediction Employing KNearest Neighbor Algorithms and Genetic Parameter Optimization. J. Chem. Inf. Model. 2006, 46 (6), 2412–2422. (23) Haswell, S. J. Practical Guide to Chemometrics; Marcel Dekker, Inc.: New York, 1992.



For TOC only:


Page 12 of 12

Printed Colorimetric Arrays for the Identification and Quantification of

Recommend Documents