Macromolecular optical sensor arrays

Linda Mitchell,† Elizabeth J. New,†§ Clare S. Mahon*‡

†The University of Sydney, School of Chemistry, Sydney, NSW 2006, Australia

§The University of Sydney Nano Institute (Sydney Nano), The University of Sydney, NSW
2006, Australia

§Australian Research Council Centre of Excellence for Innovations in Peptide and Protein
Science, The University of Sydney, NSW 2006, Australia

‡Department of Chemistry, Durham University, South Road, Durham DH1 3LE, United
Kingdom

*E-mail: [email protected]

Abstract
Chemical sensors play an important role in our understanding of chemical and biological
systems, providing sensitive and rapid detection of a variety of substrates. Array-based
sensing approaches avoid the ongoing challenge of designing and synthesizing selective
receptors for particular analytes, a labor-intensive process that can frustrate the development
of sensors. Instead, cross-reactive sensor arrays utilize multiple sensing elements that interact
uniquely with each analyte and produce a distinct pattern of responses, enabling identification.
To date, there are a variety of strategies both to gain cross-reactivity and diversity of sensors
required for array-based sensing, and to broaden the scope of analytes for detection. Sensor
arrays constructed using macromolecular components, such as polymers and nanoparticles,
offer an attractive route to the discrimination of multiple, similar analytes, particularly within
the context of biological sensing, where recognition over large areas is often required. Here,
we focus on macromolecular sensing arrays underpinned by optical detection methods, which
can enable rapid, sensitive detection of a range of analytes. We discuss the current state-of-
the art and explore the challenges to be overcome in translating exciting scientific advances
to applications beyond the laboratory.

Keywords: polymer, sensor array, fluorophore, chromophore, discrimination

1. Introduction
Chemical sensors have enabled improvements in our understanding of many biological and
natural processes.1 There are a great number of sensors that have been developed to study
cells and tissues, to understand and monitor cellular function2 and to investigate metal and
redox homeostasis,3,4 drug uptake5 and disease progression.6 Furthermore, optical sensors
are also a promising strategy for the detection of contaminants in the environment.7 Natural
receptor species such as enzymes, antibodies and other proteins are often a source of
inspiration for chemists when designing sensors capable of analyte recognition, on account of
their unparalleled selectivity.8 Chemical sensors can also be categorized based on the sensing
approach (specific lock-and-key9 vs cross-reactive sensors),10 the sensor material (e.g. small
molecule vs macromolecule) and signal output (electrochemical, thermal and optical).10–14
Judicious selection of the respective sensor elements can address the demand to create new
sensors that enable rapid, robust, inexpensive and sensitive detection. The goal of this review
is to provide a general overview of macromolecular optical sensor arrays and the current
status of the field. Initially, we present a brief overview of each focus area, namely optical
sensing techniques, array-based sensing and polymeric sensing materials, followed by a
detailed discussion of the statistical techniques which underpin array-based sensing
methodologies. In subsequent sections we examine recent studies that utilize different
strategies for array design. Finally, we discuss the future directions of macromolecular optical
arrays in chemical sensing.

1.1 Optical sensing - A key function of a sensor, in addition to analyte recognition, is the
generation of a detectable signal. In some cases, the recognition unit is attached to the
reporter element (Figure 1a), while in other examples the sensing element can play the role
of both recognition and signaling (Figure 1b).15,16 For optical sensors, the reporter is either a
chromophore or fluorophore, and detection involves the absorption and emission of infrared,
visible or UV-light, and a change in either the intensity or wavelength of light upon analyte
binding. The attractiveness of optical sensors lies in both their sensitivity, and the broad range
of readily-accessible optical elements. Optical sensors using colorimetric17 or fluorometric
responses18 generally enable rapid detection and can provide considerable flexibility as an
analytical technique.10,19 In some cases, fluorescence based-measurements allow for
detection of analytes at the single-molecule level.20 These types of responses can be used in
a variety of systems, for example in indicator displacement assays21 or analyte-directed
aggregation techniques.22 The properties of these materials have been well-studied and are
tunable, and sensors can be selected based on desired properties such as wavelength of
absorbance/emission, quantum yield, ease of synthesis or capacity for selective recognition.23
Environmental sensitivity to aspects such as ionic strength, pH and polarity can cause
aggregation and subsequent enhancement or quenching of a response.24,25 Thus, the
versatility and broad scope of optical reporters makes them ideal candidates to produce a
detectable signal upon analyte binding.

1.2 Approaches to analyte detection - Sensors can be broadly defined as operating using
one of two primary mechanisms for analyte detection: reactivity- and reaction- based
sensing.26 Reactivity-based sensors, or chemodosimeters, rely on a selective chemical
reaction between the analyte and the sensing element to produce a dose-dependent
response.27 The process of analyte recognition is irreversible, with the sensor effectively
consumed during detection.
Recognition-based sensors harness supramolecular interactions between the analyte and
the recognition domain of the sensor to facilitate detection.6 Interactions between the sensor
and the analyte are reversible, and the sensitivity of the system is determined to a significant
extent by the Kd of the sensor-analyte interaction.
Indicator displacement assays (IDAs) are a popular route to analyte detection, presenting a
convenient method to obtain an optical read-out from a receptor.28,29 In an IDA, an indicator is
reversibly bound to a receptor, and competitive binding with an analyte will displace the
indicator from the host, leading to an optical change of the indicator (Figure 1c). IDAs have
been developed for a wide range of analytes, but it may be difficult to distinguish between high
concentrations of weakly displacing analytes and low concentrations of strongly displacing
analytes. IDAs lend themselves well to array-based sensing, since multiple receptor-indicator
pairs of varying affinities can be used to detect analytes.30–32 Multivariate analysis of a system
takes into account multiple responses: for example, multiple absorbance and fluorescence
wavelengths and fluorescence anisotropy. The combination of these responses may yield
more information about displacing analytes and allow them to be discriminated.
The macromolecular optical sensor arrays reported to date have largely focussed on
recognition-based sensing or indicator-displacement, and these systems will be the primary
focus of this review.
Figure 1. Schematic representation of a) a modular optical sensor containing a receptor with
recognition ability, attached to an optical reporting element, b) a sensing element capable of both
recognition and subsequent signaling, c) an indicator displacement assay (IDA), d) a ‘lock-and-key’
fluorescent sensor with selectivity for a particular analyte and e) an array of cross-reactive fluorescent
sensors which produces a fingerprint response for each analyte.

1.3 Single sensors vs array-based sensors - The detection of analytes in complex systems
is an ongoing challenge in chemical sensing due to the presence of interfering analytes. To
date, the majority of sensing systems have been designed using the traditional ‘lock-and-key’
approach, in which a sensor is designed to be highly selective and specific for a single analyte
of interest (Figure 1d).3,6,9 Whilst presenting a valuable feature for analytical tools, highly
specific analyte recognition is not always achievable. From a design standpoint, the creation
of these highly selective sensors can often be time consuming and expensive, particularly for
application in complex solutions such as biological or environmental samples. Additionally,
when moving from analytical studies to practical applications, retaining this specificity is often
challenging due to the presence of similar, competing analytes.9 Furthermore, this sensor
design requires the preparation of a unique sensor for each analyte. Recently, differential
sensor arrays have gained attention, as they can address some of these difficulties.33,34 These
systems enable the discrimination of multiple, potentially similar, analytes through the analysis
of interactions with numerous cross-reactive sensing components.33 Rather than being
specific or selective for a particular analyte, interactions create a unique pattern, or
‘fingerprint’, of responses between all sensors and analytes (Figure 1e). If the combination of
responses is sufficiently distinct, pattern-recognition strategies can be used to identify and
distinguish multiple analytes. Thus, sensors with good cross-reactivity are beneficial in a
sensor array approach and can avoid the issues associated with designing sensing elements
with high selectivity.

Differential sensor arrays consider the simultaneous interactions of multiple analytes in an

entire system and can therefore monitor overall changes in complex mixtures, a feature which
is particularly well suited to analysis of biological systems.35,36 Cross-reactive arrays are also
advantageous in that they are hypothesis-free, meaning that individual interactions between
sensors and analytes do not need to be understood to interpret results, facilitating the rapid
development of sensor arrays to address emerging analytical challenges. Array-based
sensing systems have been reported for a broad range of systems, including analysis of
volatile organic compounds and explosives,37–40 food and drink analysis,36,41,42 environmental
monitoring43–45 and to probe biological systems.46–48

1.4 Polymeric receptors – The design of sensor arrays requires the careful selection of
sensor materials to ensure stability, system compatibility and recognition ability. A range of
sensor arrays have been reported using small molecules,43,44 proteins,47,48 nanomaterials,49–
52
quantum dots,53 and polymers18 as sensing elements. In this review we focus on the use of
synthetic polymers as recognition elements. Macromolecules, in particular polymeric
materials, often provide a robust and inexpensive method of producing sensing elements.54
Advances in the controlled synthesis of polymers have enabled good control over the
composition, dimensions and molecular architecture of the resultant materials, offering
attractive candidates to act as sensor elements.12,54,55 This synthetic control allows robust
functionalization with reporter elements, such as chromophores or electrochemical elements.
It also enables the fine-tuning of the hydrophobic and electrostatic properties of polymers,
which proves useful for modifying these principal interaction mechanisms and controlling
water solubility.56 Furthermore, robust tunability of polymer size enables the construction of
optimally-sized recognition surfaces with multiple interaction sites, presenting the necessary
cross-reactivity for array-based detection.8,57 Polymeric receptors are particularly appealing in
the context of biological sensing, as they can enable multiple recognition events over large
area scales, such as those encountered in proteins, bacteria and cell surfaces.58,59
Incorporating these multivalent interactions into the design and synthesis of polymeric
materials opens up an exciting opportunity to produce inexpensive diagnostic tools for
biological macromolecules. In addition to presenting attractive receptor species, polymers can
also offer a matrix to deposit or ‘dope’ small molecule recognition elements upon, allowing the
construction of sensor arrays with desirable solubility and lipophilicity.

2. Analysis using multivariate statistical techniques

Cross-reactive sensor arrays generally contain multiple sensing elements and analytes,
yielding large data-sets of high dimensionality.19 In principle, each additional sensor
contributes to the ability of the array to distinguish different analytes, but also leads to higher
dimensionality, creating challenges in data visualization and interpretation.60 Multivariate
statistical techniques are commonly used to identify trends and predictability in data-sets and
improve the analysis and presentation of multidimensional data-sets by reducing
dimensionality (Figure 2a).

2.1 Multivariate statistical analysis

In general, analysis of cross-reactive arrays utilizes two classes of statistical methods:
descriptive and classification techniques.10,11 Descriptive analysis generally uses
unsupervised techniques, meaning analyte class information is withheld and the analytical
Figure 2. Schematic representation of a) a high dimensional data set (3× sensors, 4× analytes, 4×
replicates) reduced to two dimensions through multivariate statistical analysis, b) a HCA bottom-up
agglomerative approach and the resulting dendrogram illustrating the connectivity of data points, c)
PCA method of determining the center of the data, projecting points onto a new vector and calculating
the maximum variance and thus best-fitting line, d) LDA method of projecting points onto a new vector
F1 that fulfils the criteria of maximizing the ratio of between-class to within-class variance, e) the SVM
optimization process: the decision boundary (solid line) is optimized in an iterative process to
discriminate classes while maximizing the size of the margin (dashed line), f) Use of an ANN for
classification: the total input for the threshold unit is the weighted sum of all inputs. A transfer function
is applied, and the output function is defined as 1 or 0 depending on whether or not it exceeds a
threshold value. The threshold unit can classify points that are separated by a hyperplane (dashed line).
Blue points correspond to data points assigned a value of 0, and red points to those assigned values
of 1 by a threshold unit.

method is tasked with identifying trends and extracting elements that best cluster the data.61
Since unsupervised techniques are unbiased towards analyte class, they use relatively simple
algorithms to describe general trends in data sets, present a qualitative evaluation of whether
classes are well-separable, and provide information about relationships amongst samples to
identify redundant sensing elements. While descriptive techniques can provide useful
information when evaluating datasets, they are less effective at predictive analysis, limiting
the ability of the array to identify unknown interactions.
In contrast, classification analysis involves supervised techniques, which utilize analyte
class information in order to distinguish data according to these classes. The ability to identify
each replicate as the correct analyte is known as the classification accuracy. An algorithm is
developed during analysis of the ‘training set’, which can then be used to assign unknown
samples into a class. The classification accuracy of both known and unknown samples
indicates both the effectiveness of the algorithm and whether the system requires further
training or additional sensing elements. Algorithms with accurate predictive power can be
difficult to achieve and require a large number of samples to create an effective training set.
Thus, classification techniques are often supplemented by initial descriptive analysis to
optimize the composition of the sensor array prior to classification analysis. The detailed
calculations involved in statistical analysis techniques for cross-reactive sensing have been
well described and we therefore include only a brief overview of the most commonly-used
techniques.11

2.2 Hierarchical Cluster Analysis (HCA)

HCA is an unsupervised descriptive technique that describes a data-set using a discrete
number of steps to forms clusters of data. 62 Cluster formation obeys a chosen metric of
dissimilarity (Euclidean distance or Mahalanobis distance, which are distance metrics in
multivariate space63), either by a top-down divisive approach which successively divides
clusters based on differences until no clusters remain, or a bottom-up agglomerative approach
which forms clusters by pairing nearest-neighbor points based on similarities until all clusters
are merged.64 The latter is more common in sensor arrays, where each data point is treated
as its own cluster with no variance, and clusters form by pairing neighboring data points whilst
minimizing the variance (Figure 2b).62 Data is often presented as a tree-like dendrogram, with
the metric for distance/dissimilarity on the x-axis, and clustering/connectivity on the y-axis. In
array-based sensing, each data point represents an analyte, with clustering indicating which
analytes are similar to one another and distance representing how similar they are to each
other.
There are, however, a number of limitations to the method, primarily the fact that HCA has
no predictive ability for unknown data. Furthermore, dendrograms need to be recalculated to
include any additional data points and can be susceptible to misinterpretation (for example,
the order of clusters already connected in a dendrogram may be swapped without any
meaningful change to the dendrogram). HCA remains a useful technique for descriptive
analysis, presenting a clear indication of similarities between data points in an easy to visualize
manner and semi-quantitatively identifying trends in large-data sets.65

2.3 Principal Component Analysis (PCA)

PCA is another descriptive technique that is used as a tool for dimension reduction of
datasets and for reducing the number of sensors in an array, whilst retaining discriminatory
power. PCA involves the calculation of a set of orthogonal eigenvectors using linear
combinations of the original dimensions. The new dimensions are known as principal
components (PCs) and are calculated to retain the maximum variance within a data-set.66 The
first PC lies in the direction of the highest variance, meaning that data vary the greatest amount
across that dimension. This process can be more easily explained in a simple two-dimensional
case with six variables (Figure 2c).67,68 Firstly, the center of the data is determined by
calculating the average measurement of all variables and the data is shifted so this average
becomes the origin. A best-fitting line is plotted that intersects the origin and each point is
projected on to this new vector. The best-fitting line contains the maximum variance and is
determined by calculating the maximum sum of the squared distances of each data point from
the origin. Each subsequent PC is orthogonal and calculated, following the same criteria, as
the dimension with the next largest variance.
PCA is a useful technique for reducing the dimensionality of multi-dimensional data-sets, as
the first few PCs concentrate the majority of information and data can therefore be presented
in lower dimensional space. Score plots of the first few PCs allow high dimensional data to be
more easily visualized and interpreted. Typically, the number of PCs extracted will be
adequate if they account for 95% of the variance.19
A closer interrogation of the PCA corresponding to a sensor array may also help to identify
redundant or highly correlated sensors. Loading plots project the contribution of each sensing
element as a vector along each PC.69 The sensing elements with the lowest contribution to
the first few PCs can potentially be removed without significantly affecting the discriminatory
power of the array. Additionally, sensing elements that contribute similar loadings to each PC
and are highly correlated may also be redundant. Re-applying PCA to the reduced array will
produce a new set of PCs and evaluating the variance across these PCs will indicate if the
analysis is still adequate without these sensing elements. This ability to pre-screen larger
libraries of sensor elements and subsequently minimize the number of elements to a simpler,
more experimentally-convenient system is a major benefit of PCA.
PCA is an unsupervised technique, meaning that PCs are extracted without any bias
towards analyte class. Thus, if the analysis shows obvious clustering of analyte classes, one
may qualitatively say that a supervised technique is likely to have high classification
accuracy.10 The inverse, however, is not necessarily true, and data that is poorly separable by
PCA may still have high classification and discrimination ability. Overall, PCA is useful for
identifying optimal sensing elements and general trends in large data sets, but a better-
optimized classification technique is preferable for further discrimination and classification
studies.

2.4 Linear Discriminant Analysis (LDA)

LDA is a classification technique, and similar to PCA, is a dimension reduction method
involving the construction of a set of orthogonal, linear combinations that represent the original
data and concentrate this information in a minimal number of dimensions. In contrast to PCA,
LDA is a supervised technique and retains information about analyte class during the
construction of the classification algorithm.70 Using an analogous two-dimensional case with
six variables, a particular vector (F1) is calculated, and each data point is projected onto this
vector (Figure 2d). This new dimension, F1, is optimized by a specific criterion using known
class information. Specifically, the dimension is a linear combination of variables selected to
maximize the distance between sample means (u1 and u2) and minimize the separation within
classes (s1 and s2) (Figure 2d).71 Each subsequent dimension or ‘factor’ is orthogonal to the
previous dimensions and similarly obeys the same criteria; to maximize the ratio of between-
class variance to within-class variance.
A major benefit of LDA as a classification technique for sensor arrays is its predictive
ability.62 Initially LDA creates a mathematical algorithm, constructed using a training set, that
can be validated with unknown samples. A set of sample variables can be left out during initial
analysis, and the accuracy of the model determined by how successfully it classifies these
removed variables. Alternatively, a cross-validation method may be used to test the predictive
power of an array. A common cross-validation technique is leave-one-out (or jack-knife) cross-
validation, where one sample is removed, and the discriminant function is recalculated.72 The
removed sample is then used to test the model, to determine whether the algorithm can
accurately identify the sample without knowing its analyte class. The process is repeated using
all samples one at a time, to determine a cross-validation accuracy. Other cross-validation
methods involve removing subsets of data and repeating a similar process.
LDA score plots allow for a visual evaluation of class clustering and the degree of
discrimination. The majority of the information within a data-set can be presented by plotting
the discriminant scores against the largest factors. The inclusion of 90-95% confidence
intervals further highlights how well analyte classes are discriminated, and are often included
in sensor array applications.43,58,73 Other useful outputs from LDA include confusion matrices
and loading scores.74 A confusion matrix presents accurate classifications on the main
diagonal, misclassifications on the off-diagonal, and allow easy identification of analytes that
are being commonly misclassified. Loading plots illustrate how each sensing element in an
array contributes to each discriminant factor. Sensing elements that have poor loading scores
for the first few discriminant factors may be unnecessary, as they are only contributing a small
amount to discrimination.75 LDA is the most commonly utilized supervised technique in recent
sensor array applications.

2.5 Support Vector Machines (SVMs)

Support vector machines are supervised machine learning models that enable classification
within datasets.10 The technique focusses on identifying decision boundaries that separate
classes. These decision boundaries, or ‘hyperplanes,’ are defined and improved through
multiple optimization cycles to maximize the distance between clusters and minimize
misclassification errors. Optimization focuses on a small subset of data points, those close to
the decision boundary, with these points defined as support vectors (Figure 2e). Data points
that are not linearly separable can be mapped to higher dimensional feature space through
the application of a weighting function, a process known as ‘kerneling’.76

SVMs can offer advantages over discriminant analysis in that there is no requirement for large
training sets in order to enable effective discrimination. SVMs can function with small sample
populations, since only data points close to the decision boundary need to be considered.
SVM models have been optimized to function with multivariate datasets, and can be used very
effectively in array-based sensing applications.

2.6 Artificial Neural Networks (ANNs)

Artificial neural networks mimic early models of sensory processing, by simulating a network
of model neurons.77 Each model neuron, or ‘threshold unit’ processes information from
multiple sources and reports a binary outcome based on the weighted sum of these inputs in
relation to a programmed threshold value (Figure 2f). Each threshold unit defines a
hyperplane, which separates input values which satisfy threshold conditions from those that
do not. Classification problems where two classes can be separated by a hyperplane can be
solved by a threshold unit, if the weighting factors and threshold have been appropriately set,
usually through a machine learning process where the threshold unit is supplied with a training
dataset for which classification is known. More complex linearly inseparable classification
problems can be solved by the introduction of more threshold units in a layered arrangement
known as a feed-forward network. These networks contain an additional layer of ‘hidden’
threshold units which partially classify inputs and direct their outputs to a third layer of
threshold units that combine partial classifications to produce a final output.
ANNs present some advantages over other classification techniques for the processing of
sensor array data. While more commonly-used methods like PCA and LDA provide descriptive
analysis or classification, ANNs can be designed to provide similar descriptive outputs, or to
provide regression results, enabling outputs consisting of continuous rather than discrete data.
The type of output is specified by the function applied at the final layer: classification tasks use
probability related functions, while regression task employ linear functions.78 ANNs can also
process non-linear input values more effectively than other techniques, by using a combination
of threshold functions. This feature presents an advantage for chemical sensing systems
where signal responses may not be linearly related to inputs (e.g. analyte concentration) over
the range of input values.10
When using ANNs, however, the user must be vigilant towards the possibility of ‘overfitting,’
arising when small datasets are fitted using functions with too many parameters.77 Overfitting
can lead to large variations in output arising from small variations in input data. Strategies to
avoid overfitting include regularisation of data, averaging over multiple networks and using
Bayesian statistical tools.79,80 ANNs should also be rigorously evaluated, either through testing
on independent datasets, or through cross-validation procedures.

3. Strategies for the production of macromolecular optical sensor arrays

In this review, we have considered examples of sensor arrays that exhibit good analyte
discrimination, as well as unique ways to achieve diversity in sensor elements and effective
utilization of multivariate statistical techniques. Four main classes of polymeric materials that
have been utilized frequently in optical sensor arrays are featured: conjugated polymers;
polymers modulated with fluorophores and chromophores; molecularly imprinted polymers;
and polymer films. Each of these polymer materials is discussed with a focus on recent studies
that utilize them effectively in sensor array systems, highlighting examples that move into
applications beyond the laboratory.

3.1 Conjugated Polymers - Conjugated polymers are macromolecules that contain

alternating single and double (or triple) bonds along their backbone, resulting in extended
delocalization of p-electrons, which gives rise to the optical properties of these polymers.81
Fluorescent conjugated polymers contain this extended conjugated chain, and interactions
with analytes at one or more sites can be amplified across the entire polymer system, resulting
in fluorescence quenching.16,82 Many conjugated polymers with varying backbones have been
utilized as sensing elements in arrays, including poly(p-phenylenevinylene)s,83 poly(p-
phenyleneethynylene)s,18 polyfluorenes84 and polythiophenes.85 Poly(p-phenylenevinylene)s
(PPVs) are synthesized via a number of methods, the predominant method being the Wessling
polymerization route, involving the polymerization of p-xylylene precursors generated by the
base-induced elimination of p-xylene derivatives.86 Poly(p-phenyleneethynylene)s (PPEs) and
poly(p-aryleneethynylene)s (PAEs) are synthesized by Sonogashira coupling,55 and
polyfluorenes are generally prepared by Suzuki coupling75 or less commonly by Yamamoto
coupling.85 Polythiophenes can be accessed by chemical oxidation polymerization and metal-
catalyzed polycondensation.87

Figure 3. a) PPE sensors P1-P9 with various pendant arms enabling diverse analyte recognition. b)
Three-dimensional LDA score plot obtained from the fluorescence responses of P1-P6 against 17
protein analytes. Reprinted (adapted) with permission from J. Am. Chem. Soc. 2007, 129 (32), 9856–
9857. Copyright (2007) American Chemical Society. c) Two-dimensional LDA score plot of the
fluorescence responses of four different cancer cell lines using fluorescent polymers P2, P5, P7 and
P8. Reprinted (adapted) with permission from J. Am. Chem. Soc. 2010, 132 (3), 1018–1022. Copyright
(2010) American Chemical Society.

Conjugated polymer sensors can produce optical responses to analytes either by the
incorporation of receptor groups or by relying on the inherent electrostatic and chemical
properties of the polymers.15,88 Cross-reactive response to analytes can be achieved by
creating conjugated polymers that incorporate various pendant arms into the polymer scaffold,
such as charged groups or polar side chains, to create diversity along the polymer backbone.89
These diverse functional groups can participate in various electrostatic and hydrophobic
interactions with analytes. This multivalent capacity of conjugated polymers makes them
particularly suitable for sensing macromolecules with large recognition surfaces, such as
proteins,84 cell surfaces73 and other biomolecules.90,91 Furthermore, functionalization at
pendant sites along the backbone can modify the solubility properties of conjugated polymers,
enabling detection in a range of media.
Elegant work by Rotello and co-workers demonstrates the potential of this approach, with
their protein-sensor array comprising six PPE sensors, P1-P6 (Figure 3a).18,73 PPEs were
functionalized with various charged residues to both increase water solubility and target
binding of protein surfaces. This array could classify 17 proteins based on differences in their
metal cofactors, molecular weight and isoelectric point (pI), with 100% accuracy using a LDA
cross-validation routine (Figure 3b).18
Later work by the group utilized a similar PPE sensing system to discriminate healthy,
cancerous and metastatic cell types.73 This array contained most of the PPE sensors from
their 2007 work, along with three additional PPE sensors with different cationic and polar side
chains, P7-P9 (Figure 3a). The environmentally-responsive polymers interact with
various lipids, proteins, and polysaccharides on the cell surface, leading to variations in
polymer fluorescence, attributed to differences in aggregation behaviour. Furthermore, by
identifying which polymers contributed to the most discrimination in LDA, the complexity of the
system could be reduced to just four polymers (P2, P5, P7 and P8). Remarkably, using this
optimized array, LDA could discriminate four phenotypically distinct cancer cell lines with
100% accuracy (Figure 3c), and 3 isogenic cell lines (normal, cancerous and metastatic) with
94% accuracy. This work highlights the versatility of array-based sensing: optimization of the
sensing system enabled the discrimination of subtle differences within protein analytes and
varied cell types.
Lavigne and co-workers utilized polythiophenes, another class of conjugated polymer, to
discriminate various structurally similar amines.22,41,92 The polythiophene sensor P10 included
a carboxylic acid functionality (Figure 4a), and produced a unique optical response across the
visible spectrum upon addition of various amines. This single-sensor array achieved
discrimination by using a combination of nine different absorption wavelengths, and initial
studies were able to discriminate six structurally similar amines.22 LDA classified each amine
at five concentrations using leave-one-out cross-validation with 99% accuracy.

Figure 4. Polythiophene sensors a) P10 with a carboxylic acid functionality used to discriminate various
amines.22,41 Reprinted in part with permission from J. Am. Chem. Soc. 2005, 127 (15), 5695–5700,
Copyright 2005 American Chemical Society; Org. Lett. 2007, 9 (17), 3217–3220, Copyright 2007
American Chemical Society; and b) P11 containing a quaternary ammonium side chain.90 c) Three-
dimensional LDA score plot of the absorbance responses of P11 with 15 nucleotide
phosphates. Reproduced with permission of the Royal Society of Chemistry, from Chem. Commun.
2009, 405 (31), 4696–4698; permission conveyed through Copyright Clearance Center, Inc. Copyright
2009 Royal Society of Chemistry.

In subsequent studies, the group extended the range of analytes to include an additional 16 amines.41
These amines included biologically relevant amines such as histamine and tryptamine, which are
proposed neurotransmitters, or structurally similar amines like aniline, pyridine and 2-aminopyridine.91
Impressively, LDA distinguished all 22 amines with 97% cross-validated accuracy. The change in
absorbance profile was attributed to intramolecular interactions between polymers causing main-chain
twisting, and analyte-directed aggregation causing scattering at longer wavelengths. These studies
highlight the benefits of using conjugated polymers in array-based sensing as they can exhibit multiple
mechanisms of interaction.
Shi and co-workers also utilized polythiophenes in a sensor array, taking advantage of electrostatic
and hydrophobic interactions of the conjugated polymer.90 Interactions with various analytes led to
changes in both the conformation and aggregation of the polymeric backbones, thus altering the
absorbance profile. In this case, a single polythiophene sensor P11 (Figure 4b), containing a quaternary
ammonium side chain, could discriminate 15 nucleotide phosphates (adenine, uracil, guanine, cytosine
and thymine with one, two and three phosphate groups.) Discrimination was proposed to arise from a
combination of ionic-self-assembly processes between ammonium side chains and phosphates,
leading to ordered phases within the poly(thiophene) structure, and aggregation facilitated by analyte
addition. LDA of the resultant absorbance profiles allowed for the differentiation of 15 nucleotide
phosphates with 100% accuracy, using a leave-one-out cross-validation strategy (Figure 4c).
Polycyclic aromatic hydrocarbons (PAHs) are persistent environmental pollutants that damage
human health, creating a need for reliable and sensitive detection methods.75 Bonizzoni and co-workers
recently reported a series of six polyfluorene copolymers to discriminate a library of PAHs. They
attributed the differential recognition between the polymers and analytes to the inner filter effect (IFE).
The IFE arises when excitation or emitted light is absorbed within the sample, either by the fluorophore
itself, e.g. in a highly concentrated solution, or by another absorber. For example, the presence of a
PAH with an absorbance spectrum that overlaps with the excitation spectrum of the polymer sensor
reduces the amount of input excited light, and therefore the emission intensity of the fluorescent polymer
(Figure 5a). Each polymer sensor in the series, P12-P17, has a unique spectral fingerprint due to the
inclusion of a 2-phenylbenzimidazole optical modifier and either phenylene or thiophene co-monomers
(Figure 5b). Each PAH differs slightly in its absorption spectrum and can act as an optically dense
absorber through the IFE, causing slight differences in the fluorescence quenching of each polymer
due to unique regions of spectral overlap. Taking advantage of this IFE, when the absorbance profile
of the 6-series polymer system was subjected to LDA, the analysis could discriminate 16 different PAHs
with 100% accuracy.

Subsequent work by the same group also used the IFE and conjugated copolymers to
discriminate 12 azo dyes.93 This system comprised 3 polyfluorene copolymers, P18-P20, with
various conjugated comonomers of ethylene, thiophene and bithiophene, creating a diverse
range of absorbance profiles (Figure 5c). The azo dyes act as dense absorbers, with
sufficiently unique spectral overlap with each conjugated copolymer to create a differential
response. Multiple absorbance and fluorescence measurements in the regions of greatest
spectral overlap were examined using LDA, enabling complete discrimination of the dyes. The
authors demonstrated the impressive sensitivity of the system by examining the dyes at a
concentration commonly encountered in wastewater of 500 nM. At this concentration, the
absorbance signals of the dyes alone are too low to effectively detect and discriminate them.
However, when utilized in the array system at this concentration with the inclusion of both
absorbance and fluorescence changes of the three polymers, all 12 dyes were successively
discriminated. The sensitivity of these measurements highlights the advantage of
fluorescence-based detection, which enables identification at concentrations beyond the limit
of a colorimetric technique.
Schanze et al. used meta- and para-linked PPEs and polythiophenes in homopolymer and
copolymer sensors to create initial structural diversity, and variant pendant arms to generate
charge diversity.84 The conjugated polymer array could discriminate seven proteins with
varying pI and molecular weight by measuring the variation in protein induced aggregation of
the six polymers, through fluorescence correlation spectroscopy. Jiang and co-workers used
combinations of PPEs and polythiophenes as sensors, using interaction-based fluorescence
responses to discriminate various nitroaromatics.94 Polcha and co-workers introduced
combinations of fluorene, vinylene, anthracene and benzothiadiazole units along the polymer
backbone to obtain the sensor diversity necessary to discriminate various explosives.83 The
clear advantage of modifying the conjugated units along the polymer backbone is the variation
in fluorescence properties of each polymer. In particular, the unique fluorescence and
aggregation properties of different polymer backbones provides a useful platform for cross-
reactivity and thus differential sensing applications.88
Figure 5. a) An illustration of the spectral overlap between analyte and a sensor necessary for the
detection of PAHs through the IFE. Reproduced with permission of the Royal Society of Chemistry,
from Chem. Sci. 2019, 10 (44), 10247–10255; permission conveyed through Copyright Clearance
Center, Inc. Copyright 2019 Royal Society of Chemistry. b) Conjugated polymer sensors P12-17 used
in an array to discriminate PAHs and c) Conjugated polymer sensors P18-P20 with various conjugated
co-monomers.93 Reprinted (adapted) with permission from ACS Sensors 2020, 5 (6), 1541–1547,
Copyright 2020 American Chemical Society.

By far the most comprehensive studies using conjugated polymer sensors in optical arrays
have been performed by Bunz and various co-workers. Since 2005, the Bunz group has
published over 20 papers in the area, both with a variety of polymer systems and a diverse
library of analyte systems. The most prevalent conjugated polymers to appear in their sensor
arrays are PPEs and PAEs. With a range of polymer analogues, array studies have focused
on sensing diverse analytes including organic acids,95 carboxylic acids,96 nitroaromatics,97
flavonoids,98 amino acids,99,100 phosphates,101 nitroarenes40 and saccharides.102 In addition to
small molecule analytes, the group have also demonstrated progress towards using
polyelectrolytes to study complex mixtures, with various sensor arrays able to discriminate
antibiotics,103 whiskies,36 anti-inflammatory drugs,56 fruit juices,104 syrups and honeys.105 It is
remarkable to see such a diverse range of studies and the broad scope of analytes identified
using small libraries of conjugated polymers. This success may be attributed to the fact that
they have exploited the numerous interaction capabilities of conjugated polymers and
balanced these interactions in finely optimized sensing libraries.
A key strategy in the work of Bunz and colleagues involves the use of a combination of
positively-charged, neutral and negatively-charged functional arms. Ionic interactions between
polymers and analytes can provide the cross-reactivity necessary for effective discrimination.
For example, in 2017 the group reported a study to discriminate different whisky brands and
blends.36 Initially they screened a library of 22 different PAEs, of which nine were positively-
charged, four were neutral and nine were negatively-charged. Using PCA, the array could be
reduced to just 3 elements: one positively-charged and two negatively-charged polymers,
P21-P23 (Figure 6a). Elimination of the charge-neutral polymers suggests that electrostatic
interactions play an important role in discriminatory capacity. Using this small 3-element
tongue, LDA was able to discriminate 24 different types of whisky with 99% cross-validation
accuracy.
Electrostatic interactions can also be employed to distinguish analytes through the formation
of complexes between different conjugated polymers. Additional work by Bunz and colleagues
illustrated this approach using five complexes, comprised of one positively-charged PAE, P24,
and five negatively-charged PAEs, P25-P29 (Figure 6b).95 Complexation of P24 with each
negative PAE caused a different degree of quenching. The addition of various organic acids
either caused quenching by further complexation or enhanced the fluorescence by disrupting
the PAE complex. With just five complexes, LDA could discriminate 13 different organic acids
with 94% cross-validation accuracy.

Figure 6. PAE sensors with various functionalities R-R4, a) P21-23 used in an array to discriminate
different whiskies;36 reprinted (adapted) from Chem 2017, 2 (6), 817–824, Copyright 2017, with
permission from Elsevier; and b) P24-P29 utilized in an array to discriminate different organic acids;95
reprinted (adapted) with permission from Chem. Eur. J. 2016, 22 (10), 3230–3233, Copyright 2016 John
Wiley and Sons.
 Another well-explored strategy by Bunz and co-workers to increase discrimination involves
creating multiple pH channels for each sensing element. Many of the analytes that exist in
biology and nature vary their behavior with pH. We can therefore also expect that sensor-
analyte interactions will also be affected by environmental pH. This effect is highlighted by
recent work in the Bunz group on distinguishing flavonoids, which are polyphenols abundant
in many foods and drinks.98 This study used three PPEs and two poly(tetraphenylethene)s
(PTPEs) with diverse pendant functionalities (Figure 7a). The emission spectra of the five
polymers were recorded at different pH values (pH 4, 7 and 10) in the presence of each
analyte, effectively giving a 15-element array. LDA confirmed successful discrimination of 11
flavonoids. Through PCA, six key elements were identified as contributing most significantly
to the discriminatory power of the array. LDA of the optimized 6-element array could effectively
discriminate all 11 flavonoids, including clustering of sub-populations of structurally similar
flavonoids, further illustrating the structural sensitivity of the array (Figure 7b). Notably, the
work also includes structure-activity relationship studies of the hydrophobic and electrostatic
behaviour of the conjugated polymers, exemplifying the important role both types of
interactions play in the discrimination of analytes. These studies highlight how increasing
understanding of the molecular recognition events underpinning complexation can enable
effective optimization of sensor arrays.

Figure 7. a) PPE and PTPE sensors P29-P33. b) Two-dimensional LDA score plot of the fluorescence
responses with 11 flavonoids. Reprinted (adapted) with permission from ACS Appl. Polym. Mater. 2019,
1 (6), 1301–130798. Copyright 2019, American Chemical Society.

The discriminatory power of conjugated polymer-based arrays can be increased by using

other non-covalent interactions in addition to electrostatic interactions. PPEs contain aromatic
units along the polymer backbone, capable of undergoing - stacking interactions.106 Bunz
and co-workers explored these interactions in their recent work discriminating 13 nitroaromatic
compounds, including known explosives or compounds used in their preparation.97 They
sought to increase recognition of these electron-poor aromatic species by developing an array
of electron-rich PPEs. Sensors P34-P37 exhibited conjugation that extended beyond the
backbone through appendage of benzylic side chains to two of the polyelectrolytes (Figure
8a). Nitroaromatic compounds caused fluorescence quenching, proposed to be due to p-
p stacking with the aromatic arms. Differential quenching responses for the 13 analytes,
enabled classification using LDA with 99% cross-validation accuracy.

Figure 8. Conjugated polymer sensors a) P34-P37 used to discriminate nitroaromatics;97 . reprinted

(adapted) with permission from Macromolecules 2017, 50 (11), 4126–4131, Copyright 2019, American
Chemical Society; and b) P38-P39 used to study anti-inflammatory drugs;56 reprinted (adapted) with
permission from ACS Appl. Mater. Interfaces 2017, 9 (1), 790–797, Copyright 2017, American Chemical
Society.

Combining the strategies described above is likely to enhance the discriminatory capacity
of a sensor array. This extra depth of discriminatory power is particularly beneficial when
identifying analytes within a large library, or within complex mixtures, where individual
interactions are difficult to discern. An elegant example of this approach was conducted by the
Bunz group in their 2017 paper on detecting anti-inflammatory drugs.56 Building on their
previous work on the discrimination of structurally similar aromatic acids,96 they selected a
highly fluorescent cationic polymer (P38) and a complex of this polymer with a weakly
fluorescent anionic polymer (P39) to form the two elements of their sensor array (Figure 8b).
Different analytes disrupt the P38-P39 complex and either enhance or quench the
fluorescence. The experimentally determined binding constants of the complex are pH
dependent, so measurements of each analyte-sensor combination were conducted at both pH
10 and pH 13. As expected, the interaction behaviour of sensors and analytes is also pH
dependent. LDA of the 4-element array revealed 100% cross-validation classification accuracy
of 11 different anti-inflammatory drugs.
Rotello and colleagues used an IDA strategy with conjugated polymers.48,50 Their 2007 study
used different nanoparticles to create six non-covalent gold nanoparticle-conjugated polymers
with various surface functionality. These nanoparticles quench the fluorescence of the PPE
polymer, which they previously reported for protein discrimination, P1.50 Interaction with
proteins disrupts the nanoparticle-polymer interaction and produces a change in polymer
fluorescence. The six sensing elements enabled discrimination of seven proteins of differing
size, charge and pI with 100% accuracy using LDA, and an additional study with 56 unknown
protein samples led to classification with 96.4% accuracy. Impressively, the study went on to
achieve protein identification at varying concentrations, to meet the requirements for real-
world applications.
More recently, Rotello, Bunz and colleagues designed an IDA-based sensor array
comprised of conjugated polymers, and the green fluorescent protein (GFP) as the indicator,
with Förster resonance energy transfer (FRET) quenching of the polymer by the protein. The
FRET-based sensor array comprised the supramolecular complexes of GFP with four
conjugated polymers, P2, P7, P8 and P33, all of which had shown effective discrimination
capacity in previous studies.18,73,98 Multivalent binding of the polymers with cell surfaces
disrupts the FRET signal of the polymer-GFP complexes to produce a ratiometric response,
and LDA was able to differentiate and identify unknown cell lines. 16 different cell types were
tested with this array, revealing 100% cross-validation accuracy of 3 different isogenic cell
lines, 100% accuracy of four site-specific metastatic cell lines, 100% accuracy of four
glycosaminoglycan (GAG)-modified cell lines and 94.5% accuracy of 128 unknown samples
of all cell types. Importantly, using the conjugated polymers alone, only 63% and 72% cross-
validated accuracy was achieved for GAG-modified cell lines and metastatic sublines
respectively, highlighting the importance of the ratiometric response arising from the pairing
between the polymers and GFP.
Another example by Bunz and colleagues utilized P1 paired with four different antimicrobial
peptides, which partially quench the polymer fluorescence, to discriminate bacteria in urine.58
The interactions between the polymer complexes and each bacterial species were found to
be dependent on the components on the bacterial surface. These interactions produced a
pattern-based fluorescence response, due to either aggregation of complexes or displacement
of the polymer, and responses for each unique bacterial species were classified using LDA
with 100% accuracy.
The work addressed in this section highlights the broad scope of analytes that can be
detected by conjugated polymers and has explored various interactions responsible for
discrimination. Many systems tune the balance between two of the major interactions
responsible for discrimination – hydrophobic interactions and electrostatic interactions – by
varying elements such as the polymer backbone, the functionalization of ionic pendant arms,
cationic/anionic complexes, pH, hydrophobicity of side chains, or various combinations of
these factors.

3.2 Polymers decorated with chromophores and fluorophores

Another common method used to create optically active polymers involves appending both
recognition motifs and optical reporters to a polymer scaffold in a modular approach. The
modular strategy involves the synthesis of parent polymer architectures that can be
functionalized with both a recognition unit and a chromophore/fluorophore in a post-
polymerization modification step, enabling both components to be modified to achieve cross-
reactivity and diversity of response. These systems have shown many successful applications
for the discrimination and identification of various biomolecules, particularly for proteins.57,59,107
Kurita and co-workers utilized two copolymers, P40-P41, functionalized with a cationic
lysine recognition site and two environmentally sensitive fluorophores (Figure 9a), to
distinguish protein post-translational modifications (PTMs).107 They postulated that the lysine
would undergo multiple electrostatic and hydrophobic interactions with analytes due the
charged amino group and the n-butyl group respectively.108,109 Proteins with different PTMs
would likely interact in a unique way to environmentally sensitive fluorophores and increase in
fluorescence due to a decrease in polarity in the microenvironment surrounding the
fluorophore upon protein binding. Both P40 and P41 demonstrated fingerprint responses to
four different serum albumins, before PTM. P40 was selected for further sensing experiments
as it displayed a markedly larger fluorescence enhancement upon analyte addition, enabling
improved precision and sensitivity for subsequent experiments. The fluorescence response of
P40 with various analytes was dependent on both the pH and ionic strength of the system,
supporting the hypothesis that electrostatic and hydrophobic interactions between P40 and
proteins contributed to complexation. The researchers therefore optimized the array to one
comprising P40 in six different buffer solutions of varying pH and ionic strength at four
wavelength channels. The array could discriminate 12 protein analytes comprising six proteins
in the presence and absence of various PTMs including phosphorylation, acetylation,
methylation and glycation. LDA presented individual proteins in well-separated classes, and
100% accuracy in a leave-one-out classification procedure (Figure 9b).
.The data was also explored using HCA, with clustering of data indicating that ionic strength
was the most important discriminating factor, followed by pH, supporting earlier observations
about the important role of electrostatic and hydrophobic interactions in differentiation. While
the system enables accurate discrimination of known protein samples, the reliance on ionic
strength may limit its applicability for identifying unknown protein solutions, as the ionic
strength of a solution is likely to be an unknown quantity.

Figure 9. a) Poly-L-lysine copolymers P40-P41 with environmentally sensitive fluorophores and b)

three-dimensional LDA score plot of the fluorescence responses to various PTMs of different proteins.
Reprinted (adapted) with permission from ACS Appl. Mater. Interfaces 2017, 9 (27), 22970–22976.
Copyright 2017, American Chemical Society

Subsequent work by the same group again utilized the dansyl-modified poly-L-lysine
copolymer scaffold P40, in this case to discriminate different human cell lines.110 As the
scaffold had already displayed cross-reactive sensitivity to proteins and PTMs of proteins, the
authors hypothesized that the sensor could also bind non-specifically to hydrophobic and
charged components of cell surfaces, in a mechanism similar to that used by Rotello and co-
workers, as described above.18,73 By measuring emission at multiple wavelengths in buffers of
varying pH and ionic strength, then employing LDA, they were able to discriminate eight
human cancer cell lines from different tissue types, with 100% cross-validation accuracy.
Applying their original protein sensor system to a more complex cell system highlights the
benefits of array-based sensing. Translation of chemical sensors from single analytes to
complex biological systems often results in considerable errors related to biocompatibility,
arising particularly through the presence of interfering and competing species. Array-based
sensing avoids this complication by characterizing the analyte based on the entire complex
system. Thus, by employing a small number of optimization steps and re-training the array in
biological fluid, Kurita and colleagues were able to avoid the vast redesign efforts often
required to apply sensors to biological systems.

Figure 10. a) PONI random copolymers containing a benzoate solubilizing group and fluorophores
pyrene (P42), dapoxyl (P43) and PyMPO (P44).35 Reproduced (adapted) from Adv. Mater. 2018, 30
(28), 2–7, under a Creative Commons CC BY licence. b) PONI random copolymers P45-P47 containing
different recognition groups (R1) and fluorophores (R2). c) Normalized fluorescence response of P45-
P47 and to five unique biofilms and d) two-dimensional LDA score plot of sensor responses to biofilm
analytes. Reprinted (adapted) with permission from ACS Appl. Mater. Interfaces 2019, 11 (12), 11202–
11208. Copyright 2019, American Chemical Society.

Rotello and colleagues have presented another approach to the design of macromolecular
sensor arrays, based upon polymer scaffolds decorated with optical elements.35 Their
approach involves a poly(oxanorborneneimide) (PONI) random copolymer produced via a
ring-opening metathesis polymerization, enabling a high level of synthetic precision. The
scaffold consists of a benzoate group that acts as a recognition element for proteins and either
a pendant pyrene, dapoxyl or PyMPO fluorescent dye, creating a library of three polymers,
P42-P44, with distinct signal responses (Figure 10a). These environmentally responsive dyes
were proposed to function as both cross-reactive recognition components and reporting
groups for serum proteins implicated in liver fibrosis. Interestingly, the array design involved
combining the three polymers in solution to enable a multichannel output from a single sample
measurement. Initially, the array was tested against a number of common serum proteins to
confirm the system was sensitive to fluctuations in protein levels in human samples. Five
human serum proteins (human serum albumin, immunoglobulin, transferrin, fibrinogen and
alpha-1-antitrypsin) were added to human serum and could be distinguished using LDA with
86% classification accuracy.
The ability of the array ability to classify fibrotic liver tissue was explored through a study of
65 clinical human serum samples. A set of 50 samples were tested and using LDA, the array
could discriminate healthy and fibrotic samples with 80% accuracy. Using this as a training
set, 15 additional samples were identified and correctly classified with 80% accuracy. Further
tests examined whether classification accuracy was correlated to biomarkers linked to liver
fibrosis, by comparing the concentration of each biomarker in the fibrotic tissue and any
misclassification results. Ultimately, it was found that no single biomarker was responsible for
classification results, indicating that multiple biomarkers were required to generate the
signature response enabling discrimination between healthy and fibrotic samples. Importantly,
the results of the sensor array were found to be comparable to other tests identifying these
biomarkers, but with a faster and robust method that does not require specialist equipment.111
Subsequent work from the Rotello group used a similar PONI scaffold, with each polymer
incorporating a cationic unit for recognition and an environmentally sensitive fluorophore as a
signaling element.57 Using various post-polymerization modification strategies, a guanidine
unit, a benzyl unit and a trimethylammonium unit (R1) were functionalized on the PONI scaffold
with pyrene, NBD and naphthalimide derivatives respectively (R2) (Figure 10b). The three
resultant cationic fluorescent polymers P45-P47 were again combined in solution, generating
six distinct wavelength channels due to excimer formation and FRET-based interactions
between different polymers. In this instance, the sensor array was used to identify different
species of bacteria known to be present in biofilms. The polymers interact with the surfaces
of biofilms, consisting of live and dead bacterial cells, proteins, DNA, polysaccharides and
other biomolecules, depending on the species on bacteria present. Interaction with different
biofilm matrices created a distinct fingerprint response for five different biofilm models
containing a single species of bacteria (Figure 10c), and LDA produced 100% correct
classification accuracy (Figure 10d).
The array was further validated with 40 ‘blind’ samples of biofilms which were correctly
classified with 95% accuracy. Additional studies revealed the array was also able to
discriminate biofilm models consisting of mixed species of bacteria and these were LDA-
classified with 100% accuracy. The strategy of mixing all polymers together for biofilm
classification not only simplifies sample preparation and measurement, but also increases the
discriminatory power of the array. The two wavelength channels from the FRET-based
interactions resulted from the mixing of polymers, and when LDA was performed without these
channels, the prediction accuracy dropped to below 80%.
In a recent paper, Rotello and colleagues created an assembly between P45 and GFP to
create a FRET-based sensor array system to discriminate different macrophage polarization
phenotypes.112 Similar to their previous PONI polymer scaffolds, a single well containing both
components of the sensor provided five unique fluorescence channels based on different
interactions. The fluorescence responses provided enough discriminating power to identify
five different macrophage polarization states using LDA with 100% classification accuracy.
In the context of new diagnostics, it is important to consider straightforward and robust
sensor systems that are advantageous when compared to common protocols.34 The PONI
polymer systems discussed achieve this, as rather than including an individual measurement
for each sensor, they mixed sensors together and decreased sample preparation and
measurement time to produce more rapid and robust sensor systems. Despite this success of
mixing polymer sensors and exploiting intermolecular interactions, this one-pot sensor array
concept has not been explored far beyond these key examples, presenting ample scope for
the development of next-generation diagnostics for a wide range of diseases.
Recent work by Albrecht and co-workers has also explored the use of pyrene-labelled
polymers, combining emission responses for monomer and excimer species to enable
discrimination of proteins.59 A co-polymer, P48, consisting of (2-(dimethylamino)ethyl
methacrylate) (DMAEMA) and a pyrene-functionalized methacrylamide was synthesized via
RAFT polymerization (Figure 11a), and presented an optical fingerprint response to various
metallo- and non-metalloproteins (Figure 11b). The combination of hydrophobic pyrene and
hydrophilic DMAEMA functionalities results in intermolecular assembly and aggregation. At
neutral pH, hydrophilic PDMAEMA segments are partially protonated and form an outer shell
that is able to interact with biomolecules, such as the charged surfaces of proteins. Pyrene
units produce two distinct emission bands at 384 nm and 394 nm, and an excimer band at
488 nm, suggesting the presence of domains of aggregated pyrenes as well as isolated
pyrenes. The molecular characteristics of each protein influenced its interactions with the
polymer assembly and resulted in structure dependent disassembly of aggregates and either
a fluorescence turn-on or quenching effect. These interactions resulted in an optical fingerprint
of the monomer and excimer wavelengths for eight different proteins. Negatively-charged non-
metalloproteins caused disassembly of aggregates due to strong Coulomb interactions
between the analyte and partially protonated DMAEMA segments, resulting in disruption of
the hydrophobic interactions in the pyrene domains and subsequent quenching of excimer
fluorescence. Positively-charged non-metalloproteins were observed to have minimal effect
on fluorescence, presumably due to their limited interactions with the partially positively-
charged polymer subunits. Finally, metalloproteins quenched both the monomer and excimer
fluorescence of the polymer aggregates, an observation which was attributed to electron or
energy transfer effects.

Figure 11. a) Copolymer P48 b) Schematic representation of the interaction of P48 with various non-
metallo and metalloproteins.59 Reproduced with permission of the Royal Society of Chemistry, from J.
Mater. Chem. B 2018, 6 (41), 6599–6606; permission conveyed through Copyright Clearance Center,
Inc. Copyright 2018 Royal Society of Chemistry.

Despite establishing a fingerprint response towards multiple proteins, conditions and

polymer wavelengths, the study did not include any multivariate analysis to classify them. In
a simple, one-protein per well model with a specific pH and ionic strength, this is an adequate
system to distinguish proteins. However, any further studies in more complex mixtures will
likely encounter issues with interfering species and system variation and some of these effects
could be addressed by multivariate statistical analysis.

Bonizzoni and colleagues have reported a number of IDA-based array systems based on
polymers decorated with various dyes.113–116 Early work employed a water-soluble poly-
(amidoamine) dendrimer (PAMAM) that incorporated a fluorescein dye to discriminate
physiological phosphates.113 The fluorescence of 5(6)-carboxyfluorescein was quenched upon
binding to the dendrimer host, and displacement of the dye by various phosphates revived the
intense fluorescence. PCA resolved the four biological phosphates in well-separated clusters,
which was not possible by a univariate approach. Including additional dyes in IDA systems is
likely to increase variability within a dataset and thus improve discrimination capacity.
Subsequent work by Bonizzoni and co-workers demonstrated the advantage of
incorporating additional dyes, by including both fluorescein and 4-methylumbelliferyl
phosphate dyes in their dendrimer IDA to discriminate organophosphates, including
environmental contaminants and compounds used in the preparation of chemical warfare
agents.114,117 The degree of displacement of each dye from the PAMAM dendrimer depends
on the relative affinities of the dye and the analyte guest to the dendrimer host. The inclusion
of a second guest dye with the dendrimer host increased the number of distinct displacement
interactions between analytes and host, and therefore increased the discriminatory power of
the array. The multi-dye IDA was able to discriminate four organophosphates and phosphate
at 800 µM using LDA with a 100% cross-validation accuracy (Figure 12a). Remarkably, the
array achieved additional concentration dependent discrimination of three phosphates from
10 µM to 2 µM with 96% jack-knifed cross-validation accuracy (Figure 12b).
Carbohydrates play important roles in biology as fuel sources,118 and in cellular
recognition,119 and there is therefore much interest in the detection and distinction of sugars.
This task is challenging, as most mono- and disaccharide species have minimal structural
differences beyond the configuration of selected stereocentres. Sensing of carbohydrates is
commonly achieved through a boronic acid-diol displacement strategy, as aromatic boronic
acids have a strong affinity for the vicinal diols found in sugars.120 This affinity can by
harnessed in an IDA strategy, in which a dye is covalently bound to a boronic acid and
subsequently displaced by competitive binding of the boronic acid to a carbohydrate, with
release of the dye resulting in an optical signal change.
Bonizzoni and colleagues utilized this boronic acid-diol IDA approach in the development of
a sensor array for sugars, comprising two boronic acid modified PAMAM dendrimers and two
fluorescent dyes, 4-methylesculetin and alizarin red S.115 The displacement of each dye from
the two dendrimer complexes at two pH values (7.4 and 10) was monitored by absorbance
and fluorescence spectroscopy and fluorescence anisotropy. The combined results were
processed using LDA, discriminating four sugars into well-separable clusters.
Another recent example of sugar-sensing by Bonizzoni and colleagues is a polymer-dye
complex involving a copolymer of poly(methacrylic acid) and 3-(acrylamido)phenylboronic
acid (PMAA-co-AAPBA).116 Both a hematoxylin and cyanidin chloride dye illustrated suitable
binding to the boronic acid functionality on the polymer and were examined as sensing units
in an IDA approach. The dyes and their polymer-dye complexes were exposed to eight
common sugars, and spectra were collected across multiple absorbance and emission
wavelength channels. LDA revealed distinct responses to all eight sugars. Further
examination of the data indicated that the cyanidin chloride dye and complex presented a low
contribution to the discriminatory capacity of the array. Re-analysis using LDA of the eight
sugars with only the hematoxylin dye complex successfully resolved all analytes. The authors
noted that whilst clustering is slightly tighter in the multi-dye array, the single dye array
presents a more practical option. After reducing to a single sensor system, another factor to
consider is whether multivariate analysis was necessary for analyte identification in this case,
or if closer analysis of absorbance and emission profiles would have been adequate to identify
analytes.
Figure 12. a) Two-dimensional LDA score plot for the response of five organophosphates, phosphate,
glyphosate (GlyP), pinacolyl methylphosphonate (PMP), methylphosphonate (MPA) and ethyl
methylphosphonate (EMP) at 800 µM. b) Two-dimensional LDA score plot for the analysis of three
phosphates over a concentration range from 10 µM to 2 mM. Reprinted (adapted) with permission from
J. Am. Chem. Soc. 2014, 136 (40), 14223–14229. Copyright 2014, American Chemical Society.

The broad scope of analytes detected by the sensor arrays in this section highlights the
benefit of using a modular approach between polymer and chromophore/fluorophore to create
sensor diversity. A wide range of polymer architectures and optical elements are available,
presenting an opportunity to produce large libraries of sensors by modifying either partner.
The ability to choose different optical elements provides the flexibility to select and fine tune
optimal features relating to wavelength, quantum yield and functionalization sites.
Furthermore, the various polymer structures, namely dendrimers, homopolymers and
copolymers, create different interaction surfaces, sizes and structures for sensors, allowing
the detection of biomolecules as large as proteins and bacteria all the way down to small
molecules such as phosphates and saccharides.

3.3 Molecularly imprinted polymer sensor arrays - Molecularly imprinted polymers121

(MIPs) have, in recent years, been used as artificial receptors in a variety of sensing
applications.122,123 The molecular imprinting process involves the assembly of building
blocks/monomers around a target template molecule through supramolecular interactions,
and subsequent cross-linking polymerization to fix the spatial arrangement of these
monomers.8 Removal of the template yields a matrix containing cavities with specific
functionality, geometry and size to enable selective recognition of the template molecule. MIP
sensors provide access to the detection of analytes that may be difficult to sense selectively
due to non-specific binding interactions. By including additional specifications regarding
orientation, shape and size, MIPs narrow the opportunity for non-selective binding. Upon initial
consideration, it seems that MIP sensors would be poor candidates in a sensor array, where
optimal sensing elements are preferentially cross-reactive rather than selective. It is therefore
important to note that despite their successes, MIPs suffer from a few key limitations. In
general, highly cross-linked materials have fairly rigid structures, which decrease the number
of available binding sites.122 Furthermore, these binding sites may not be uniform throughout
the material and often consist of more lower affinity sites rather than high-affinity sites,
because templating proceeds under kinetic rather than thermodynamic control.8,124
Additionally, imprinted sites created by larger structures, such as proteins, may also have
affinity to smaller biomolecules with similar binding sites.123 These limitations generally lead
to reduced selectivity and it is these shortcomings that often redirect these sensors for use in
cross-reactive arrays.
Shimizu and colleagues conducted early work incorporating MIPs as sensing elements in
cross-reactive arrays (Figure 13a).17,21,124 Initially, they synthesized an eight polymer array,
using seven different aryl amines as template molecules and one in the absence of a
template.17 The aryl amines were both pharmaceutically and biologically important amines
such as propranolol, ephedrine and pseudoephedrine, as well as other structurally similar
analytes (Figure 13b). The affinity of each polymer to each amine was examined by measuring
the ratio of amine absorbance at 258 nm before and after shaking in acetonitrile. The imprinted
polymers showed higher affinities than the non-imprinted polymer and also showed a
significant change in absorbance not only for their imprinted analyte but also other structurally
similar aryl amines. Responses for the first six aryl amines were analyzed using LDA, revealing
94% correct classification of analytes using a leave-one-out cross-validation method (Figure
13c). The study illustrated the effectiveness of using MIPs in the array, as each analyte
showed the best response to its corresponding imprinted polymer, suggesting that selectivity
within the array was likely due to the imprinting process. Whilst effective at discriminating
these amines, each analyte required absorbance at the appropriate wavelength (258 nm),
limiting the studies to UV-active analytes. Typically, MIPs contain no intrinsic signaling
component, which is why this study utilized the absorbance of the unbound analyte and its
modified response upon binding for discrimination of each analyte. An attractive strategy to
address this limitation involves incorporating a chromophore as a signaling element within the
MIP. Analyte binding will ideally cause a change in the spectral properties and signal response
of the chromophore.124
Subsequent work by Shimizu and co-workers moved beyond sensing of UV-active analytes,
by incorporating an IDA strategy into their MIP-based sensor array.21 Similar to their previous
study, they synthesized seven polymers, six imprinted with various aryl amines and one non-
imprinted polymer. The final additional step in preparing the sensors involved the incorporation
of a benzofurazan dye in each polymer scaffold. Benzofurazan is a small dye with similar
functionality to aryl amines, so has sufficient binding affinity with each polymer, but strong
absorption in a different region of the visible spectrum, reducing any potential interference
from analyte absorbance. Each polymer was tested with seven amine analytes and the relative
dye displacement response was measured. LDA was used to separate these responses into
seven distinct clusters for each analyte, with 94% classification accuracy using leave-one-out
cross-validation (Figure 13d).
Figure 13. a) Schematic representation of MIP sensor arrays. b) Structures of seven aryl amines utilized
as templates for MIPs. c) Two-dimensional LDA score plot of the responses of six amines tested against
the MIP array. a)- c) Reproduced with permission of the Royal Society of Chemistry, from Chem.
Commun. 2004, 1172–117; permission conveyed through Copyright Clearance Center, Inc. Copyright
2004, Royal Society of Chemistry. d) Three-dimensional LDA score plot of the responses of seven
amines against the MIP array. Reprinted (adapted) with permission from J. Am. Chem. Soc. 2005, 127
(15), 5695–5700. Copyright 2005, American Chemical Society.

Yan and co-workers have constructed a number of sensor arrays using molecularly
imprinted mesoporous silica as a sensing matrix.125,126 They reported a metal ion sensor array
system consisting of an 8-hydroxyquinoline (8-HQ) monomer covalently attached to the
mesoporous silica, in the presence of two templates, Zn(II) and Cd(II), and in the absence of
template during cross-linking.125 8-HQ was chosen as the fluorescent receptor as it forms
fluorescent chelates with a number of metal ions. Unsurprisingly, the two imprinted sensors
showed a higher affinity and fluorescence turn-on with each corresponding metal template
than the non-imprinted sensor. The sensor array was subsequently tested with three additional
metal analytes, Mg(II), Ca(II) and Al(III), at two different concentrations (10-4 M and 10-5 M).
The characteristic response pattern for each metal ion was then analyzed using PCA to
evaluate the discriminatory capabilities of the array. Each metal ion was clustered into a
distinct group, as well as apparent sub-clustering of the two concentrations of each metal ion.
The authors noted that the array was able to discriminate non-templated ions, together with
templated ions, highlighting that using molecularly imprinted materials with high levels of
cross-reactivity increases the number of analytes the array is capable of discriminating.
Subsequent work by the same group utilized a similar mesoporous silica matrix and an IDA
strategy for saccharide discrimination.126 The materials were constructed by covalently
attaching a phenylboronic acid moiety to the mesoporous silica for saccharide discrimination,
using D-fructose and D-xylose as templates, along with a nonimprinted material prepared
without a carbohydrate template.126 The two sugars selected as templates varied in size and
structure, with different binding affinity to boronic acid, to create binding sites distinct enough
Figure 14. a) Structurally similar amine templates and analytes and b) metalloporphyrin dye structures
with the metals Zn(II), Co(II) and Co(III). c) Two-dimensional LDA score plot for the response of 4
polymer channels combined with Zn(II)-porphyrin. d) Two-dimensional LDA score plot for the response
of 4 polymer channels and Zn(II)-, Co(II)- and Co(III)- metalloporphyrins. Reprinted (adapted) from
Tetrahedron Lett. 2013, 54 (22), 2890–2893, Copyright 2013, with permission from Elsevier.

to capture a more diverse library of carbohydrates. Employing an IDA, the fluorescent dye
Alizarin Red S binds to phenylboronic acid through adjacent diols present on the scaffold,
resulting in a fluorescence response which is quenched upon displacement by various
carbohydrates. Subsequent studies investigated a library of 10 saccharides: four five-carbon
saccharides; five six-carbon saccharides and one disaccharide. PCA analysis illustrated tight
non-overlapping clusters of all 10 saccharides, indicating both good reproducibility and great
discriminatory power of the sensor array. The authors then tested their sensor array with three
brands of orange juice, of which the major ingredients included fructose, sucrose and glucose.
After confirming that other components of juice such as citric acid, sodium citrate and
carotene, had no effect on the fluorescence response, the array was run and analyzed by PCA
to confirm successful discrimination of the three juices. Translation of this saccharide
imprinted sensor array to real-world samples illustrates the advantages and feasibility of this
type of system, particularly for future sensing systems.
Hong and co-workers have reported a system involving doping MIPs with metalloporphyrins
in an array to discriminate a variety of structurally similar amines.127 These amines comprised
both primary and secondary amines and amines with other functional groups including a
pharmaceutical agent. The four MIPs are synthesized using three amines as templates,
phenethylamine, tyramine and (+)-norephedrine as well as a non-imprinted version (Figure
14a). The three metalloporphyrin dyes were prepared using 5,10,15,20-tetrakis-(3,5-di-tert-
butylphenyl)-21H, 23H-porphine and the metals Zn(II), Co(II) and Co(III) (Figure 14b). The
array was constructed by mixing amine analytes with each MIP, allowing the mixture to
equilibrate, removing the MIPs and testing the supernatant containing all unbound amines
with a solution of a metalloporphyrin. Absorbance measurements with different concentrations
of MIP established that the spectral change of the dyes was a consequence of binding to
analytes remaining in the supernatant, and therefore presented an indirect measure of the
binding affinity of each analyte to the MIP. An initial test array utilized all four polymer channels
combined with Zn(II)-porphyrin, against the three imprinted amines and two additional amines,
L-tyrosine methyl ester and 4-methylbenzyl amine. LDA enabled accurate classification of all
analytes, including the two amines not used during the imprinting process (Figure 14c). A 12-
channel array consisting of the four polymers and all three metalloporphyrins was also tested
with the five amines and resulted in a higher degree of visual discrimination using LDA,
specifically tighter clusters and smaller confidence intervals (Figure 14d). The combination of
MIPs and metalloporphyrins in this study capitalized upon the advantages of both materials,
without the addition of costly design and synthesis processes. Specifically, the pattern
recognition capabilities of MIPs and the useful absorbance properties of metalloporphyrins
upon binding to analytes were together able to discriminate structurally similar amines, with
improved distinguishability compared to one system alone.
It is evident that molecularly imprinted sensors are useful in array-based sensing
applications on account of their capacity for molecular recognition, and the facile production
of sensors. MIPs have the benefit of presenting large interaction surfaces to produce general
differential responses to analytes, as well as containing regions tailored towards selective
recognition created during the imprinting process.21 Furthermore, this optimal combination of
general and specific recognition properties in MIP arrays is accessed through minimal
synthetic effort compared to other types of receptor, and large sensor libraries can be
generated relatively quickly through modification of the template molecule only.
Since first reported in the early 1970s by Sarhan and Wulff,128 MIPs have been extensively
explored as receptor species in the academic literature, but their commercial exploitation has
rarely been demonstrated. This gap in their development may be a consequence of limitations
in production capabilities, such as challenges associated with preparing MIPs on a large scale,
or issues with reproducibility arising from the kinetically-controlled nature of the templating
process. More recent research developments, including ‘smart’ molecular imprinting
approaches may accelerate the development of MIP-based receptors and sensors for
commercial use.122

3.4 Sensor arrays constructed on polymer films

In addition to functioning as macromolecular receptor species, polymers can also present a
convenient matrix upon which small molecule sensors can be deposited, or ‘doped’ to facilitate
convenient fabrication of sensor arrays. A solid-state array is generally prepared by inclusion
of sensing dyes within a polymer matrix such as polyurethane, or by the deposition of a
pigment onto a porous polymer film, such as polyethylene terephthalate (PET). These polymer
film sensor arrays are beneficial as they can improve the compatibility of aqueous systems,
and in some cases are more user-friendly as they often do not require the dissolution or
pipetting of sensors.129
Anzenbacher and colleagues have demonstrated the success of a doping strategy in an
anion sensor array.130 Their system incorporated a fluorescent calix[4]pyrrole probe (Figure
15a) with affinity to various anions, into ten poly(ether-urethane) hydrogel copolymer matrices
(Figure 15b). The proportions of ethylene glycol ether and butylene glycol ether were varied
to create different hydrophilic environments, with the aim to exploit the effect this would have
on the transport of anions within the gel after exposure to water. Eight anions were tested with
the ten polymer films doped with the fluorescent probe, namely acetate, benzoate, chloride,
fluoride, hydrogen sulfide, cyanide, hydrogen phosphate, and hydrogen pyrophosphate. LDA
of the fluorescence response illustrated 100% cross-validated classification accuracy. The
sensor array was also applied to eight urine samples, providing a complex multi-electrolyte
system to validate the array’s versatility. The 10-sensor array was able to correctly classify the
eight urine samples with 100% accuracy.

a)
N
H
CN
NH HN
H
N
CN

b)

O O
O O
O N N O
H H m
n

Figure 15 a) Structure of calix[4]pyrrole fluorescent probe. b) General structure of the 10 poly(ether-

urethanes) hydrogel copolymer films used. Reprinted (adapted) with permission from Chem. Eur. J.
2013, 19 (26), 8497-8506, Copyright 2013 John Wiley and Sons.

Another example utilizing a polymer matrix by Anzenbacher and colleagues is their sensor
array for metal ion discrimination.131 The system involves six sensors, P49-P54, made up of
an extended conjugated fluorenes and a pyrene moiety attached to an 8-hydroxyquinoline (8-
HQ) metal receptor (Figure 16), which are immobilized in a polyurethane film. The conjugated
chromophores are partially quenched by 8-HQ, depending on the length of the fluorene
bridges, with emission changes resulting from coordination to various metal ions. The
hydrophilic film draws in the aqueous solution of metal ions, assisting complexation to the 8-
HQ receptor and overcoming issues with solubility and lipophilicity. The responses to 9 metal
cations were investigated with the six sensors and LDA classified all replicates with 100%
accuracy. The authors performed an additional study to reduce the number of sensors
required for metal ion discrimination.132 PCA identified the key sensors responsible for the
most variance within the dataset, and the original set of six sensors could be reduced to two
sensors (P50 and P52) that enabled accurate discrimination of 10 metal ions using LDA. The
array’s versatility was explored by determining its ability to discriminate different complex
mixtures based on different concentrations of metal cations. Further studies of electrolyte
drinks identified the best combination of sensors (P50, P51 and P53) to achieve 100%
classification accuracy in these complex samples.
 N N

HO HO

P49 P50

N
N
HO
OH
R R

P51

R1 R1 N
N

HO OH
R R n
R R

P52, n = 1
P53, n = 3
P54, n = 4
R= R1 =

Figure 16. Structures of fluorene and pyrene sensors, P49-P54, modified with 8-
hydroxyquinoline. Reproduced with permission of the Royal Society of Chemistry, from Chem.
Commun. 2007, 7345 (36), 3708–3710; permission conveyed through Copyright Clearance Center, Inc.
Copyright 2007 Royal Society of Chemistry.

Suslick and colleagues have developed numerous colorimetric sensor arrays based on
metalloporphyrins and other colorimetric dyes to discriminate diverse libraries of analytes.10 A
notable strategy involves the immobilization of a range of chemically responsive dyes onto an
ormosil matrix and printing onto polyethylene terephthalate (PET) films.32,133–135 The printed
array is exposed to each analyte and a before and after image taken using a flatbed scanner.
A difference map using RGB values can be generated from the difference between these
images, creating a fingerprint response for each analyte. This approach has been
demonstrated to discriminate a wide range of analytes, using a combination of different dye
classes, namely metalloporphyrins, pH indicators, vapochromic dyes and redox responsive
metal salts. A 36-element sensor array using these dye classes was printed and tested against
19 different toxic industrial chemicals (Figure 17).133,134 HCA of the color-difference responses
correctly classified all 19 toxins with 100% accuracy. The sensitivity of the array system was
further investigated in a study testing a similar library of toxic, retaining 100% classification
accuracy with estimated limits of detection in the ppb range.135
Figure 17. Example of a colorimetric sensor array with various dyes and pigments printed on a
polyethylene terephthalate (PET) film. Reprinted (adapted) with permission from Anal. Chem. 2010, 82
(22), 9433–9440. Copyright 2010, American Chemical Society.

Suslick and colleagues also used their sensor array system to discriminate various natural sugars
and artificial sweeteners.32,136 In this case, an IDA strategy was utilized, harnessing boronate ester
formation between carbohydrate diols and boronic acids, which reduces the pH of the solution. 16
chemically responsive dyes were printed onto a PET film, and 3-nitrophenylboronic acid was included
in the buffer solution. The addition of sugars altered the pH of the solution, causing changes in the
colors of pH indicators in the array. The responses of 15 different sugars and sweeteners were
processed using HCA and all replicates were classified correctly with 100% accuracy.136 A subsequent
study was also able to accurately discriminate different teas infused with sweeteners, demonstrating a
potential real-world application for the sensor array.32
Recent work by Suslick and colleagues has yielded a hand-hand device reader containing
colorimetric sensor array cartridges to improve the speed and sensitivity of gaseous analyte
detection.137 Sensors are printed onto a polypropylene (or similar) film, mounted to the cartridge and
sealed. The cartridge is designed to be compatible for gaseous analytes by allowing a low-volume flow
path for exposure. A cartridge of this design was subsequently utilized for the discrimination of
aldehydes and ketones.138 Exploiting the rapid reactivity between gaseous analytes and the solid-state
sensor array elements, three amine-containing colorimetric dyes were printing onto the sensor array
under various acidic conditions. The reaction of the printed anilines and phenylhydrazines with
aldehydes and ketones led to imine formation and a subsequent change in UV-absorption. Color
difference maps illustrated a cross-reactive response between the dyes and analytes. Seven aldehydes
and eight ketones were measured at concentrations of both 25 and 0.5 ppm, and were easily
discriminated using HCA, with 99.4% classification accuracy. Analysis also showed that aliphatic
aldehydes were clustered separately from aromatic aldehydes. The same sensor array was also tested
with various complex alcoholic liquid samples to demonstrate its potential in food inspection and
security screening. After pre-oxidation of the alcohol analytes and exposure to the array, SVM analysis
was performed, demonstrating effective discrimination of 6 liquor samples with 100% cross-validated
accuracy. Additional work by Suslick and colleagues utilizing their rapid, gaseous detecting colorimetric
sensor array includes discrimination of air pollutants at risk of damaging sensitive artworks, other
alcoholic liquors and volatiles emitted during fruit ripening.139–141
Bueno et al. have recently developed a colorimetric sensor array immobilized on cellulose acetate
film to discriminate volatile amines.142 The amines studied were identified as present during food
spoilage and have the potential to give early warning signs for rotting food that may lead to food
poisoning. Five pH sensitive dyes were immobilized by mixing with cellulose acetate and drying, before
a small sample of each polymer-dye sensor was placed on a dish in a closed chamber. Exposure to
each analyte caused a pattern-based colorimetric response of the sensor array and a before and after
color map was captured. RGB values generated a unique pattern to discriminate between three amines
(triethylamine, isobutylamine, isopentylamine). PCA and HCA of these responses revealed distinct
clusters of each amine and no misclassifications, with successful discrimination achieved of samples
at concentrations of 5 ppm, 2.5 ppm and 1 ppm. The applicability of the system for food quality
applications was also demonstrated by testing in meat samples contaminated with each amine, again
illustrating separate clusters of each amine contaminant.
Polymer films provide clear benefits for the development of optical sensor arrays. Primarily, the
polymer matrix allows properties such as hydrophilicity and lipophilicity to be easily adjusted to suit the
application without affecting the choice of indicator. Additionally, solid-state arrays enable convenient
detection of both solution-phase and gaseous analytes and often avoid time-consuming solvent
handling and pipetting of sensors required for traditional solution-based sensor systems. Finally,
compared to previous polymetric materials, polymer film sensor arrays have seen the greatest progress
towards the development of sensor array cartridges and portable and accessible devices to detect and
analyze responses.

4. Conclusion and future directions

Whilst still a developing field, macromolecular optical sensor arrays have already been utilized in a
number of successful sensing applications. In particular, significant progress has been made in sensing
challenging biological systems such as bacteria, proteins and cellular surfaces. This review has
highlighted the benefits of the cross-reactive array technique for these systems, specifically the different
strategies that have been used to achieve cross-reactive recognition and the interpretation of these
results using multivariate statistical analysis. We have focused on the use of polymers as a sensing
element to emphasize the benefits of polymeric materials, both for their robust functionalization and
their tunability of size, allowing for large diverse recognition surfaces and thus a capacity for good cross-
reactivity. Finally, we have explored the optical detection strategies utilized in these systems, through
either the inherent absorption or emission properties of the material, or the covalent attachment of
chromophores/fluorophores. Importantly, it is the optimal combination of these aspects that we believe
is responsible for the successful discrimination capacity within these sensor systems and the broad
scope of analytes scrutinized in this review.
Unsurprisingly, in most cases the addition of more sensors gave better discrimination of analytes.
However, a new generation of cross-reactive arrays has begun to emerge, involving other strategies to
achieve diverse recognition capabilities. A common strategy involves using the same sensor in multiple
channels and altering the pH, solvent, ionic strength and wavelengths of absorbance, excitation and
emission. These array simplification strategies often improve the applicability of the system, by reducing
the sensor-to-analyte ratio and decreasing the time spent designing and synthesizing additional
sensors. Whilst a seemingly desirable strategy, it is crucial that these simplification steps do not limit
the compatibility of the system in end-user applications. For example, a benefit of optical sensors within
array systems lies in the ability to select a number of wavelengths for a single sensing element and has
been successfully demonstrated in a number of systems. However, if the absorbance and emission
wavelengths of these arrays do not match the common filters in optical instruments, it will be difficult to
translate these systems beyond high-specification laboratory instruments. Similarly, altering the ionic
strength of a solution to identify different proteins may be useful for training a sensor array, but would
be of limited use for identifying unknown analytes in the likely event that the ionic strength of the protein
solution is also unknown. IDAs are a popular approach in sensing applications and the examples
addressed in all sections of the review highlight the benefits of including them in an array-based sensing
approach. In particular, the marriage of the techniques provides added advantages for IDAs by
discriminating between strongly and weakly displacing analytes at multiple concentrations.
While good discrimination was achieved using purely polymeric materials as sensing elements, there
are numerous examples of systems that incorporate composite materials such as proteins,
nanoparticles, peptides, mesoporous silica and photonic crystals, together with polymeric materials. In
many of these instances, very good discrimination was achieved and allowed for more flexibility during
the design process. The success of these systems suggests that polymer composite materials present
an exciting future direction for macromolecular sensor arrays.
The field of optical sensing, both selective sensors and sensor arrays, is dominated by the use of
sensors decorated with only one recognition/reporter system. More recent work has highlighted the
promise of tethering multiple sensors to a single scaffold, with small molecule systems reported that
distinguish different drug molecules,143 a broad range of inorganic and organic analytes144 and β-
amyloid aggregates.145 Since macromolecular systems lend themselves to multifunctionalization, this
strategy is likely to be promising for the future development of sensor arrays that comprise fewer
discrete sensing elements.
Multivariate statistical techniques are the standard approach to analyze and interpret array responses
to analyte addition. PCA, and more recently LDA, are commonly used to evaluate systems and classify
analytes. Most examples display data in reduced dimensionality 2D and 3D score plots, allowing for
easy interpretation of the results. To date, there are few examples of macromolecular sensor arrays
harnessing more sophisticated multivariate techniques such as SVMs and ANNs, presenting exciting
opportunities for progress in array development. A few key examples have presented further
interrogation of the analysis by looking at loading plots, building an understanding of how sensors are
responding to analytes. Whilst the hypothesis-free approach to the generation of sensing arrays allows
for a more general interpretation of results, further examination of the data output and better
understanding of sensor-analyte interactions may pave the way for the next generation of sensors.
Despite the benefits of macromolecular optical sensing arrays, there remain other challenges and
limitations in the studies discussed above that must to be addressed to access the true potential of
these systems, as is the case for all sensor arrays. Primarily, when designing and evaluating an array
system, it is crucial that the end-user and application of the system are considered. Whilst all the studies
we examined identified a key application for the sensing system, array systems which move out of the
lab, and conduct analysis in clinical and environmental samples are the most impressive and highlight
the necessity for the successful translation of these systems into the real world. Generally, sensing
platforms based on optical detection strategies may require multiple sample preparation or clean-up
steps, which increase the complexity of the analytical protocol, and may be sensitive to background
interferants. Systems that rely on colorimetric responses have the advantage of easy interpretation by
untrained users, as demonstrated by the success of lateral flow testing for pregnancy. Other optical-
based detection strategies, such as fluorescence, typically display far greater sensitivity but require a
more complex experimental setup and an increased level of user skill in interpretation of results.
Consideration of these factors is important if the platform is to be used beyond the confines of the
laboratory.
A great challenge of detection within clinical samples is the presence of different background
interferents. For systems to be practically useful, it is important to explore discrimination beyond ‘spiked’
lab samples in a controlled environment and investigate if the system is robust enough to function in
the presence of these contaminants and across the variable nature of clinical samples.
It is key that the sensing ability matches the application, for example having adequate limit of
detection that is relevant to real samples, such as concentration levels of toxic metal contaminants.89
Furthermore, the ability to distinguish not only different analytes, but high concentrations from low
concentrations is often relevant, and a few examples impressively illustrate this kind of testing.50,114
More broadly, the development of functional sensing platforms requires a broad range of expertise,
from fundamental (bio)chemical research to device and interface design, along with the need to
establish rigorous quality analysis and ensure compliance with regulatory frameworks. An interesting
discussion on the challenges of bringing biosensing devices to the market can be found in an excellent
article by Sia and coworkers.146
Finally, improvements in technology will certainly play a role in the future of array-based sensing. The
development of smart phone technology may provide a scope for improving the accessibility of sensor
array technology, however it is important to ensure the system is compatible with constantly evolving
technology. Instead, the development of portable devices for testing methods may be more applicable
for array-based sensing and have already shown some promise for colorimetric sensor arrays.

Acknowledgements

The authors would like to acknowledge the Westpac Scholars Trust for a Future Leaders Scholarship
(LM) and a Research Fellowship (EJN), the University of Sydney for a SOAR Fellowship (EJN) and the
Australian Government for a Research Training Program Scholarship (LM).
References:

(1) Hare, D. J.; New, E. J. On the Outside Looking in: Redefining the Role of Analytical Chemistry
in the Biosciences. Chem. Commun. 2016, 52 (58), 8918–8934.

(2) Pendin, D.; Greotti, E.; Lefkimmiatis, K.; Pozzan, T. Exploring Cells with Targeted Biosensors.
J. Gen. Physiol. 2016, 149 (1), 1–36.

(3) Carter, K. P.; Young, A. M.; Palmer, A. E. Fluorescent Sensors for Measuring Metal Ions in
Living Systems. Chem. Rev. 2014, 114 (8), 4564–4601.

(4) Kaur, A.; New, E. J. Bioinspired Small-Molecule Tools for the Imaging of Redox Biology. Acc.
Chem. Res. 2019, 52 (3), 623–632.

(5) Ong, J. X.; Lim, C. S. Q.; Le, H. Van; Ang, W. H. A Ratiometric Fluorescent Probe for Cisplatin:
Investigating the Intracellular Reduction of Platinum(IV) Prodrug Complexes. Angew. Chem. Int. Ed.
2019, 58 (1), 164–167.

(6) Wu, D.; Sedgwick, A. C.; Gunnlaugsson, T.; Akkaya, E. U.; Yoon, J.; James, T. D. Fluorescent
Chemosensors: The Past, Present and Future. Chem. Soc. Rev. 2017, 46 (23), 7105–7123.

(7) Ibañez, G. A.; Escandar, G. M. Fluorescence and Phosphorescence Chemical Sensors Applied
to Water Samples. In Smart Sensors for Real-Time Water Quality Monitoring. Smart Sensors,
Measurement and Instrumentation; Springer Berlin Heidelberg, 2013; pp 45–65.

(8) Mahon, C. S.; Fulton, D. A. Mimicking Nature with Synthetic Macromolecules Capable of
Recognition. Nat. Chem. 2014, 6 (8), 665–672.

(9) New, E. J. Harnessing the Potential of Small Molecule Intracellular Fluorescent Sensors. ACS
Sensors 2016, 1 (4), 328–333.

(10) Li, Z.; Askim, J. R.; Suslick, K. S. The Optoelectronic Nose: Colorimetric and Fluorometric
Sensor Arrays. Chem. Rev. 2019, 119 (1), 231–292.

(11) Anzenbacher, P.; Lubal, P.; Buek, P.; Palacios, M. A.; Kozelkova, M. E. A Practical Approach
to Optical Cross-Reactive Sensor Arrays. Chem. Soc. Rev. 2010, 39 (10), 3954–3979.

(12) You, L.; Zha, D.; Anslyn, E. V. Recent Advances in Supramolecular Analytical Chemistry Using
Optical Sensing. Chem. Rev. 2015, 115 (15), 7840–7892.

(13) Huynh, T. P.; Kutner, W. Molecularly Imprinted Polymers as Recognition Materials for
Electronic Tongues. Biosens. Bioelectron. 2015, 74, 856–864.

(14) Ramanathan, K.; Danielsson, B. Principles and Applications of Thermal Biosensors. Biosens.
Bioelectron. 2001, 16 (6), 417–423.

(15) Tan, C.; Pinto, M. R.; Schanze, K. S. Photophysics, Aggregation and Amplified Quenching of
a Water-Soluble Poly(Phenylene Ethynylene). Chem. Commun. 2002, 2 (5), 446–447.

(16) Alvarez, A.; Costa-Fernández, J. M.; Pereiro, R.; Sanz-Medel, A.; Salinas-Castillo, A.
Fluorescent Conjugated Polymers for Chemical and Biochemical Sensing. Trends Anal. Chem. 2011,
30 (9), 1513–1525.

(17) Greene, N. T.; Morgan, S. L.; Shimizu, K. D. Molecularly Imprinted Polymer Sensor Arrays.
Chem. Commun. 2004, 1172–1173.

(18) Miranda, O. R.; You, C. C.; Phillips, R.; Kim, I. B.; Ghosh, P. S.; Bunz, U. H. F.; Rotello, V. M.
Array-Based Sensing of Proteins Using Conjugated Polymers. J. Am. Chem. Soc. 2007, 129 (32),
9856–9857.
 (19) 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.

(20) Gooding, J. J.; Gaus, K. Single-Molecule Sensors: Challenges and Opportunities for
Quantitative Analysis. Angew. Chem. Int. Ed. 2016, 55 (38), 11354–11366.

(21) Greene, N. T.; Shimizu, K. D. Colorimetric Molecularly Imprinted Polymer Sensor Array Using
Dye Displacement. J. Am. Chem. Soc. 2005, 127 (15), 5695–5700.

(22) Nelson, T. L.; O’Sullivan, C.; Greene, N. T.; Maynor, M. S.; Lavigne, J. J. Cross-Reactive
Conjugated Polymers: Analyte-Specific Aggregative Response for Structurally Similar Diamines. J. Am.
Chem. Soc. 2006, 128 (17), 5640–5641.

(23) New, E. J. Harnessing the Potential of Small Molecule Intracellular Fluorescent Sensors. ACS
Sensors 2016, 1, 328–333.

(24) Albert, K. J.; Lewis, N. S.; Schauer, C. L.; Sotzing, G. A.; Stitzel, S. E.; Vaid, T. P.; Walt, D. R.
Cross-Reactive Chemical Sensor Arrays. Chem. Rev. 2000, 100 (7), 2595–2626.

(25) Svechkarev, D.; Sadykov, M. R.; Bayles, K. W.; Mohs, A. M. Ratiometric Fluorescent Sensor
Array as a Versatile Tool for Bacterial Pathogen Identification and Analysis. ACS Sensors 2018, 3 (3),
700–708.

(26) Aron, A. T.; Ramos-Torres, K. M.; Cotruvo, J. A.; Chang, C. J. Recognition- and Reactivity-
Based Fluorescent Probes for Studying Transition Metal Signaling in Living Systems. Acc. Chem. Res.
2015, 48 (8), 2434–2442.

(27) Bruemmer, K. J.; Crossley, S. W. M.; Chang, C. J. Activity-Based Sensing: A Synthetic Methods
Approach for Selective Molecular Imaging and Beyond. Angew. Chem. Int. Ed. 2020, 59 (33), 13734–
13762.

(28) Nguyen, B. T.; Anslyn, E. V. Indicator-Displacement Assays. Coord. Chem. Rev. 2006, 250
(23–24), 3118–3127.

(29) Sedgwick, A. C.; Brewster, J. T.; Wu, T.; Feng, X.; Bull, S. D.; Qian, X.; Sessler, J. L.; James,
T. D.; Anslyn, E. V.; Sun, X. Indicator Displacement Assays (IDAs): The Past, Present and Future.
Chem. Soc. Rev. 2020, DOI:10.1039/C9CS00538B.

(30) Buryak, A.; Severin, K. A Chemosensor Array for the Colorimetric Identification of 20 Natural
Amino Acids. J. Am. Chem. Soc. 2005, 127 (11), 3700–3701.

(31) Schiller, A.; Wessling, R. A.; Singaram, B. A Fluorescent Sensor Array for Saccharides Based
on Boronic Acid Appended Bipyridinium Salts. Angew. Chem. Int. Ed. 2007, 46 (34), 6457–6459.

(32) Musto, C. J.; Lim, S. H.; Suslick, K. S. Colorimetric Detection and Identification of Natural and
Artificial Sweeteners. Anal. Chem. 2009, 81 (15), 6526–6533.

(33) Lavigne, J. J.; Anslyn, E. V. Sensing a Paradigm Shift in the Field of Molecular Recognition:
From Selective to Differential Receptors. Angew. Chem. Int. Ed. 2001, 40, 3118–3130.

(34) Geng, Y.; Peveler, W. J.; Rotello, V. M. Array-Based “Chemical Nose” Sensing in Diagnostics
and Drug Discovery. Angew. Chem. Int. Ed. 2019, 5190–5200.

(35) Peveler, W. J.; Landis, R. F.; Yazdani, M.; Day, J. W.; Modi, R.; Carmalt, C. J.; Rosenberg, W.
M.; Rotello, V. M. A Rapid and Robust Diagnostic for Liver Fibrosis Using a Multichannel Polymer
Sensor Array. Adv. Mater. 2018, 30 (28), 2–7.

(36) Han, J.; Ma, C.; Wang, B.; Bender, M.; Bojanowski, M.; Hergert, M.; Seehafer, K.; Herrmann,
A.; Bunz, U. H. F. A Hypothesis-Free Sensor Array Discriminates Whiskies for Brand, Age, and Taste.
Chem 2017, 2 (6), 817–824.
 (37) Keshri, G.; Magan, N. Detection and Differentiation between Mycotoxigenic and Non-
Mycotoxigenic Strains of Two Fusarium Spp. Using Volatile Mycotoxigenic Strains of Two Production
Profiles and Hydrolytic Enzymes. J. Appl. Microbiol. 2000, 89 (5), 825–833.

(38) Severin, E. J.; Doleman, B. J.; Lewis, N. S. An Investigation of the Concentration Dependence
and Response to Analyte Mixtures of Carbon Black/Insulating Organic Polymer Composite Vapor
Detectors. Anal. Chem. 2000, 72 (4), 658–668.

(39) Huang, W.; Smarsly, E.; Han, J.; Bender, M.; Seehafer, K.; Wacker, I.; Schröder, R. R.; Bunz,
U. H. F. Truxene-Based Hyperbranched Conjugated Polymers: Fluorescent Micelles Detect Explosives
in Water. ACS Appl. Mater. Interfaces 2017, 9 (3), 3068–3074.

(40) Huang, W.; Bender, M.; Seehafer, K.; Wacker, I.; Schröder, R. R.; Bunz, U. H. F. A
Tetraphenylethene-Based Polymer Array Discriminates Nitroarenes. Macromolecules 2018, 51 (4),
1345–1350.

(41) Maynor, M. S.; Nelson, T. L.; O’Sullivan, C.; Lavigne, J. J. A Food Freshness Sensor Using the
Multistate Response from Analyte-Lnduced Aggregation of a Cross-Reactive Poly(Thiophene). Org.
Lett. 2007, 9 (17), 3217–3220.

(42) Arrieta, Á. A.; Rodríguez-Méndez, M. L.; de Saja, J. A.; Blanco, C. A.; Nimubona, D. Prediction
of Bitterness and Alcoholic Strength in Beer Using an Electronic Tongue. Food Chem. 2010, 123 (3),
642–646.

(43) Smith, D. G.; Mitchell, L.; New, E. J. Pattern Recognition of Toxic Metal Ions Using a Single-
Probe Thiocoumarin Array. Analyst 2019, 144 (1), 230–236.

(44) Bowyer, A. A.; Shen, C.; New, E. J. A Fluorescent Three-Sensor Array for Heavy Metals in
Environmental Water Sources. Analyst 2020, 145 (4), 1195–1201.

(45) Mallet, A. M.; Davis, A. B.; Davis, D. R.; Panella, J.; Wallace, K. J.; Bonizzoni, M. A Cross
Reactive Sensor Array to Probe Divalent Metal Ions. Chem. Commun. 2015, 51 (95), 16948–16951.

(46) Kim, I. B.; Dunkhorst, A.; Bunz, U. H. F. Nonspecific Interactions of a Carboxylate-Substituted

PPE with Proteins. A Cautionary Tale for Biosensor Applications. Langmuir 2005, 21 (17), 7985–7989.

(47) De, M.; Rana, S.; Akpinar, H.; Miranda, O. R.; Arvizo, R. R.; Bunz, U. H. F.; Rotello, V. M.
Sensing of Proteins in Human Serum Using Conjugates of Nanoparticles and Green Fluorescent
Protein. Nat. Chem. 2009, 1 (6), 461–465.

(48) Rana, S.; Elci, S. G.; Mout, R.; Singla, A. K.; Yazdani, M.; Bender, M.; Bajaj, A.; Saha, K.; Bunz,
U. H. F.; Jirik, F. R.; Rotello, V. M. Ratiometric Array of Conjugated Polymers-Fluorescent Protein
Provides a Robust Mammalian Cell Sensor. J. Am. Chem. Soc. 2016, 138 (13), 4522–4529.

(49) Sener, G.; Uzun, L.; Denizli, A. Colorimetric Sensor Array Based on Gold Nanoparticles and
Amino Acids for Identification of Toxic Metal Ions in Water. ACS Appl. Mater. Interfaces 2014, 6 (21),
18395–18400.

(50) You, C. C.; Miranda, O. R.; Gider, B.; Ghosh, P. S.; Kim, I. B.; Erdogan, B.; Krovi, S. A.; Bunz,
U. H. F.; Rotello, V. M. Detection and Identification of Proteins Using Nanoparticle-Fluorescent Polymer
“Chemical Nose” Sensors. Nat. Nanotechnol. 2007, 2 (5), 318–323.

(51) Fahimi-Kashani, N.; Hormozi-Nezhad, M. R. Gold Nanorod-Based Chrono-Colorimetric Sensor

Arrays: A Promising Platform for Chemical Discrimination Applications. ACS Omega 2018, 3 (2), 1386–
1394.

(52) Bigdeli, A.; Ghasemi, F.; Golmohammadi, H.; Abbasi-Moayed, S.; Nejad, M. A. F.; Fahimi-
Kashani, N.; Jafarinejad, S.; Shahrajabian, M.; Hormozi-Nezhad, M. R. Nanoparticle-Based Optical
Sensor Arrays. Nanoscale 2017, 9 (43), 16546–16563.
 (53) Abbasi-Moayed, S.; Hormozi-Nezhad, M. R.; Maaza, M. A Multichannel Single-Well Sensor
Array for Rapid and Visual Discrimination of Catecholamine Neurotransmitters. Sensors Actuators, B
Chem. 2019, 296 (June), 126691.

(54) Anslyn, E. V. Supramolecular Analytical Chemistry. J. Org. Chem. 2007, 72 (3), 687–699.

(55) Bunz, U. H. F. Poly(Aryleneethynylene)s: Syntheses, Properties, Structures, and Applications.

Chem. Rev. 2000, 100 (4), 1605–1644.

(56) Han, J.; Wang, B.; Bender, M.; Kushida, S.; Seehafer, K.; Bunz, U. H. F.
Poly(Aryleneethynylene) Tongue That Identifies Nonsteroidal Anti-Inflammatory Drugs in Water: A Test
Case for Combating Counterfeit Drugs. ACS Appl. Mater. Interfaces 2017, 9 (1), 790–797.

(57) Ngernpimai, S.; Geng, Y.; Makabenta, J. M.; Landis, R. F.; Keshri, P.; Gupta, A.; Li, C. H.;
Chompoosor, A.; Rotello, V. M. Rapid Identification of Biofilms Using a Robust Multichannel Polymer
Sensor Array. ACS Appl. Mater. Interfaces 2019, 11 (12), 11202–11208.

(58) Han, J.; Cheng, H.; Wang, B.; Braun, M. S.; Fan, X.; Bender, M.; Huang, W.; Domhan, C.; Mier,
W.; Lindner, T.; Seehafer, K.; Wink, M.; Bunz, U. H. F. A Polymer/Peptide Complex-Based Sensor Array
That Discriminates Bacteria in Urine. Angew. Chem. Int. Ed. 2017, 56 (48), 15246–15251.

(59) Kapf, A.; Albrecht, M. Discrimination of Proteins through Interaction with Pyrene-Labelled
Polymer Aggregates. J. Mater. Chem. B 2018, 6 (41), 6599–6606.

(60) Smith, D. G.; Topolnicki, I. L.; Zwicker, V. E.; Jolliffe, K. A.; New, E. J. Fluorescent Sensing
Arrays for Cations and Anions. Analyst 2017, 3549–3563.

(61) Cunningham, P.; Cord, M.; Delany, S. J. Machine Learning Techniques for Multimedia,
Springer, Berlin, Heidelberg, 2008.

(62) Hair, J. F.; Black, B.; Babin, B.; Anderson, R. E.; Tatham, R. L. Mutivariate Data Analysis, 6th
ed.; Pearson, 2006.

(63) Flury, B. K.; Riedwyl, H. Standard Distance in Univariate and Multivariate Analysis. Am. Stat.
1986, 40 (3), 249–251.

(64) Frades, I.; Matthiesen, R. Overview on Techniques in Cluster Analysis; Humana Press, 2010.

(65) Landau, S.; Chis Ster, I. Cluster Analysis: Overview. Int. Encycl. Educ. 2010, No. December,
72–83.

(66) Jollife, I. T.; Cadima, J. Principal Component Analysis: A Review and Recent Developments.
Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2016, 374 (2065), 20150202.

(67) Wold, S.; Esbensen, K.; Geladi, P. Principal Component Analysis. Chemom. Intell. Lab. Syst.
1987, 2 (1–3), 37–52.

(68) Ringnér, M. What Is Principal Component Analysis? Nat. Biotechnol. 2008, 26 (3), 303–304.

(69) Stewart, S.; Ivy, M. A.; Anslyn, E. V. The Use of Principal Component Analysis and Discriminant
Analysis in Differential Sensing Routines. Chem. Soc. Rev. 2014, 43 (1), 70–84.

(70) Izenman, A. J. Linear Discriminant Analysis. In Modern Multivariate Statistical Techniques:

Regression, Classification, and Manifold Learning; Springer New York: New York, NY, 2008; pp 237–
280.

(71) Webb, A. R.; Copsey, K. D. Linear Discriminant Analysis. In Statistical Pattern Recognition;
John Wiley & Sons, 2011; pp 221–273.

(72) Molinaro, A. M.; Simon, R.; Pfeiffer, R. M. Prediction Error Estimation: A Comparison of
Resampling Methods. Bioinformatics 2005, 21 (15), 3301–3307.
 (73) Bajaj, A.; Miranda, O. R.; Phillips, R.; Kim, I. B.; Jerry, D. J.; Bunz, U. H. F.; Rotello, V. M. Array-
Based Sensing of Normal, Cancerous, and Metastatic Cells Using Conjugated Fluorescent Polymers.
J. Am. Chem. Soc. 2010, 132 (3), 1018–1022.

(74) Leech, N. L.; Barrett, K. C.; Morgan, G. A. SPSS for Intermediate Statistics: Use and
Interpretation.; Psychology Press, 2005.

(75) Tropp, J.; Ihde, M. H.; Williams, A. K.; White, N. J.; Eedugurala, N.; Bell, N. C.; Azoulay, J. D.;
Bonizzoni, M. A Sensor Array for the Discrimination of Polycyclic Aromatic Hydrocarbons Using
Conjugated Polymers and the Inner Filter Effect. Chem. Sci. 2019, 10 (44), 10247–10255.

(76) Kung, S. Y. Kernel Methods and Machine Learning; Cambridge University Press: Cambridge,
2014.

(77) Krogh, A. What Are Artificial Neural Networks? Nat. Biotechnol. 2008, 26 (2), 195–197.

(78) Burns, J. A.; Whitesides, G. M. Feed-Forward Neural Networks in Chemistry: Mathematical

Systems for Classification and Pattern Recognition. Chem. Rev. 1993, 93 (8), 2583–2601.

(79) Priddy, K. L.; Keller, P. E. Artificial Neural Networks: An Introduction; SPIE press: Belingham,
Washington USA, 2005.

(80) Bishop, C. M. Neural Networks for Pattern Recognition; Oxford University Press: Oxford, 1995.

(81) Pei, Q. Light-Emitting Polymers. Mater. Matters 2007, 2.3 (26), 1–9.

(82) Thomas, S. W.; Joly, G. D.; Swager, T. M. Chemical Sensors Based on Amplifying Fluorescent
Conjugated Polymers. Chem. Rev. 2007, 107 (4), 1339–1386.

(83) Woodka, M. D.; Schnee, V. P.; Polcha, M. P. Fluorescent Polymer Sensor Array for Detection
and Discrimination of Explosives in Water. Anal. Chem. 2010, 82 (23), 9917–9924.

(84) Wu, D.; Schanze, K. S. Protein Induced Aggregation of Conjugated Polyelectrolytes Probed
with Fluorescence Correlation Spectroscopy: Application to Protein Identification. ACS Appl. Mater.
Interfaces 2014, 6 (10), 7643–7651.

(85) Freudenberg, J.; Hinkel, F.; Jänsch, D.; Bunz, U. H. F. Chemical Tongues and Noses Based
upon Conjugated Polymers. Top. Curr. Chem. 2017, 375 (4), 67.

(86) Blayney, A. J.; Perepichka, I. F.; Wudl, F.; Perepichka, D. F. Advances and Challenges in the
Synthesis of Poly(p-Phenylene Vinylene)-Based Polymers. Isr. J. Chem. 2014, 54 (5–6), 674–688.

(87) Perepichka, I. F.; Perepichka, D. F.; Meng, H.; Wudl, F. Light-Emitting Polythiophenes. Adv.
Mater. 2005, 17 (19), 2281–2305.

(88) Lee, S. H.; Kömürlü, S.; Zhao, X.; Jiang, H.; Moriena, G.; Kleiman, V. D.; Schanze, K. S. Water-
Soluble Conjugated Polyelectrolytes with Branched Polyionic Side Chains. Macromolecules 2011, 44
(12), 4742–4751.

(89) Wu, Y.; Tan, Y.; Wu, J.; Chen, S.; Chen, Y. Z.; Zhou, X.; Jiang, Y.; Tan, C. Fluorescence Array-
Based Sensing of Metal Ions Using Conjugated Polyelectrolytes. ACS Appl. Mater. Interfaces 2015, 7
(12), 6882–6888.

(90) Yao, Z.; Feng, X.; Hong, W.; Li, C.; Shi, G. A Simple Approach for the Discrimination of
Nucleotides Based on a Water-Soluble Polythiophene Derivative. Chem. Commun. 2009, 405 (31),
4696–4698.

(91) Cooper, J. R.; Bloom, F. E.; Roth, R. H. The Biochemical Basis of Neuropharmacology; Oxford
University Press, USA, 2003.

(92) Nelson, T. L.; Tran, I.; Ingallinera, T. G.; Maynor, M. S.; Lavigne, J. J. Multi-Layered Analyses
Using Directed Partitioning to Identify and Discriminate between Biogenic Amines. Analyst 2007, 132
(10), 1024–1030.
 (93) Tropp, J.; Ihde, M. H.; Crater, E. R.; Bell, N. C.; Bhatta, R.; Johnson, I. C.; Bonizzoni, M.;
Azoulay, J. D. A Sensor Array for the Nanomolar Detection of Azo Dyes in Water. ACS Sensors 2020,
5 (6), 1541–1547.

(94) Wu, J.; Tan, C.; Chen, Z.; Chen, Y. Z.; Tan, Y.; Jiang, Y. Fluorescence Array-Based Sensing
of Nitroaromatics Using Conjugated Polyelectrolytes. Analyst 2016, 141 (11), 3242–3245.

(95) Han, J.; Bender, M.; Hahn, S.; Seehafer, K.; Bunz, U. H. F. Polyelectrolyte Complexes Formed
from Conjugated Polymers: Array-Based Sensing of Organic Acids. Chem. - A Eur. J. 2016, 22 (10),
3230–3233.

(96) Han, J.; Wang, B.; Bender, M.; Seehafer, K.; Bunz, U. H. F. Water-Soluble Poly(p-
Aryleneethynylene)s: A Sensor Array Discriminates Aromatic Carboxylic Acids. ACS Appl. Mater.
Interfaces 2016, 8 (31), 20415–20421.

(97) Wang, B.; Han, J.; Bender, M.; Seehafer, K.; Bunz, U. H. F. Array-Based Sensing of Explosives
by Water-Soluble Poly(p-Phenyleneethynylene)s. Macromolecules 2017, 50 (11), 4126–4131.

(98) Huang, W.; Seehafer, K.; Bunz, U. H. F. Discrimination of Flavonoids by a Hypothesis Free
Sensor Array. ACS Appl. Polym. Mater. 2019, 1 (6), 1301–1307.

(99) Wang, B.; Han, J.; Bojanowski, N. M.; Bender, M.; Ma, C.; Seehafer, K.; Herrmann, A.; Bunz,
U. H. F. An Optimized Sensor Array Identifies All Natural Amino Acids. ACS Sensors 2018, 3 (8), 1562–
1568.

(100) Wang, B.; Han, J.; Ma, C.; Bender, M.; Seehafer, K.; Herrmann, A.; Bunz, U. H. F. A Simple
Optoelectronic Tongue Discriminates Amino Acids. Chem. - A Eur. J. 2017, 23 (51), 12471–12474.

(101) Kim, I. B.; Han, M. H.; Phillips, R. L.; Samanta, B.; Rotello, V. M.; Zhang, Z. J.; Bunz, U. H. F.
Nano-Conjugate Fluorescence Probe for the Discrimination of Phosphate and Pyrophosphate. Chem.
- A Eur. J. 2009, 15 (2), 449–456.

(102) Bojanowski, N. M.; Bender, M.; Seehafer, K.; Bunz, U. H. F. Discrimination of Saccharides by
a Simple Array. Chem. - A Eur. J. 2017, 23 (50), 12253–12258.

(103) Han, J.; Wang, B.; Bender, M.; Pfisterer, J.; Huang, W.; Seehafer, K.; Yazdani, M.; Rotello, V.
M.; Rotello, C. M.; Bunz, U. H. F. Fingerprinting Antibiotics with PAE-Based Fluorescent Sensor Arrays.
Polym. Chem. 2017, 8 (17), 2723–2732.

(104) Han, J.; Wang, B.; Bender, M.; Seehafer, K.; Bunz, U. H. F. Poly(p-Phenyleneethynylene)-
Based Tongues Discriminate Fruit Juices. Analyst 2017, 142 (3), 537–543.

(105) Bojanowski, N. M.; Hainer, F.; Bender, M.; Seehafer, K.; Bunz, U. H. F. An Optical Sensor Array
Discriminates Syrups and Honeys. Chem. - A Eur. J. 2018, 24 (17), 4255–4258.

(106) Kim, J.; Swager, T. M. Control of Conformational and Interpolymer Effects in Conjugated
Polymers. Nature 2001, 411 (6841), 1030–1034.

(107) Tomita, S.; Ishihara, S.; Kurita, R. Environment-Sensitive Turn-On Fluorescent Polyamino Acid:
Fingerprinting Protein Populations with Post-Translational Modifications. ACS Appl. Mater. Interfaces
2017, 9 (27), 22970–22976.

(108) Kim, J.; Mosior, M.; Chung, L. A.; Wu, H.; McLaughlin, S. Binding of Peptides with Basic
Residues to Membranes Containing Acidic Phospholipids. Biophys. J. 1991, 60 (1), 135–148.

(109) Mislick, K. A.; Baldeschwieler, J. D. Evidence for the Role of Proteoglycans in Cation-Mediated
Gene Transfer. Proc. Natl. Acad. Sci. U. S. A. 1996, 93 (22), 12349–12354.

(110) Sugai, H.; Tomita, S.; Ishihara, S.; Kurita, R. One-Component Array Based on a Dansyl-
Modified Polylysine: Generation of Differential Fluorescent Signatures for the Discrimination of Human
Cells. ACS Sensors 2019, 4 (4), 827–831.
 (111) Castera, L.; Yuen Chan, H. L.; Arrese, M.; Afdhal, N.; Bedossa, P.; Friedrich-Rust, M.; Han, K.
H.; Pinzani, M. EASL-ALEH Clinical Practice Guidelines: Non-Invasive Tests for Evaluation of Liver
Disease Severity and Prognosis. J. Hepatol. 2015, 63 (1), 237–264.

(112) Geng, Y.; Hardie, J.; Landis, R. F.; Mas-Rosario, J. A.; Chattopadhyay, A. N.; Keshri, P.; Sun,
J.; Rizzo, E. M.; Gopalakrishnan, S.; Farkas, M. E.; Rotello, V. M. High-Content and High-Throughput
Identification of Macrophage Polarization Phenotypes. Chem. Sci. 2020, 10, 8231–8239.

(113) Mallet, A. M.; Liu, Y.; Bonizzoni, M. An Off-the-Shelf Sensing System for Physiological
Phosphates. Chem. Commun. 2014, 50 (39), 5003–5006.

(114) Liu, Y.; Bonizzoni, M. A Supramolecular Sensing Array for Qualitative and Quantitative Analysis
of Organophosphates in Water. J. Am. Chem. Soc. 2014, 136 (40), 14223–14229.

(115) Liang, X.; Bonizzoni, M. Boronic Acid-Modified Poly(Amidoamine) Dendrimers as Sugar-

Sensing Materials in Water. J. Mater. Chem. B 2016, 4 (18), 3094–3103.

(116) Liang, X.; Trentle, M.; Kozlovskaya, V.; Kharlampieva, E.; Bonizzoni, M. Carbohydrate Sensing
Using Water-Soluble Poly(Methacrylic Acid)-Co-3(Acrylamido)Phenylboronic Acid Copolymer. ACS
Appl. Polym. Mater. 2019, 1 (6), 1341–1349.

(117) Sambrook, M. R.; Notman, S. Supramolecular Chemistry and Chemical Warfare Agents: From
Fundamentals of Recognition to Catalysis and Sensing. Chem. Soc. Rev. 2013, 42 (24), 9251–9267.

(118) Sun, X.; Zhai, W.; Fossey, J. S.; James, T. D. Boronic Acids for Fluorescence Imaging of
Carbohydrates. Chem. Commun. 2016, 52 (17), 3456–3469.

(119) Bernardi, A.; Jiménez-Barbero, J.; Casnati, A.; De Castro, C.; Darbre, T.; Fieschi, F.; Finne, J.;
Funken, H.; Jaeger, K. E.; Lahmann, M.; Lindhorst, T. K.; Marradi, M.; Messner, P.; Molinaro, A.;
Murphy, P. V; Nativi, C.; Oscarson, S.; Penadés, S.; Peri, F.; Pieters, R. J.; Renaudet, O.; Reymond, J.
L.; Richichi, B.; Rojo, J.; Sansone, F.; Schäffer, C.; Bruce Turnbull, W.; Velasco-Torrijos, T.; Vidal, S.;
Vincent, S.; Wennekes, T.; Zuilhof, H.; Imberty, A. Multivalent Glycoconjugates as Anti-Pathogenic
Agents. Chem. Soc. Rev. 2013, 42 (11), 4709–4727.

(120) Oshovsky, G. V.; Reinhoudt, D. N.; Verboom, W. Supramolecular Chemistry in Water. Angew.
Chem. Int. Ed. 2007, 46 (14), 2366–2393.

(121) Wulff, G. Molecular Imprinting in Cross-Linked Materials with the Aid of Molecular Templates -
A Way towards Artificial Antibodies. Angew. Chem. Int. Ed. 1995, 34 (17), 1812–1832.

(122) Chen, L.; Wang, X.; Lu, W.; Wu, X.; Li, J. Molecular Imprinting: Perspectives and Applications.
Chem. Soc. Rev. 2016, 45 (8), 2137–2211.

(123) Selvolini, G.; Marrazza, G. MIP-Based Sensors: Promising New Tools for Cancer Biomarker
Determination. Sensors (Basel) 2017, 17 (4), 718.

(124) Zimmerman, S. C.; Lemcoff, N. G. Synthetic Hosts via Molecular Imprinting- Are Universal
Synthetic Antibodies Realistically Possible? Chem. Commun. 2004, 4 (1), 5–14.

(125) Tan, J.; Wang, H. F.; Yan, X. P. A Fluorescent Sensor Array Based on Ion Imprinted
Mesoporous Silica. Biosens. Bioelectron. 2009, 24 (11), 3316–3321.

(126) Tan, J.; Wang, H. F.; Yan, X. P. Discrimination of Saccharides with a Fluorescent Molecular
Imprinting Sensor Array Based on Phenylboronic Acid Functionalized Mesoporous Silica. Anal. Chem.
2009, 81 (13), 5273–5280.

(127) Lee, J. D.; Hong, J. I. Two-Dimensional Sensor Array for Discrimination of Amines. Tetrahedron
Lett. 2013, 54 (22), 2890–2893.

(128) Wulff, G. The Role of Binding-Site Interactions in the Molecular Imprinting of Polymers. Trends
Biotechnol. 1993, 11 (3), 85–87.
 (129) 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.

(130) Anzenbacher, P.; Liu, Y.; Palacios, M. A.; Minami, T.; Wang, Z.; Nishiyabu, R. Leveraging
Material Properties in Fluorescence Anion Sensor Arrays: A General Approach. Chem. Eur. J. 2013,
19 (26), 8497–8506.

(131) Palacios, M. A.; Wang, Z.; Montes, V. A.; Zyryanov, G. V.; Hausch, B. J.; Jursíková, K.;
Anzenbacher, P. Hydroxyquinolines with Extended Fluorophores: Arrays for Turn-on and Ratiometric
Sensing of Cations. Chem. Commun. 2007, 7345 (36), 3708–3710.

(132) Palacios, M. A.; Wang, Z.; Montes, V. A.; Zyryanov, G. V.; Anzenbacher, P. Rational Design of
a Minimal Size Sensor Array for Metal Ion Detection. J. Am. Chem. Soc. 2008, 130 (31), 10307–10314.

(133) 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.

(134) Kemling, J. W.; Suslick, K. S. Nanoscale Porosity in Pigments for Chemical Sensing. Nanoscale
2011, 3 (5), 1971–1973.

(135) Feng, L.; Musto, C. J.; Kemling, J. W.; Lim, S. H.; Zhong, W.; Suslick, K. S. Colorimetric Sensor
Array for Determination and Identification of Toxic Industrial Chemicals. Anal. Chem. 2010, 82 (22),
9433–9440.

(136) Lim, S. H.; Musto, C. J.; Park, E.; Zhong, W.; Suslick, K. S. A Colorimetric Sensor Array for
Detection and Identification of Sugars. Org. Lett. 2008, 10 (20), 4405–4408.

(137) Askim, J. R.; Suslick, K. S. Hand-Held Reader for Colorimetric Sensor Arrays. Anal. Chem.
2015, 87 (15), 7810–7816.

(138) Li, Z.; Fang, M.; LaGasse, M. K.; Askim, J. R.; Suslick, K. S. Colorimetric Recognition of
Aldehydes and Ketones. Angew. Chem. Int. Ed. 2017, 56 (33), 9860–9863.

(139) LaGasse, M. K.; McCormick, K.; Li, Z.; Khanjian, H.; Schilling, M.; Suslick, K. S. Colorimetric
Sensor Arrays: Development and Application to Art Conservation. J. Am. Inst. Conserv. 2018, 57 (3),
127–140.

(140) Li, Z.; Suslick, K. S. A Hand-Held Optoelectronic Nose for the Identification of Liquors. ACS
Sensors 2018, 3 (1), 121–127.

(141) Li, Z.; Suslick, K. S. Colorimetric Sensor Array for Monitoring CO and Ethylene. Anal. Chem.
2019, 91 (1), 797–802.

(142) Bueno, L.; Meloni, G. N.; Reddy, S. M.; Paixão, T. R. L. C. Use of Plastic-Based Analytical
Device, Smartphone and Chemometric Tools to Discriminate Amines. RSC Adv. 2015, 5 (26), 20148–
20154.

(143) Rout, B.; Unger, L.; Armony, G.; Iron, M. A.; Margulies, D. Medication Detection by a
Combinatorial Fluorescent Molecular Sensor. Angew. Chem. Int. Ed. 2012, 51 (50), 12477–12481.

(144) Sarkar, T.; Selvakumar, K.; Motiei, L.; Margulies, D. Message in a Molecule. Nat. Commun.
2016, 7 (May), 1–9.

(145) Hatai, J.; Motiei, L.; Margulies, D. Analyzing Amyloid Beta Aggregates with a Combinatorial
Fluorescent Molecular Sensor. J. Am. Chem. Soc. 2017, 139 (6), 2136–2139.

(146) Chin, C. D.; Linder, V.; Sia, S. K. Commercialization of Microfluidic Point-of-Care Diagnostic
Devices. Lab Chip 2012, 12 (12), 2118–2134.