According to their applications, the planar protein microarrays have been classified in three main categories: analytical microarrays, reverse phase arrays (RPA) and functional microarrays (Figure 1) [7,14,21]. On the other hand, microspheres bead based systems should also be considered, which use different size or color beads as a support of the capture agent to analyze the sample. In such microarray format, flow cytometry is coupled in order to support the identification of each specific binding according to the size, color and mean fluorescence intensity of conjugated fluorochromes [14].

Protein microarray technology has been successfully applied in different biomedical areas (Table 1).

Regarding planar arrays, capture agents are immobilized in rows and columns creating a set of spots onto the microarray ready to be exposed to the test sample. The binding between the capture agent and the analyte can be detected by label or label-free detection methods such as fluorescence, chemiluminescence, mass spectrometry, radioactivity or electrochemistry [9,11,12].

Although there are several different capture agents, antibodies are the most common. However, new approximations are being developed, e.g. phage display or mRNA display are the most promising techniques, but also, highly specific oligonucleotides and aptamers (Table 2) [12]. Employing monoclonal antibodies in routine array production incurs excessive costs due to the hybridoma technology and their production. Therefore, phage display is an appropriate substitute according to the easier and cheaper methods to obtain and purify the antibody fragments (Table 2) [23,39,40].

On the other hand, highly purified recombinant proteins are needed to satisfy the huge demand for capture agents. Because of that, new high-throughput techniques have to be developed, as well as, improving the existence ones. Moreover, these specific proteins allow generating microarrays which are analyzed faster, with high affinity bindings, permitting more efficient screening and also, avoiding or, at least decreasing, the cross-reactivity [12,17].

The slides used to immobilize the capture agent are usually made of glass, but can also be made of plastic, metal or polymer membranes [12,41] Since microarrays were first developed, surfaces have evolved from polyvinylidene fluoride (PVDF) membranes to glass slides. The most significant challenge is obtaining a surface which can take a variety of protein structures and compositions while preserving the function, structure and binding of each single protein. Immobilization is important, but also the capability of accurately detecting the protein-protein interaction. That is why three types of surfaces have been developed, each with a specific immobilization protocol. The first category is two-dimensional (2D) plain glass slides, which are activated with a diversity of coupling chemistries such as aldehyde, epoxy or carboxylic esters and they bind proteins or antibodies through electrostatic interactions or the generation of covalent bonds. The strength of the binding and the low variation enables rapid evaporation of the liquid environment and it is suspected that this affects the three-dimensional structure due to the close protein surface contact. On the other hand, there are three-dimensional (3D) gels or membrane-coated surfaces, for instance, polyacrylamide, agarose and nitrocellulose. They bind the protein by adsorption and seem to better safeguard the initial conformation; nevertheless, they present variations in signal intensity. Finally, the third type is a mixture of the previous ones: showing a supra-molecular structure at the surface which is why it is not a 2D slide, nor has it a visible 3D structure [39].

The immobilization of proteins is usually completed using non-covalent protein surface interactions with hydrophobic (e.g., nitrocellulose and polystyrene) or positively charged (e.g., poly-lysine and aminosilane) surfaces [12]. This kind of method attachment determines a random orientation of the proteins onto the slide because of the passive adsorption [22]. Covalent attachment achieves employing chemically activated surfaces (e.g., aldehyde, epoxy or ester functional groups), also the attachment is achieved by specific bimolecular interactions (e.g., streptavidin-biotin, His-tag-nickel-chelates) [12]. The uniform orientation of the different proteins onto the chip surface can be achieved using nickel coated slides for the His-tag use or Streptavidin slides [22]. The tiny microspots on the slide surface are made using contact printing arrayers with tiny needles placing sub-nanoliter sample volumes directly on the surface. However, non-contact deposition technologies are used that apply capillaries or ink-jet technology to deposit nanoliter-picoliter droplets onto the surface [12]. Immobilizing a protein onto a slide necessitates preservation of the conformation and the function of the protein, as well as, the binding capacity [22].

Finally, it is worth mentioning that thanks to flow cytometry, a new type of array has been developed, called the color-coded microsphere array. Each microsphere is covered with different antibodies, and each antibody can be detected through different fluorochromes; thereby, monitoring the binding antibody-antigen. The latter is attained from the sample of interest, can be fast, accurate, low cost and highly-sensitive. A new system of magnetic microspheres has been designed recently with high-reproducibility and sensibility, low background noise and price, and having a wide dynamic range [9,11].

In summary, a color-coded suspension microsphere array presents the following advantages over current antibody arrays: (i) High level of multiplexing. (ii) A sensitive, accurate and wide dynamic range signal readout. (iii) Information about protein-protein interactions. (iv) Flexibility in array composition. (v) Automatic preprocessing (gating, QC) in a short time. (vi) Clinical applications. All due to the integration of flow cytometry, antibody array detection of size-resolved lysates and computer-assisted data processing [42].

Many different detection techniques were developed over time that enable reliable, sensitive, and specific detection of arrays in a high-throughput manner. Some of them are based label tools and others are label-free techniques. The first type involved a fluorochrome or radioisotope tag molecule for the query element [7,9,11,39] or some new tags recently have been developed, such as quantum dots (QDs), gold nanoparticles (NPs), Raman-dye label carbon nanotubes or silica NPs [7,11]. However, these techniques can interfere with the probe’s capacity to bind to the target protein [22]. To solve these problems, label-free techniques have been developed. They include surface plasmon resonance (SPR), carbon nanotubes, microelectromechanical cantilevers and surface-enhanced laser desorption ionization (SELDI)-TOF-MS. These techniques can measure the mass, dielectric or optical properties of the query molecule [9,11,22]. Currently, only SPR is a label-free technique generally available in many laboratories. Although the other tools need to be developed to become suitable high-throughput techniques, they show huge potential in protein microarrays [9].

The fluorochrome labeling technique, directly incorporated from the DNA-chip technology, has become widely used due to its effectiveness and its compatibility with different systems of laser scanning [9,22]. Both, antibodies and antigens are labeled with two different fluorochromes, mixed and concurrently incubated on the same array. Then, dual color detection systems allow the detection of both signals and measure their ratios. Finally, the analysis of data reveals whether the targets are present in different or similar concentrations on each spot (Figure 2) [12,14] and allowing differentiate between two separate samples [12,39]. The fluorochrome is bound to the antibody or to antigen depending on the type of array [12,14]. A key advantage is the possibility of comparing two samples without the need of a second independent array, since both assays are performed on the same slide [39].

Achieving a more powerful signal is possible using indirect labeling, which is the most common system with serum samples, antibodies and in sandwich assays. However, cross-reactivity can be caused; hence the assortment of antibodies able to be employed is limited [39]. Anyway, these problems can be solved using different strategies, such as labeling procedures based on enzymatic signal amplification. In this way, Schweitzer et al. developed a method which requires enzymatic extension. This can be done through rolling circle amplification (RCA) which is based on the enzymatic extension of a primer-antibody conjugate followed by hybridization of labeled probes [43]. Also, tyramide signal amplification (TSA) system can be utilized [44]. On the other hand, Huang et al. have employed chemiluminescence in order to achieve the required purpose [45]. However, sometimes, these strategies are not sufficient to solve the cross-reactivity problems. When these problems appear, together with others, data analysis is the tool which allows obtaining correct conclusions from experiments, by taking the problems derived from the hybridization method into account and solving them. However, there are no simple and uniform strategies for the analysis of data obtained by different kinds of protein microarrays. Thus, the development of user-friendly and accurate algorithms still needs to be developed.

Antibodies immobilized on the microarray surface can be detected through direct labeling, requiring only a single capture antibody specific for each target protein. Alternatively, a sandwich approach can be carried out, which consists of two sets of antibodies, the first one is specific for the target protein, and the second one for the first antibody [47]. Then, the signal is detected by a colorimetric reaction or a fluorescent dye. This last alternative enables a dual-color layout that is based on labeling each sample with different fluorescent dyes (e.g., Cy3 and Cy5), which competes for the binding sites of the antibodies immobilized on the array. After the incubation, intensity signals are measured for each dye using fluorescence image scanners. Dual-color assays typically display better reproducibility and discriminative power compared to single-color assays [47].

All the steps mentioned above are commonly used in protein arrays due to the difficulty of quantifying protein expression in a multiplexed manner. Multiple causes may lead to the occurrence of such artifacts, including the effect of electric charges, hydrophobic interaction of proteins, artifacts due to differences in protein sizes and antibody/antigen binding kinetics. For all these reasons, microarray data frequently require normalization. Several methods can be used for data normalization, such as: (i) house-keeping probes, (ii) inclusion of spike-in controls and (iii) use of algorithms to define sets of probes. In the following, these methods are described [48].

Different microarray designs may be considered (Table 3, adapted from [48]) which differ in the number of samples and type of array employed, as described below:

Balanced-block and loop designs allow correct dye effect normalization, which are required for the correct analysis of two-dye systems. The first one is used for comparison studies, whereas the loop design is useful for discovery studies [48]. The table below shows the relations between samples and dyes used in the experimental designs indicated above [49].

Stuchlý et al. developed a protein-profiling tool which allows computational feasibility, hands-on time, standardization and reproducibility, quality control feedback loops, data normalization and presentation of the results in an appropriate way to be analyzed [42].

Size exclusion chromatography-resolved microsphere-based affinity proteomics (Size-MAP) is a new tool that permits obtaining information about protein sizes, protein complexes and protein profile changes through flow cytometry detection. The statistical method used was based on modifications of Partitioning Around Medoids (PAM). Previously, they have tried to resolve individual color-coded microsphere types by using k-means clustering, model-based clustering, minimum spanning tree clustering, and hierarchical clustering. But, finally, PAM was seen as the best method. In this way, they adopted the standard approach of sequential two-dimensional gating. Kernel density was used to analyze the distribution of fluorescent signal [42].

Initially, in order to analyze antibody microarrays, strategies from cDNA arrays were implemented. Fluorescent dyes (e.g., Cy3 and Cy5) were used to label each sample and, then, log ratio of the intensities was calculated for every feature on the array. Nevertheless, this design was not the optimal one [48]. Therefore, two new methods were developed: balanced-block design and loop design. The first technique considers the arrays as a block, in such a way that samples hybridized are balanced with respect to dyes. In the second strategy, each sample is hybridized onto two different arrays, each with a different dye. Both designs allow suitable dye normalization. However, using two arrays for each sample supposes a considerable disadvantage [48].

The normalization is based on an internally normalized ratio (INR). A reference sample is labeled with fluorescent dye (e.g., Cy5), whereas the sample of interest is labeled with another dye (e.g., Cy3). In other arrays, the same samples are labeled with the opposite dyes (Cy5 for the sample of interest and Cy3 for the reference sample). Then, the ratio of both fluorescent dyes is calculated for each array. The INR is the geometric mean of the two ratios [48]. Also, ANOVA models, mixed ANOVA models and within-print tip local regression smoothing methods have been developed for the removal of systemic effects typically found in cDNA arrays [48].

RPA analysis is based on the construction of a serial dilution curve, which is characterized by two main advantages [51]: first, the signals in successive dilutions can be related to each other. In this way, protein concentration and signal intensity can be accurately established. Second, data quality can be checked due to raw data display. In RPA analysis several steps must be carried out, as follows:

In the algorithm displayed above, a corresponds to the background noise; b is the response rate in the linear range; M is the maximum or saturation level and x is the concentration of the protein. Also, the equation can be modified in order to avoid data about protein concentration. In this way, y can vary. Generally, y≠ 1 applies to condition in which there is some heterogeneity in the solute molecules or the surface receptors. y approaches to 1 when the range of the free energy of binding shrinks to a singular point. In this last case, the equation is equivalent to the conventional Langmuir model.

where M/r is a threshold value in which M is the saturation level and r should be >1. K refers to the K^th dilution step.

In summary, the Sips model presents physically meaningful parameters and has the optimal conditions for RPA experiments [51].