3. Results and Discussion 3.1. Live Cell Imaging and Controls To validate the assay, integrin αvβ3-positive M21 cells (positive control) labeled with the 700 nm NIR fluorophore ESNF10 and integrin αvβ3-negative M21-L cells (negative control) labeled with the 800 nm NIR fluorophore IR786 were panned over the surface of our SMM (Figure 1A). PAAm, a “sticky” cationic polymer showing no specificity to cell surfaces was used as a positive ligand control, which bound all cell types. Using dual-channel NIR fluorescence microscopy, the number of individual cells binding each spot could be counted (Figure 1B). Thus the readout of our assay was number of cells bound per spot, with the theoretical maximum number of bound cells (i.e., the dynamic range of the assay) being defined by the PAAm control spots (≈300 cells per spot for all cell lines tested). Results of the assay using the integrin-binding peptide cRGDyK as the ligand spot are shown in Figure 1B. Specificity was defined in one of two ways. In the absence of negative control cells, specificity was the number of receptor-positive cells binding a ligand spot divided by the number of these same cells binding inter-spot blank space on the slide. In the presence of negative control cells, specificity was the number of receptor-positive cells binding a ligand spot minus the number of receptor-negative cells bound to that same spot. Sensitivity was defined as the absolute number of receptor-positive cells bound to a particular spot. Of note, pseudo-coloring of 700 nm fluorescence in red and the 800 nm fluorescence in green permitted rapid visual assessment of specificity as demonstrated in Figure 1. Figure 1 Dual-channel screening strategy and controls. (A) Living integrin αvβ3-positive M21 cells (target cells; stained with ESNF10 and pseudo-colored in red) and integrin αvβ3‑negative M21-L cells (control cells; stained with IR786 and pseudo-colored in green) prior to dissociation from their respective plates. Scale bars = 100 μm. (B) The same cells mixed together and panned over PAAm positive control spots (top row) or cRGDyK ligand spots (bottom row). The yellow color indicates co-localized M21 and M21-L cells. Scale bars = 300 μm. 3.2. Optimization of SMM Screening Using Living Cells In order to optimize screening parameters of our SMM using living cells, cRGDyK spots were arrayed, and a mixture of M21 and M21-L cells were applied while systematically varying motion, incubation time, ligand spotting concentration, and number of panned cells (Figure 2A). Notably, cell binding was greatly improved in the absence of motion, a result that might be explained by a boundary layer of shear stress created immediately above the surface of the slide in the presence of motion. Once stationary, cell binding increased linearly with incubation time up to 120 min, at which time saturation occurred. Cell binding also increased as a function ligand spotting concentration, with saturation occurring above 0.25 mM. The number of panned cells also increased binding, with saturation occurring at 4 × 106 cells per slide. To maximize the specificity of binding, we compared incubation in the presence or absence of 10% serum, and with or without negative control cells (Figure 2B). Both serum and competing cells improved specificity, likely by blocking non-specific interactions. Figure 2 Optimization of screening parameters: (A) Maximizing sensitivity through the effect of motion, incubation time, ligand spotting concentration, and the number of panned cells using a cRGDyK array and a mixture of M21 and M21-L cells. Shown are mean ± SD for each data point from 4 randomly chosen spots on the slide. (B) Maximizing specificity through the use of serum or competing receptor-negative cells. 3.3. Screening of Diverse Chemical and Cellular Interactions To explore the usefulness of our SMM screening system, we tested three cell-ligand interactions, which all differed in terms of cell type, receptor type, ligand type, and Bmax (i.e., the number of receptors per cell) and are shown in Figure 3 and summarized in Table 1. M21 cells have approximately 5 × 104 integrin αvβ3 receptors per cell (i.e., Bmax) on their surface, while M21-L cells have no detectable integrin αvβ3 receptors (1 × 103) [27]. Integrin αvβ3 has a type I transmembrane topology and binds the cyclic peptide ligand cRGDyK with an affinity of approximately 50 nM. LNCaP cells have approximately 2 × 105 prostate-specific membrane antigen (PSMA) receptors per cell on their surface, while PC3 cells have no detectable PSMA [28]. PSMA has a type II transmembrane topology and binds the small molecule KUE with an affinity of approximately 15 nM. B16 cells have approximately 7 × 103 melanocortin 1 receptors (MC1R) receptors per cell on their surface, while LNCaP cells have no detectable MC1R [29]. MC1R is a G protein-coupled receptor with 7 transmembrane domains and binds the peptide α-melanocyte stimulating hormone (α-MSH) with an affinity of approximately 0.4 nM. Using the optimized parameters from Figure 2 (no motion, 60 min incubation time, 1 mM ligand spotting concentration, and 4 × 106 panned cells per slide), all three cell-ligand interactions are detectable with our SMM screening system with relatively high specificity and with sensitivity proportional to Bmax. It should be noted, however, that many of the system parameters are inter-dependent and should be re-optimized for a particular model system. For example, in the presence of extremely high affinity and Bmax, motion might not only be possible but could improve specificity by reducing non-specific interactions. Similarly, we only explored the presence or absence of 10% serum, but in some model systems, a higher or lower concentration could be optimal. microarrays-04-00053-t001_Table 1 Table 1 Selected small molecule ligands and cell lines used for SMM screening. β-AG, beta Ala-Gly; cRGDyk, cyclo Arg-Gly-Asp-D-Tyr-Lys; GPI, 2[(3-amino-3-carboxypropyl)(hydroxy)(phosphinyl)-methyl]pentane-1,5-dioic acid; KUE, 2-(3-(5-amino-1-carboxypentyl)ureido)pentanedioic acid; MC1R, melanocortin 1 receptors; α-MSH, α-melanocyte stimulating hormone; M.W., molecular weight; N.A., not applicable; PAAm, polyallrylamine; PSMA, prostate-specific membrane antigen. 3.4. Exploring the Relationship between Ligand Affinity and Bmax The results from Figure 3 suggested that ligand affinity (KD), and therefore the ratio of affinity to Bmax, might have a profound impact on the sensitivity of the SMM screening assay. To explore this relationship we utilized a series of PSMA ligands previously reported by our group (Figure 4A), which spanned a wide range of affinity [22,31]: β-AG (KD = 2 µM), GPI (KD = 9 nM), β-AG trimer (KD = 60 nM), and GPI trimer (KD = 0.4 nM). The SMM was probed with PSMA-positive cells (LNCaP) and PSMA-negative cells (PC3), using the optimized parameters described for Figure 3. As shown in Figure 4, the SMM assay was able to perform well over 3 logs of affinity space, with the number of cells bound per spot being proportional to affinity. These results reinforce the importance of defining any cell-bound ligand spot as “positive” during initial screening of diverse chemical libraries because low affinity ligands might have only a few cells bound. And, if Bmax is low, even high affinity interactions may result in only a few cells bound per spot [22,31]. Figure 3 Robustness of the SMM Screening Assay: Three different model systems, varying in ligand chemical structure, cell type, receptor transmembrane topology, Bmax, and ligand affinity were tested as described in the text. Shown are mean ± SD for each data point from four randomly chosen spots on the slide. Receptor-positive and receptor-negative cells were labeled with 700 nm and 800 nm NIR fluorophores and pseudo-colored red and green, respectively, during microscopy. Scale bars = 200 μm. Figure 4 The effect of affinity on cell-ligand spot binding: PSMA-positive LNCaP cells and PSMA-negative PC3 cells were labeled with 700 nm and 800 nm NIR fluorophores and pseudo-colored in red and green, respectively. (A) Chemical structures of targeting ligands employed, (B) 4× microscopy images, and (C) 20× microscopy images. Scale bars = 200 μm. 4.