|Home | About | Journals | Submit | Contact Us | Français|
Genetic findings have suggested that neuregulin-1 (Nrg1) and its receptor v-erb-a erythroblastic leukemia viral oncogene homologue 4 (ErbB4) may play a role in neuropsychiatric diseases. However, the downstream signaling events and relevant phenotypic consequences of altered Nrg1 signaling in the nervous system remain poorly understood. To identify small molecules for probing Nrg1−ErbB4 signaling, a PC12-cell model was developed and used to perform a live-cell, image-based screen of the effects of small molecules on Nrg1-induced neuritogenesis. By comparison of the resulting phenotypic data to that of a similar screening performed with nerve growth factor (NGF), this multidimensional screen identified compounds that directly inhibit Nrg1−ErbB4 signaling, such as the 4-anilino-quinazoline Iressa (gefitinib), as well as compounds that potentiate Nrg1−ErbB4 signaling, such as the indolocarbazole K-252a. These findings provide new insights into the regulation of Nrg1−ErbB4 signaling events and demonstrate the feasibility of using such a multidimensional, chemical-genetic approach for discovering probes of pathways implicated in neuropsychiatric diseases.
Chemical genetics aims to discover small molecules that can be used as probes to alter protein function. This approach provides an important path both to decipher the molecular circuitry that regulates complex biological phenotypes and to potentially identify new targets for therapeutic intervention. There has been a dramatic increase in the application of chemical genetics to a variety of biological systems and disease contexts (1). However, to date this approach has not been widely used to dissect the function of candidate disease genes and pathways implicated in neuropsychiatric disorders.
Genetic analysis of several neuropsychiatric disorders has led to the identification of several potential risk genes and has opened up the possibility of testing their functional significance. In the case of schizophrenia (OMIM 181500), a highly heritable and devastating neuropsychiatric disorder that affects between 0.5% and 1.0% of the world’s population, the genes encoding neuregulin-1 (Nrg1) (2−4) and its 180-kDa transmembrane tyrosine kinase receptor ErbB4 (v-erb-a erythroblastic leukemia viral oncogene homologue 4) of the epidermal growth factor receptor (EGFR) family have been identified as susceptibility genes (5−11). Nrg1 and ErbB4 have been implicated in a variety of neuronal development processes, including neuritogenesis (the formation of extended processes that become axons and dendrites), neuronal migration, myelination and synapse formation, as well as multiple forms of synaptic plasticity (12,13), many of which have been implicated in the pathogenesis of schizophrenia as well as other psychiatric diseases. While Nrg1 and ErbB4 are attractive susceptibility genes, functional variation in these genes has yet to link clearly Nrg1−ErbB4 signaling to the pathogenesis of schizophrenia. In fact, contradictory data exist that suggest decreases or increases in Nrg1−Erb4 signaling may account for disease pathogenesis. For instance, decreased Nrg1 signaling has been proposed to contribute to altered brain development, neurotransmission, and cortical function, while an alternative gain-of-function hypothesis suggests increased levels of Nrg1 and ErbB4 (14) and high Nrg1−ErbB4 signaling (15) exist in the prefrontal cortex of schizophrenia patients. Despite this intriguing progress, clear functional evidence connecting these risk genes to schizophrenia still does not exist (13).
The difficulty of establishing clear functional association between Nrg1−Erb4 signaling and psychiatric disease is partially due to the diversity of Nrg1 isoforms and family members and the diversity of ErbB family of transmembrane receptors. The Nrg1 gene encodes multiple proteins containing an EGF-like domain that binds to the extracellular domain of ErbB family receptors, either in a paracrine or in a juxtacrine signaling manner, and causes receptor dimerization (16). ErbB receptor dimerization stimulates the intrinsic tyrosine kinase activity of these receptors leading to autophosphorylation and subsequent recruitment of a variety of signaling molecules. In addition, the ErbB4 gene is known to undergo alternative splicing to encode either a metalloprotease cleavable (JM-a) or cleavage-resistant (JM-b) extracellular domain and a cytoplasmic domain (Cyt-1), a phosphotidylinositol-3 kinase (PI3K) binding site, or Cyt-2 (17−19). Among the four members of the ErbB family, ErbB3 and ErbB4 are known to have Nrg1-binding ability. The former lacks a functional kinase domain and is believed to be a kinase-dead ErbB isoform. ErbB3 is known to play an active role in regulating the function of other ErbB family members. EGFR (ErbB1) and ErbB2 do not bind Nrg1, but these receptors can be activated by heterodimerizing with ErbB3 and ErbB4 upon Nrg1 induction (13,20). In addition to Nrg1 and EGF, the ErbB family receptors can also respond to other growth factors, such as transforming growth factor alpha (TGFα) and HB-EGF (21). Thus, dissecting the complexity of Nrg1−ErbB4 poses a great challenge especially since no specific ErbB4 inhibitor is currently available.
Besides understanding how genetic variation influences nervous system function, a major focus of the field of neurobiology is to understand the molecular machinery and mechanisms that enable neuronal networks to be generated and remodeled throughout development and in response to neural activity. Neuritogenesis, the initial stage of neuronal differentiation, involves the generation of processes termed neurites that emerge from the cell body of postmitotic cells. These processes extend steadily until one (the future axon) starts growing more rapidly, inducing morphological polarization. In primary neurons, both in cell culture and in vivo, the extension of neurites leads to the establishment of polarity in which one process becomes an axon and the remaining processes become dendrites, and then they eventually establish synapses. Besides initiating differentiation, neuritogenesis plays an important role in the initiation of neuronal migration and patterning that gives rise to the intricate networks of neuronal connections in the adult brain. Consistent with the fundamentally important role of neuritogenesis, a growing number of neuropsychiatric disease risk genes, such as Nrg1 and DISC1, have been shown to alter neurite formation (22). These findings suggest that by using neurite outgrowth as a phenotype, it may be possible to develop small-molecule probes that can be used to target and discover new properties of the underlying signaling networks that are integral to the etiology and pathophysiology of severe mental illnesses. To initiate systematic efforts to test this notion, we describe herein the development and characterization of a model PC12 cell system expressing ErbB4 and the results of a phenotype-based screen for differential modulators of Nrg1- versus NGF-induced neuritogenesis.
To develop a cellular model that would enable chemical-genetic characterization of Nrg1−ErbB4 signaling and discovery of new small-molecule probes, we chose the neuroendocrine cell line PC12 (23), derived from a rat pheochromocytoma, as a model neuronal system for several reasons. First, PC12 cells do not naturally express ErbB4 but are known to express EGFR (ErbB1), ErbB2, and ErbB3, so this would allow us to express exogenous genetically altered ErbB4 receptors. Second, PC12 cells engineered to express human ErbB4 have been reported to differentiate and undergo neuritogenesis upon treatment with recombinant Nrg1 (24), which we reasoned could provide a phenotype that could be quantified using automated microscopy and image analysis. Finally, NGF-induced differentiation has been extensively studied in PC12 cells, and much is known about the downstream signaling pathways, which we reasoned would assist in comparing the selectivity of any compounds that we identify.
To create a system for chemical-genetic study, we prepared a stable PC12 cell line that coexpresses green fluorescent protein (GFP) and the human ErbB4 isoform JMa-Cyt2 and a control cell line PC12-GFP that stably expresses GFP but lacks ErbB4. The PC12-ErbB4-GFP cells were examined by fluorescence imaging for the effects of NGF and Nrg1 on neurite outgrowth. As expected, both the PC12-ErbB4-GFP and the PC12-GFP cell lines exhibited the characteristic morphological changes indicative of neuronal differentiation involving neurite outgrowth when treated with NGF (Figure (Figure1A).1A). Under low magnification (<10×), the distribution of coexpressed GFP is uniform throughout the cell body and neurite projections, and the fluorescent image reliably represents the whole cell body and attached processes (Figure (Figure1B).1B). Nrg1 stimulation of PC12-ErbB4-GFP cells, but not PC12-GFP, for 2 days resulted in significant neurite outgrowth similar to the effects of NGF. This result indicated that although PC12 cells express ErbB3, which can also bind to Nrg1 and heterodimerize with other ErbB family members, the presence of the ErbB4 receptor is necessary for Nrg1 signaling to stimulate neuritogenesis. In addition, EGF treatment failed to stimulate neurite outgrowth in both cell lines (Figure (Figure1A),1A), suggesting that although EGFR-activation might also activate other ErbB receptors via heterodimerization, it is not capable of triggering neuritogenesis in the presence or absence of ErbB4.
ErbB4 is known to activate downstream effectors of neurotrophic factor pathways such as mitogen-activated protein kinase (MAPK). To verify that PC12-ErbB4-GFP cells activate MAPK pathway components in response to Nrg1, we examined the ability of NGF and Nrg1 treatment to activate of the MAPK pathway as measured by phosphorylation of extracellular signal-regulated kinase (Erk1/2). In both the PC12-GFP and PC12-ErbB4-GFP cell lines, NGF induced a rapid phosphorylation of Erk1/2, which peaked as early as 5 min after treatment. In contrast, Nrg1 induced a strong phosphorylation of Erk1/2 only in PC12-ErbB4-GFP cells (Figure (Figure1C).1C). In addition, it has been reported that Nrg1 increases the release of the intracellular domain of ErbB4, ErbB4-ICD, in neural precursor cells (25). However, we did not observe this effect in PC12 cells (Figure (Figure1C).1C). We also found that a γ-secretase inhibitor that is known to inhibit the cleavage of ErbB4 did not inhibit Nrg1-induced neurite outgrowth in PC12-ErbB4 cells (data not shown).
Having validated that stimulation of PC12-ErbB4-GFP cells with Nrg1 could induce neuritogenesis, we sought next to determine whether the outgrowth phenotype was suitable for use with automated imaging and image analysis. The expression of soluble GFP in PC12 cells allowed the use of automated microscopy to acquire fluorescent images for the quantitative analysis of cell number and morphological changes associated with neuronal differentiation over a developmental time course. To minimize cell clumping and intersection of neurites in 96- or 384-well plates for time periods of up to 4 days, we seeded cells at a low density of ~4000 cells/cm2. Imaging with an ImageXpress 5000A automated microscope equipped with 4× objective enabled the acquisition of the entire well of a 384-well plate in one image with sufficient resolution to accurately detect and quantify various properties of neurites and cell bodies. A typical pixel map of the segmentation mask generated by the MetaXpress software is shown in Figure Figure2A.2A. Cell bodies were identified as pixel blocks with minimum area of 200 μm2 and maximum width of 40 μm, and the neurites were subsequently identified as line objects longer than 10 μm and connected to each cell body. The mean neurite length per cell for each well was quantified as a single parameter (see the Methods section for a more detailed description of imaging and analysis parameters), represented as “mean outgrowth per cell (μm)”. We accounted for variation in cell density due to seeding or to the effect of antiproliferative compounds by quantifying neurite outgrowth on a per-cell basis. The use of automated image analysis methods enabled the accurate assessment of morphological properties of hundreds of cells per well without human bias of which cells to measure the properties.
We studied the robustness of our automated neurite detection by comparing the dose response of PC12-ErbB4-GFP cells to Nrg1 and NGF as a function of time with image acquisition every 24 h. Both NGF and Nrg1 stimulated a continuous increase in mean neurite length over a four- day time course. NGF-treated cells appeared to differentiate more slowly than Nrg1-treated cells in the first 24 h, but after four days, the mean length of neurites per cell in both treatments was similar (Figure (Figure2B).2B). This delayed response to NGF, but similar overall effect after four days, might be explained by an up regulation of the expression levels of the tyrosine kinase receptor TrkA (26) or a secondary receptor of NGF, such as p75NTR27, which is known to potentiate the activity of TrkA through the formation of a high-affinity NGF receptor. The mean neurite length exhibited a strong correlation to the dose of Nrg1 or NGF added to the cells especially under concentrations of 10 ng/mL at day two (Figure (Figure2C)2C) and under 20 ng/mL at day four (Figure (Figure2D).2D). These data suggested that we could use automated microscopy to measure neurite length as a phenotype for screening.
Having established methods for quantitatively measuring Nrg1-induced neuritogenesis in PC12-ErbB4-GFP cells, before embarking on a screen for chemical modulators, we sought to better understand the signaling events associated with Nrg1−ErbB4 signaling to assist in the eventual downstream characterization of any probes that might be identified. Besides activating ErbB4 homodimers, Nrg1 can mediate the formation of ErbB dimers consisting of ErbB3 (21), as well as EGFR (20). Thus, although ErbB4 is required for Nrg1 to promote neurite outgrowth, since all ErbB family members are expressed in the PC12-ErbB4-GFP cells, it remained unclear whether the other ErbB receptors were also playing a role. To address specifically the contribution of other ErbB family members to Nrg1−ErbB4 dependent neurite outgrowth, we individually reduced the expression of ErbB1/EGFR, ErbB2, and ErbB3 with pools of small inhibitory RNAs (siRNAs). The efficacy and specificity of each ErbB receptor siRNA pool was verified by Western blotting (Figure (Figure3A).3A). The Nrg1-driven neurite outgrowth was diminished only when ErbB4 expression was reduced (Figure (Figure3B,C),3B,C), suggesting that ErbB1/EGFR, ErbB2, and ErbB3 do not contribute significantly to Nrg1-dependent neuritogenesis in PC12 cells and that ErbB4 is necessary for the observed effects of Nrg1 on neuritogenesis. These results are consistent with the observation that PC12 cells only extended neurites in response to Nrg1 when the ErbB4 receptor was expressed. Surprisingly, we observed that reducing ErbB3 expression enhanced Nrg1-induced neurite outgrowth, suggesting that ErbB3 inhibits some aspect of neuritogenesis in our PC12 cell system. While the molecular basis for this observation is not understood, it suggests that ErbB3, although lacking a functional kinase domain, may contribute to the signaling of kinase-active ErbB receptors as well as other proteins important for neuritogenesis (28).
To investigate the downstream components of Nrg1−ErbB4 signaling that triggers neuritogenesis, we specifically examined two key kinase cascades coupled to receptor tyrosine kinase signaling, MEK and PI3K pathways. The ERK kinase (MEK) has long been known to mediate NGF-induced PC12 differentiation (29). In our system, two specific MEK inhibitors, PD098059 and U0126, attenuated both NGF- and Nrg1-induced neurite outgrowth (Figure (Figure3D,E).3D,E). These results further validated that both Nrg1- and NGF-induced neuritogenesis are dependent on the Erk1/2 cascade. PI3K is another important component of many neurotrophic factor signal transduction pathways. Although we used the human ErbB4 isoform JMa-Cyt2, which lacks the PI3K binding domain, it has been demonstrated that all ErbB4 isoforms, including Cyt2 isoforms, associate with and activate PI3K (30). Consistent with this finding, we observed an up-regulation of phosphorylated Akt (Ser473), an indicator of PI3K activation, when cells were exposed to Nrg1 (data not shown). To determine whether both NGF- and NRG1-induced neuritogenesis requires PI3K signaling, PC12-ErbB4-GFP cells were treated with two structurally distinct PI3K inhibitors, LY294002 and wortmannin. Both of these PI3K inhibitors caused an inhibitory effect on NGF-induced neurite outgrowth as expected (31). In contrast, neither LY294002 nor wortmannin were capable of inhibiting Nrg1-induced neurite outgrowth at doses ranging from 0.1 to 10 μM (Figure (Figure3B).3B). Collectively, these results suggest that although both Nrg1 and NGF stimulate Erk1/2 and PI3K cascades, activation of PI3K is not required for neuritogenesis induced by Nrg1 in our PC12 cellular system.
Having shown that our cellular model activated neuritogenesis in an Nrg1−ErbB4-dependent manner and retained the ability to respond to NGF, we next screened for small molecules that could specifically modulate Nrg1−ErbB4 signaling without affecting NGF-induced neuritogenesis. We initially tested 400 known bioactive small molecules at a single dose (~10 μM) for their ability to modulate neurite outgrowth induced by the addition of Nrg1 and NGF (see Supplementary Tables 1 and 2, Supporting Information, for the complete list of compounds tested and resulting high-content imaging data). A total of three 384-well plates (one DMSO control plate and two compound plates each containing 200 bioactives and 184 DMSO control wells) were screened (Figure (Figure44A).
Cell morphological features measured from each image using MetaXpress software included (1) percent of cells with significant neurite growth, (2) number of cells, (3) total neurite outgrowth, (4) mean neurite outgrowth length per cell, (5) normalized mean neurite outgrowth per cell, (6) mean number of processes per cell, (7) mean branches per cell, and (8) mean cell body area. Each feature is described in the Methods section in more detail, and the complete data set for Nrg1- and NGF-treated wells is provided in Supplementary Tables 1 and 2, respectively, in the Supporting Information. Supplementary Tables 3−10, Supporting Information, provide a global statistical analysis of the eight cellular features, an assessment of the degree to which the cellular feature is normally distributed, and the observed relationships of each feature to each other in the form of Pearson correlation coefficient. Based upon these global analyses, as shown in Supplementary Figure 3, Supporting Information, a graphical representation of the two-dimensional Pearson correlation map between the eight cellular features in the Nrg1 and NGF high-content imaging screens reveals that many of these cellular features are highly correlated across the 400 compound treatments and there existed differences in the global feature profiles between Nrg1 and NGF. However, we also noticed that there were potentially informative relationships among the cellular and neurite features. For instance, in the case of Nrg1-treated cells (see Supplementary Figure 4, Supporting Information), the mean neurite length feature was correlated with the percentage of cells with significant neurite outgrowth (r = 0.93) and the number of processes per cell (r = 0.95) but was more moderately correlated with the number of branches measured per cell (r = 0.70). In contrast, the mean neurite length per cell correlated less with cell body size (r = 0.5) suggesting that Nrg1-induced neuritogenesis can be separated from the control of soma size.
Since we found that the mean neurite length per cell feature was strongly positively correlated to the dose and duration of Nrg1 and NGF treatment (Figure (Figure2),2), we chose this feature as a surrogate of Nrg1 and NGF signaling for use in image-based screening while recognizing that analysis of other features may lead to different types of modulators of Nrg1 signaling. With this feature, in our assay, the 752 DMSO control wells for each treatment exhibited consistent background levels of neurite outgrowth and changes in cell number (see Supplementary Tables 3 and 4, Supporting Information). The mean neurite length values from each set of DMSO control treatments was taken as the baseline value. For subsequent data visualization, the mean neurite length value in each well upon Nrg1 or NGF induction was normalized to the respective baseline value (Figure (Figure4B).4B). Within the library of 400 known bioactives, 51 compounds led to a significant reduction in cell number in either the Nrg1 or NGF treatment conditions as defined by a threshold of having less than 100 cells after two days of incubation. While these compounds may have additional phenotypes at lower concentrations, they were not considered further in the studies reported here. The remaining 349 compounds were categorized into nine classes based on their relative activities compared with the DMSO controls and their specificities toward Nrg1- and NGF-induced neurite outgrowth (Figure (Figure4B)4B) using a simple fold-change cutoff of 2-fold for molecules that potentiated neurite outgrowth and 0.5-fold for molecules that inhibited outgrowth. The numbers of bioactives in each category based upon this classification are summarized in Figure Figure44C.
To identify specific inhibitors of Nrg1−ErbB4 signaling, we first focused our analysis on those compounds that inhibited Nrg1-induced neurite outgrowth but had no effect on NGF-induced neurite outgrowth. From our original screen, we noted that two 4-anilino-quinazoline-containing compounds, WHI-P180 and CL-387,785, satisfied our selection criteria (Figure (Figure4C)4C) and had no effect on the cell number, indicating that they did not alter cell proliferation over the two-day time course.
While both WHI-P180 and CL-387,785 were previously shown to inhibit EGFR (32,33), little was known about the ability of these two compounds to inhibit Nrg1 signaling. 4-Anilino-quinazoline-based compounds are known to reversibly and competitively bind to the ATP pocket of EGFR (34). To further explore the ability of 4-anilino-quinazoline-based compounds to inhibit Nrg1-signaling, we tested four commonly used small molecules of this structural class: AG1478, PD158780, Iressa (gefitinib), and Tarceva (erlotinib) (Figure (Figure5A),5A), along with nine additional 4-anilino-quinazolines also classified as EGFR inhibitors (Supplementary Figure 2, Supporting Information). Iressa and Tarceva are Food and Drug Administration (FDA)-approved drugs used for the treatment of non-small-cell lung cancer through a mechanism thought to involve the inhibition of the tyrosine kinase activity of EGFR and have been optimized for their pharmacological properties and safety in humans. All four 4-anilino-quinazolines inhibited Nrg1-induced outgrowth in a dose-dependent manner, while PD158780 appeared to have a lower potency (at ~2 μM) against Nrg1 stimulation and decrease the mean length of neurites induced by NGF at the same concentration. On the other hand, AG1478, Iressa, and Tarceva had no significant effect on NGF-induced neurite outgrowth but all inhibited Nrg1-induced neurite outgrowth with EC50’s of ~500 nM (Figure (Figure5B).5B). The two irreversible EGFR inhibitors, CL-387,785 and PD168393, were both superior inhibitors of Nrg1-induced neurite outgrowth (EC50 ≈ 100 nM), while PD168393 exhibited the strongest inhibition of NGF-induced neurite outgrowth as well (Supplementary Figure 2, Supporting Information). CP-724,714, reported to be an inhibitor ErbB2 selective over EGFR (35), was the weakest compound to inhibit Nrg1 (EC50 ≈ 4 μM) (Supplementary Figure 2, Supporting Information). Taken together, most of the 4-anilino-quinazoline-based compounds in this study inhibit Nrg1-induced neurite outgrowth with various efficacies and specificities over NGF-induced neurite outgrowth.
While Iressa was originally developed to selectively target EGFR (36), our results described here suggest that Iressa also inhibits ErbB4-dependent neuritogenesis. To characterize the interaction between Iressa and ErbB4 in greater detail and to test whether Iressa inhibits the activation of ErbB4 by Nrg1, ErbB4 was immunoprecipitated from PC12-ErbB4-GFP cells after a short exposure to Nrg1 with or without Iressa treatment. The phosphorylation status of ErbB4, a measure of receptor activity, was examined using a phosphotyrosine specific antibody. Indeed, the phosphorylation of ErbB4 receptors induced by Nrg1 was inhibited when cells were treated with Iressa (2 μM), and the subsequent phosphorylation of the downstream Erk1/2 was also diminished. In contrast, Iressa did not affect NGF-induced activation of Erk1/2, thereby confirming the selectivity observed for the 4-anilino-quinazolines in the initial small-molecule screen (Figure (Figure5C).5C). In addition, when cells were treated with Nrg1 (20 ng/mL), the phosphorylation levels of ErbB4 and Erk1/2 were diminished by Iressa (0.2−5 μM) in a dose-dependent manner as determined by Western blotting with phospho-ErbB4 and phospho-Erk1/2 antibodies (Figure (Figure5D,E),5D,E), respectively. These results indicate that Iressa treatment inhibits ErbB4 receptor activation and its downstream signaling.
Although we demonstrated that ErbB4 activation is necessary and sufficient for Nrg1-induced neuritogenesis (Figure (Figure1A)1A) and that the activation of ErbB4 is inhibited by Iressa (Figure (Figure5D,E),5D,E), it remained possible that the inhibition is indirect through inhibition of trans-phosphorylation by other ErbB family members or other targets. Since ErbB3 lacks kinase activity, we ruled this ErbB receptor out as a direct target. It is also known that Iressa inhibits EGFR (IC50 ≈ 30 nM) more potently than ErbB2 (IC50 > 3.7 μM) (36). Furthermore, the selective ErbB2-inhibitor, CP-724,714, poorly inhibited Nrg1-induced neurite outgrowth (Supplementary Figure 2, Supporting Information), suggesting that inhibiting ErbB2 is not critical. Thus, we focused our efforts on determining whether the inhibition of ErbB4 is due to an indirect inhibition of EGFR.
To determine whether Iressa acts through ErbB4 to inhibit neurite outgrowth, to complement the chemical treatments described above, we used RNAi-mediated silencing to reduce the levels of ErbB family members and then treated with Iressa. Iressa was found to still effectively diminish neurite outgrowth in cells where expression of EGFR, ErbB2, or ErbB3 was reduced by siRNAs (Figure (Figure5F),5F), suggesting that none of these three ErbB family members are required for Iressa’s effects on Nrg1 signaling and that their loss-of-function does not potentiate the effect of Iressa. Of note, even though ErbB3 silencing potentiated the effects of Nrg1, the enhanced neurite outgrowth observed upon ErbB3 knock down was still blocked by Iressa treatment. This result suggests either that the neuritogenesis signaling caused by ErbB3 depletion is transduced through ErbB4 signaling itself or that Iressa causes a dominant inhibition of an alternative signaling pathway that mediates the alterations of neuritogenesis caused by loss of ErbB3.
Based on the cellular results described above, we used three other lines of investigation to further test the hypothesis that the 4-anilino-quinazolines identified here act as direct inhibitors of ErbB4 kinase activity. First, to demonstrate that Iressa interacts with full-length ErbB4 receptor in a physiologically relevant setting, we created a new chemical tool, “iTrap”, consisting of Iressa immobilized on an agarose solid support and performed affinity chromatography (Figure (Figure6A).6A). iTrap was able to affinity-capture full-length ErbB4 and ErbB4-ICD (intracellular domain containing the kinase domain) in PC12-ErbB4-GFP but not the parental PC12-GFP cells lacking ErbB4 expression (Figure (Figure6B).6B). Most importantly, the levels of ErbB4 captured by iTrap were diminished by addition of Iressa (50 μM) as a soluble competitor revealing the specificity of the iTrap reagent.
To determine whether Iressa directly bound to ErbB4, we used surface plasmon resonance (SPR) binding assays. We carried out SPR binding assays with the kinase domains of ErbB4 and EGFR using Iressa at concentrations ranging from 0 to 20 μM (Figure (Figure6C,D).6C,D). We found that Iressa bound both EGFR and ErbB4 with different affinities (Figure (Figure6E,F),6E,F), while the specific GSK-3β inhibitor, CHIR-99021, showed no interaction (Figure (Figure6G,H).6G,H). Kd's were determined from equilibrium binding measurements and by fitting these equilibrium measurements with a 1:1 interaction model using global parameters. Kd's for Iressa were determined to be approximately 30 and 150 nM for EGFR and ErbB4, respectively.
Finally, the effects of Iressa on the in vitro kinase activity of recombinant ErbB4 and EGFR were measured. Iressa was found to inhibit ErbB4 kinase domain activity in vitro with an IC50 ≈ 1 μM (compared with 50 nM against EGFR), consistent with its EC50 for inhibition of Nrg1-induced neurite outgrowth (Figure (Figure6I).6I). Thus, in agreement with the iTrap affinity reagent studies and SPR binding assays, these biochemical findings provide support for the potential of direct interaction between Iressa and ErbB4 leading to a block of Nrg1-induced neuritogenesis.
Overall, our screen revealed that among the negative regulators of Nrg1−ErbB4 signaling, anilino-quinazolines are a rich source of inhibitors with diverse levels of efficacy and intra-ErbB family class specificity. Over the past decade, tremendous effort has been invested in ErbB receptor inhibition, especially targeting EGFR and ErbB2, because of their long-recognized role in cancer (42). As a result, a growing number of ErbB inhibitors have been identified. However, the specificity of these inhibitors has mostly been annotated by comparing EGFR and ErbB2, and no small molecules that are selective inhibitors of ErbB4 are currently available. Based on the close homology among ErbB family members in their kinase domain, several EGFR inhibitors, such as AG1478 and PD158780, have been considered as pan-ErbB inhibitors and used against ErbB4. Previously, these two inhibitors were shown to inhibit Nrg1-signaling and downstream biological consequences such as neurite outgrowth in hippocampal neurons (43), inhibition of NMDA receptor currents in pyramidal neurons from rodent prefrontal cortex (44), inhibition of long-term potentiation at Schaffer collateral-CA1 synapses in the hippocampus (45) and glutamatergic synapse maturation and plasticity (46). The identification of some of these compounds in our screen suggests that the cell-based imaging assay we developed may provide a surrogate system for identifying compounds that modulate Nrg1−ErbB4 regulated synaptic plasticity. However, dissecting ErbB4-specific inhibition from pan-ErbB inhibition poses a new challenge. We also noticed that, unlike Iressa or Traceva, PD158780 has an inhibitory effect on NGF-induced neurite outgrowth, which confounds the interpretation of results when this compound is used in physiological conditions where other neurotrophic factors might interfere. Thus, caution must be taken when these compounds are used because of potential off-target or indirect effects that might be attributed to inhibition of other hererodimerizing ErbB receptors instead of ErbB4 itself.
While this manuscript was in preparation, elegant studies by Krivosheya et al. (41) demonstrated that treatment of rat hippocampal neurons with soluble Nrg1 resulted in enhanced dendritic arborization through activation of the tyrosine kinase domain of ErbB4 and that RNAi-mediated silencing of ErbB4 decreased the number of primary neurites. These findings are consistent with our findings using RNAi toward ErbB4 in PC12 cells engineered to express this receptor and again provide evidence supporting the role of the kinase activity of ErbB4 in mediating neuritogenesis. However, our results differ in some aspects, as treatment of neurons with the PI3 kinase inhibitor LY294002, but not the MAPK inhibitor PD980059, blocked neurite remodeling upon Nrg1 treatment. We speculate that these differences are due to differences in cell type and culture conditions.
In addition to identifying inhibitors such as Iressa, our small-molecule screen also identified small molecules that had no effect on NGF-induced neurite outgrowth but potentiated Nrg1-induced neurite outgrowth. One compound, the indolocarbazole, K-252c (Figure (Figure7A),7A), satisfied our selection criteria and furthermore had no effect on cell death or proliferation in the concentration range tested. Since K-252c is structurally similar to K-252a, a potent TrkA inhibitor that is widely used for inhibition of NGF-induced processes (e.g., refs (37−39), we speculated that K-252a may also have effects on Nrg1−ErbB4 signaling. To test this hypothesis, we first treated the PC12-ErbB4-GFP cells with K-252a and NGF or Nrg1. As expected, K-252a completely inhibited NGF-induced neurite outgrowth at concentrations as low as 50 nM. In contrast, however, similar to K-252c, K-252a significantly potentiated Nrg1-induced neurite outgrowth at the same concentration that inhibited NGF-induced neurite outgrowth. Furthermore, both NGF inhibition and Nrg1 potentiation are dose-dependently modulated by K-252a (Figure (Figure7B,C).7B,C). Though we have yet to identify the specific target of K-252a that is responsible for mediating its effect on Nrg1−ErbB4 signaling, we and others have found that small modifications to the scaffold can afford remarkable selectivity (47,48). Functionally, the early Erk1/2 phosphorylation in response to Nrg1 is not dramatically affected by K-252a treatment. On the other hand, NGF-induced Erk1/2 phosphorylation was diminished by K-252a (Figure (Figure7D).7D). These findings, and the potentiation of neuritogenesis phenotype, suggest that K-252a affects Nrg1 signaling in a manner distinct from its effects on Trk receptor mediated signaling.
Overall, the finding that Nrg1-induced neuritogenesis can be potentiated by both K-252a and NGF suggests that ErbB4 signaling in the brain can be enhanced by removing an inhibitory signal or by activating potentially intersecting or parallel signaling networks. It is possible that K-252a acts as a potent modulator of a downstream component shared by all the neurotrophic factors; however in the case of NGF signaling, its inhibitory effect on the TrkA receptor is dominant. K-252a has been shown previously to have a neuroprotective effect in several cell types through a mechanism reportedly due to inhibition of Trk family receptors (49,50). The detailed mechanism for K-252a’s ability to potentiate Nrg1-induced signaling as observed here for the first time remains a challenge for future studies to address. While we speculate that the relevant target is a kinase, additional potential targets include other ATP-binding proteins such as ATPases involved in chromatin remodeling (e.g, SWI/SNF family) and cytoskeletal dynamics (e.g., myosin).
The ability of neurotrophic factors such as Nrg1 and NGF to regulate neuritogenesis, neuronal survival, differentiation, and aspects of synaptic plasticity is of fundamental importance to brain function and development. Yet our understanding of the underlying molecular mechanisms through which these factors operate is incomplete. A growing number of candidate neuropsychiatric disease risk genes and pathways, including Nrg1−ErbB4 characterized here, alter neurite formation (22). This suggests that in vitro cell culture systems that fulfill the needs of high-throughput screening (HTS), both with engineered systems and primary neurons or neural stem cells, can be used as surrogate systems to discover small-molecule probes that target signaling networks integral to the etiology and pathophysiology of severe mental illnesses.
The findings described here provide new insights to the regulation of neuritogenesis by the tyrosine kinase activity of ErbB4. They demonstrate the feasibility of using such a multidimensional, chemical-genetic approach for discovering probes of pathways implicated in neuropsychiatric disease. In our particular case, the ability to either potentiate or inhibit signaling with small-molecule probes will provide a means for testing the importance of Nrg1-ErbB4 signaling to psychiatric disease pathogenesis.
Collectively, our cellular and biochemical findings described above support the hypothesis that Iressa directly interacts with ErbB4 to inhibit Nrg1-induced neuritogenesis rather than through an indirect interaction with ErbB receptors that heterodimerize and trans-phosphorylate ErbB4. The use of the iTrap affinity reagents, SPR assays, and kinase assay highlight that Iressa is not selective within the ErbB family. From this latter finding, in combination with the RNAi-mediated gene silencing data, we conclude that the inhibition of EGFR signaling is neither detrimental nor beneficial for Nrg1-induced signaling involved in neurite outgrowth.
For the purpose of the present work, we focused on the feature of mean neurite outgrowth per cell because it was sensitive to the dose−response of Nrg1 and NGF treatment and correlated with many other cellular features. However, we recognize that further analysis of other features may lead to other modulators of Nrg1−ErbB4 signaling. In addition, expanded screening of libraries of known bioactives, purified natural products, FDA-approved drugs, and products of diversity-oriented synthesis using the system described here could also yield other useful chemical tools and improve in-depth understanding of Nrg1-mediated neurotrophic processes. An example is our recent description of a potent pyridine-containing molecule (Cpd-52), which potently (EC50 = 300 nM) inhibited Nrg1-induced neurite outgrowth (51). Further investigation of the electrophysiological and biochemical effects of the compounds identified in this study, along with target identification and more in depth exploration of the underlying structure−activity relationships, would provide important insight into the role of Nrg1 in regulating neural circuitry.
It will also be possible to extend these screening efforts to include other signaling pathways implicated in neuropsychiatric disorders, including brain-derived neurotrophic factor/TrkB. We anticipate that chemical genetics will provide a wealth of novel small-molecule probes for dissecting the neural circuitry implicated in neuropsychiatric diseases both in cells and in vivo in animal models. Iressa (gefitinib) and Tarceva (erlotinib) are under clinical investigation for the treatment of glioblastoma multiforme (GM). Since Iressa and Tarceva are relatively well tolerated, and can cross the blood−brain barrier (in patients with GM), these results suggest a translational paradigm in which the small molecules identified in our cell-based assays may provide a means to test the hypothesis that Nrg1−ErbB4 signaling is associated with abnormal behavioral states in animal models and potentially humans.
PC12 cells (subclone Neuroscreen-1) were obtained from Cellomics (Now ThermoFisher Scientific, Pittsburgh, PA). pcDNA3-ErbB4 (52) was kindly provided by Dr. Steven R. Vincent (University of British Columbia, Vancouver). Antibodies used were rabbit anti-ErbB4 c-18 and rabbit anti-phospho-ErbB4 (Santa Cruz Biotechnology, Santa Cruz, CA, nos. SC283 and SC33040), rabbit anti-phospho-p42/44 MAPK and rabbit anti-p42 MAPK (Cell Signaling Technology), and mouse anti-phosphotyrosine 4G10 (Upstate, Charlottesville, VA). The EGF domain of Nrg1β1 (corresponding to amino acid residues 176−246 of neuregulin-1β1), was expressed and purified from Escherichia coli (R&D Systems; no. 396-HB) and reconstituted in phosphate-buffered saline (PBS) with 0.1% bovine serum albumin as a nonspecific carrier and frozen in aliquots at −20 °C. Murine 2.5S nerve growth factor (Promega; G5141) was reconstituted in PBS 0.1% bovine serum albumin as a nonspecific carrier and frozen in aliquots at −20 °C. siRNA SmartPools for ErbB1, ErbB2, ErbB3, ErbB4, nontargeting, and DharmaFECT2 were purchased from Dharmacon, Chicago, IL (L080049, L090224, L088799, L080170, D001810, and T-2002-03, respectively).
PC12 cells were maintained in RPMI 1640 media (Gibco; 22400) containing 10% heat inactivated horse serum (Gibco; 26050), 5% heat inactivated fetal bovine serum (Gibco; 16140), and 1% penicillin/streptomycin (Gibco; 10378) referred to here as RPMI+. For PC12-ErbB4-GFP and PC12-GFP, 1% penicillin/streptomycin was replaced with 750 μg/mL gentamicin (Gibco; 15750). Cells were passaged at 80−90% confluency and incubated at 37 °C in 5% CO2. Media was changed every 3 days. PC12 cells were cotransfected with pcDNA3-ErbB4-neomycin or pcDNA3-neomycin and pcDNA-GFP using FuGene 6 transfection reagent (Roche Diagnostics; 11814443). Cells that express the neomycin resistant gene were selected and maintained in same culture media with substitution of 750 μg/mL G418. After 2 weeks of G418 selection, cells were further selected by fluorescence-activated cell sorting (FACS) using a MoFlo Cell Sorter (Dako, Denmark) of the top 5% most strongly GFP expressing cells. The expression of GFP in the resulting cell populations, PC12-ErbB4-GFP and PC12-GFP, was observed to be stable for at least 50 passages.
Cells were lysed with RIPA buffer (Pierce Technology, Rockford, IL; 89901) containing 1 tablet/10 mL protease inhibitor cocktail Complete Mini (Roche Applied Science; 11836153). For phosphoprotein analysis, Halt Phosphatase Inhibitor Cocktail (Pierce Technology; 78415) was also included. Cell lysates were cleared by centrifugation at 15000 rpm for 30 min at 4 °C followed by addition of LDS sample buffer (Invitrogen; NP008) for direct analysis or were immunoprecipitated with specific primary antibodies and Protein A/G agarose (Pierce Technology 20421) following the manufacturer’s protocol. Samples were separated using a 4−12% gradient gel (Invitrogen) and SDS−PAGE and transferred to a polyvinylidene difluoride (PVDF; Schleicher & Schnell 10413096) membrane in 25 mM Tris, 192 mM glycine, and 20% methanol. The membrane was probed with specific primary antibodies according to specified recipes provided by their venders and then horseradish peroxidase-conjugated secondary antibody to mouse or rabbit IgG (GE Healthcare, Piscataway, NJ; NA934V and NA931V). Target protein bands were detected with SuperSignal West Femto Max Sensitivity Substrate (Pierce Technology; 34095).
Cells were typically seeded at a density of 4000 cells/cm2 and incubated for 12 h. siRNAs were prepared as 1 μM solution in serum-free medium and DharmaFECT2 was diluted 2:100 (v/v) with serum-free medium. Both solutions were incubated at room temperature for 5 min before mixing thoroughly and incubating for additional 20 min. The transfection mix was then added to cell culture at 1:5 (v/v). The cells were incubated for 24 h at 37 °C in 5% CO2 and treated with neurotrophic factors or directly lysed for Western blot analysis.
Cells were seeded in black, clear bottom, tissue culture-treated 96-well (Corning; 3904) or 384-well (Corning; 3712) plates at typical density of 4000 cells/cm2 corresponding to approximately 1200 cells per well of a 96-well plate in 100 μL of media and 300 cells per well of a 384-well plate in 40 μL of RPMI medium. Even distribution was achieved by a quick centrifugation at 500 rpm using a tabletop centrifuge (Sorvall, LegendRT) and multiwell plate adaptors shortly after seeding. Cells were then incubated for 12 h followed by treatment with growth factors or compounds as indicated. At specified time points, fluorescent images were taken using an ImageXpress 5000A or ImageXpress Micro automated microscopy (Molecular Devices) either manually or laser-based, autofocus with a Nikon 4× objective (ELWD S Fluor/0.20 NA) and an image acquisition time of 150 ms or as specified. Transmitted light images were taken using an ImageXpress Micro (Molecular Devices) with an attached transmitted light device with a Nikon 4× objective (ELWD S Fluor/0.20 NA). Neurite detection and analysis were performed with MetaXpress (Molecular Devices) using the “Neurite Detection” analysis module. Cell bodies were specified as pixel blocks of maximum width 40 μm, minimum area 200 μm2, and pixel intensities 1000 units above local background. Neurites were specified as linear objects with maximum width 3 μm and pixel intensities 500 units above the local background of the object being measured. Fluorescent images shown were imported as tagged image file format (TIFF) files into Adobe Photoshop (San Jose, CA) and in specified cases overlaid with transmitted light images that were processed in the same manner. After more than 4 days of incubation, significant cell detachment, cell clumping, and decreased GFP signal were observed and eventually caused aberrant detection of neurites. Although a longer exposure is potentially achievable by changing the cellular medium, we concluded that our automated neurite detection methods with up to 4 days incubation can reliably report the effects of Nrg1 and NGF on neurite induction and is sufficient to study quantitatively the kinetics of neurite outgrowth.
Cells were seeded into black, clear bottom, tissue culture-treated 96-well (Corning; 3904) or 384-well (Corning; 3712) plates at typical density of 4000 cells/cm2 corresponding to approximately 1200 cells per well of a 96-well plate in 100 μL of RPMI media and 300 cells per well of a 384-well plate in 40 μL of RPMI media. Even distribution was achieved by a quick centrifugation at 500 rpm using a tabletop centrifuge (Sorvall, LegendRT) and multiwell plate adaptors shortly after seeding. Cells were then incubated for 12 h to allow attachment. A total of 400 compounds, along with a total of 752 DMSO controls, were pin-transferred into wells of 384-well plates containing PC12-ErbB4-GFP cells prior to treatment of Nrg1 or NGF. After pinning compounds, either Nrg1 (20 ng/mL) or NGF (20 ng/mL) was added robotically to each well that received compound. At time points as specified, images were taken using ImageXpress 5000A (Molecular Devices) or ImageXpress Micro (Molecular Devices) automated microscopy systems using either manual or laser-based autofocus with a Nikon 4× objective (ELWD S Fluor/0.20 NA) and an image acquisition time of 150 ms using a xenon light source and 483/536 nm filter sets for measuring GFP fluorescence. Transmitted light images were taken using an ImageXpress Micro (Molecular Devices) with an attached transmitted light device with a Nikon 4× objective (ELWD S Fluor/0.20 NA). Neurite detection and analysis were performed with MetaXpress (Molecular Devices) using the “Neurite Detection” analysis module. Cell bodies were specified as pixel blocks of maximum width 40 μm, minimum area 200 μm2, and pixel intensities 1000 units above local background. Neurites were specified as linear objects with maximum width 3 μm and pixel intensities 500 units above the local background of the object being measured. Cell morphological features measured using MetaXpress software included: “% Cells Significant Growth” (the percentage of cells that have neurites of at least 10 μm in length), “Number of Cells” (the total cell count in the well of 384 well plate), “Total Outgrowth” (the total length of neurite within the well in micrometers), “Mean Outgrowth Per Cell” (the average total length of neurite per cell in micrometers), “Mean Processes Per Cell” (the average number of neurites per cell), “Normalized Mean Neurite Length” (the average length of neurite per cell in micrometers normalized to the values of DMSO treated wells), “Mean Branches Per Cell” (the average number of neurite branches per cell), “Mean Cell Body Area” (the average size of the cell in square micrometers). Processed screening data were visualized using Spotfire DecisionSite software (Somerville, MA) and in Microsoft Excel (Seattle, WA).
A custom library of 400 known bioactive compounds was assembled from commercial sources (Calbiochem, Sigma, Biomol, and Tocris) and stored in DMSO at −80 °C in custom dryboxes (see Supplementary Tables 1 and 2, Supporting Information, for a full list of compounds screened in the two conditions). For primary screening, compounds were robotically pin-transferred from 10 mM stocks in DMSO to a final concentration of ~10 μM using a CyBio CyBi-Well vario equipped with a 50 nL pin head. Follow-up characterization was performed on reordered stocks or compounds purified from expired pharmaceutical-grade tablets (Iressa and Tarceva).
The EGFR/ErbB4 surface plasmon resonance assays were conducted on a Biacore T100 instrument using Biacore CM5 sensor chips. Ethanolamine, EDC, NHS, and P-20 surfactant were all obtained from GE Lifesciences. An anti-GST antibody (GST capture kit, GE LifeSciences) was directly immobilized through primary amines using standard EDC/NHS chemistry according to the manufacturer’s instructions. Either GST alone or GST-ErbB4 (Cell Signaling) or GST-EGFR (Millipore) was captured to generate the ErbB4 or EGFR sensor chips, respectively. GST-fusion proteins were thawed immediately before use and kept at 4 °C during sample preparation. Binding assays were performed at 25 °C. The EGFR/ErbB4 assays were carried out in 1× PBS, 2% DMSO, 0.005% P-20 surfactant (PBS−P20). Compounds were injected at a flow rate of 30 μL/min into the flow cell for 60 s followed by 60 s of buffer without compound. Iressa and CHIR-99021 were stored in 100% DMSO and diluted into PBS−P20 with 2% DMSO for binding assays. CHIR-99021 activity was verified by binding to GST-GSK3β (BPS Biosciences) under identical conditions. Sensorgram data was analyzed using both Scrubber 2 software (BioLogic Software Pty) and Biacore T100 Evaluation software. Data was GST-reference subtracted and corrected for protein capture, DMSO concentration, and analyte molecular weight. Kd values were determined from steady-state binding values (Req) measured at 56 s of a 60 s injection and averaged over a 5 s window. Steady-state binding values were plotted against concentration values and fit using a model assuming 1:1 analyte to ligand binding.
See Supporting Information for detailed synthetic method and characterization of the iTrap affinity reagent. PC12-ErbB4-GFP cells were lysed with a modified RIPA buffer (50 mM Tris-HCl, pH 7.4, 1% NP40, 250 mM NaCl, 1 mM EDTA, 1 mM NaF, 1 mM Na3VO4, supplemented with an EDTA-free protease inhibitor cocktail) at 4 °C for 10 min, and the cell lysate was cleared by centrifugation. The cleared lysate (0.4 mL) was tumbled with Iressa or DMSO at indicated concentration at 4 °C for 30 min before addition of the iTrap resin (10 μL). The resulting mixture was tumbled at 4 °C for 12 h. The suspension was centrifuged, and the supernatant was discarded. The resin was washed with the above modified RIPA buffer (1 mL) for four times. After the final wash, the supernatant was removed, and SDS sample buffer (20 μL) was added to the resin. The affinity purified proteins were heat denatured and separated by SDS−PAGE. Western immunoblotting experiment was then performed using anti-ErbB4 antibody (BD Biosciences, Cat. No. 610808) following the manufacturer’s protocol.
We thank Pamela Sklar, Tracey L. Petryshen, Martha Constatine-Paton, Steve A. Carr, and Shao-En Ong for helpful comments throughout this work. Steven R. Vincent (University of British Columbia, Vancouver) is thanked for providing the pcDNA-ErbB4 cDNA clone. Ralph Mazitschek and James E. Bradner are thanked for providing Iressa and Tarceva. Daniel M. Fass is thanked for helping generate cell lines. We wish to thank the National Cancer Institute’s Initiative for Chemical Genetics (contract no. N01-CO-12400), who provided support for this publication, and the Chemical Biology Platform of the Broad Institute of Harvard and MIT for their assistance in this work. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Service, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.
EDC, (1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride; EGFR, epidermal growth factor receptor; ErbB4, v-erb-a erythroblastic leukemia viral oncogene homologue 4; GFP, green fluorescent protein; MAPK, mitogen-activated protein kinase; NGF, nerve growth factor; NHS, N-hydroxysuccinimide; Nrg1, neuregulin-1; PI3K, phosphotidylinositol-3 kinase.
National Institutes of Health, United States
Letian Kuai, Xiang Wang, and Jon M. Madison performed the experimental work and data analysis described herein. Stuart L. Schreiber, Edward M. Scolnick, and Stephen J. Haggarty provided guidance and advice. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.
The authors report no conflicts of interest.
L.K. and S.J.H were supported by funds from the Stanley Medical Research Institute and the Broad Institute SPARC program. S.J.H. was also supported in part by Grants 1R21MH076146-01 (NIMH) and 1R21MH087896-01 (NIMH). The National Cancer Institute’s Initiative for Chemical Genetics (Contract No. N01-CO-12400) provided support for the Broad Chemical Biology Platform.
Description of instruments and materials, figures showing synergistic effect of cotreatment of PC12-ErbB4-GFP cells with NGF and Nrg1, the effect of additional 4-anilino-quinazolines on Nrg1 and NGF-induced neuritogenesis, graphical representation of two-dimensional correlation map between eight cellular features in the Nrg1 and NGF high-content imaging screens, and comparison of relationships among eight cellular features from high-content imaging screen of Nrg1-induced neuritogenesis, and tables showing Nrg1 and NGF data sets from complete screen, Nrg1 and NGF data sets from DMSO controls only, Nrg1 and NGF data sets from compounds only, Nrg1 and NGF descriptive statistics and correlation analysis of data sets from compounds only. This material is available free of charge via the Internet at http://pubs.acs.org.