- Top of page
- Conflicts of interest
For more than half a century, synthetic glucocorticoids (GCs) have been extensively used to treat chronic inflammatory diseases such as rheumatoid arthritis, asthma and inflammatory bowel diseases (Barnes, 2006; Hillier, 2007). The basis of their therapeutic action is the impairment, in most cell types, of the expression of pro-inflammatory genes. Various unpredictable and occasionally life-threatening side effects of GCs have been documented since their earliest clinical use (Schacke et al., 2002). These include osteoporosis and diabetes mellitus, atrophy of skin and muscle, hypertension and increased susceptibility to infection. Nevertheless, GCs are still the cornerstone of treatment for many diseases. At the same time, major research initiatives attempt to separate the desired anti-inflammatory effects of GCs from their side effects (Schacke et al., 2007; Hudson et al., 2008; De Bosscher et al., 2010).
GCs modulate gene expression via the GC receptor (GR), a transcription factor belonging to the nuclear hormone receptor superfamily (Newton, 2000; Tuckermann et al., 2005). Lipophilic ligands such as the endogenous GC cortisol or the synthetic GC dexamethasone (dex) diffuse across the cell membrane and bind to GR in the cytoplasm. This promotes the release of GR from a large complex of chaperone proteins and its migration to the nucleus. In most but not all cases, transcriptional activation by GR (transactivation) is dependent on homodimerization, which is mediated by a short motif adjacent to the first of two zinc finger DNA-binding motifs. GR homodimers recognize sequences related to the idealized, palindromic consensus AGAACAnnnTGTTCT (GC response element or GRE).
A second physiologically important function of GR is to inhibit transcription via a mechanism known as transrepression (Kassel and Herrlich, 2007; De Bosscher et al., 2010; Glass and Saijo, 2010). In this case, GR does not bind directly to DNA but instead is recruited to DNA via direct or indirect interactions with other transcription factors, notably members of the activating protein 1 (AP-1) and NF-κB families, both of which play important roles in the expression of pro-inflammatory genes. The presence of GR at AP-1 or NF-κB binding sites is thought to inhibit transcriptional activation by impairing recruitment of transcriptional co-activators, or by promoting recruitment of co-repressors.
It is often stated that the anti-inflammatory effects of GCs are largely mediated by transrepression, whereas side effects are largely mediated by transactivation. If this is correct, it may be possible to improve upon classical GCs by identifying novel ligands of GR that selectively promote its transrepressive function rather than its transactivating function (Schacke et al., 2007; Berlin, 2010; De Bosscher et al., 2010; Newton et al., 2010). Such compounds are known as dissociated GR ligands, selective GR agonists (SEGRAs) or modulators (SGRMs). They are predicted to retain anti-inflammatory effects of classical GCs like dex but cause fewer or less severe side effects. Typically, SGRMs have been identified from drug libraries first on the basis of affinity for GR and second on the basis of effects on reporter constructs. Transactivation is tested against well-known GC target genes such as tyrosine aminotransferase (TAT), or against constructs that contain well-characterized GC responsive promoters or multimerized GR binding sites. Transrepression is tested using promoters that contain AP-1 and/or NF-κB binding sites and are activated by pro-inflammatory stimuli. Alternatively, reporters containing multimerized AP-1 or NF-κB binding sites may be used. As reviewed elsewhere (Schacke et al., 2007; Berlin, 2010; De Bosscher et al., 2010), a number of interesting compounds have been identified using this basic approach. Recently described examples include ZK216348 and LGD-552, which are non-steroidal GR ligands (Schacke et al., 2004; Humphrey et al., 2006; Miner et al., 2007; Lopez et al., 2008).
As well as directly inhibiting expression of pro-inflammatory genes by means of transrepression, GCs can exert indirect therapeutic effects, via the up-regulation of several anti-inflammatory genes (Clark, 2007; Newton and Holden, 2007). For example, many cell types respond to GCs by expressing dual specificity phosphatase (DUSP1), an enzyme that dephosphorylates and inactivates both p38 MAPK and JNK (Abraham and Clark, 2006; Owens and Keyse, 2007). The up-regulation of DUSP1 has been suggested to contribute to destabilization of pro-inflammatory mRNAs (Lasa et al., 2001; 2002; Quante et al., 2008) and to the inhibition of AP-1 and NF-κB function (Diefenbacher et al., 2008; Bladh et al., 2009; Cho and Kim, 2009; King et al., 2009). Correspondingly, many of the anti-inflammatory effects of GCs are impaired in macrophages derived from Dusp1−/− mice, or cells in which DUSP1 has been down-regulated using RNA interference (Abraham et al., 2006; Furst et al., 2007; Issa et al., 2007; Kang et al., 2008; Quante et al., 2008; King et al., 2009). In vivo anti-inflammatory effects of dex were dependent on DUSP1 in experimental models of acute inflammation (Abraham et al. 2006; Wang et al., 2008), asthma (Li et al., 2010) and rheumatoid arthritis (our unpublished results). There are therefore problems with a paradigm that equates anti-inflammatory effects of GCs with transrepression and not transactivation. In fact, such a model is not strongly supported by experimental evidence. For example, there is no genetically modified mouse strain that clearly demonstrates a separation between side effects and transactivation on one hand, and transrepression and anti-inflammatory effects on the other. A knock-in mouse strain expressing a dimerization defective mutant of GR (known as GRdim) was initially thought to provide evidence of just such a mechanistic separation between therapeutic and harmful effects (Tuckermann et al., 1999), but emerging complexities in the phenotype of the GRdim mouse now undermine rather than support the paradigm (Kleiman and Tuckermann, 2007; Frijters et al., 2010; Rauch et al., 2010). Recent results also show that, although SGRMs were identified on the basis of impaired transcriptional activation, they may be quite capable of inducing expression of DUSP1 and other genes with anti-inflammatory roles (Chivers et al., 2006; Janka-Junttila et al., 2006; Lopez et al., 2008; Newton et al., 2010). We therefore asked whether anti-inflammatory effects of SGRMs may actually be dependent on the induction of DUSP1. The answer to this question will have an important impact on how novel GR ligands with improved therapeutic indices are discovered, and how their properties are to be understood.
- Top of page
- Conflicts of interest
Novel compounds that preferentially mediate transrepression have been predicted to cause fewer side effects than classical GCs (Newton and Holden, 2007; Schacke et al., 2007; Berlin, 2010; De Bosscher et al., 2010). The notional mechanistic uncoupling of therapeutic and harmful consequences of GR activation suggested a straightforward course of action. Safer GR ligands might be discovered through screening strategies based on constructs that contain multimerized binding sites for GR itself (as a reporter for transactivating function) or for NF-κB or AP-1 (as reporters for transrepressing function). However, this idea has its roots in relatively simplistic and now outdated conceptions of how GR controls gene expression. Transcriptional activation by GR is now known to be a remarkably diverse process. Binding sites for GR can extensively vary from the idealized consensus AGAACAnnnTGTTCT, only five or six positions within this sequence being strongly constrained (So et al., 2007; Reddy et al., 2009; John et al., 2011). Single nucleotide variations in binding site sequence can have profound effects on the conformation adopted by GR and the downstream consequences (Meijsing et al., 2009). Only about 0.4% of possible binding sites are recognized by GR in one cell type, this repertoire being dictated by cell type-specific modulation of chromatin accessibility (John et al., 2011). GR also co-operates with a large number of other transcription factors to control transcription (Clark, 2007; Kassel and Herrlich, 2007). Some of these accessory factors are likely to be required for the establishment of domains of open chromatin structure within which GR can bind to DNA. At individual GC-regulated genes, and probably at individual cis-acting elements of one gene, GR displays different requirements for transcriptional cofactors (Chen et al., 2006; Galliher-Beckley et al., 2008; John et al., 2011). Finally, GR is extensively post-translationally modified, and GC-responsive elements may display differential requirements for different GR modifications (Beck et al., 2009; Galliher-Beckley and Cidlowski, 2009).
To stand as representative of all GR-mediated transcriptional activation events is therefore an unreasonably large burden for one highly simplified reporter, or even for one or two endogenous genes (Clark, 2007). Some of the problems of extrapolating from simple reporters are well illustrated by the present study. Even within one cell type (A549), the dose-dependence of induction of Dusp1 gene expression by the two SGRMs did not resemble the dose dependence of activation of the GRE reporter (compare Figures 2A and 4A). The predictive value of the reporter became even poorer when other cell types were considered (Figure 4B–D). Individual response elements of the Dusp1 locus were not necessarily better predictors of the behaviour of the endogenous gene. For example, a previous study identified a powerfully GC-responsive region located 4.6 kb upstream of the Dusp1 transcription start site (Tchen et al., 2010). In HeLa cells, this element was quite weakly activated by Cpd1 and Cpd2 compared with dex (Figure 3), whereas the endogenous gene was similarly activated by all three GR ligands at the relevant dose of 10−8 M (Figure 4B). At least some of this variation in response of reporter constructs and endogenous GC-regulated genes is probably explained by differential cofactor requirements and variable expression of cofactors in different cell types.
It is hard to escape the conclusion that idealized reporters containing tandem GR binding sites are of little practical help when trying to determine the transactivating properties (and hence dissociated nature) of GR ligands. RU24858, an earlier example of a supposedly dissociated GR ligand, was later found to be capable of up-regulating a subset of GR-regulated genes (Chivers et al., 2006; Janka-Junttila et al., 2006). LGD-5552, a near-relative of Cpd2, was also found to be capable of transcriptional activation but differed from the classical GC prednisolone in the profile of genes activated (Lopez et al., 2008). When a compound is described as being dissociated or as having poor capacity to activate transcription, it should therefore be asked exactly what this means and how it has been demonstrated. It is unclear whether genuine and consistent separation between transactivation and transrepression properties of GR can be demonstrated using simple reporters, or whether it can be achieved. We and other investigators have questioned whether such separation is even desirable, given that GCs up-regulate a number of anti-inflammatory factors, and depend on these factors for at least some of their anti-inflammatory effects (Smoak and Cidlowski, 2006; Clark, 2007; Newton and Holden, 2007).
Both Cpd1 and Cpd2 have emerged from deliberate efforts to selectively promote the transrepression rather than transactivation function of GR. With respect to transcriptional activation of GRE reporters and endogenous Dusp1 genes, both compounds behaved like partial agonists of GR. However, Cpd1 did not significantly block transcriptional activation by the full agonist dex (Figure 2D). This surprising finding is reminiscent of the first generation SGRM RU24858, which was also found not to inhibit transcriptional activation by the full agonist dex (Vayssiere et al., 1997). A possible conclusion is that dex and Cpd1 recognize different surfaces of GR and do not bind the receptor in a mutually exclusive manner. This appears unlikely because of the ability of RU486 to antagonize transcriptional activation by either compound (Figure 2C). A second possibility is that Cpd1 has relatively low affinity for GR in intact cells and is therefore not an effective competitor for binding. It should also be pointed out that demonstration of the expected partial antagonism is quite challenging in the case of Cpd1. At the respective concentrations of 10−7 and 10−6 M, dex and Cpd1 differ by only about 40% in transactivation of the GRE reporter (Figure 2D). This is a relatively small window in which to demonstrate competitive inhibition by Cpd1 of the response to dex. We cannot conclude that partial antagonism does not occur, only that we have been unable to demonstrate it. Cpd2 being a weaker activator of transcription, partial antagonistic behaviour was more straightforward to demonstrate (Figure 2D). To fully understand the partial agonist/antagonist properties of the two SGRMs requires biochemical and crystallographic studies that are beyond the scope of the present study. In any case, this issue does not effect our major conclusions.
Most importantly, the anti-inflammatory effects of the SGRMs were in direct proportion to their capacity to induce DUSP1 expression in a number of cell types and demonstrably dependent on DUSP1 in mouse macrophages. Cpd2, which had the higher dissociation indices, was the weaker inducer of DUSP1 and the poorer anti-inflammatory agent in all experimental settings used. The importance of DUSP1 as a mediator effect of GR was particularly clearly demonstrated in the case of IL-6. Neither dex, Cpd1 nor Cpd2 was capable of decreasing expression of IL-6 protein in Dusp1−/− macrophages. It is ironic that IL-6 is a well-characterized NF-κB target, whose inhibition by GCs has usually been interpreted in terms of transrepression.
Cpd1 is closely related to ZK 245186 (otherwise known as BOL-303242-X or Mapracorat), which is being tested as a novel drug for dermatological or opthalmological indications such as atopic dermatitis, dry-eye syndrome and postoperative eye inflammation (Schacke et al., 2009; Zhang et al., 2009; Cavet et al., 2010; Pfeffer et al., 2010; Shafiee et al., 2010). ZK 245186 was found to inhibit JNK and p38 MAPK phosphorylation in corneal epithelial cells subjected to hyperosmolar stress (Cavet et al., 2010). Inhibition of the MAPKs is thought to contribute to the therapeutic effect of the SGRM and may be mediated by up-regulation of DUSP1. The classical GC dex induces expression of Dusp1 in primary human lens epithelium (Gupta et al., 2005). In the present study, we were not able to assess responses to ZK 245186 itself, and we acknowledge the danger of extrapolating too far from in vitro studies. Nevertheless, we consider it quite possible that the anti-inflammatory efficacy and safety of SGRMs like ZK 245186 have little to do with whether or not these compounds are ‘dissociated’. This raises important questions about how the properties of the current generation of SGRMs should be interpreted and how the next generation of SGRMs might best be identified.