|Home | About | Journals | Submit | Contact Us | Français|
Xenoestrogens can affect the healthy functioning of a variety of tissues by acting as potent estrogens via nongenomic signaling pathways or by interfering with those actions of multiple physiological estrogens. Collectively, our and other studies have compared a wide range of estrogenic compounds, including some closely structurally related subgroups. The estrogens that have been studied include environmental contaminants of different subclasses, dietary estrogens, and several prominent physiological metabolites. By comparing the nongenomic signaling and functional responses to these compounds, we have begun to address the structural requirements for their actions through membrane estrogen receptors in the pituitary, in comparison to other tissues, and to gain insights into their typical non-monotonic dose-response behavior. Their multiple inputs into cellular signaling begin processes that eventually integrate at the level of mitogen-activated protein kinase activities to coordinately regulate broad cellular destinies, such as proliferation, apoptosis, or differentiation.
Endocrine-disrupting chemicals such as xenoestrogens are known to contaminate our environment and affect the reproductive health of animals and probably humans (Singleton and Khan, 2003). We and others (e.g., Otto et al., 2008) have studied a variety of subclasses of these compounds to explore understudied mechanistic pathways and receptors that they might engage to mediate their effects. These compounds either may act as inappropriate estrogens and/or could interfere with the actions of endogenous estrogens. Many disease susceptibilities (e.g., involving heart, brain, bone, or joints) worsen or change in women after menopause (Benedetti et al., 2001; Compton et al., 2002; Dluzen and Mickley, 2005; Foltynie et al., 2005; Kurlan, 1992; Mao et al., 2009; Quinn, 2005; Yoon et al., 2007) or at other life stages with different estrogen metabolite profiles, suggesting a differential protective or vulnerability effect due to different physiological estrogen metabolite levels. Therefore, interference by any of the xenoestrogens with these life stage–specific hormonal profiles may cause or alleviate stage-specific diseases in women. Because men also have these estrogen metabolites and receptors to bind to them, xenoestrogens are also likely to alter male physiology (Carreau and Levallet, 2000; Delbes et al., 2006).
Because many xenoestrogenic compounds bioaccumulate in fat tissues, resulting in prolonged and escalating human exposures, the exposure levels causing possible deleterious health effects are somewhat difficult to determine and are actively debated (Myers et al., 2009; Whitten and Patisaul, 2001). Although physiological estrogens can normally influence the growth and functioning of both female and male reproductive, skeletal, and cardiovascular systems (Cornwell et al., 2004), prolonged estrogen exposures have also been linked to the development of cancer in tissues, such as breast, colon, and pituitary (Brownson et al., 2002; Fritz et al., 1998; Mueller and Gooren, 2008). It is thus important to determine which of these estrogenic attributes are shared by xenoestrogens, and via which cellular mechanism(s) they operate, so that their suspected effects on health can be predicted, prevented, perhaps remediated, or even used therapeutically, such as in the case of phytoestrogens (Adlercreutz, 1995).
In the past, xenoestrogenic compounds have undergone extensive testing for actions via nuclear estrogen receptor–mediated gene transcription but have largely been found to act very weakly, if at all, via these genomic pathways (generally at least 1000-fold more weakly than estradiol [E2]). We have instead probed the ability of different classes and structurally variant estrogens to trigger signal cascades initiated at the plasma membrane via membrane estrogen receptors (α, β, and GPR30). To link these effects to the presence of specific receptor subtypes, our laboratory has demonstrated the importance of a membrane form of the estrogen receptor-α (mERα) in clonal rat pituitary cell lines that naturally express high or low receptor levels (Pappas et al., 1994); mERα was also the predominant receptor through which effects were mediated in neuronal cells, although the other estrogen receptors mERβ and GPR30 (Filardo and Thomas, 2005; Revankar et al., 2005) had inhibitory roles (Alyea and Watson, 2009a; Alyea et al., 2008). We also linked E2’s and xenoestrogens’ abilities to mediate membrane-initiated signaling via Ca++ elevation, prolactin (PRL) release, and various kinase activations to complex cellular events, such as cell proliferation and apoptosis (Jeng and Watson, 2009; Zivadinovic et al., 2005). Others have likewise found that specific estrogen receptor subtypes can mediate effects of xenoestrogens (Kuiper et al., 1998; Nadal et al., 2009). Xenoestrogen-altered responses could explain a variety of exposed tissue malfunctions, including both short-term functional deficits and long-term changes in cell and tissue activities (such as teratogenesis and cancer).
Human exposures to xenoestrogens have been associated with a variety of reproductive and neurological impairments (reviewed in Colborn, 2004; Hotchkiss et al., 2008; McKinlay et al., 2008). The actions of estrogens that we have compared in our own laboratory were largely in cultured cells representing anterior pituitary lactotrophs (Pappas et al., 1994). In the pituitary, estrogens facilitate both synthesis and regulated secretion of PRL (Dannies, 1985) and other peptide hormones, but we have focused on those actions that happen rapidly in response to estrogens—the secretory response. To understand to what kinds of pathologies these actions may be related, we must review the numerous roles of PRL. PRL coordinates the female hormonal cycle with preparation of tissues (e.g., mammary gland) for reproduction and the control of reproductive behavior. Hyperprolactinemia is a recognized cause of infertility as well as behavioral illnesses. Exaggerations in pregnancy behaviors and also in pseudopregnancy (where PRL levels rise without a pregnancy) include maternal behavior aggressiveness and sexual dysfunction. PRL overstimulation can also be correlated with depression, changed affect, and abnormal responses to stress (Sobrinho, 2003).
Estrogen-induced cell proliferation is part of the normal response of the pituitary but can also produce pituitary tumors (Gorski et al., 1997). PRL is believed to be a growth factor for many target tissues, including breast and prostate (Adams, 1992; Nevalainen et al., 1997), so its overproduction may lead to pathologies or tumors of these tissues (Clevenger et al., 2003; Gutzman et al., 2004; Rose-Hellekant et al., 2003). If any xenoestrogens act differently than the normal timing and levels of endogenous estrogens, then imbalances of PRL secretion and stimulation could occur; these could be developmental stage specific or gender specific and might cause unanticipated harmful responses. Our studies demonstrated how this could occur mechanistically for different estrogens and xenoestrogens via membrane-initiated signaling pathways (Bulayeva and Watson, 2004; Jeng and Watson, 2009; Jeng et al., 2009; Kochukov et al., 2009; Watson et al., 2008).
Estrogen mimetics and antagonists have long been noted to have selective estrogen receptor modulation effects, such as those responding differently in different tissues but via the same receptor (Azuma and Inoue, 2004), usually explained by alternative associations with such transcription factor co-modulators as histone acetyl transferases and histone deacetylases (Liu and Bagchi, 2004). Xenoestrogens appear to be SmERMs, i.e., they are selective membrane estrogen receptor modulators. We and others have demonstrated that these selective effects could result from alternative partnering with other signaling proteins in different tissues or regulatory circumstances (Alyea and Watson, 2009b; Boonyaratanakornkit et al., 2001; Song et al., 2005). As different tissue and cell types are explored for activities in the nongenomic signaling pathway with these different xenoestrogens, tissue-selective profiles will emerge for nongenomic responses such as those we have seen for pituitary cancer cells (Bulayeva and Watson, 2004; Jeng et al., 2009; Wozniak et al., 2005), breast cancer cells (Zivadinovic and Watson, 2005; Zivadinovic et al., 2005), cells of the immune system (Narita et al., 2007), and neuronal cell types (Alyea and Watson, 2009a), and others have seen for bone versus mammary tissues (Otto et al., 2008). This should help to explain their toxicities.
Over the past decade or so, our laboratory has examined several subgroups of estrogens with different potentials for promoting or interfering with estrogenic actions via signaling at the cell membrane. The 16 compounds that we have investigated so far (see Fig. 1) represent either structurally related subgroups (endogenous hormone metabolites, alkylphenols, chlorinated pesticides, soy isoflavones) or are related via their use and exposure route (dietary vs. endogenous metabolites vs. environmental contaminants from industry, agriculture, or consumer products). Considering them together, let us begin exploring the structural requirements for nongenomic estrogenic signaling via different pathways. Collectively, we and others have learned that estrogens originally deemed “weak” for their actions in the nucleus can potently activate nongenomic signaling pathways (Alyea and Watson, 2009a,b; Bouskine et al., 2009; Bulayeva and Watson, 2004; Jeng and Watson, 2009; Jeng et al., 2009; Kochukov et al., 2009; Nadal et al., 2009; Otto et al., 2008; Watson et al., 2007, 2008). It is also clear that the rules for engagement via membrane-initiated estrogen actions for various signaling cascades differ.
While E2 is the physiological estrogen most often studied and associated with reproductive function during the reproductive years, other endogenous estrogenic compounds can be more prevalent during other life phases. These other estrogens may have significant effects on tissue development, function, and disease states (such as the development of cancers in reproductive tissues). Estrone (E1) is a significant estrogenic hormone contributor in both reproductive (~0.5–1nM) and postmenopausal (150–200pM) women and in men (~100pM); estriol (E3) levels are significantly higher in pregnant women (~10–100nM) than in nonpregnant women (< 7nM) (Greenspan and Gardner, 2004). Lowered E3 levels in pregnancy have been associated with complications of eclampsia (Shenhav et al., 2003) and the incidence of Down’s syndrome in offspring (Chard and Macintosh, 1995). These estrogenic metabolites are also produced by aromatases in a number of nonreproductive tissues where their effects may extend beyond reproductive functions (Meinhardt and Mullis, 2002). One example is that E3 has protective effects against the development of arthritis in certain experimental models (Jansson and Holmdahl, 2001), as has been known previously for E2. Effects in brain, bone, cardiovascular system, and many other tissues may be affected differentially by these three endogenous estrogenic compounds during different life stages; therefore, loss or enhancement of these effects due to interference by xenoestrogenic compounds could affect human health in a large number of tissues. These metabolites also present an interesting structure-activity study group as their modifications are simple variations at only two positions on their D-rings (see Fig. 1). Scant previous information about the actions of physiological concentrations of E1 and E3 via nongenomic steroid signaling mechanisms (Morley et al., 1992; Selles et al., 2005) have now been augmented by our recent studies (Alyea and Watson, 2009b; Watson et al., 2008). We saw that E1 and E3 were similarly potent with E2 in some responses for both pituitary cells (increasing the number of Ca++-responding cells and evoking extracellular-regulated kinase [ERK] phosphorylation) and neuronal cells (evoking dopamine efflux). However, in neuronal cells, E1 and E3 (inhibitory) had opposite effects from E2 (stimulatory) on the activation of dopamine efflux by the dopamine transporter (DAT), and these physiological hormones achieved this by differentially causing rapid trafficking of the estrogen receptors (α, β, and GPR30) and DAT to and from the plasma membrane. Further exploration of these potent and differential effects of physiological estrogen metabolites, and interference with their activities by xenoestrogens, could illuminate other life stage–specific changes in estrogen-related disease vulnerabilities.
Alkylphenols represent a group of ubiquitous environmental estrogens that are highly related in structure, although somewhat different from E2. These compounds are surfactants or monomer byproducts of plastic manufacturing or product breakdown. They have been found at surprisingly high concentrations in human fluids (Lakind and Naiman, 2008; Stahlhut et al., 2009) and at environmental sites (Kolpin et al., 2002; Talsness et al., 2009; Thomas and Doughty, 2004). Our laboratory compared several members of this class with either different lengths of carbon side-chain modifications at Position 4 on the phenol ring (nonyl-, octyl-, propyl-, and ethylphenol [NP, OP, PP, EP]; see Fig. 1), or instead an added phenolic group (bisphenol A [BPA]). These compounds were active at very low doses in our studies (Kochukov et al., 2009), a potency confirmed by others studying other endocrine tissues (Alonso-Magdalena et al., 2008; Nadal et al., 2009). These comparisons represent low environmentally common concentrations (femtomoles to nanomoles). Overall, the alkylphenols are quite potent in several of our assays for nongenomic responses, including PRL release, cell proliferation, calcium (Ca++) influx, and in the activation of mitogen-activated protein kinases (MAP kinases) (Bulayeva and Watson, 2004; Kochukov et al., 2009; Wozniak et al., 2005). These activities are summarized for all our publications to date in Figure 2.
By comparison, BPA and nonylphenol have shown very low potency in nuclear transcription assays for estrogen-responsive genes (Gaido et al., 1997; Gutendorf and Westendorf, 2001; Kloas et al., 1999; Sheeler et al., 2000; Singleton et al., 2004; Steinmetz et al., 1997). The long carbon side-chain alkylphenols were previously shown to have weak estrogenic activity in genomic assays, and the shorter side-chain versions were even less active (Kwack et al., 2002; Routledge and Sumpter, 1997; Tabira et al., 1999). In contrast, their nongenomic activities are quite robust, and short or long carbon chain variants are more effective in different responses (Kochukov et al., 2009). Therefore, inactivity in genomic assays does not predict inactivity in nongenomic mechanisms. In addition, this class of xenoestrogens is becoming increasingly important to consider for further modification by chlorination in manufacturing and waste water treatment plants (Fukazawa et al., 2001; Gallard et al., 2004; Gross et al., 2004; Hu et al., 2002; Petrovic et al., 2003), so structure-activity knowledge about their estrogenic effects will become increasingly important.
Our laboratory has also performed nongenomic signaling studies with several chlorinated pesticides known to be estrogenic—dieldrin, dichlorodiphenyldichloroethylene, and endosulfan. These compounds break down slowly, and so persist in the soil even though their use has largely been banned (U.S. Department of Health and Human Services, 2005). Plants and animals that are part of the food supply become exposed, subsequently passing on these exposures to humans. Because many xenoestrogens bioaccumulate in fat tissues, resulting in prolonged and escalating human exposures, the exposure levels causing deleterious health effects are actively debated (Myers et al., 2009). In our studies, these chlorinated xenoestrogen pesticides were quite effective in eliciting all the responses examined, including ERK activation (Bulayeva and Watson, 2004), PRL release, and Ca++ influx (Wozniak et al., 2005). The studies of others have also demonstrated rapid signaling actions of these compounds on endocrine cells (Wu et al., 2006).
Phytoestrogens, another category of nonphysiological estrogens, have diverse estrogenic biological activities due in part to their ability to act as either estrogen agonists or antagonists depending on the dose and the specific tissue involved. These abilities have caused a lot of attention to be focused on these compounds as potential safe, effective, and inexpensive estrogen replacement medications. Coumestrol, first reported to be estrogenic when it was associated with disrupting reproduction in livestock (Bickoff et al., 1957), is found in such dietary sources such as legumes, clover, and sprouts of soybeans and alfalfa. The reported serum concentration resulting from ingesting these foods in humans is approximately 0.01μM (Mustafa et al., 2007). Isoflavones are represented in our studies by daidzein and genistein, and their major source is soy-based foods. In Asia, the intake of soy is high, and plasma concentrations of genistein from 0.1 to 10μM have been measured (Mustafa et al., 2007; Whitten and Patisaul, 2001); Western diets usually contain about 10-fold lower concentrations (Adlercreutz et al., 1993). Some isoflavones, such as genistein, have also been shown to act predominantly via estrogen receptor-β in genomic responses (Kuiper et al., 1998). Trans-resveratrol, a stilbene (Gehm et al., 1997) that has recently attracted significant attention as a potential anti-aging agent, is found in high quantities in foods such as red grapes (or wine) and peanuts and has peak serum concentrations estimated to be close to 2μM in humans (Walle et al., 2004).
To rank the effectiveness of all these compounds together and to examine one chemical feature of xenoestrogens thought to facilitate their behavior as estrogens, we graphed their responses according to each compound’s lipophilicity (Fig. 2). We chose an octanol-water partition coefficient to numerically represent this value for graphing. We combined the results on compounds from different classes of estrogens and xenoestrogens, gleaned from a number of our studies, so as to compare their lipophilicity to their estrogenicity at multiple end points in the nongenomic pathway. Depending on the signaling or functional end point being assessed, the lipophilicity value positively influenced (PRL release, Ca++ oscillation frequency, and p38 activity), negatively influenced (Jun-Kinase [JNK] activity), or did not influence (proliferation and ERK activity) a response parameter. Also, not influenced by lipophilicity (data not shown) was the total amount of Ca++ influx (combined peak areas, correlation coefficient of r = 0.0015). Therefore, lipophilicity is one characteristic of xenoestrogens that can partially predict some aspects of estrogenicity. There are undoubtedly other aspects of these chemicals’ structures that will need to be evaluated in the future for their contributions to such predictions. It is not surprising that for different end points, estrogens can have positive influence, negative influence, or no influence. Estrogen receptors liganded by a given estrogen will create specific shape changes in the receptor, resulting in a different constellation of interaction surfaces (Pike et al., 1999) to which other proteins can bind. Partner proteins may be activated or further recruit other proteins, leading to a given functional response.
Nongenomic estrogenic responses often display non-monotonic dose-response characteristics (Weltje et al., 2005) (dose curves that do not follow the principle of low-dose effects rising from a threshold and plateauing at higher concentrations). Response decay at higher steroid concentrations (∩-shaped curves) has been a curious feature long noted for genomic and whole-animal functional responses (Welshons et al., 2003). Non-monotonic dose-response curves studied in our laboratory for rapid nongenomic responses to estrogens not only reverse direction at higher doses but sometimes rise and decline multiple times with increasing dose over a wide concentration test range (perhaps representing two ∩-shaped curves put together). Such dose relationships make it impossible to extrapolate high-dose effects to predict low-dose effects. Since much toxicological regulatory policy is based on putative monotonic dose relationships, toxicants that mimic the non-monotonicity characteristic of hormone effects have important ramifications for the regulation of compounds such as environmental estrogens. We have probably been able to observe these responses to very low levels of hormones and mimetics in our studies because of our ability to manipulate cells in culture to exclude endogenous hormone sources entirely (defined media and exhaustively steroid-stripped media). Examples of two such responses to environmental estrogens demonstrating this non-monotonic characteristic from our own work are shown in Figure 3. EP, an alkylphenol, is shown here to cause rapid PRL release at very low concentrations and at higher concentrations but reverses this response at intermediate concentrations. BPA elicits a similar behavior in one example of MAP kinase responses—ERK phosphorylation. The shapes of these dose-response curves are similar to those we have observed for both physiological and a variety of nonphysiological estrogens (Alyea and Watson, 2009a; Bulayeva and Watson, 2004; Jeng and Watson, 2009; Jeng et al., 2009; Kochukov et al., 2009; Watson et al., 1999, 2008; Wozniak et al., 2005; Zivadinovic et al., 2005).
We have speculated that such typical bimodal dose responses for these membrane-initiated mechanisms could result from different receptor subpopulations present in unique compartments of the plasma membrane. For example, membrane forms of steroid receptors have also been shown to reside in membrane caveolae by us (Zivadinovic and Watson, 2005) and others (Chambliss and Shaul, 2002; Lu et al., 2001; Norman et al., 2002), where it is well known that lipid content and other signaling molecule and scaffolding protein availability are quite different from non-raft membranes. Basolateral versus apical or endocytosed membrane compartments represent compartments of different accessibility for hormones to their receptors (Cao et al., 1998), although this is less likely to be the case for small lipophilic molecules like steroids or their mimetics than for peptide hormones. Subcellular location–based availabilities could also dictate different physical associations with other proteins by altering hormone-binding and -partnering opportunities. In addition, differences in lipid content simulated in artificial membranes are known to affect the functioning of proteins imbedded therein (Wu and Gorenstein, 1993), likely causing alteration of ligand-binding pockets and protein partner interaction interfaces. Therefore, characteristics of receptors that target to the membrane or membrane subcompartments may affect signaling responses.
We also know from our own work that estrogens, including xenoestrogens, can signal via several different pathways simultaneously although differentially and that these signals traverse their pathways at variable speeds (Bulayeva and Watson, 2004). Different phasing of pathway travel, along with feedback or feedforward regulation or crossing over to parallel paths, can result in complex contributions to dose-response changes, not obvious when examined at a single time point. An example could be the estrogenic activation of a phosphatase, which then inactivates another protein, such as a kinase; if the response being monitored is the kinase activation, then one would see an unexpected decline in the response whenever the phosphatase has been activated. We have some evidence for this effect on response curves in breast cancer cells (Zivadinovic and Watson, 2005 and Banga and Watson, unpublished data). For nuclear receptor actions, it is known that there are different dose-response sensitivities to the same level of nuclear steroid receptor for different genes in the same tissue (Catterall et al., 1985; Simons, 2008), probably also involving receptors partnering with other receptors or coregulators that are target gene specific. Ligand-induced conformations of nuclear receptors can vary due to the structure and level of the hormones bound to them, likely altering their interaction interfaces being presented to other regulatory molecules, thus initiating different responses (Pike et al., 1999). Such mechanisms are also likely to account for associations that regulate membrane-initiated signaling cascades.
Many more examples of nonconventional dose responses will likely be found owing to advances in cell culture and assessment methods for such studies. Our increased understanding of responses to very low concentrations of these ligands demonstrates that animal cells can be extraordinarily sensitive to estrogens. We now use more effective methods for removing other active molecules from culture media (Cao et al., 2009; Wilkinson, 1993) without disturbing the osmolality and protein content milieu needed for normal cell signaling and regulation (Campbell et al., 2002). Completely defined media are now becoming much more available so that cells can be treated with very small known quantities of signaling ligands without interference from endogenous serum-resident steroids and other molecules that can mask responses. Quantitative assays for phosphorylated signaling proteins make detection of small changes at low concentrations more easily defined as significant compared to immunoblot technologies (Bulayeva et al., 2004; Howells et al., 2008; Versteeg et al., 2000). These advances made it possible to do much more comprehensive dose-response analyses, encompassing the low physiological or disease-relevant concentration ranges of steroids and their mimetics. Single-cell analysis methods (e.g., for Ca++ responses) allow detection of responses that would otherwise be masked by unresponsive cells in the same culture; cell expression levels for mERα that control their responsiveness to estrogens are heterogenous at any given time point (Kochukov et al., 2009; Wozniak et al., 2005) due to cell cycle changes and other forms of cellular regulation (Campbell et al., 2002). The “lower hump” of the “camelback” non-monotonic dose-response curve (see Fig. 3) is only recently being examined and appreciated.
In the past, we have most often studied signaling mechanisms one pathway at a time. However, an overview of the accumulated data and powerful new genome-wide assessment technologies have taught us that the regulation of cellular function is more realistically depicted by a convergence of information delivered via many pathways, yet with pathway-selective use by some ligands (Michel and Alewijnse, 2007). Recently, the study of the nongenomic actions of multiple xenoestrogens has made a particularly striking example of this new point of view as these compounds activate many parallel pathways at once, yet selectively with respect to timing and predominant use of these different signaling avenues (Bulayeva and Watson, 2004). These considerations also are important as we contemplate the multiplicity of estrogens we are exposed to at any given time and the effect of the combination of them on signaling outcomes.
The changing circumstances to which a cell must respond are first presented by the ligands that it encounters at its surface, of which there are many that arrive simultaneously. The engagement of surface receptors by these ligands sets in motion coordinated actions, eventually leading to major cellular decision points: proliferation, differentiation, or death. Cells must integrate all these incoming signals and parallel pathways that can eventually contribute to a final common pathway, such as those involving MAP kinases (see Fig. 4). These enzymatic “signal-receiving stations” tally up many inputs from multiple signaling cascades and sum them toward establishing a level of end point of MAP kinase (ERK, JNK, and p38) activities. The final count dictates a decision about how to regulate major cellular responses that require coordination so that the whole cell “is on the same page” during major cellular responses to change. Acting via their membrane receptors, steroids can be one class of input signals to the MAP kinase signal integrator, the cellular rheostat that is dialed up or down according to which pathways feed into it. Not all estrogens elicit identical responses (in level or timing) along these pathways (Bulayeva et al., 2004). Thus, different endogenous metabolites (representing different life-stage challenges) or xenoestrogen mimetic exposures will cause a different tally and resulting cellular response. Estrogens provide a very illustrative example of these cellular strategies because of the availability of many different and medically/environmentally important estrogens that make differential use of the same pathways. This example thus gives new insights into how such signaling webs can lead to variant outcomes depending on which estrogens engage them.
New layers of detail in the workings of estrogen mimetics through nongenomic pathways are being revealed, including characteristics of their unusual dose responses, and specific subclasses of xenoestrogens whose structure-activity relationships can be compared. We are learning how these compounds are different or similar to physiological estrogens when acting at membrane estrogen receptors. While the lipophilicity of these compounds can predict some of their signaling capabilities, there are clearly other structural features that dictate other aspects of the diverse signaling responses these compounds elicit. We have presented examples that show how such understanding of estrogenic or xenoestrogenic ligands may allow us to determine their estrogenic or antiestrogenic potential and use this knowledge to design or prioritize intervention opportunities.
National Institute of Environmental Health Sciences (R01 ES015292 to C.S.W.); NIDA (P20 DA024157 to C.S.W.); American Institute for Cancer Research (06A126 to C.S.W.).
We gratefully acknowledge Dr David Konkel for expert editorial assistance for this review.