Connections matter − how viruses use cell–cell adhesion components

Cell adhesion is a central mechanism that drives the development of multicellular organisms. Indeed, cells use adhesion to move, communicate and differentiate, which ultimately leads to the formation of epithelia and highly organized organs. Adhesion occurs through specific cell-adhesion molecules, depending on the type of interaction. Cell adhesion to the extracellular matrix is mostly supported by integrins. Cell–cell adhesion is ensured by the apical junction complex (AJC) that contains tight junctions (TJs) and adherens junctions (AJs) (Fig. 1A) (Farquhar and Palade, 1963). The AJC is important for the permeability and polarization of epithelia. Whereas AJs are essential to anchor cells to each other and ensure epithelium integrity, TJs seals the space between the cells (Shen, 2012), only allowing selective passage of ions and size-selective diffusion of non-charged molecules (Steed et al., 2010). In addition to the AJC, desmosomes participate in cell–cell adhesion along the basolateral membrane.

There are several classes of virus; either with RNA or DNA as genetic material, and either with or without an envelope. Viral replication cycles differ but always begin with the recognition of a specific protein or oligosaccharide on the cell surface, to be named a receptor, and cell entry. Entry leads to release of the viral genome in the cytoplasm or its transport to the nucleus, depending on the virus class. Expression of the proteins that are necessary for viral replication and particle assembly follows. New virus particles are released, and a replication cycle starts anew when these particles enter the next cell displaying a specific receptor.

Several viruses use proteins of the AJC as receptors for entry in their target epithelium, even if most of these proteins are not readily accessible. As shown in Fig. 1B (viral entry strategy 1), some viruses directly contact their receptor in the junction, possibly by taking advantage of compromised epithelium. Other viruses (viral entry strategy 2) first attach to co-receptors exposed on the apical surface that further transport the particle to the AJC, where it enters through its receptor. Viruses that are already in an organism approach the basolateral side of the AJC (viral entry strategy 3). But how do these viruses get there? The remarkable strategy – developed by the measles virus and the other morbilliviruses – to first bypass the respiratory epithelium and then invade it from its basolateral side has only recently been understood (Box 1). Morbilliviruses include important animal pathogens, such as canine distemper (infecting dogs and other carnivores) and rinderpest (infecting cows and other ungulates).

In this Commentary, we describe how several viruses use AJC proteins for entry and propagation in epithelia. Whereas some viruses immediately disrupt epithelium integrity, thereby bypassing the epithelial barrier, others adopt a more sophisticated strategy. They do not cause immediate harm but take advantage of the AJC-cytoskeletal connections to spread very rapidly without causing any apparent cytotoxicity. Hidden viral spread in epithelia may be key not only to virulence but also to the success of cancer therapies that are based on viruses that replicate selectively in transformed cells and use AJC proteins as receptors.

Most organs are lined by columnar epithelial cells with an apical side in contact with the lumen and a basolateral side in contact with the lamina propria (Fig. 1A). The TJ is the most apical junction in columnar epithelial cells and is composed of several integral and peripheral proteins (Steed et al., 2010). Whereas the respiratory and enteric epithelial cells – which allow replication of most viruses – have a columnar structure, liver epithelial cells have a different organization (Fig. 1C). The lateral side of hepatocytes is in contact with sinusoidal vessels and the apical cell membranes of two neighboring hepatocytes form the bile canaliculus. A single hepatocyte participates in the formation of two canaliculi and two AJCs are found on each side of a canaliculus.

TJ integral membrane proteins interact with partners on the neighboring cell membrane to form the intercellular seal. In addition, TJs contain peripheral proteins that connect the integral membrane proteins to the underlying actin cytoskeleton, as well as to signaling proteins [in Fig. 2A, the connecting zonula occludens (ZO) proteins are shown in yellow]. Several additional TJ proteins have been identified that can confer important tissue-specific properties to certain TJ. However, we focus here only on the main TJ components.

Two types of integral protein are found in TJs: multipass transmembrane proteins, such as claudin and occludin, that trans-interact with their partners through their external loops (Fig. 2A), and single-span transmembrane proteins of the immunoglobulin (Ig) superfamily. Ig superfamily proteins include the coxsackievirus and adenovirus receptor (CAR; officially known as CXADR) and a junctional adhesion molecule (JAM) (Fig. 2A).

Although occludin was the first TJ transmembrane protein identified (Furuse et al., 1993), it is not required for the formation of TJs (Saitou et al., 1998) but rather regulates their permeability (Steed et al., 2010). Claudin, the most essential transmembrane TJ component, seals the TJs (Van Itallie and Anderson, 2013). Strikingly, there are more than 23 members of the claudin family in humans, and the diversity of their trans-interactions may account for the distinct sealing properties of different epithelia. Homotypic trans-interactions with other claudins on adjacent cells involve both extracellular loops and are instrumental for TJ formation.

JAM-A, JAM-B and JAM-C (officially known as F11R, JAM2 and JAM3, respectively) are components of the TJs that form between epithelial and endothelial cells (Liu et al., 2000; Williams and Barclay, 1988). JAM-A and JAM-C are also expressed by leukocytes, which take advantage of trans-interactions between JAMS to migrate across endothelial and epithelial barriers (Del Maschio et al., 1999; Woodfin et al., 2011). Expression of CAR is exclusively restricted to epithelial cells (Fig. 2A, top) (Cohen et al., 2001; Raschperger et al., 2006). Homotypic trans-interactions of JAMs and CAR (see Box 2 for details) contribute to the formation and stability of the TJ (Chiba et al., 2008; Prota et al., 2003; van Raaij et al., 2000).

Binding motifs in the cytosolic tails of claudin, CAR and JAM (Bonazzi and Cossart, 2011) interact with PDZ domains present in the zonula occludens proteins ZO-1, ZO-2 and ZO-3 (officially known as TJP1, TJP2 and TJP3, respectively) (Coyne and Bergelson, 2005; Ebnet et al., 2004; Van Itallie and Anderson, 2013). In the case of occludin, a C-terminal coiled-coil domain interacts with ZO-1 (Cummins, 2012). ZO proteins are central scaffolding components of the TJ. ZO-1 interacts with ZO-2 and ZO-3, and these interactions may contribute to the clustering of the integral TJ proteins. In addition, ZO proteins link the integral proteins to the actin cytoskeleton and to intracellular signaling molecules (Steed et al., 2010) (Fig. 2A). These multiple interactions connect the filamentous actin (F-actin) belt with the AJC (Fig. 1A) (Ivanov, 2008), meaning that actin belts of adjacent cells interact through the AJC. Those are the connections that matter for viruses.

Remarkably, all four main TJ components are viral receptors. The hepatitis C virus (HCV) uses both occludin and claudin to enter hepatocytes (Fig. 3A). Entry of HCV in the liver epithelium begins with the recognition of two basolateral co-receptors – SRBI (officially known as SCARB1) and CD81 – by viral particles that circulate in the bloodstream (Reynolds et al., 2008). These co-receptors may transport particles to the TJ through a mechanism resembling that used by coxsackieviruses to access CAR in the TJ (Evans et al., 2007) (see Viruses targeting airway TJs, below).

After reaching the TJ, binding to occludin and claudin is required for HCV cell entry (Evans et al., 2007; Ploss et al., 2009). It is possible that virus binding to these receptors disrupts their homotypic trans-interactions: the HCV attachment protein E2 may contact the second extracellular loop of occludin (Liu et al., 2010) and the first extracellular loop of claudin-1, claudin-6 or claudin-9 (Evans et al., 2007; Zheng et al., 2007). Consistently, when trans-interactions between TJ proteins are impaired, downregulation of TJ components at the cell surface and increased permeability are observed after infection with HCV (Benedicto et al., 2008).

Initial entry of HCV to the liver epithelium and spreading of HCV through cell-free particles – that are released by infected cells – both require SRBI and CD81. However, cell-to-cell spread in absence of cell-free virus particles mainly relies on claudin-1 (Timpe et al., 2008) through a mechanism that might require – yet unknown – co-factors (Catanese et al., 2013). The recent development of primary liver cultures should help to answer important questions regarding the cell-to-cell spread of HCV in relevant model systems (Andrus et al., 2011; Sourisseau et al., 2013). It is worth noting that HCV infectivity is strongly cell-associated and that extracellular particles are barely detectable (Vieyres and Pietschmann, 2013).

The integral TJ protein CAR owns its name to two viruses; it was originally identified as the coxsackievirus and adenovirus receptor (Bergelson et al., 1997). In particular, CAR is a membrane receptor for group B coxsackieviruses – that are members of the Picornaviridae family, and also for subgroup C adenoviruses (serotypes 2 and 5) – two species C members of the Adenoviridae family. The two viruses are very different; the coxsackievirus is a small positive strand RNA virus, whereas the adenovirus is a medium to large double-stranded DNA virus (Fig. 3A). Thus, individual members of two virus families have evolved to use the same AJC protein. However, the strategies used by these viruses to gain access to CAR are distinct.

Coxsackievirus B initially binds to a co-receptor, the complement decay-accelerating factor (DAF; officially known as CD55) that is located on the apical surface of polarized epithelial cells (Bergelson et al., 1995). Remarkably, DAF engagement sets in motion a signaling system. It activates Abl kinase and Rac GTPase, which – in turns – leads to actin cytoskeleton rearrangements and transport of the DAF–virus complex to the TJ, where the capsid interacts with CAR (Bergelson, 2009; Coyne and Bergelson, 2006) (Fig. 1B).

How adenoviruses access CAR is less clear. Polarized epithelial cells are only susceptible to basolateral infections with adenoviruses (Grubb et al., 1994; Pickles et al., 1998). The initial apical infection may occur only when epithelium integrity is compromised and CAR proteins are exposed to the lumen (Lütschg et al., 2011). Interestingly, adenovirus replication leads to excess production and extracellular release of the attachment protein Fiber knob that can directly disrupt the TJ by impairing CAR-CAR interactions or inducing an inflammatory response that then leads to reorganization of the AJC (Coyne and Bergelson, 2005; Coyne et al., 2002; Walters et al., 2002).

Both coxsackievirus and adenovirus bind CAR (Bewley et al., 1999; He et al., 2001) with nM-range affinity (Coyne and Bergelson, 2005; Goodfellow et al., 2005), whereas the homophilic CAR–CAR interaction has an affinity that is in the µM range (van Raaij et al., 2000). Thus, after entry from the apical side, both coxsackieviruses and adenoviruses may disrupt homotypic CAR-CAR interactions at the TJ and, therefore, overall TJ integrity.

The integral TJ protein JAM-A is a receptor for reoviruses (Barton et al., 2001) but how these viruses gain access to JAM-A is not clearly understood. It has been proposed that sialic acid facilitates transport of reoviruses to the TJ in a way that is similar to DAF supporting the transport of coxsackieviruses (Bergelson, 2009).

The prefix ‘reo’ is derived from respiratory enteric orphan viruses, which have a double-stranded RNA genome. Distinct reoviruses infect many species but rarely cause disease. Interestingly, the initial entry of reovirus particles into the host occurs independently of JAM-A and, instead, involves transcytosis of the virus through the enteric epithelium (Wolf et al., 1981). Indeed, recent work with JAM-A-deficient mice suggests that this protein is not critical for virus replication in epithelia but, rather, for the virus crossing the endothelium and spreading into the bloodstream (Antar et al., 2009). Moreover, it is unclear whether reoviruses directly replicate in the endothelium or just cross it within infected leukocytes that use JAM-A to migrate in the bloodstream (Bergelson, 2009).

How reoviruses bind to JAM-A is understood at the atomic level (Kirchner et al., 2008). The attachment protein σ1 (officially known as S1) contacts the adhesive surface of JAM-A (Kirchner et al., 2008) with an affinity that is ∼150× greater than the JAM-A homodimer interaction (Prota et al., 2003; Guglielmi et al., 2007). Thus, binding of reoviruses to JAM-A may disrupt the homotypic trans-interactions of the receptor as is the case for adenovirus and CAR.

Remarkably, viruses interact with both CAR and JAM through the analogous interface of these two Ig-superfamily proteins. This interface includes three loops of the respective membrane-distal V domains (Box 2). Interestingly, antibodies recognize viruses by using a similar interface (Dermody et al., 2009). Moreover, we discuss below how other viruses, including herpes simplex and measles viruses, also appropriate this interface.

Despite similarities in the way they contact their Ig-superfamily receptors, viruses that target TJs have different ways to spread in epithelial cells. Adenoviruses, coxsackievirus and reoviruses are released at the apical surface (Excoffon et al., 2008; Walters et al., 2002) (Fig. 4; strategy 1). Apical release facilitates access of coxsackievirus to DAF on the neighboring cells (Bergelson et al., 1995). However, for adeno- and reoviruses, access to CAR and JAM might be facilitated by lysis of neighboring infected cells.

The AJ is located right underneath the TJ, at the apical side of the epithelium (Fig. 1A). It comprises two types of single-pass integral receptor, cadherins and nectins, which are linked to the actin cytoskeleton by two types of peripheral protein, catenins and afadin (Fig. 2B). In humans, the cadherin protein family contains more than 100 members that can be separated into three main families: classic cadherins involved in AJ formation, desmosomal cadherins involved in desmosome formation and protocadherins that have a role in development (Tepass et al., 2000).

The extracellular domain of classic cadherins is composed of five cadherin-like domains, EC1, EC2, EC3, EC4 and EC5, whose conformation is dependent on the presence of Ca²⁺ (Troyanovsky, 1999). The adhesive properties of classic cadherins are supported by trans-interactions and lateral clustering. Trans-interactions of cadherins are strictly homophilic and mediated by the membrane-distal EC1 domain (Fig. 2B, center). Lateral clustering may be mediated by both, the extracellular contacts between the ectodomains in cis and by interaction of the cadherin cytosolic tails with catenins (Troyanovsky, 1999).

The other AJ constituents, nectins (Fig. 2B) and nectin-like (Necl) proteins, were originally identified as virus receptors (Mendelsohn et al., 1989). When searching the human genome or other databases, nectins 1–4 are listed as poliovirus receptor-related 1–4 (PVRL1 to PVRL4), for Necl1–4 they are cell adhesion molecule 1–4 (CADM1 to CADM4) and for Necl5 it is poliovirus receptor (PVR). Nevertheless, we have adopted here the nectin nomenclature (Takai et al., 2008b) because it is more meaningful in the context of the cell.

Nectins are responsible for Ca²⁺-independent homophilic and heterophilic trans-interactions at cell–cell junctions. The nectin family comprises four members (nectin-1, nectin-2, nectin-3 and nectin-4, whereas the Necl family has five members (Necl1, Necl2, Necl3, Necl4 and Necl5). Nectins have a specialized function in regulating cell–cell adhesions, whereas Necl proteins participate in several cellular functions (Takai et al., 2008b); and nectins interact with F-actin through afadin, whereas Necl proteins do not have known cytoskeletal interactions.

The nectin ectodomain is composed of three Ig-like domains, one membrane-distal V-type and two membrane-proximal C2-types. Trans-interactions occur through a canonical adhesive interface (Harrison et al., 2012) (Box 2). Notably, heterophilic interactions are much stronger than homophilic ones (Harrison et al., 2012). Nectins exhibit different tissue distributions. Nectin-1, nectin-2 and nectin-3 are ubiquitously expressed, with nectin-1 showing a preferential expression in the brain (Satoh-Horikawa et al., 2000). Nectin-4 is expressed in the placenta and among somatic tissues mainly in the respiratory epithelium (Reymond et al., 2001).

Similar to the TJ, the AJ is connected to the actin belt through adapter proteins. The actin belt forms a ring with contractile properties (Fig. 1A), and is associated with myosin II and α-actinin (Ivanov, 2008). Cadherins are linked to the actin belt through catenins; β-catenin directly binds to cadherins and interacts with the actin-binding α-catenin (Fig. 2B, top). In addition, the cytosolic tail of cadherin interacts with p120 catenin (also known as CTNND1), which participates in actin remodeling at the AJ (Hartsock and Nelson, 2008).

Afadin (also known as MLLT4) links nectins to F-actin (Takai et al., 2008a). Afadin is a large protein containing multiple protein-interaction domains (Fig. 1D). In addition to binding to nectins through its PDZ domain, and F-actin through a C-terminal binding domain, afadin also interacts with α-catenin, thus connecting nectins to cadherins (Fig. 2B) (Indra et al., 2013). Afadin also contacts the TJ protein ZO-1, therefore clustering TJ and AJ in the AJC.

Although the AJC is crucial to keep the epithelium tight, TJs and AJs are very dynamic structures that are constantly remodeled (Indra et al., 2013; Ivanov, 2008). During remodeling, the rearrangement of the actin cytoskeleton is mainly mediated by the Rho and Ras families of the small GTPases and the Arp2/3 complex (Hartsock and Nelson, 2008; Ivanov, 2008; Perez-Moreno et al., 2003; Takeichi, 2014; Zihni et al., 2014).

Perhaps surprisingly, no virus has yet been found to use cadherins as receptors. However, some oncogenic viruses degrade cadherins – notably Epstein Barr virus (Krishna et al., 2005), hepatitis B and C viruses (Arora et al., 2008; Liu et al., 2006) and human herpesvirus 8 (HHV8), which is associated with Kaposi's sarcoma (Qian et al., 2008). Such mechanism is thought to favor formation of metastases.

Several viruses use nectins and Necl proteins as receptors (Fig. 3B). Poliovirus, a small non-enveloped RNA virus of the Picornaviridae family, uses Necl5 (Mendelsohn et al., 1989). Large enveloped double-stranded DNA viruses of the Herpesviridae family, including human herpes simplex virus (HSV), take advantage of nectin-1 and nectin-2 (Geraghty et al., 1998; Lopez et al., 2000). Finally, enveloped negative-stranded RNA viruses of the morbillivirus genus in the Paramyxoviridae family, such as the measles virus (Box 1), enter cells through nectin-4 (Mühlebach et al., 2011; Noyce et al., 2011). Although these viruses are not evolutionary related, they all contact their respective nectin receptors with high affinity and in a similar fashion by binding the canonical adhesive interface that is involved in homo- and hetero-dimerization, and located at the tip of Ig-V domain (Box 2) (Di Giovine et al., 2011; Harrison et al., 2012; Mateo et al., 2014a; Mateo et al., 2013; Mateo et al., 2014b; Zhang et al., 2008; Zhang et al., 2013).

Necl5 is required for poliovirus entry and is broadly expressed. Because poliovirus mainly replicates in oropharyngeal and enteric epithelia, and rarely propagates in neurons, its replication must be blocked post-entry in most tissues (Racaniello, 2006). Surprisingly, poliovirus spread in epithelia has not been investigated in much detail; however, it is known that budding occurs from the apical side of intestinal epithelial cells, while the virus propagates in absence of cytotoxicity and while maintaining a functional TJ (Tucker et al., 1993).

In principle, transmission of viruses through zones of tight cell–cell contact has at least two advantages: first, it allows them to escape neutralization by antibodies, which are unable to penetrate the AJC. Second, viral proteins can be easily concentrated and assembled at the AJC by using the cellular infrastructure built for this purpose. Interestingly, two distinct families of enveloped virus localize their membrane fusion apparatus to the AJ by using a nectin as their receptor. We suggest here a new mechanism for their hidden transmission: HSV and measles virus may not form extracellular particles. Instead, they may induce the formation of ‘canals’ at the AJ. These canals are fusion pores elicited by the viral membrane fusion apparatus and stabilized by the cytoskeleton that surround the AJ.

HSV replicates in oral and genital epithelia, and establishes latency in neurons (Johnson and Huber, 2002). During replication in epithelia, components of the HSV fusion apparatus, namely the envelope glycoproteins E, I and D (gE, gI and gD) accumulate near cell–cell junctions (Johnson and Huber, 2002; Krummenacher et al., 2003). In particular, gE–gI heterodimers accumulate at the TJ in polarized cells (Polcicova et al., 2005), whereas gD (Spear et al., 2006) accumulates at the AJs where it trans-interacts with nectin-1 (Krummenacher et al., 2003). Thus, the components of the HSV membrane fusion apparatus are concentrated at the AJ, which may allow pore formation and virus spread without particle formation (Dingwell et al., 1994). However, whether encapsidated HSV genomes cross the AJC within enveloped particles (Fig. 4, strategy 2) or through the postulated canals (Fig. 4, strategy 3) remains unknown. Remarkably, the existence of fusion pores between infected neurons has been previously postulated to explain their synchronized firing during infections with an animal herpes virus that uses nectin-1 as receptor (McCarthy et al., 2009).

The other human virus that may use the AJ for hidden transmission is the measles virus, whose infectivity is always strongly cell-associated (Cathomen et al., 1998) and that is delivered to nectin-4-expressing epithelial cells through infected immune cells (Box 1). Whereas in vitro replication of the measles virus induces the fusion of many cells yielding one multinucleated giant cell (syncytium), in the trachea of infected monkeys no syncytia are detected (Frenzke et al., 2013; Ludlow et al., 2013). Moreover, in well-differentiated primary human airway epithelial cells, rapid lateral spread occurs without the formation of syncytia: cytotoxicity is minimal and trans-epithelial resistance remains constant (Leonard et al., 2008; Mühlebach et al., 2011; Sinn et al., 2002). Thus, fusion pores might form in these epithelia, allowing rapid intercellular spread of viral genomes. However, because syncytia do not form and trans-epithelial resistance does remain constant, the pores do not expand – probably because the cytoskeletal connections of the AJC constrain them. Importantly, interference with this pore-widening effect occurs in natural host cells and in well-differentiated sheets of airway epithelial cells but is amiss in non- or poorly polarized epithelial cells. In these systems the pores expand and syncytia are formed.

Thus, both HSV and the measles virus (1) target nectin receptors, (2) compete with nectins for the trans-interactions at the AJ and (3) can spread without disrupting epithelium integrity. Nectins are not crucial to maintain the AJ integrity once cadherin trans-interactions have been established (Indra et al., 2013), which may explain why viruses that seek to maintain epithelial integrity to facilitate their spread have evolved to target nectins instead of cadherins.

Remarkably, except HCV, all the viruses that have exploited junctional proteins as receptors are currently used as experimental cancer therapeutics. Current virotherapy clinical trials are based on Adenoviridae, Herpesviridae, Paramyxoviridae, Picornaviridae and Reoviridae and, in particular, on members of these families that use the junctional proteins discussed here (Miest and Cattaneo, 2014). This is surprising, because considerations about virus spread within epithelia did not play a role in the development of the oncolysis field.

Rather, the concept of cancer virotherapy originates from clinical reports that described cancer regression that coincided with naturally occuring viral infections (Kelly and Russell, 2007). Cellular mutations that increase cell proliferation rates and reduce innate immune defenses (Hanahan and Weinberg, 2011) offer a favorable environment for efficient viral replication in cancer cells. Another hallmark of most cancerous tissues is a deregulation in the expression of cell-adhesion molecules (Cavallaro and Christofori, 2004), and oncolytic viruses might take advantage of any overexpression of cell-adhesion molecules in order to efficiently initiate infection.

Clinical trials of oncolysis have been performed for decades but strategies that engineer viruses have only recently been developed in order to closely monitor virus replication and to address clinically relevant challenges, such as efficient systemic delivery, tight tumor specificity and improved efficacy in combination with current cancer therapies (Miest and Cattaneo, 2014). Certain virus strains are further targeted through genetic modifications for selective entry and enhanced replication in cancer cells; and clinical trials reported that all viruses that have been tested so far are safe.

Although the direct effect of CAR and JAM expression levels on the replication of coxsackievirus and reovirus in tumor cells has not been clearly defined, expression levels of CAR and nectins do, indeed, correlate with the oncolytic efficiency of adenovirus, HSV and the measles virus (Jiang et al., 2009; Sugiyama et al., 2013; Yu et al., 2007). Cell-associated viruses, such as HSV and the measles virus, can destroy tumors owing to efficient receptor-dependent lateral spread in cancerous tissues. Basically, to reach tumors and metastases, oncolytic viruses must be able to either traverse epithelia or to compromise their barrier function, and this may be the viral property that consistently favors effective viral oncolysis.

All viruses need to traverse or break epithelial barriers. Many of them have selected proteins of the AJC as their receptors, which is counterintuitive because these proteins are not readily accessible. Whereas some viruses quickly disrupt epithelium integrity, others carefully preserve it. Perhaps the most interesting recent insight is that – for rapid viral spread in epithelia – the cytoskeletal connections of the junctional proteins might be as important as those of their ectodomains.

Another important observation is that, in their natural hosts, viruses tend to not interfere with homeostasis, even when they spread rapidly. However, virologists naturally tend to prefer cellular systems that document striking cytopathic effects. As these effects might be irrelevant to the host, it is important to operate with well-differentiated primary cellular systems. Genetically modified viruses whose location and intensity of replication can be monitored will be of paramount importance to fully understand viral spread in the epithelia of living hosts (Miest and Cattaneo, 2014).

We thank Matthew Taylor (Montana State University, Bozeman, MT) and Urs Greber (University of Zurich, Switzerland) for helpful comments on the manuscript.