By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Dengue is the most common arthropod-borne viral infection in humans with ∼50 million cases annually worldwide. In recent decades, a steady increase in the number of severe dengue cases has been seen. Severe dengue disease is most often observed in individuals that have pre-existing immunity against heterotypic dengue subtypes and in infants with low levels of maternal dengue antibodies. The generally accepted hypothesis explaining the immunopathogenesis of severe dengue is called antibody-dependent enhancement of dengue infection. Here, circulating antibodies bind to the newly infecting virus but do not neutralize infection. Rather, these antibodies increase the infected cell mass and virus production. Additionally, antiviral responses are diminished allowing massive virus particle production early in infection. The large infected cell mass and the high viral load are prelude for severe disease development. In this review, we discuss what is known about the trafficking of dengue virus in its human host cells, and the signalling pathways activated after virus detection, both in the absence and presence of antibodies against the virus. This review summarizes work that aims to better understand the complex immunopathogenesis of severe dengue disease.
Dengue virus (DENV) infection is the most common arthropod-borne viral infection in humans with approximately 50 million cases annually worldwide. DENV is an enveloped (+)ssRNA virus of the family Flaviviridae, genus flavivirus, with four serotypes (DENV1–DENV4) sharing 70–80% amino acid sequence homology. The genus flavivirus also includes yellow fever virus and West Nile virus (WNV). Currently, there is no prophylaxis available against DENV . DENV infection can results in diseases ranging from the relatively mild dengue fever (DF) to the severe, life-threatening dengue haemorrhagic fever (DHF) or dengue shock syndrome (DSS).
The question why some patients develop DF and others DHF or DSS is continuously under investigation and debate. Epidemiologic research has identified pre-existing humoral immunity against DENV as a predisposing factor for severe disease [2-4]. Indeed, heterotypic antibodies, with sub-neutralizing properties, or waning concentrations of homotypic antibodies have been found to enhance DENV infectivity in vitro and in vivo [5-9]. Therefore, depending on the infection history of an individual, two distinct infection mechanisms can be distinguished: infection in the absence and infection in the presence of DENV antibodies. In this review, we will discuss the trafficking of DENV and the signal transduction pathways activated during infection in the absence of antibodies and during infection under conditions of antibody-dependent enhancement (ADE).
DENV Infection in Absence of Antibodies
Following the bite of a DENV-infected mosquito, the resident skin dendritic cells (DC), Langerhans cells, are among the first cells to be infected with DENV . Specifically, these cells spread the infection by transferring the virus from the skin to the lymph nodes , containing other DENV target cells such as resident DC [12-14], monocytes [6, 15] and macrophages [15-17]. The latter two cell types may become the preferred host cells for viral replication as DC-derived virions appear to be much less efficient in infecting new DC than insect-derived virions . Additionally, hepatocytes can serve as host cells [16, 19, 20].
Cell binding is mediated by the DENV envelope protein (E) and can occur through a wide range of attachment factors, whose affinity for DENV binding can be serotype-specific . Identified attachment factors for DENV include heat-shock protein 90 , heat-shock protein 70 , heparan sulphate [23, 24], CD14 , GRP8/BiP  and a 37/67 kDa high-affinity laminin receptor . Also, DENV has been found to bind to several C-type lectin receptors , including DC-SIGN [13, 21], L-SIGN , mannose receptor (CD206)  and C-type lectin domain family 5, member A (CLEC5A / MDL-1) , as well as αvβ3 integrins [29, 30]. Currently, many factors are known to serve as attachment factors for DENV infection, but none of them have been identified as an entry receptor for DENV infection of human cells . For example, when present, DC-SIGN is an important factor for DENV infection, however, mutating DC-SIGN internalization motifs abolished DC-SIGN internalization but not DENV cell entry .
DENV entry depends on clathrin-mediated endocytosis [32-35]. DENV has been shown to diffuse along the cell surface and to associate with pre-existing clathrin-coated pits prior to cell entry . Subsequently, the internalized virions are delivered to early endosomes (EEs) [32, 33]. Depending on the virus strain, serotype and host cell type, DENV fuses either from the EE (pH ∼6.0) or in a later stage of the endocytic pathway, after maturation of the EE into a late endosome (LE; pH 5.0–6.0) [32, 33].
Endosome maturation is pivotal for DENV fusion and escape, as depletion of v-ATPase, or addition of lysosomotropic drugs, which inhibit the acidification of endosomes, impair DENV infection [13, 21, 32]. In addition to acidification, microtubule integrity and the transport proteins kinesin and dynein  are important factors in endosome maturation. Disruption of microtubules in ECV306 cells did not appear to inhibit infection , whereas in BHK-21 cells DENV2 was found to associate with dynein . Using single-particle tracking, both microtubule-dependent and -independent intracellular transport behaviour of DENV was seen in BS-C-1 cells prior to membrane fusion. Addition of nocodazole to these cells during infection resulted in a ∼30% drop in viral infectivity suggesting that microtubule-dependent movement is important but not essential for the initiation of DENV infection .
Furthermore, Zaitseva and colleagues showed that DENV fusion depends not only on low pH, but also on the presence of anionic lipids in the target membrane . In mammalian cells, specific anionic lipids are localized within the late endosomal compartments [41, 42], which may explain why DENV fuses predominantly from within late endosomal compartments.
Membrane fusion is facilitated by the E glycoprotein and is triggered by the low pH and lipid environment of endosomes [40, 43, 44]. First the E homodimer rafts on the viral membrane dissociate into E monomers. The E monomers subsequently interact with the target membrane and this interaction facilitates the formation of E trimers. The energy released by these conformational changes is believed to drive the fusion process .
Post escape: replication, assembly and secretion
Once introduced into the cytoplasm, the positive sense RNA genome encodes for a single polyprotein, which is subsequently processed by autoproteases and cellular proteases to yield seven non-structural proteins (NS1, 2A, 2B, 3, 4A, 4B and 5), as well as three structural proteins: C (capsid), prM (precursor membrane protein) and E . The non-structural proteins assemble in a sequential manner to initiate RNA replication . The prM and E proteins form heterodimers that are oriented into the lumen of the endoplasmic reticulum (ER).
In the infected cell, large membrane rearrangements are seen, including the formation of vesicles by ER membrane invagination . ER-remodelling was found to be independent of the unfolded protein response , correlating with the extent of viral replication [49, 50], and to require transport from the rough ER to the Golgi . For DENV, expression and processing of NS4A is sufficient to induce ER-remodelling . Both ER-remodelling and production of viral particles require additional lipids. For this, flaviviruses induce both relocalization of cellular cholesterol [49, 51, 53, 54], as well as increased lipid production on the ER [53, 55, 56]. Furthermore, at a late stage during infection, flaviviruses induce autophagy  in order to liberate additional fatty acids for the continuation of replication . Indeed, replication is negatively affected by depletion of cholesterol  or inhibition of autophagy  and cholesterol synthesis .
An elegant study by Welsch and co-workers showed that the ER-derived vesicles contain the viral replication complex and have an open connection with the cytosol. The authors hypothesized that the pore serves as an exit for progeny viral RNA . Progeny RNA then associates with C proteins to form a nucleocapsid which buds into the ER and thus acquires a lipid membrane containing heterodimers of E and prM proteins .
Structural studies revealed that newly assembled immature particles have 60 hetero-oligomeric spikes, a single spike consisting of a trimer of prM/E heterodimers [60, 61]. Assembled particles are then transported to the Golgi in an ADE-ribosylation factors 4 and 5 (Arf-4 and -5)-dependent pathway , which is possibly also microtubule- and dynein-mediated . Transit through the Golgi and the trans-Golgi apparatus is required for maturation  and secretion of DENV particles . Within the Golgi, the prM protein is cleaved by furin into a soluble pr-peptide and the M protein . The pr-peptide, however, remains associated with the E protein during exocytosis  to prevent premature fusion of the virion within the acidic compartments of the Golgi network. Lastly, DENV particles are secreted into the extracellular space.
Once in the pH-neutral extracellular space, the pr-peptide dissociates and the virion becomes mature and fully infectious [64-66]. Furin cleavage is not very efficient; cells infected with DENV produce a high fraction of prM-containing particles (∼45%), which can be either fully immature or partially immature [67, 68]. Fully immature DENV particles require furin-mediated cleavage upon cell entry to acquire infectivity [65, 67, 69]. Dengue particles which are predominantly mature are likely infectious in a furin-independent manner. The minimal level of prM cleavage that is required for infection is as yet unknown. The viral life cycle, as discussed above, has been depicted in Figure 1.
Upon DENV internalization in endosomes, the virus triggers innate antiviral responses , particularly expression of interferons (IFNs). Both type I (α,β) and type II (γ) IFNs have been recognized as important cytokines in protection against DENV infection [71-73].
Induction of IFN expression occurs after recognition of pathogens by pattern recognition receptors, which detect pathogens through highly conserved molecular motifs. Recognition of DENV occurs through the endosomal receptors: toll-like receptor (TLR)-3 (dsRNA) , TLR8 (G-rich oligonucleotides)  and TLR7 (ssRNA) [70, 74], as well as the cytoplasmic RNA helicases RIG-I (retinoic-acid inducible gene 1) and MDA5 (melanoma differentiation-associated gene 5) . The latter two recognize cytoplasmic viral RNA, i.e. after escape from endosomes, whereas the TLR3 and TLR7 are endosome-associated and will detect DENV within endosomes. Furthermore, C-type lectins, like DC-SIGN and CLEC5A, are not only important for cell binding, but also induce expression of pro-inflammatory cytokines after DENV binding [28, 76]. Interestingly, DENV also alters expression patterns of PRRs. For example, DENV replication has been found to induce upregulation of TLR3, TLR4 and TLR7 , as well as the TLR-independent RIG-I and MDA5 in the monocytic THP-1 cell line . Activation of the TLR-dependent and -independent pathways has been shown to induce expression of pro-inflammatory cytokines, IL8, IL12, IFNα and IFNγ, in THP-1 cells  and primary monocyte-derived macrophages . Upregulation of IL8 expression, mediated by nuclear factor κB (NF-κB), has also been detected in primary monocytes . IFN expression subsequently activates STAT1 , and upregulates IRF1 (IFN regulatory factor 1) expression, resulting in a strong production of nitric oxide radicals (NOs) . The combined action of IFNs and NOs results in an antiviral state in bystander cells [83, 84] and limits replication in infected cells, respectively . We modelled these effects in Figure 2A.
Clearance of infection
Once an efficient antiviral state has been established, cells like macrophages and natural killer cells can clear infection . Additionally, DENV-specific B and T cells are generated approximately 6 days post-infection and these will help to completely control the infection. Dengue virions will be recognized by antibodies directed against the structural DENV proteins E and prM [87-90]. A recent article suggested that the majority of the antibodies generated in humans target the prM protein , although others have found a dominant response against E [87-89, 91, 92]. Antibody-mediated neutralization of viral infectivity may occurs at two levels: (i) at the level of cell binding, through inhibition of the interaction of the virus with cell-surface receptors [93, 94] and (ii) at the level of viral fusion, through binding of antibodies to e.g. the E protein fusion loop [95, 96], or through blocking of the conformational changes of the E glycoprotein that are required for membrane fusion [97-101]. Several factors determine the neutralizing potency of an antibody; these include the strength of binding and the accessibility of the epitope on the virus surface [99, 102-104], the latter being strongly dependent on the maturation state of the virus .
Infection Under ADE-Conditions
Upon a secondary, heterologous, DENV infection, pre-existing plasma cells are triggered to rapidly produce antibodies but these are predominantly directed against the initial DENV serotype, a process often referred to as ‘original antigenic sin’ [106-108]. While most of the produced antibodies will bind to the heterologous virus serotype, they are more likely to have non-neutralizing properties against the new serotype than towards the original serotype [87, 89, 90]. Surprisingly, these non-neutralizing antibodies have been found to enhance infection, a phenomenon called ‘ADE’ [5, 6, 9, 107]. Enhancement is facilitated through efficient interaction of the virus-antibody complex with Fc receptors, which will be discussed in more detail below. Given the lack of knowledge on the steps after escape of the genome from the endosome, the current hypothesis is that all cellular changes associated with ADE are induced prior to escape out of the endosome. Thus, we will discuss only these steps.
All E antibodies tested to date have been shown to enhance standard DENV preparations when the concentration of the antibody is below the threshold required for neutralization. Interestingly though, the enhancing properties of some E antibodies appear to be dependent on the maturation status of the virus . The structural organization of the E protein in mature versus immature particles is very distinct and therefore affects the accessibility of specific epitopes and thus the threshold for antibody neutralization . Indeed, E antibodies were identified that preferentially enhance infectivity of immature virions . Other E antibodies were found to interact with immature particles but could not rescue the infectious properties of these particles whereas enhancement was seen in standard preparations . Antibodies directed against prM target specifically (partially) immature DENV via the pr peptide. Furthermore, prM antibodies were found to be highly cross-reactive to all DENV serotypes, and generally have poor neutralizing capacity [63, 90]. In fact, recent studies showed that prM antibodies enhance the infectivity of essentially non-infectious immature DENV particles [69, 90]. A few studies indicate that during secondary infection, in particular, the prM antibody repertoire is amplified [90, 92].
The current hypothesis is that DENV particles employ ADE-specific pathways to enter and infect cells, leading to a higher number of infected cells (extrinsic ADE), as well as altered immune responses, and subsequently increased virus production per infected cell (intrinsic ADE). In this article, we will review these three phenomena separately, starting with what is known on the entry pathway of opsonized DENV particles.
As indicated above, during DENV ADE, cell binding occurs by the Fcγ receptor (FcγR) [111, 112] expressed on monocytes, macrophages and DC . Thus, DENV targets the same host cells in the presence or in the absence of DENV antibodies. The FcγR family consists of three classes (I–III), with decreasing affinity for antibodies going from I to III . Yet, FcγRIIA appears to be more permissive to ADE than FcγRI [77, 111] suggesting that binding affinity is not the only determinant of infectivity. An elegant study by Boonnak and colleagues showed that immature DC (iDC) express FcγRIIa to levels similar to those found in mature DC (mDC). However, iDC do not support ADE [12, 14, 113]. The authors observed an inverse relationship between DC-SIGN expression levels and DENV ADE . DC-SIGN expression levels decrease going from iDC to mDC to monocytes. Reduction of DC-SIGN expression reduces cell permissiveness to DENV infection in the absence of antibodies, and thus augments the relevance of virus entry through FcγR during ADE of DENV infection in mDC.
Antibody-opsonization has been found to facilitate cell binding of immature DENV particles by interaction with FcRII [69, 90]. Interestingly, under optimal conditions of ADE, a near-wildtype binding efficiency was seen for antibody-opsonized immature particles and in combination with intracellular furin cleavage, as discussed in detail below, immature DENV particles turn highly infectious [69, 90].
FcγR facilitate not only virus-cell binding, but also cell entry, during DENV ADE, as disruption of the FcγR cytoplasmic tail or activation motifs within the tail abolishes ADE mediated by FcγRI  or FcγRIIa . Similar effects were observed when antibodies against these receptors were studied [69, 77].
As mentioned before, FcγRIIA appears to be the most permissive FcγR for DENV ADE. This property may be dependent on the ratio of expression levels of FcγRI and FcγRIIA, but may also depend on the internalization pathways followed by FcγRI- or FcγRIIA-ligands . For example, different pathways may deliver the opsonized particles into more beneficial environments for the virus to initiate infection. FcγR-mediated phagocytosis has been found to be negatively regulated by FcγRIIB . A high antibody density on the viral particle ligates FcγRIIB , and can inhibit phagocytosis, and infection, irrespective of whether the antibodies themselves possess neutralizing properties. This also suggests that FcγR-mediated phagocytosis is an important entry mechanism involved in DENV ADE, as has been long hypothesized .
In the phagocytosis entry model, the presence of DENV-antibodies will enhance cell binding and entry by phagocytosis. In line with the role of phagocytosis in DENV-cell entry, there seems to be a relationship between the phagocytic activity of cells and DENV-infectivity [111, 112]. Yet, two different entry mechanisms have been identified for WNV in the presence of antibodies: first, single opsonized particles seem to enter into coated pits similar to the primary infection [117, 118]; second, antibody-mediated aggregates of multiple particles appear to be phagocytized . Based on this closely related virus, both entry mechanisms may be important for DENV entry under ADE-conditions. Furthermore, it is not known whether FcγR-mediated entry delivers opsonized DENV into the same entry route used during infection in the absence of antibodies or side-tracks opsonized DENV into a different pathway.
Immature particles are likely to be delivered to endosomes after binding and FcγR-mediated entry. It is proposed that the acidic conditions in the endocytic pathway induce conformational changes in the prM/E heterodimers similar to the maturation of virions in the secretory pathway [60, 61, 65, 119]. These conformational changes facilitate cleavage of prM by host cell furin . It is unclear how the pr-peptide is released, but one could speculate that the complex is first recycled back to the pH-neutral cell membrane or that pr dissociates in endosomes because of the more acidic environment compared to trans-Golgi network . Furin-mediated cleavage and prM/E dissociation are required to enable the E protein to initiate fusion of the viral membrane with the endosomal membrane [69, 119-121]. The route and trafficking dynamics of antibody-opsonized immature DENV are as yet unknown.
Modulation of the immune response
Entry of DENV via antibodies appears to remodel and suppress the innate immune response thereby favouring virus particle production in infected cells. This phenomenon has been called intrinsic ADE and can occur along both TLR-dependent and -independent pathways. However, the exact mechanisms remain to be elucidated. Below, we will address this topic, and in Figure 2B we have modelled the current knowledge in this area.
Adaptation of the innate immune system
Using the THP-1 cell line, Modhiran and colleagues have shown that DENV ADE targets the TLR signalling pathway by upregulating the negative TLR-expression regulators SARM (Sterile-alpha and Armadillo Motif containing protein) and TANK (TRAF family member-associated NF-κB activator), and subsequent downregulation of TLR3, TLR4, TLR7 expression, and of TLR-signalling molecules , using an unknown pathway . TLRs signal through two distinct pathways, either by activating the NF-κB and subsequent IRF1 through canonical IKKs (Inhibitor of κ B Kinase) , or by activating IRF3 through IKK-related kinases . A recent article suggested that TANK negatively regulates activity of canonical IKKs by facilitating complex formation between IKK-related kinases and canonical IKKs , signifying that DENV ADE shifts TLR signalling via NF-κB to signalling along a non-NF-κB pathway. Also TANK functions to limit TLR signalling along the MyD88 adaptor protein . Taken together, DENV ADE appears to reduce TLR expression and signalling to facilitate ongoing virus replication in the host cell.
Furthermore, Ubol et al. found that ADE enhances expression of DAK and Atg5–Atg12, which subsequently reduces expression of RIG-I and MDA5 in THP-1 cells . However, in monocyte-derived-macrophages, RIG-I is unaffected , and MDA5 might be modestly decreased under ADE conditions in both cell types [78, 128]. Thus, the modification of the TLR-independent genes appears to be less pronounced than the expression of those genes that are TLR-dependent.
DENV ADE has been shown to be accompanied by reduced IFNβ expression [127, 128], indicating that ADE indeed suppresses induction of an antiviral state. Rather, induction of type I IFNs appears to be dependent on high, neutralizing, concentrations of antibodies , which is in line with reduced phagocytosis after FcγRIIB ligation . In contrast, a recent report of Kou and colleagues measured type I IFN by DENV-infected PBMCs, using an elegant VSV infection-inhibition assay. At first glance, ADE appeared to result in higher and longer secretion of type I IFNs, visualized by complete inhibition of VSV infection . However, as VSV infection is very sensitive to type I IFNs , the enhanced infection under ADE conditions relative to DENV infection in absence of antibodies may actually have resulted in a higher number of DENV-infected, IFNα/β-producing cells at any time point , with an associated higher inhibition of VSV. Thus type I IFN production during DENV ADE still may be lower on a per-cell basis.
Type I IFNs induce an antiviral state by activating the JAK/STAT pathway, which subsequently activates STAT-1/2, IRF-1/-3 and production of NOs . NOs are potent inhibitors of DENV replication  and viral protease activity . In line with the work by Ubol et al. , Chareonsirisuthigul et al.  found that ADE results in reduced levels of NOs in THP-1 cells, due to blocked activation of STAT-1 and expression of IRF-1. As described above, the affected IRF-1 levels may also be due to upregulation of TANK expression and an associated shift in TLR-signalling along canonical IKKs to IKK-related kinases.
Inhibition of the JAK/STAT pathway was found to be mediated through upregulation of SOCS3 (suppressor of cytokine signalling 3)  in both monocyte-derived-macrophages  and THP-1 cells . SOCS3 upregulation in THP-1 was mediated through increased levels of IL10 [79, 127], an immunosuppressive cytokine, the expression of which would be induced by FcγR ligation, as previously observed for opsonized amastigotes  and Ross River virus . By contrast, in primary monocyte-derived-macrophages , IL6 upregulation is responsible for SOCS3 upregulation, and this is in line with the finding of Boonnak and colleagues that IL6 expression is maximal at peak ADE in primary monocytes and monocyte-derived-macrophages . Enhanced IL6 levels were also found in THP-1 cells under ADE conditions, although less pronounced than for IL10 and its role was not further investigated . Therefore, it remains to be investigated whether SOCS3 upregulation is induced by IL10 and IL6 independently, or mainly by IL6.
As described above, DENV infection in the presence of antibodies may skew the innate antiviral response towards a virus-tolerant state within the cell, so-called intrinsic ADE . Suppression of the innate immune system during DENV ADE likely facilitates longer survival of infected cells and thus increased total virus output [78, 79, 113, 129].
Kou et al. noted that, under ADE conditions the number of infected human PBMCs is doubled, but that the viral output increased 10-fold . Similar observations have been made by Boonnak et al. in monocytes (4-fold and 1500-fold, respectively), mDC (2.5-fold and 3000-fold, respectively), as well as monocyte-derived-macrophages (7-fold and 2000-fold, respectively) . These results suggest that ADE activates an intrinsic pathway resulting in enhanced virus production per infected cell. It should be noted that the increased number of infected cells in DENV ADE infected samples had not been compensated. When compensation was applied to yield similar numbers of infected cells, no significantly increased viral output was observed between normal infection and infection under ADE conditions using wildtype DENV in mDC  and immature virus in the K562 cell line  suggesting that enhanced cellular virus production may be more related to extrinsic ADE rather than alterations in intracellular pathways (intrinsic ADE).
Conclusions and Challenges
The development of severe dengue disease is a multifactorial process in which the presence of pre-existing heterologous antibodies plays an important role. Severe disease is often preceded by high circulating virus titres and already in the early 1970s it was hypothesized that non-neutralizing heterologous antibodies stimulate the infectious properties of the virus thereby causing an increased viral load and enhanced disease. Recent studies in this area, as highlighted in this review, revealed that ADE of infection is driven by two main elements: (i) an increased number of infected cells, due to increased antibody-mediated cell binding and entry of both mature and (partially) immature DENV particles, also known as extrinsic ADE; (ii) enhanced virus production per infected cell due to suppression of the innate antiviral response. The latter process is called intrinsic ADE and appears to mainly stem from suppressed TLR-signalling, although more research is required to fully understand this phenomenon.
A future challenge for ADE research is to characterize the cell entry pathway used by antibody-opsonized DENV. It will be of importance to determine whether antibody-opsonized DENV is internalized through a different cell entry pathway, compared to non-opsonized virus, thereby providing a more favourable environment for cytosolic escape of the DENV genome, or assisting the virus to avoid detection by PRRs at the site of genome escape .
The current literature suggests that the reduced antiviral response observed under ADE conditions is triggered during the early stages (binding, entry) of the viral life cycle. Yet, many effects, e.g. cytokine expression levels, are measured at later time points, up to 48 h post-infection, near the end of the second round of infection. One could speculate that the effects at later time points are associated with transcription and translation of the DENV genome [137-139] rather than specific steps during viral entry. Indeed, UV-inactivated DENV fails to affect cytokine production, even under ADE conditions [14, 77, 113]. To further understand the phenomenon of intrinsic ADE it would be interesting to investigate the differences in gene expression patterns within the first hours post-infection rather than 18–24 h or even days post-infection.
Continued research on this topic is required to fill important gaps in our knowledge in the immunopathogenesis of severe dengue disease and may guide the development of safe and efficacious antiviral drugs and vaccines.
We thank the reviewers, I. A. Rodenhuis-Zybert and M. J. Schuijs for reading the manuscript. J. F. and J. M. S. acknowledge support from the Dutch Organization for Scientific Research (NWO-VIDI).