Metalloproteinase production from macrophages – a perfect storm leading to atherosclerotic plaque rupture and myocardial infarction

Introduction

Production of matrix metalloproteinases (MMPs) from macrophages contributes to destruction of the extracellular matrix (ECM) in a broad range of chronic inflammatory diseases. Atherosclerosis is a special case because lipoprotein particles trapped in the artery wall recruit monocytes that convert to foam-cell macrophages by engorging oxidized and other modified forms of these lipoproteins (Williams & Tabas, 1995). In advanced atherosclerosis, plaques consisting of amorphous lipid deposits with overlying, expanded connective tissue can obstruct the coronary and other conduit arteries, leading to stable ischaemic syndromes, including angina pectoris. Moreover, depletion of collagen and other ECM molecules from the core and fibrous cap overlying plaques can lead to loss of mechanical competence, culminating in rupture of the cap, thrombus formation on the exposed thrombogenic core and partial or complete occlusion of the lumen (Libby, 2013). Plaque rupture underlies the majority of myocardial infarctions (MIs) and strokes (Virmani et al. 2006), which together constitute the principal cause of death in many advanced societies. Inhibiting MMP activity (Newby, 2012, 2015) or the mechanisms responsible for production of MMPs from macrophages (reviewed here) therefore represent viable targets for therapies to prevent MIs and strokes. The earlier literature relating to this topic was previously discussed exhaustively (Newby, 2008); hence, this article seeks to provide an update by emphasizing findings during the last 7 years. These new insights suggest that multiple inflammatory mediators need to act in concert to raise a ‘perfect storm’ that provokes net destruction of the ECM leading to MIs and strokes.

Involvement of matrix metalloproteinases in atherosclerosis

There are at least 23 MMP enzymes, most of which are secreted, except the six membrane-type MMPs that are inserted into or attached to the external membrane surface. The catalytic sites of MMPs may be blocked by all or at least some of the four tissue inhibitors of MMPs (TIMPs; reviewed in detail by Nagase et al. 2006). A structurally similar active catalytic domain occurs also in some members of the disintegrin metalloproteinases (ADAMs) and in the ADAMs with thrombospondin domains (ADAM-TSs). Matrix metalloproteinases have the ability to degrade a variety of ECM components but also many other cell surface, secreted or ECM-sequestered substrates, many of which regulate inflammation (Khokha et al. 2013).

As summarized previously (Newby, 2012, 2015), the evidence that MMPs play pathological roles in atherosclerosis comes partly from rabbit and especially from mouse models. However, the expression pattern of MMP mRNAs in human blood and mouse bone marrow macrophages isolated and classically activated in very similar conditions is quite divergent, with far more MMP-1, -7 and -9 and TIMP-3 in human macrophages but much less MMP-12 and -13 compared with mice (Newby, 2015). This conclusion is reinforced by other studies of mouse bone marrow macrophages and Raw264.7 cells (Hald et al. 2012; Murray et al. 2013). Moreover, a comparison of unstimulated mouse and human blood monocytes and macrophages shows the same similarities and differences (Fig. 1). Differentiation of blood monocytes to macrophages greatly increases expression of MMPs in both mice and humans (Fig. 1), but levels of MMP-1, -7 and -9 and TIMP-1 and -3 are much higher in man (Huang et al. 2012), whereas MMPs-12, -13 and -23 are much higher in mice (Tsaousi et al. 2016). The high levels of MMP-12 and -13 expression in mouse macrophages correspond to dramatic effects on atherosclerosis (Johnson et al. 2011; Quillard et al. 2011). However, MMP-12 (Scholtes et al. 2012) and MMP-13 (Molloy et al. 2004) have restricted expression in human atherosclerotic plaques, which invites caution over the clinical translation of the mouse studies. In the case of MMP-12, there are genome-wide association studies (GWASs) supporting a causative role in strokes (Traylor et al. 2014), but this is not the case for MMP-13. Conversely, MMP-7 is hardly expressed in mouse macrophages and has a modest impact on atherosclerosis (Johnson et al. 2005) but could be more important in man (Fig. 1). Most recently, MMP-28 was shown to affect macrophage functions in mice (Ma et al. 2013), but MMP-28 is not expressed actively in human monocytes or macrophages (Bar-Or et al. 2003). Furthermore, the profound morphological differences and the need for high-level transfer of fully active forms of MMPs to provoke plaque rupture in mice (Gough et al. 2006; Liang et al. 2006) discourage extrapolation to the human disease.

Genome-wide association studies provide convincing evidence of a pathogenic role for MMP-12 (Traylor et al. 2014) and the distantly related ADAMTS-7 (Reilly et al. 2011). For other MMPs and TIMPs, only correlative evidence is available so far. For example, many MMPs and TIMPs are overexpressed in human atherosclerotic plaques compared with normal tissues (reviewed in detail by Newby, 2005). More persuasively, MMP-8, -9, -12 and -14 have been shown in biobank studies to associate with plaque morphologies suggesting vulnerability to rupture, whereas MMP-2 and TIMP-3 show negative association (Sluijter et al. 2006; Peeters et al. 2011; Scholtes et al. 2012; Johnson et al. 2014). Furthermore, at least MMP-8 and MMP-12 levels in plaques are risk factors for subsequent adverse cardiovascular events (Peeters et al. 2011; Scholtes et al. 2012). In future, it may be possible to combine biochemical and genetic analyses, for example in Mendelian randomization studies or by the identification of rare null mutations. In the meantime, a causative role for MMPs in human plaque rupture is highly plausible but still a hypothesis.

Monocyte and macrophage diversity in atherosclerosis

Production of monocytes and macrophages from myeloid precursors relies on the trophic effects of macrophage colony-stimulating factor (CSF-1). Deletion of CSF-1 or blocking its receptor in mice prone to atherosclerosis greatly reduces plaque formation (Di Gregoli & Johnson, 2012). Likewise, depletion of monocytes and macrophages in the early stages of mouse atherosclerosis abolishes foam cell formation and reveals the accumulation of lipoprotein deposits in susceptible sites (Paulson et al. 2010). These experiments establish that macrophages derived from circulating monocytes are required to clear lipoprotein deposits retained in the ECM and that this leads to foam cell formation. Some contribution from macrophage proliferation (Robbins et al. 2013) or from expansion of resident stem cell populations has also been debated (Nguyen et al. 2012); and additional foam cells can be generated by transdifferentiation of resident vascular smooth muscle cells (Shankman et al. 2015). As reviewed previously (Newby, 2005), both macrophages and vascular smooth muscle cells elaborate MMPs and TIMPs. Moreover, MMPs and other proteases promote vascular smooth muscle cell migration and proliferation so as to establish the fibrous cap of plaques. In contrast, the high levels of many MMPs produced by macrophages may provoke destruction of the ECM, causing plaque rupture.

At least two phenotypes of mouse monocytes (Ly6Chi, CCR2hi and Ly6CloCXCR3hi) and three phenotypes of human monocytes (CD14hiCD16lo, CD14dimCD16lo and CD14dimCD16hi) have been characterized (Ziegler-Heitbrock et al. 2010). Despite performing different functions in relation to acute inflammation and patrolling behaviour, both monocyte phenotypes appear to contribute to atherosclerosis in mouse models (Combadière et al. 2008). Moreover, they do not seem to give rise to different macrophage populations in plaques (Tacke et al. 2007).

Differentiated macrophages adopt a host of different phenotypes. These were initially divided into pro-inflammatory (so-called classically activated or M1 type) or anti-inflammatory (so-called alternatively activated or M2 type). However, the M1–M2 dichotomy has more recently been replaced with more nuanced descriptions of phenotypes (Murray et al. 2014) based on the activating mediators and their related signalling pathways, some of which are illustrated in Fig. 2.

Regulation of MMP and TIMP production from monocytes and macrophages

Binding of transcription factors of the activator protein-1 (AP-1) family to regulatory elements in the proximal promoters of many MMPs appears to be of central importance for their transcriptional regulation and certainly contributes to their increased production during inflammation (Clark et al. 2008). However, not all MMP promoters contain proximal AP-1 sites or even a TATA box, which is necessary for induced transcription of most genes (Clark et al. 2008). Moreover, a plethora of other proximal transcription factor binding sites, including sites for specificity protein-1 (SP-1), nuclear factor-κB (NF-κB) and signal transducer and activator of transcription-1 (STAT-1), mediate inflammatory activation of several MMPs (Clark et al. 2008). Synergy between activation of AP-1 and NF-κB is responsible for induction of several MMPs in a variety of cell types (reviewed by Newby, 2005), including macrophages. This may depend on a signalosome that brings together widely separated transcriptional activators, including distal enhancer or suppressor elements (Glass & Natoli, 2015).

In order to influence MMP and TIMP expression therapeutically in plaques and other inflammatory foci, it would be valuable to identify the key mediators and the underlying mechanisms. This review will attempt to synthesize the available information, in part by providing searchable databases for monocytes (Supplementary information Table S1) and macrophages (Supplementary information Table S2).

Prostaglandin E₂ and the cAMP pathway

Prostaglandin E₂ (PGE₂) mediates upregulation of at least MMP-1, -7, -9, -10 and -14, as well as TIMP-1 in undifferentiated human monocytes, as previously reviewed (Newby, 2008). Prostaglandin E₂-dependent MMP upregulation has also been observed in human alveolar macrophages, mouse peritoneal macrophages and RAW264 cells (see Tables S1 and S2). As shown in Fig. 2, the action of inflammatory mediators or integrin-mediated binding to various ECM components activates phospholipase C, which releases arachidonic acid. This is transformed by the sequential activity of cyclo-oxygenase (COX) and PGE₂ synthase (PGES-1) to PGE₂. Cyclo-oxygenase-1 is constitutively expressed in human monocytes, and COX-2 is rapidly upregulated by adherence or lipopolysaccharide (Reel et al. 2011) or by tumour necrosis factor-α (TNFα) together with granulocyte–macrophage colony-stimulating factor (GM-CSF, CSF-2) (Zhang & Wahl, 2015). PGE₂ acts specifically on EP4 receptors to stimulate cAMP formation, which then activates transcription through direct binding of cAMP response element binding protein (CREB) to the MMP-1 promoter or by enhancing the binding NF-κB to the MMP-9 promoter (Lai et al. 2003). Work using other cell types also identifies cross-talk with the mitogen-activated protein kinases (MAPKs; Gerits et al. 2008) that could promote AP-1 binding (Fig. 2). Other activators of PGE₂-dependent MMP production include extracellular MMP-1 or MMP-3, which can cleave active TNFα from the surface of mouse peritoneal macrophages, leading to MMP-9 secretion (Steenport et al. 2009). Furthermore, TNFα generated in this way upregulates early growth response protein 1 (EGR-1), which induces PGES-1 expression (Khan et al. 2012). Exposure to Mycobacterium tuberculosis infection can also upregulate MMP-1, but not MMP-7, in a PGE₂-dependent manner (Rand et al. 2009).

Differentiation of monocytes to macrophages

Matrix metalloproteinase-2, -7, -9, -11, -12 and -14 and TIMP-2 and -3 are selectively upregulated in human macrophage-CSF-1-differentiated macrophages, independently of COX, MAPKs or NF-κB (Reel et al. 2011), and MMP-8, -13, -19, -23 and -25 are also increased in mouse macrophages (Tsaousi et al. 2016). Increased expression of MMP-14 has been ascribed to upregulation of the SAF-1 transcription factor (reviewed by Newby, 2008), but the other mechanisms remain to be clarified.

Foam cell formation

The properties of foam cells formed from differentiated macrophages in vitro depends on the source of lipid [e.g. platelets, acetylated low-density lipoprotein (LDL), minimally or extensively oxidized LDL]; hence, widely different results have been obtained, suggesting stimulation, inhibition or little effect on levels of MMPs (Tables S1 and S2). Stimulation could result from (weak) action on Toll-like receptors (TLRs; Lundbery & Yan, 2011), whereas inhibition of MMP-9 expression in U937 cells (Table S1) resulted from formation of peroxisome proliferator-activator receptor (PPAR) γ ligands. We found no effect of oxidized LDL on mRNA levels of MMPs and TIMPs, but foam cells expressed more MMP-14 and less TIMP-3 protein, which implicated epigenetic mechanisms in part mediated by microRNA24 (Johnson et al. 2014). In vivo results are also divergent. Rabbit granuloma foam cells showed increased MMP-1, -3, -12 and -14 and decreased TIMP-3 expression (Table S2), although mouse granuloma foam cells showed no changes (Thomas et al. 2015), and peritoneal foam cells had decreased MMP-13 expression, owing to activation of the LXR nuclear receptor by the cholesterol pathway intermediate desmosterol (Spann et al. 2012). So far, none of these in vivo studies has investigated protein levels, which might vary despite similar mRNA expression if epigenetic mechanisms intervene (Johnson et al. 2014).

Classical macrophage activators and the AP-1/NF-κB pathway

The pro-inflammatory mediators TNFα, interleukin (IL)-1β, CD40L and pathogen-associated molecular patterns that act at several Toll-like receptors have been observed to stimulate MMP expression in both monocytes and macrophages (Tables S1 and S2). Moreover, activation of TLR-2 was implicated directly in MMP-1 and MMP-3 production from isolated human plaque-derived cells examined ex vivo (Monaco et al. 2009). These inflammatory mediators share the ability to activate the MAPKs, extracellular signal-related kinases 1/2 (ERK1/2), p38 MAPK and c-jun N-terminal kinase (JNK), as well as phophoinositide-3 kinase (PI3 kinase) and the inhibitor of κB kinase2 (IKK2) that leads to activation of NF-κB (Fig. 2). Not surprisingly, therefore, inhibitors of one or more these kinases generally reverse the effects of this broad class of inflammatory mediators (Tables S1 and S2). However, the precise identity of the activating kinases seems to depend on the MMP and the source of cells (Tables S1 and S2). One complicating factor is the participation to a greater or lesser degree of PGE₂ derived from COX-2 (Fig. 2), induction of which requires all these kinases (Huang et al. 2012). For example, inhibition of p38 reduced MMP-1 expression in a PGE₂-dependent manner in human monocytes, whereas inhibition of ERK1/2 decreased both MMP-1 and MMP-9 expression independently of PGE₂ (Zhang & Wahl, 2015). In the absence of PGE₂, upregulation of MMP-1 and -10 in human blood-derived monocytes depended on ERK1/2, JNK and IKK2 but not p38 MAPK (Reel et al. 2011). The same was true for lipopolysaccharide induction of MMP-1, -3, -10, -12 and -14 in human macrophages, but induction of MMP-25 required p38 (Huang et al. 2012). Furthermore, specificity for the various activating kinases may depend on the specific inflammatory signal. For example, induction of MMP-1 in human alveolar macrophages by M. tuberculosis depends selectively on p38 MAPK (Rand et al. 2009). Both basal and induced expression of many MMPs, especially MMP-1, -3, -10 and -13, and TIMP-3 is reduced by inhibitors of PI3 kinase in human macrophages (Huang et al. 2012). However, the basis for these effects is still not clarified. Other inflammatory mediators, such as clusterin (Shim et al. 2011) or complement component C5a (Speidl et al. 2011), or homophylic interactions of CD147 (EMMPRIN) or interactions with its ligand cyclophylin A (Yang et al. 2008) also employ MAPKs, PI3 kinase, IKK2 and the resultant activation of AP-1/NF-κB signalling to upregulate MMPs (Fig. 2). To confirm that these signalling pathways contribute to increased MMP expression in human atherosclerotic plaques, we demonstrated co-localization of activated NF-κB with MMP-1 and MMP-10 (Huang et al. 2012).

Interferons and the JAK/STAT pathway

Earlier studies demonstrated profound inhibitory effects of interferon-γ (IFNγ) on MMP-1, -3, -9, and -12 and TIMP-1 production from human monocytes and macrophages (Tables S1 and S2). In contrast, IFNγ acting through the JAK/STAT pathway can upregulate MMP-12, -14 and -25 and suppress TIMP-3 mRNA expression in human macrophages (Huang et al. 2012). The effect on MMP-25 may be especially interesting in view of its ability to modulate the activity of several chemokines (Marco et al. 2013). Given that many human plaques contain IFNγ, induction of MMP-14 and suppression of TIMP-3 could promote the invasive and destructive MMP14⁺TIMP-3⁻ macrophage phenotype that we detected in rabbit and human foam cells (Johnson et al. 2008). The effects of IFNγ appear to be very different in human and mouse macrophages (Hayes et al. 2014), which complicates the interpretation of the mouse models. Nevertheless, inhibition of mouse macrophage MMP-9 production by IFNγ correlated with slower ECM degradation and thrombus resolution in wild-type compared with IFNγ knockout mice in a model of deep vein thrombosis (Nosaka et al. 2011). Deletion of transforming growth factor-β (TGFβ) receptors in T lymphocytes, which promotes polarization to the T-helper-1 phenotype that releases IFNγ, also decreased MMP-9 expression in atherosclerotic mouse aortas (Ovchinnikova et al. 2009). In contrast, MMP-13 was increased, suggesting that IFNγ from T-helper-1 cells can promote as well as inhibit the expression of different MMPs. In other experiments, deletion of all T and B cells (Hayes et al. 2014) or only T-helper-1 cells (Tsaousi et al. 2016) did not affect MMP or TIMP expression in mouse foam cells from subcutaneous granulomas or in atherosclerotic plaques. Consequently, the evidence for stimulatory effects of IFNγ on MMP expression is stronger in humans than in mice.

Interleukin-6 and GM-CSF

As illustrated in Fig. 2, IL-6 activates JAK1 and STAT-3, MAPKs and PI3 kinases (Schaper & Rose-John, 2015), which may account for its upregulation of MMPs (Tables S1 and S2). Interestingly, induction of MMP-9 in mouse macrophages by IL-6 is independent of COX-2 (Kothari et al. 2014).

GM-CSF signals through the CSFR2 complex to activate JAK2 and STAT-5 as well as MAPKs and PI3 kinases (Broughton et al. 2012); hence, the transcriptional programme initiated by GM-CSF is unique, although it replicates some aspects of both the classical and alternative paradigms. GM-CSF is especially associated with upregulation of MMP-12 (Table S1), which occurs through activation of the proximal AP-1 site. Why this direct action of GM-CSF is selective for MMP-12 over other MMPs with proximal AP-1 sites is unclear. GM-CSF can also induce TNFα secretion, leading to the upregulation of other MMPs (Zhang et al. 1998). Given this and the fact that GM-CSF can be upregulated by oxidized LDL and several inflammatory mediators (Di Gregoli & Johnson, 2012), GM-CSF-stimulated and classically activated macrophage phenotypes probably overlap in vivo. Despite this, in mice exposed to cigarette smoke, neutralization of GM-CSF selectively decreases MMP-12 but not MMP-9 activity in lung macrophages (Vlahos et al. 2010). GM-CSF increases MMP-14 protein expression and activity independently of changes in mRNA expression but because micro-RNA24 is decreased, which relieves an inhibitory effect on protein translation (Di Gregoli et al. 2014). These observations are particularly interesting because there is evidence for distinct populations of macrophage-CSF and GM-CSF macrophages in human plaques that may make different contributions to plaque stability (Di Gregoli & Johnson, 2012). Indeed, the action of GM-CSF might also account for the harmful MMP14⁺TIMP-3⁻ macrophage phenotype (Johnson et al. 2008).

Hypoxia

Most macrophages in atherosclerotic plaques are in a chronic state of hypoxia (Sluimer et al. 2008). Hypoxia increases expression of MMP-7 (Table S2). Transcriptomic data from hypoxic macrophages indicates that MMP-1, -3, -10 and -12 are also significantly upregulated, perhaps secondarily to increased production of IL-1α,β (Fang et al. 2009). Pathways through hypoxia-inducible factor-1α (Lee et al. 2012), hypoxia-inducible factor-2α (Yang et al. 2010) and JAK2/STAT-3 have been implicated (Gao et al. 2015).

Anti-inflammatory pathways

Priming with IL-4 inhibits expression of MMP-1, MMP-9 and TIMP-1 in monocytes and macrophages (Tables S1 and S2), perhaps owing to overexpression of suppressor of cytokine signalling (SOCS) proteins. However, consistent with previous work in mouse macrophages (Table S2), we found that IL-4 selectively increases MMP-12 in human monocyte-derived macrophages (Huang et al. 2012). Matrix metalloproteinase-25 and TIMP-3 were also upregulated (Huang et al. 2012), but the mechanisms responsible remain unclear. Interleukin-10 also antagonizes the upregulation of MMP-1 and MMP-9, but unlike IL-4 it increases expression of TIMP-1 (Tables S1 and S2). Again, the intermediary action of SOCS proteins appears reasonable but remains to be fully documented. Interleukin-10, in particular, is abundant in atherosclerotic plaques and is therefore most likely to exert a physiological dampening effect on MMP activity. Transforming growth factor-β inhibits MMP-12 production in human monocytes (Table S1). However, TGFβ can both stimulate and inhibit MMP-2 and MMP-9 secretion from mouse peritoneal macrophages (Ogawa et al. 2011). Upregulation of MMP-9 by TGFβ has recently been ascribed to stimulation of PI3K leading to activation of AP-1 transcription factors (Haidar et al. 2015). Activation of several anti-inflammatory nuclear hormone receptors inhibits MMP production (see Tables S1 and S2). For example, PPARα selectively inhibits IL-1β-induced MMP-12 production by direct binding to components of the AP-1 complex (Souissi et al. 2008), whereas both PPARα and PPARγ inhibit MMP-9 secretion from human macrophages (Table S2). PPAR γ agonists protect against the macrovascular complications of diabetes (Dormandy et al. 2005), and inhibition of MMP activity could play an important part in this action. Statins, the mainstay of atherosclerosis prevention, have also been shown to inhibit the expression of a broad range of MMPs by both transcriptional and post-transcriptional mechanisms (reviewed by Newby, 2008).

Conclusion: the combined action of multiple mediators causes MMP upregulation and plaque rupture

Animal and human data support the concept that an excess of MMP over TIMP production from macrophages and foam cells contributes to atherosclerotic plaque growth and rupture. In rabbit and mouse models, several MMPs promote plaque progression and affect plaque morphology in ways consistent with greater vulnerability to rupture. Furthermore, foam cell macrophages in subcutaneous granulomas or atherosclerotic plaques actively express several MMPs that are also secreted by non-foamy macrophages. Adaptive immunity seems to have little impact on macrophage polarization and increasing levels of MMPs in mice, implying a more prominent role for innate immune mechanisms, including the production of CSFs, inflammatory cytokines and Toll-like receptor ligands. Even so, MMP activity must be tightly regulated because overexpression of high levels of fully activated MMPs is needed to provoke plaque disruption in mice. The importance of specific MMPs may be over- or underemphasized in mice, where they are more or less abundant, compared with man (see Fig. 1); hence, studies in human cells and tissues should be given primary importance, especially if genetic approaches at a population level (such as that for MMP-12) can be developed to give clearer indications of causality.

Longitudinal imaging studies lead to the striking conclusion that most vulnerable plaques go on to heal rather than rupture (Van Mieghem et al. 2006). Hence, ulceration of human plaques is a relatively rare outcome that, exactly like any other accident, most probably occurs because of an unusual combination of adverse circumstances. Plaque rupture is most likely to result from a ‘perfect storm’ caused by the synergistic local effects of multiple inflammatory mediators acting together in a hypoxic environment combined with the loss of inhibitory signals from nuclear hormone receptors, TGFβ and IL-10. The potential mediators of MMP overproduction include IL-1β, which can be produced in plaques in response to oxidized LDL (Williams & Tabas, 1995) and as a result of inflammasome activation by cholesterol crystals (Duewell et al. 2010; Rajamäki et al. 2010). The ongoing CANTOS clinical trial will examine the causal role of IL-1 in unstable coronary disease (Dinarello et al. 2012). Other pro-inflammatory mediators, including TNFα, GM-CSF and IL-6, which stimulate macrophages through different signalling pathways (Fig. 2), have the potential to induce MMPs synergistically. Suppression of TIMP-3 expression by foam cell formation, INFγ or GM-CSF could be a further significant factor. Toll-like receptor ligands, the most effective stimulators of MMP production in vitro, are also present in the atherosclerotic plaques (Lundberg & Yan, 2011). Conversely, anti-inflammatory treatments, including, importantly, the use of statins, currently provide the best approach to reducing MMP activity in plaques and therefore preventing plaque rupture. In future, it is likely that more selective treatments will be developed. These should be aimed at inhibiting excess production of specific MMPs, especially the collagenases MMP-1 (Libby, 2013) and MMP-8 (Ye, 2015) and MMP-12 (Traylor et al. 2014), whilst preserving the activity of those MMPs, including MMP-9, that are primarily involved in vascular repair (Newby, 2005). The widely different regulation of different MMPs in human macrophages that recent studies have so clearly emphasized (Huang et al. 2012) provides strong encouragement for such an approach.

Competing interests

None declared.

Funding

The author's work is supported by the British Heart Foundation grant CH95001 and the National Institute for Health Research Bristol Cardiovascular Biomedical Research Unit.