Identification and analyses of inhibitors targeting apolipoprotein(a) kringle domains KIV-7, KIV-10, and KV provide insight into kringle domain function

Increased plasma concentrations of lipoprotein(a) (Lp(a)) are associated with an increased risk for cardiovascular disease. Lp(a) is composed of apolipoprotein(a) (apo(a)) covalently bound to apolipoprotein B of low-density lipoprotein (LDL). Many of apo(a)'s potential pathological properties, such as inhibition of plasmin generation, have been attributed to its main structural domains, the kringles, and have been proposed to be mediated by their lysine-binding sites. However, available small-molecule inhibitors, such as lysine analogs, bind unselectively to kringle domains and are therefore unsuitable for functional characterization of specific kringle domains. Here, we discovered small molecules that specifically bind to the apo(a) kringle domains KIV-7, KIV-10, and KV. Chemical synthesis yielded compound AZ-05, which bound to KIV-10 with a Kd of 0.8 μm and exhibited more than 100-fold selectivity for KIV-10, compared with the other kringle domains tested, including plasminogen kringle 1. To better understand and further improve ligand selectivity, we determined the crystal structures of KIV-7, KIV-10, and KV in complex with small-molecule ligands at 1.6–2.1 Å resolutions. Furthermore, we used these small molecules as chemical probes to characterize the roles of the different apo(a) kringle domains in in vitro assays. These assays revealed the assembly of Lp(a) from apo(a) and LDL, as well as potential pathophysiological mechanisms of Lp(a), including (i) binding to fibrin, (ii) stimulation of smooth-muscle cell proliferation, and (iii) stimulation of LDL uptake into differentiated monocytes. Our results indicate that a small-molecule inhibitor targeting the lysine-binding site of KIV-10 can combat the pathophysiological effects of Lp(a).

vivo studies, transgenic animals, and ex vivo cell experiments (1,(9)(10)(11)(12). These studies imply that Lp(a) has prothrombotic as well as proatherogenic properties. The simplest model for the prothrombotic effect of Lp(a) states that inhibition of plasminogen activation to plasmin is a result of competition between apo(a) and plasminogen for binding sites on fibrin (13). However, apo(a), when bound to fibrin, can itself bind plasminogen, which can subsequently be activated (converted to plasmin) by tissue plasminogen activator (14), implying a more complex mechanism. In addition, most human studies have failed to show a correlation between apo(a) concentrations and fibrinolysis using various clot lysis assays (15), but Månsson et al. (16) could show that altered fibrin clot properties ex vivo were dependent on Lp(a) concentrations in samples from the "Diabetes and Impaired Glucose Tolerance in Women and Atherosclerosis" study. It was also recently shown that apo(a) is a substrate for coagulation factor XIIIa and is present in the fibrin clot (17). In support of proatherogenic effects, it was shown in vitro that recombinant apo(a) and Lp(a) isolated from plasma promote smooth muscle cell (SMC) migration and proliferation (18). Lp(a) may also alter the function of endothelial cells (19,20) as well as stimulate IL-8 expression (21). In addition, it has been shown that mutation of residues in the lysine-binding site (LBS) in KIV-10 reduces atherosclerosis in apo(a) transgenic mice (22). Whether Lp(a) is thrombogenic, atherogenic, or both remains open for continued research. High plasma concentration of Lp(a) is linked to disease, and it has been postulated that an absolute reduction of 100 mg/dl may be required to produce a clinically meaningful reduction in risk for cardiovascular disease (8). Treatments that lower Lp(a) plasma concentrations include apheresis (23,24), PCSK-9 inhibitors (25)(26)(27)(28), and apoB-degrading mipomersen (29,30) and, as the only approach directly aimed at lowering Lp(a), antisense oligonucleotides targeting apo(a) (31,32). Apheresis has been shown to effectively reduce Lp(a) by 73% (from 112 to 30 mg/dl) (23) or 68% (from 108 to 29 mg/dl) (24), but it is an invasive technique for the patient. Antisense oligonucleotides targeting apo(a) were reported to decrease Lp(a) levels in the same range as apheresis (32). It is unclear whether any other treatment can give such an efficient reduction in plasma Lp(a) concentration.
There are no reports in the literature of intervention with small molecules directed toward functional features of apo(a), with one exception. In a small ad hoc study, Frank et al. (33) demonstrated a decrease of apo(a) plasma concentration in humans after tranexamic acid (TX) treatment as well as TX effects on Lp(a) assembly. The lack of interest in small molecules modulating the properties of Lp(a) is surprising, as many of the proposed pathological properties of Lp(a) have been attributed to apo(a)'s main structural domains, the kringles, and to being mediated by their LBS. A kringle is a protein domain that consists of around 80 amino acids, stabilized by three internal disulfide bridges ( Fig. 1) (34). Each kringle has an LBS that interacts with lysine residues in fibrin and cell membrane proteins, for example (13). Apo(a) is the result of duplication of the PLA gene encoding plasminogen. The C-terminal kringles 4 and 5 (K4 and K5) and the protease domain of plasminogen have been transferred to apo(a), whereas the N-terminal kringles 1-3 (K1-K3) of plasminogen have been lost (Fig. 2).
In apo(a), the protease domain has lost its catalytic activity by mutation, and kringle 4 (K4) has differentiated into 10 subtypes named KIV-1 to KIV-10 (35). Kringle 5 (K5) of plasminogen has been retained as a single copy, named KV in apo(a). Kringle KIV-2 exists in multiple copies ranging from 2 to Ͼ40 in different individuals (Fig. 2). The size of the apo(a) isoform is inversely correlated to the plasma concentration of Lp(a) and therefore to the risk of cardiovascular disease (6).
Several functions have been associated with different kringles in apo(a). KIV-9 possesses an unpaired cysteine residue that forms the covalent bond between apo(a) and apoB of LDL (36). KIV-6 to KIV-8 have been shown to be important for the association of apo(a) with apoB in LDL that precedes the formation of the covalent bond in the Lp(a) particle (37)(38)(39). The lysinebinding sites of KIV-6 to KIV-8 are identical in all lysine-coordinating residues, and the LBS of KIV-5 differs at a single residue. A high-affinity LBS has been described in  and linked to atherosclerosis in apolipoprotein(a) transgenic mice (22). KV has been suggested to be the kringle of apo(a) that stimulates IL-8 production by monocytes (21).
We hypothesized that small molecules with sufficient selectivity and affinity for the LBS of apo(a) kringles could be valuable chemical probes to further decipher the functional importance of different kringle domains and to investigate the pathophysiology of Lp(a).
Others have shown that ⑀-aminocaproic acid (EACA) and TX bind to the LBS of apo(a) kringles. These ligands show promiscuous binding to kringle domains and fairly low affinity. EACA has a K d of 230 and 33 M for KIV-7 and KIV-10, respectively (40), and TX has a K d of 63 M for KIV-7, 5 M for KIV-10, and 1 M for plasminogen K1 (Table 1). TX is used for the treatment of bleeding conditions and has been shown to be beneficial in bleeding trauma patients, for example (41). Its main mechanism of action is inhibition of the binding of plasminogen to fibrin by blocking the LBS of plasminogen K1, thereby preventing the formation of plasmin and subsequently slowing down fibrinolysis (42,43). However, as mentioned above, TX has also been shown to affect Lp(a) (33).
Previous work (37) suggests that high-affinity apo(a) kringleselective small molecules could lead to drugs that specifically address the needs of individuals with high Lp(a) levels. These would either lower Lp(a) and/or interfere with the pathological actions of Lp(a) that are mediated by the LBS in apo(a) kringles. We therefore set out to test a library of small molecules for binding to apo(a) kringles KIV-7, -10, and KV. Sequence alignment showed that the anionic bidentate motif (DXD or DXE), which is characteristic of a strong lysine-binding site, is present in these kringles.
Here, we describe a lead generation strategy with the aim to identify and study potent and selective compounds targeting the LBS of kringles of apo(a) and their functional effects. We also report crystal structures of KIV-7, KIV-10, and KV in complex with small molecules. Although there are kringle-binding compounds known, such as lysine analogs and TX, this is the first report of a selective apo(a) KIV-10 LBS-binding compound as well as compounds with a higher affinity for KIV-7 or KV than for KIV-10. To our knowledge, these are the first chemical entities with this unique selectivity profile, and therefore, we

Inhibitors for apo(a) kringle domains
have studied their effect on the functions of the apo(a) component of the Lp(a) particle.

Selection of screening library
A set of 430 compounds with affinity for plasminogen K1 from a plasminogen-focused AstraZeneca project (42,43) was analyzed to identify properties typical for compounds binding to the LBS. We found that all compounds binding to the kringles of plasminogen were small, with a molecular weight between 150 and 300 g/mol, and rather polar with a calculated log D of less than 3. In addition, analysis of the compounds showed that all ligands had a basic functionality, as defined by the ACD/Labs software (Advanced Chemistry Development, Inc.).
AstraZeneca's compound collection and the compound collections from external vendors were filtered to identify small molecules fulfilling the criteria derived from plasminogen kringle ligands above. The resulting compounds were docked into plasminogen K1 and K5 (PDB codes 1CEA and 5HPG) using the Glide software from Schrodinger (version 5.8, Schrödinger, LLC, New York) and the 1500 compounds with best-docking score in plasminogen K1 and the 1000 best-scoring compounds in plasminogen K5 were kept. Subsequent dockings were performed in plasminogen kringles.
An additional 2000 compounds were selected from fingerprint similarities (44) to compounds binding to plasminogen kringle K1. 1100 compounds generated in the K1 project were also included in the screening set adding up to a total of 5600 compounds.

Screening of compound library
The set of 5600 compounds was tested for binding to KIV-7, KIV-10, and KV using the Epic biosensor platform. Apo(a) and plasminogen have opposing actions in fibrinolysis (42,43). Therefore, K1 of plasminogen was included as a selectivity screen because compounds binding to this kringle may inhibit fibrinolysis and lead to stimulation of thrombosis. All compounds effective in the Epic dose-response assays (hit compounds) were tested for binding against the same four kringles using nuclear magnetic resonance (NMR). The NMR experiments confirmed specific binding of the hit compounds to the different kringle LBS by studying the displacement of a reporter molecule of known affinity. Any compound that could not be confirmed to be binding specifically to the LBS of at least one of the kringles was excluded from further testing. The NMR measurements were also used to determine the K d values for each compound and kringle (Table 1 and Fig. S1). Most of the hit compounds had at least a 5-fold selectivity over one of the other

Inhibitors for apo(a) kringle domains
apo(a) kringles. This was somewhat unexpected given the high degree of sequence identity between apo(a) kringles, especially KIV-7 and KIV-10 (81% identical residues), with two amino acids differing in the LBS (Figs. 1 and 3). Sequence alignment of all apo(a) kringles shows that the LBS of KIV-7 is identical to those of KIV-6 and KIV-8. The LBS of KIV-5 differs at a single residue (Ser-34), but the interaction with the ligand is mediated by the main-chain nitrogen (Fig. 1). Therefore, compounds binding to KIV-7 are expected to bind to KIV-5 to KIV-8 with similar affinities.

Crystal structures of kringle domains
To facilitate chemistry design with respect to selectivity and affinity, we solved crystal structures of ligand complexes of KIV-10, KIV-7, and KV. The selection of compounds for crystallization was primarily based on high affinity for one apo(a) kringle and selectivity toward plasminogen K1 ( Table  1). The results showed that all compounds had a basic nitrogen binding to the bidentate acidic motif at one end of the LBS (DXD or DXE, Fig. 1) and a negatively-charged entity forming hydrogen bonds on the cationic side of the LBS, which is in agreement with the literature (45,46). The KIV-10 crystal structures revealed that compounds that displayed at least a 5-fold selectivity for KIV-10 against K1 had multiple polar interactions in the KIV-10 LBS, including hydrogen bonds to Arg-71, Arg-35, and the main-chain nitrogen of Gln-34. Superposition of the KIV-10 and the K1 crystal structures (PDB code 1CEB) (47) showed that Arg-35 is replaced by a phenylalanine in K1 and that Gln-34 is replaced by an arginine pointing into the binding site. We refer to this part of the pocket as the "selectivity region" (Fig.  3). The position of the Arg-35 side chain differs in the four KIV-10 crystal structures we report here.
Comparison of the crystal structures of KIV-7 and KIV-10 showed that the overall fold of the kringles was similar and that the differences in the binding pocket were limited to residues Glu-57 KIV-7 /Asp-57 KIV-10 and Tyr-64 KIV-7 /Phe-64 KIV-10 (Figs. 1 and 3). The AZ-04 compound was selected for crystallization in KIV-7, despite its lack of selectivity, based on its small size and relatively high affinity ( Table 1). The carboxyl group of the compound formed hydrogen bonds to Arg-71 and Tyr-62 as well as to a crystallographic symmetry-related molecule. The guanidinium group of Arg-35 was stacking against the aromatic ring of AZ-04 (Fig. 3).
We also report the first crystal structure of apo(a) KV. The structure was solved with the protein domain in complex with the antibiotic ciprofloxacin (AZ-07, Table 1 and Fig. 3) which was a low-affinity hit from the screen. The carbonyl of the quinolinone scaffold formed a hydrogen bond to Tyr-64, and the basic nitrogen of the piperazine was within hydrogen-bonding distance of Asp-55 and Asp-57. The fluorine atom was positioned in the cleft formed between the side chains of Trp-62 and Phe-72, with a distance of 3.4 Å to each. Unexpectedly, no interactions were observed between the , and the C-terminal serine protease domain (SP). In apo(a), the protease domain has been changed by mutation in critical residues, destroying its plasmin activity, and the kringle K5 of plasminogen was retained as a single copy (KV), whereas K1-K3 were lost. K4 of plasminogen has differentiated into 10 subtypes of KIV in apo(a) (KIV-1 to KIV-10). Kringle KIV-2 exists in multiple copies and can range from 2 to Ͼ40 copies. B, Lp(a) consists of an LDL particle to which apo(a) is covalently linked by a disulfide bond. The LDL particle is composed of a central lipid core and a single molecule of apoB. The disulfide bond is formed between apoB and KIV-9 of apo(a).

selection of hit compounds
An asterisk next to a K d value indicates the highest affinity and therefore primary apo(a) kringle selectivity for each compound. The K d values for plasminogen K1 and selectivity against K1 are also reported. PDB IDs for kringle/compound complexes are reported in the right-hand column and in brackets are the kringle proteins in which it has been co-crystallized.

Inhibitors for apo(a) kringle domains
carboxyl group of compound AZ-07 and the protein or ordered water molecules. The cyclopropyl group was pointing out of the LBS toward the solvent (Fig. 3). The crystal structure was highly similar to plasminogen K5 (PDB accession code 5HPG (48)), with a root mean square deviation for all C␣ of 0.76 Å 2 .

Structure-based design
The KIV-7-and KV-selective hits, AZ-06 and AZ-07, showed moderate affinities toward their target kringles but had affinities for K1 above the detection level of the NMR assay, which corresponds to at least 6-and 3-fold K1 selectivity (Table 1). For the compounds directed toward KIV-10, AZ-01 was 50 times selective over K1, whereas AZ-02 and AZ-04 showed considerable affinity for K1. We therefore decided to try to improve the selectivity for KIV-10 versus K1 by design and synthesis of new compounds.
Overlay of two crystal structures of KIV-10 in complex with AZ-02 or AZ-03, respectively, suggested that hybridization of the two compounds would be beneficial for affinity and selectivity (Fig. 4). Three compounds were synthesized and screened by NMR to test this hypothesis. This resulted in compound AZ-05 that was 5 and 16 times more potent than AZ-03 and AZ-02, respectively. AZ-05 was similar to AZ-01 with respect to affinity for KIV-10, but AZ-05 showed more than 120 times selectivity over K1 (Table 1).

In vitro testing of kringle-selective compounds
The most potent and selective KIV-10 inhibitor (AZ-05, K d ϭ 0.8 M), and two compounds of moderate affinity for KIV-7 (AZ-06, K d ϭ 17 M) and KV (AZ-07, K d ϭ 30 M) ( Table 1) were chosen as chemical probes to characterize the role of kringle domains of apo(a) in in vitro assays. The assays were designed to link the effect of the compounds on isolated kringles to the effect on full-length apo(a) as well as mimicking potential pathophysiological mechanisms of Lp(a) mediated by apo(a) in vitro. For this purpose, a recombinant human apo(a) containing 17 copies of the KIV-type kringle domains followed by one KV and the protease-like domain was expressed (referred to as fl_apo(a)). The Lp(a) mechanisms acting through apo(a) studied here include the following: (i) binding to fibrin; (ii) stimulation of smooth muscle cell proliferation; (iii) stimulation of LDL uptake into differentiated monocytes; and (iv) assembly of Lp(a) from apo(a) and LDL. The KV-selective compound AZ-07 was excluded from the surface plasmon resonance (SPR)-based assays, as it bound nonspecifically to the sensor surfaces used.
Specific binding of compounds to LBS on fl_apo(a) was assessed by NMR using the same reporter molecule displacement setup as was used for the individual kringle domains. The results confirmed that AZ-05, AZ-06, and AZ-07 competed for binding with a reporter molecule targeting the LBS of kringles (Fig. S1B). In contrast to the data generated using the isolated kringle domains, none of the three ligands completely displaced the reporter in the full-length protein (Fig. S1B). This might reflect that the selective inhibitors compete for binding to one or a few kringles, whereas the reporter binds to several LBS in fl_apo(a). Also, we cannot rule out unspecific binding of the reporter to fl_apo(a), in addition to the specific LBS interaction.  (47)) is shown. The "selectivity region" is circled. Hydrogen bonds between AZ-01 and KIV-10 are indicated by dashed lines, and distances vary between 2.6 and 3.0 Å. C, crystal structure of the apo(a) KV/AZ-07 complex. Hydrogen bonds are illustrated as dashed lines, and distances vary between 2.8 and 3.0 Å. The piperazine of AZ-07 is forming hydrogen bonds to Asp-55 and Asp-57, and the carbonyl of the AZ-07 quinolinone is within hydrogen-bonding distance of Tyr-64. The F o F c omit map calculated using the refined model is contoured at 3.2 . The figures were prepared with PyMOL (Schrö dinger).

Inhibitors for apo(a) kringle domains
To make it easier for the reader, we will from now on add the most pronounced apo(a) kringle selectivity as superscripts to the compounds resulting in the nomenclature AZ-05 KIV-10 , AZ-06 KIV-7 , and AZ-07 KV .
The KIV-10 -selective AZ-05 KIV-10 is expected to have selectivity over all other apo(a) kringles, even though only KIV-7 and KV have been experimentally tested. This is based on the high degree of similarity of the LBS of KIV-5, -6, and -8 to the KIV-7 LBS, which does not bind AZ-05  . Furthermore, the KIV-10 LBS contains the bidentate DXD motif that is lacking in kringles KIV-1 to KIV-4 and KIV-9 (Fig. 1).

Apo(a) binding to fibrin
The binding of Lp(a) to fibrin mediated by apo(a) has been postulated to be the main mechanism behind the prothrombotic effects of Lp(a) (49). For this reason, we set up an SPR assay to monitor the interaction between fl_apo(a) and fibrin in real time. Fibrin was immobilized on the biosensor surface, and the response was monitored when fl_apo(a) in solution was injected with and without compound. The apo(a)-fibrin interaction was inhibited by TX with a K i of 30 M (Fig. 5 and Fig.  S2A). We then tested the apo(a) kringle-selective compounds, AZ-05 KIV-10 and AZ-06 KIV-7 , up to a concentration of 200 M (Fig. 5). AZ-06 KIV-7 reached a maximum of 30% inhibition at 25 M but lost its effect at higher concentrations, and at 200 M no inhibition was observed. AZ-05 KIV-10 , however, showed an increase in inhibition of apo(a) binding to fibrin reaching 69% inhibition at 200 M (50% inhibition at 20 M). We also per-formed experiments where the two compounds were combined to assess whether there were additive effects. AZ-05 KIV-10 , when combined with AZ-06 KIV-7 , reached the same maximum level of inhibition but at 1/8th the concentration. A tendency to loss of effect at higher concentrations was observed, similar to that seen for AZ-06 KIV-7 in the single compound experiment (Fig. 5). We speculate that this apparent stimulation of apo(a) binding to fibrin at a high concentration (Ͼ25 M) of AZ-06 KIV-7 is related to conformational changes of apo(a) as shown by Becker et al. (50) and Fless et al. (51). Alternative explanations for the apparent mass increase at higher concentrations would be ligand binding to the fibrin surface or compound-compound interactions. However, no interaction between the compound and the fibrin surface was observed in control experiments. To rule out the compound-compound interaction, we verified the solubility and nonaggregation for both compounds at 200 M using NMR and the same buffer conditions as used in the SPR assays. These results imply that KIV-10 is involved in fibrin binding and that this interaction can be inhibited by blocking the KIV-10 LBS with a small molecule.

Lp(a) assembly
In vitro evidence suggests that before formation of the covalent bond between KIV-9 and apoB, a noncovalent complex is formed by apo(a) kringle domains binding to lysine residues on apoB on LDL (37-39). To mimic this crucial step in Lp(a) formation, we developed an Lp(a) assembly assay using SPR that To improve the affinity for KIV-10, the pyrazole ring of AZ-03 was replaced by the more flexible linker from AZ-02, and a carbonyl was added in the position of the AZ-02 carboxyl group, which enabled an additional hydrogen bond with Arg-71. We predicted that by keeping the sulfone of AZ-03 with its polar interactions, the K1 selectivity could be retained. B, crystal structure of the KIV-10/AZ-05 complex. Three hybrid compounds were synthesized and the best, AZ-05, showed a 6-fold increase in affinity to KIV-10 and a 10-fold increase in selectivity against K1 compared with AZ-03. Water molecules are shown as spheres, and hydrogen bonds are indicated by dashed lines in the same colors as the protein complex to which they belong.

Inhibitors for apo(a) kringle domains
allowed us to monitor the binding of apo(a) to apoB in a native immobilized LDL particle in real time. Biotinylated LDL particles were attached to the chip surface; fl_apo(a) was injected; and the binding of fl_apo(a) to LDL was recorded (Fig. S2B). Addition of TX to fl_apo(a) before injection inhibited the formation of the noncovalent apo(a)/LDL complex with a K i of 30 M (Fig. 5). No binding was observed between fl_apo(a) and a reference surface with LDL particles where all surface-exposed lysines had been modified by acetylation. These experiments support a specific kringle-lysine interaction.
We then tested the kringle-selective compounds, separately and in a combination experiment (Fig. 5). Addition of AZ-05 KIV-10 to fl_apo(a) prior to injection resulted in a dosedependent increase of inhibition of fl_apo(a) binding to LDL up to 200 M reaching 50% inhibition. Unlike the TX experiment, the dose-response curve of AZ-05 KIV-10 is not sigmoidal, implying a more complex inhibitory mechanism. Addition of AZ-06 KIV-7 to fl_apo(a), in contrast, had no inhibitory effect. However, the combination of AZ-06 KIV-7 and AZ-05 KIV-10 resulted in a modest increase in maximum inhibition from 50 to 55% and also reached this effect at lower compound concentrations (Fig. 5). These results suggest that the first step in the fl_apo(a)-induced formation of the Lp(a) particle can be blocked by a compound directed toward KIV-10 of apo(a). In addition, increased efficacy may be achieved by combining inhibitors with different kringle selectivity.

Apo(a) induced SMC proliferation
The initial steps of atherosclerosis include the proliferation of resident intimal smooth muscle cells (SMC) and the development of foam cells via the maturation of monocytes into macrophages and their subsequent uptake of lipid, for example lipoprotein particles (52). We investigated the effect of fl_apo(a) with and without inhibitors using in vitro model systems that mimic these events.
Addition of 1 M fl_apo(a) significantly stimulated SMC proliferation compared with cells grown in media without fl_apo(a). This effect could be completely reversed by 6 mM TX (the estimated IC 50 was Ͼ180 M) (Fig. 6). We added selective compounds and TX together with fl_apo(a) and determined the dose-dependent effects of compounds on the apo(a)-induced proliferation up to 180 M compound concentration. The addition of AZ-06 KIV-7 and AZ-05 KIV-10 resulted in IC 50 values of 80 and 40 M, respectively (Fig. 6). However, the interpretation of these numbers is complex because multiple binding events may be involved as AZ-06 KIV-7 is expected to also interact with KIV-5, KIV-6, and KIV-8 in the full-length protein. In addition, AZ-05 KIV-10 is a more potent inhibitor than AZ-06 KIV-7 . None of the tested compounds by themselves, including TX, had an effect on the proliferation or the viability of the SMC in the absence of fl_apo(a) up to the highest concentrations tested.

LDL aggregation and macrophage uptake
Lp(a) particles are formed by apo(a) binding to LDL (Fig. 2B), and our SPR results demonstrated that TX and AZ-05 KIV-10 prevented apo(a) binding to LDL (Fig. 5B). Xu (53) showed by scanning force microscopy that apo(a) binds to the LDL sphere at two distant sites and that, occasionally, apo(a) could bridge two LDL particles, bringing them together. We therefore hypothesized that addition of fl_apo(a) could accelerate aggregation of LDL. Indeed, we could show that addition of fl_apo(a)

Inhibitors for apo(a) kringle domains
to isolated human LDL induced formation of aggregates visible in a light microscope within a few hours, whereas no aggregation occurred in the absence of apo(a). Addition of 6 mM TX together with fl_apo(a) reduced the formation of aggregates (Fig. S3). Similarly, when fl_apo(a) was added to 5% fetal calf serum (FCS), aggregates were formed, and this aggregation was inhibited when TX was added together with fl_apo(a). No aggregate formation occurred when lipoprotein-depleted serum was used. Taken together, these data suggest that the aggregates were formed from LDL and promoted by apo(a).
In the experiments, we also observed that macrophages (differentiated THP-1 monocytes) engulfed the formed lipoprotein aggregates (Video S1 and Fig. S4). Consumption of LDL or Lp(a) aggregates by macrophages is a known biological phenomenon (54, 55). To study the effects of apo(a) on LDL/Lp(a) incorporation, we added fluorescently-labeled human LDL (DiI-LDL) to the macrophages and determined the cellular content of fluorescence (Fig. 7). The presence of fl_apo(a) resulted in an increased uptake of DiI-LDL by the macrophages, and addition of TX with fl_apo(a) reduced the DiI-LDL incorporation in a dose-dependent manner (Fig. 7). We therefore hypothesized that the increased uptake of fluorescent LDL by macrophages was a consequence of the apo(a)-induced aggregation of LDL. The kringle-selective compounds were tested in the DiI-LDL assay in a concentration response up to 180 M. TX and AZ-05 KIV-10 showed inhibition of the apo(a)-mediated increase of DiI-LDL uptake, whereas AZ-06 KIV-7 and AZ-07 KV had a limited effect ( Table 2). This suggests that blocking the kringle KIV-10 LBS can reduce apo(a)-induced LDL uptake into macrophages, although we cannot rule out other mechanisms. By themselves, none of the compounds influenced the viability of the macrophages. In summary, addition of fl_apo(a) induced aggregation of LDL particles and subsequent uptake of the aggregates by macrophages. These effects could be inhibited by TX or a KIV-10 -selective small molecule inhibitor.

Discussion
In this study, we report the identification of compounds binding to the apo(a) LBS with selectivity profiles different from TX and lysine analogs. Most of the hit compounds from the initial screen showed some selectivity toward one of the three apo(a) kringles for which they were tested. This was also true for KIV-10 and KIV-7, despite the high degree of sequence identity between the two kringles. Only two amino acids differ between the respective LBS, and comparison of crystal structures show that all other residues are conserved in position and in sequence ( Figs. 1 and 3). The acidic Asp-57 in KIV-10 is replaced by a glutamic acid in KIV-7, making the KIV-7 pocket slightly smaller. The exchange of Phe-64 from KIV-10 to a tyrosine in KIV-7 changes the pattern of hydrogen bonds and that leads to small but significant changes in binding modes for the same ligand. These two differences taken together may well account for the KIV-10/KIV-7 selectivity observed for the hit compounds. Inhibition of plasminogen K1 will lead to decreased fibrinolysis and thereby increase the risk of clinical thrombosis. This is not a desirable effect for a patient group that may benefit from an apo(a) inhibitor, and therefore obtaining compounds that are selective over K1 is of great importance. Especially, selectivity between KIV-10 and plasminogen K1 will be the main focus for future compound design. The KIV-10 crystal structures revealed that all compounds that displayed selectivity against K1 (Table 1) had multiple polar interactions in the KIV-10 LBS "selectivity region," including hydrogen bonds to Arg-71, Arg-35, and the main chain nitrogen of Gln-34 (Fig. 3). Superposition of the crystal structures of K1 (47) and KIV-10 ( Fig. 3) showed that the hydrogen bonds between the ligand and Arg-35 in KIV-10 would not be possible in K1, because K1 has a phenylalanine (Phe-36) in this position. In addition, the shape of the LBS in this region is changed because the backbones of Arg-34 K1 and Gln-34 KIV-10 are shifted relative to each other, with the side chain of Arg-34 K1 pointing into the LBS (Fig. 3) (47). The compounds AZ-01 and AZ-05 have a sulfonamide moiety bound in this region (Table 1 and Fig. 3), forming hydrogen bonds to Arg-35 and Gln-34 in the KIV-10 LBS. AZ-01 showed a 50-fold selectivity for KIV-10 over K1. In the case of AZ-05, the selectivity is improved to more than 120-fold, possibly due to increased steric hindrance introduced by the trifluoromethoxy substituent (Table 1). K1, like KIV-7, has a tyro-

Inhibitors for apo(a) kringle domains
sine in the place of Phe-64 of KIV-10, which may provide an alternative opportunity for a hydrogen bond.
We report here the first crystal structure of apo(a) KV. Plasminogen K5 and apo(a) KV share 84% sequence identity, and their lysine-binding sites are highly similar with a two-residue difference, Phe-72 KV /Tyr-72 K5 and Thr-35 KV /Ile-35 K5 . However, KV is quite divergent from KIV-7 and KIV-10 with its larger lysine-binding site and different charge distribution, suggesting that it should be feasible to develop KV-specific compounds with selectivity over other apo(a) kringles. During protein purification, KV was observed to not adhere to a lysine-Sepharose matrix. The likely reason for this is that the cationic region of the LBS present in all other apo(a) kringles (Arg-71 and Arg-35 in KIV-10) is replaced by neutral residues in K5 and KV (Fig. 1). Removal of the cationic center that coordinates the carboxylic acid of C-terminal lysines may be enough to abolish lysine affinity. In addition, Rahman et al. (56) reported that mutation of Trp-72 (the KIV-10 numbering used in this paper, see Fig. 1) to phenylalanine in the binding pocket of KIV-10 reduced the affinity for lysine and fibrin. This tryptophan is replaced in both K5 and KV by tyrosine and phenylalanine, respectively. In contrast, the anionic bidentate motif is still present in KV and K5, and we cannot rule out that there is still some affinity for the positively-charged lysine side chain. It is tempting to speculate that the lack of binding by KV to lysine-Sepharose translates to a lack of binding to lysines on cell membranes and proteins, which could suggest that KV does not mediate the pathophysiological actions of Lp(a).
The access to crystallization systems for all three kringles has enabled structure-based design. Overlay of crystal structures has in this case been used to hybridize two compounds of mod-

Inhibitors for apo(a) kringle domains
erate affinity, AZ-02 and AZ-03, into compound AZ-05 KIV-10 with a 5-fold increase in affinity for KIV-10 compared with AZ-03 and a 16-fold increase compared with AZ-02. AZ-05 KIV-10 also showed a 10-fold increase in selectivity compared with AZ-03, and a more than 120-fold increase in selectivity compared with AZ-02 was obtained in a single design step. The hybridization of AZ-02 and AZ-03 into AZ-05  showed that our models can be used to improve the compound selectivity for the different kringles and gave us a tool to specifically inhibit KIV-10. By studying the effects of the chemical probes in in vitro assays with fl_apo(a), we have gained insight into the roles of KIV-10 and KIV-7 of apo(a). These results could potentially be extrapolated to better understand the pathophysiological effects of Lp(a) mediated by apo(a). NMR experiments, supported by X-ray crystal structures, unambiguously showed that the compounds bound to the LBS of the isolated kringles KIV-10, KIV-7, and KV. NMR experiments also confirmed that AZ-05 KIV-10 and AZ-06 KIV-7 bound specifically to fl_apo(a) but did not provide information on which kringle(s) in the full-length protein with which they interacted. None of the selective compounds tested resulted in complete displacement of the reporter molecule. This is consistent with the reporter binding to all lysine-binding sites in fl_apo(a), whereas the selective inhibitors only compete for binding to one or a few of the kringles (Fig. S1). The affinity of AZ-05 KIV-10 for fl_apo(a) could be estimated under the assumption that AZ-05 KIV-10 only binds to KIV-10 in fl_apo(a) and that the reporter has the same affinity for KIV-10 in the full-length protein as when measured in the isolated kringle. Under these assumptions a K d for AZ-05 of 3 M for fl_apo(a) could be calculated, which is in a similar range as for isolated KIV-10 (0.8 M). The fact that the K d values agree could suggest that AZ-05 binds exclusively to KIV-10 also in the full-length protein in this experimental setup, but this conclusion is speculative and need to be assessed by other methods, e.g. mutational studies. It is also not known which kringle LBS is available in the full-length apo(a) protein. As a comparison, plasminogen has been shown to exist in a form where four of its five kringle domains are inaccessible, and a conformational change is required for all lysine-binding sites to be exposed (57,58). It has been shown that a similar conformational change takes place in apo(a) (50,51), and potentially only one or a few kringle lysinebinding sites are accessible for ligand binding in a more compact form of the apo(a). Another line of work (59) shows that apo(a) that has been produced the same way as fl_apo(a) used in our assays can contain certain oxidized phospholipids as posttranslational modifications that also can be found in vivo and might have a pro-inflammatory and pro-atherogenic effect. For this modification, the intact LBS of KIV-10 is needed, but which amino acid(s) in fl_apo(a) that are modified is not known.
However, for this discussion of the effects seen by our compounds in assays, we assume that the compounds bind primarily to the same kringles in fl_apo(a) as shown for the isolated domains.
The interaction between the ligands and fl_apo(a) was also established in the SPR assays where fl_apo(a) was incubated with compounds prior to testing for binding to fibrin or LDL.
Sequence alignment of the apo(a) kringle domains (Fig. 1) showed that the critical acidic bidentate motif (DXD or DXE) of the LBS is missing in KIV-1 to -4 and KIV-9, and therefore, we do not expect our compounds to bind to these kringles with high affinity. The LBS in KIV-7 very much resembles those in KIV-5, KIV-6, and KIV-8; according to sequence alignment, all amino acids with side chains involved in ligand binding are identical (Fig. 1). It is therefore likely that effects of AZ-06 KIV-7 could be mediated by any, or several, of the KIV-5 to KIV-8 kringles. KIV-7 has been described to have an ϳ10-fold lower affinity, for L-lysine and its analogs, than . This is consistent with our findings that a compound blocking the LBS in KIV-10 was the most effective inhibitor of apo(a) binding to C-terminal lysines on fibrin (Fig. 5), highlighting the importance of KIV-10 for this interaction. However, the lack of effect of AZ-06 KIV-7 might also be the result of lower affinity. Surprisingly, blocking the LBS in KIV-7 only had an inhibitory effect at lower concentrations, although the effect was reversed at higher concentrations (Fig. 5). It has been suggested that apo(a) is maintained in a compact state in Lp(a) through interactions between its weak LBS and multiple lysines on apo(a) and/or apoB (51). Binding of lysine analogs induces a global conformational change that is brought about by the saturation of a small number of weak lysine-binding sites, probably in KIV-5 to KIV-8, at high ligand concentrations (50,51). One could speculate that AZ-06 KIV-7 has a similar effect at high concentrations and thereby increases the accessibility of the LBS in KIV-10 for lysine residues on fibrin. The second SPR assay we have developed gave us the opportunity to follow the primary step in the assembly of Lp(a). For the interaction between apo(a) and apoB of LDL, we also found additive effects of AZ-06 KIV-7 and AZ-05 KIV-10 ( Fig. 5) even though AZ-06 KIV-7 alone did not have an effect in this assay. This is surprising considering that KIV-5 to KIV-8 have been reported to be important for the association of apo(a) with apoB in LDL (37)(38)(39). Helmhold et al. (38) concluded from a KIV-10 deletion mutant of apo(a) that KIV-10 is not required for Lp(a)-like particle formation. In contrast, our experiments showed that blocking the LBS of KIV-10 with a small molecule inhibitor can prevent the association of apo(a) and apoB and ultimately hinder Lp(a) formation. This effect can be increased by combining KIV-10 and KIV-7 inhibition. The same result has previously been observed with TX hindering Lp(a) formation (33). Taken together, this suggests that several kringles of apo(a) play a role in binding to apoB and can potentially substitute for each other, supporting the observation by Xu (53) that apo(a) was bound to the LDL sphere at two distant sites. In contrast, for the apo(a) interaction with fibrin, KIV-10 seems to be the most important apo(a) component. AZ-05 KIV-10 is inhibiting the apo(a)-fibrin interaction more efficiently than the apoB on LDL. In the fibrin-binding assay, 50% inhibition is reached at 20 M compound concentration whereas in the Lp(a) association assay 50% inhibition is achieved at 10 times higher concentration of AZ-05  . TX, however, has the same apparent K i for both effects because it affects several kringles. The next step would be to test the effect of the selective inhibitors on Lp(a). It is likely that some apo(a) kringle lysine-binding sites are occupied by lysines from apoB in the Lp(a) particle and that the interaction with apoB

Inhibitors for apo(a) kringle domains
affects the conformation of apo(a). Because blocking the LBS of KIV-10 in free apo(a) is efficiently inhibiting the association between fibrin and apo(a), it is tempting to speculate that this kringle is accessible and able to bind tightly to fibrin in the Lp(a) particle as well. However, this remains to be tested experimentally.
Apo(a) has been shown to increase the proliferation of SMC in cell culture (18,60), and the presence of kringle KIV-9 is required (18). The mechanism by which fl_apo(a) stimulates SMC proliferation is not clear, but our experiments indicate involvement of both KIV-10 and KIV-5 to KIV-8. This could mean that an association of apo(a) with cells, involving kringles KIV-7 or KIV-10, is necessary for further effects that are then mediated by KIV-9. Xu (53) showed that apo(a) was bound to more than one site on LDL, and occasionally apo(a) was bridging two LDL particles, bringing them close together. Phillips et al. (61) reported that the bulk of apo(a) extended away from the LDL surface where it may be available for other ligands. Also, apo(a) is often found dissociated from LDL in the arterial wall (62). All this adds up to a scenario where apo(a) can induce accelerated aggregation of LDL, which is subsequently taken up by macrophages leading to foam cell formation. This adds another hypothesis to the proposed pathophysiological mechanisms of Lp(a) that it can act, by itself or via free apo(a), as an accelerator of LDL aggregation and ultimately foam cell formation. Our experiments support this hypothesis because apo(a) increased LDL uptake into THP-1 monocytes differentiated into macrophages. This increase could be inhibited by TX and by our compound AZ-05 KIV-10 , suggesting that KIV-10 is required for connecting LDL particles together and inducing their aggregation.
Our work supports the idea that blocking the LBS of KIV-10 with a small molecule can interfere with pathophysiological effects of Lp(a) and that this may be a way forward as treatment for Lp(a)-associated cardiovascular disease. This is supported by the initial work of Frank et al. (33,63), but the same group later reported a study where feeding TX to transgenic mice expressing human apo(a) or Lp(a) resulted in a significant increase in plasma concentration of both Lp(a) and apo(a) (63). This is the opposite of the effect we are aiming for with our compounds. However, Frank et al. (63) suggested that the plasma concentration of TX in the mice did not reach the concentrations needed for effect. This needs to be further investigated using compounds that are more potent and selective than TX. We now have compounds with a novel selectivity profile, and the next step would be to test a KIV-10 -selective compound with high bioavailability in vivo to assess the compound efficacy in the context of Lp(a) assembly, anti-thrombosis and anti-atherogenesis. Such an assessment would further probe the potential of KIV-10 potent compounds as future therapies for Lp(a)-related diseases. There are therapeutics available that reduce Lp(a) levels, but not enough for clinical benefit (8,(25)(26)(27)(28)(29)(30). Modulation of kringle function by a small molecule might be a synergistic additive to such treatments or sufficient as a stand-alone therapy.
In summary, we have generated compounds with selectivity toward the different kringles of apo(a). In vitro assessment of these chemical tool compounds suggests a prominent role of kringle KIV-10 in the four tested functions of Lp(a)/apo(a) binding to fibrin, Lp(a) assembly, apo(a)-induced proliferation, and LDL aggregation with subsequent macrophage uptake ( Table 3) because addition of AZ-05 KIV-10 has an effect in all in vitro assays tested. These data support that a KIV-10-specific small molecule inhibitor may interfere with the pathophysiological effects of Lp(a). However, verification of which kringle domains in fl_apo(a) that interact with our KIV-10 specific inhibitor and if there is a time dependence remain to be investigated. Nevertheless, the current compounds provide starting points for developing small molecule drugs against Lp(a)-associated cardiovascular disease.

Experimental procedures
Expression and purification of human apo(a)-Human apo(a) was expressed in HEK293-6E cells. The protein (fl_apo(a)) comprised 17 copies of the apo(a) KIV-type kringle domains followed by one KV domain and the protease-like domain, identical to the protein described by Koschinsky et al. (64).
Apo(a) kringles KIV-7, KIV-10, KV, and plasminogen K1 were expressed in Pichia pastoris. A detailed description of the expression and purification of human apo(a) and kringle domains is available in the supporting information.
Protein crystallization and structure determination-Crystals of ligand complexes of KIV-10, KIV-7, and KV were obtained by co-crystallization. Data processing, structure solution, and refinement were performed using programs from the CCP4 suite (65). The crystal structures were solved to 1.6 -2.1 Å resolution. A detailed description of protein crystallization and structure determination is available in the supporting information. Details on data collection and refinement are summarized in Table S1. The coordinates and structure factors have been deposited in the Protein Data Bank, and the accession codes are listed in Table 1.

Inhibitors for apo(a) kringle domains
injection of ethanolamine, thrombin, and plasmin, the flow channels were washed with two 20-s pulses of 2 M guanidinium hydrochloride (GdmHCl). Immobilization levels of fibrin after plasmin treatment were around 5300 RU. Compounds were analyzed at 20°C with 9.6 nM apo(a) in 10 mM HEPES, pH 7.4, 0.15 M NaCl, 0.05% (v/v) polysorbate 20, 2% DMSO (v/v) by 1-min injections at 30 l/min, and a report point was recorded 45 s after injection start. The chip surface was regenerated between sample injections by a 30-s injection of 2 M GdmHCl . Lp(a) assembly SPR assay-Biotinylated native LDL (Biotrend) at 5 g/ml was coupled to streptavidin on the sample channel of an SA-chip sensor surface (GE Healthcare) in 10 mM HEPES, pH 7.4, 0.15 M NaCl, 0.05% (v/v) polysorbate 20, 3 mM EDTA, 10 M butylated hydroxytoluene (BHT) using the immobilization wizard in the Biacore T200 software (GE Healthcare). In a similar fashion, biotinylated, acetylated LDL (Biotrend) was immobilized on the sensor chip surface of the control channel. Immobilization levels were around 2500 RU. Compounds were analyzed at 20°C with 100 nM apo(a) in 10 mM HEPES, pH 7.4, 0.15 M NaCl, 0.05% (v/v) polysorbate 20, 3 mM EDTA, 10 M BHT, 2% DMSO (v/v) by 80-s injections at 30 l/min, and a report point was recorded directly after the injection. The chip surface was regenerated between samples by a 1-min injection of 10 mM tranexamic acid.
Proliferation of smooth muscle cells-Primary human coronary artery smooth muscle cells were from Lonza. 1000 cells/ well in Medium 231 supplemented with growth supplement (Invitrogen) in 100 l/well were plated in View Plates (black/ clear 96, PerkinElmer Life Sciences) and incubated at 37°C, 5% CO 2 overnight. Then the media were replaced. 50 l of RPMI 1640 medium supplemented with proteins and/or compounds was added to the cells, and then 50 l of RPMI 1640 medium ϩ 10% FCS (Invitrogen) were added to each well in the cell plates followed by incubation for 72 h in 37°C, 5% CO 2 . Subsequently, the proliferation was determined using the Cy-QUANT NF cell proliferation assay kit (Invitrogen).
Lipoprotein aggregation assay-Different combinations of 5% FCS, 1 M fl_apo(a), 6 mM TX, and 5% lipoprotein-depleted serum were mixed in 96-well View Plates (black/clear 96, PerkinElmer Life Sciences) without cells. The plates were incubated overnight in a normal CO 2 cell incubator at 37°C and 5% CO 2 . Images were taken in a Nikon Eclipse TE2000-U light microscope using ϫ20 magnification.
PMA differentiated THP-1 monocytes and aggregate uptake-After differentiation of monocytes (described above), the cells were treated with RPMI 1640 medium containing 5% FCS with and without 1 M fl_apo(a) present. Images were captured over time in Cell-IQ equipment (Chipman Technologies).
Fluorescent LDL uptake assay-The medium was removed, and addition of compound and/or apo(a) was made in RPMI 1640 medium with Glutamax ϩ 10% FCS up to a volume of 100 l. Fluorescence-marked DiI-LDL (Intracel) was present at a final concentration of 2.5 g/ml. Compounds had a final concentration ranging from 180 to 0.74 M in a dilution series. For the TX experiment, the compound concentration varied from 6 mM to 25 M. In the apo(a)-supplemented experiments, the protein concentration was 1 M. The plates were incubated for 4 days at 37°C and 5% CO 2 . The cells were then fixated by adding an equal volume of 7.4% formaldehyde (Sigma) directly to the cell medium without washing. After incubation at room temperature for 10 -20 min, the cells were washed three times with 100 l of PBS, and the nuclei were stained by adding 50 l of Hoechst (Invitrogen) diluted in 1:10,000 in PBS. After incubation at room temperature for 30 min, the cells were washed three times with 100 l of PBS. The DiI-LDL uptake into the cells was quantified using an imaging cytometer (Arrayscan VTI, Thermo Fisher Scientific Cellomics). Isolated human LDL was from Intracel. The synthesis or origin of all compounds (AZ-01 to AZ-07) and the NMR reporter compounds are described in detail in the supporting information.