A G protein–biased S1P1 agonist, SAR247799, protects endothelial cells without affecting lymphocyte numbers

Endothelial dysfunction is a hallmark of tissue injury and is believed to initiate the development of vascular diseases. Sphingosine-1 phosphate receptor-1 (S1P1) plays fundamental physiological roles in endothelial function and lymphocyte homing. Currently available clinical molecules that target this receptor are desensitizing and are essentially S1P1 functional antagonists that cause lymphopenia. They are clinically beneficial in auto- immune diseases such as multiple sclerosis. In patients, several side effects of S1P1 desensitization have been attributed to endothelial damage, suggesting that drugs with the opposite effect, namely, the ability to activate S1P1, could help to restore endothelial homeostasis. We found and characterized a biased agonist of S1P1, SAR247799, which preferentially activated downstream G protein signaling to a greater extent than β-arrestin and internalization signaling pathways. SAR247799 activated S1P1 on endothelium without causing receptor desensitization and potently activated protection pathways in human endothelial cells. In a pig model of coronary endothelial damage, SAR247799 improved the microvascular hyperemic response without reducing lymphocyte numbers. Similarly, in a rat model of renal ischemia/reperfusion injury, SAR247799 preserved renal structure and function at doses that did not induce S1P1-desensitizing effects, such as lymphopenia and lung vascular leakage. In contrast, a clinically used S1P1 functional antagonist, siponimod, conferred minimal renal protection and desensitized S1P1. These findings demonstrate that sustained S1P1 activation can occur pharmacologically without compromising the immune response, providing a new approach to treat diseases associated with endothelial dysfunction and vascular hyperpermeability.

INTRODUCTION

Endothelial dysfunction typically manifests in frank atherosclerosis requiring percutaneous or surgical revascularization, both of which are associated with further endothelial damage, restenosis, and repeat revascularizations (1, 2). Newer diagnostic tests have led to an increased recognition that endothelial dysfunction also affects the microcirculation (blood vessels with a diameter of <100 µm) (3), which, in contrast to larger arteries, can expand in response to physiological and pharmacological stimuli by increasing several fold the blood flow and oxygenation to organs such as the heart, kidneys, and limbs. Although traditional arterial vasodilators [such as nitrates and angiotensin converting enzyme (ACE) inhib- itors] reduce blood pressure through their vascular smooth muscle effects, they are poorly effective or ineffective on the microcircu- lation. They cause the “steal effect,” a retrograde (reversed) blood flow in an artery due to proximal stenosis and/or occlusion (4). Such an effect causes paradoxical ischemia and explains why vasodilators worsen certain ischemic conditions. Endothelial dysfunction is believed to cause the progression of vascular diseases including coronary and peripheral artery disease, diabetic complications, heart failure, microvascular angina, myocardial infarction, stroke, sickle cell disease, vascular dementia, chronic rheumatic disorders, acute lung injury, acute kidney injury (AKI), and vascular leak and septic shock associated with severe viral infectious diseases (1). There is therefore a need to identify novel agents that can restore endothelial function, particularly in the microcirculation, without causing non- specific vasodilation.

Sphingosine-1 phosphate (S1P) elicits its effects through the activation of five heterotrimeric GTP-binding protein (G protein)– coupled receptors (GPCRs), S1P1–5 (5). S1P1 is the most highly expressed in endothelial cells, in which it has a prominent role (6, 7), and has no functional role on vascular smooth muscle, where vascular reactivity is controlled by the opposing actions of S1P2 and S1P3 (8). S1P1 activation is necessary to maintain vascular barrier function through the formation of adherens and tight junctions induced by Rac activation after Gai coupling (7, 9). Mice that are genetically deficient for S1P-producing enzymes have increased vascular leakage under both basal and pathological settings (10). Endothelial S1P1 is essential for endothelial barrier integrity because genetic deletion of endothelial S1P1 is embryonically lethal due to massive vascular hemorrhage in utero (11). Short-term treatment with S1P or S1P1 agonists prevents vascular leakage in animal models of lung and kidney injury (12, 13), whereas prolonged exposure to S1P1-desensitizing agents decreases survival and causes lung injury, fibrosis, and vascular leakage in models of lung injury (14, 15).

Thus, S1P1 activation is endothelial protective, whereas S1P1 in- activation is endothelial damaging.

The role of S1P1 in lymphocyte homing has been extensively studied. Fingolimod, an immunosuppressive prodrug approved for the treatment of relapsing-remitting multiple sclerosis (MS) (16), becomes phosphorylated in vivo to FTY-720-P, a high-affinity ligand for all S1P receptors except S1P2 (17). Although FTY-720-P is a potent S1P1 agonist, it causes long-lasting S1P1 desensitization that is responsible for inhibiting lymphocyte egress from secondary lym- phoid organs and causing peripheral blood lymphopenia (17, 18). Bradycardia and atrioventricular block are side effects of fingolimod treatment in humans (18). In atrial myocytes, FTY-720-P activates an inwardly activating K+ channel through S1P3 activation (19), suggesting that at least some of the cardiovascular side effects may be attributed to S1P3. Consequently, clinical development activities are dominated by numerous S1P1 functional antagonists that are S1P3 sparing, the most advanced of which, siponimod (an S1P1/5 agonist) (20), has been approved to treat MS. Other clinical side effects of fingolimod, mainly at supratherapeutic doses, include macular edema and respiratory and renal dysfunction (18, 21, 22). Although these side effects are manageable in patients with MS receiving fingolimod, the similarity to endothelial-damaging effects (lung vascular leakage) demonstrated preclinically with S1P1 selec- tive functional antagonists (14, 15) provides human evidence impli- cating S1P1 in endothelial homeostasis.

Although there is considerable interest in developing endothelial- protective agents, the on-target lymphopenic and S1P1-desensitizing effects of all known S1P1 agents developed for MS limit their thera- peutic potential on the endothelial axis. However, the emerging concept of ligand bias at GPCRs (23) broadens the possibility for obtaining varying agonist profiles. In addition to classical G protein signaling, GPCRs can also activate parallel and sometimes distinct signaling pathways. Principal among these is signaling mediated by β-arrestins, which bind to activated receptors to uncouple and desensitize G protein signaling, promote receptor internalization, and sometimes initiate distinct G protein–independent signal trans- duction cascades. A “G protein–biased” ligand may have equivalent or better efficacy in activating a G protein relative to a reference agonist but has lowered efficacy on or can block β-arrestin recruit- ment and activation. A key advantage of G protein–biased ligands is that they escape arrestin-dependent desensitization, and clinical results with a central nervous system µ-opioid agonist (24) raise the possibility that G protein–biased molecules that target other receptors can be clinically useful.

In this study, we introduce an S1P1 agonist, SAR247799, which activates S1P1 signaling events unlike the S1P1 functional antagonists fingolimod and siponimod. The profile of SAR247799 illustrates that biased ligands can overcome the on-target therapeutic limitations of previous S1P1 agonists, providing opportunities to develop thera- peutics for different medical benefits. G protein–biased S1P1 activation by SAR247799 yields a profile preferred for endothelial protection that could be used in multiple settings of tissue injury.

RESULTS

Discovery of a G protein–biased S1P1 agonist, SAR247799

A high-throughput screening and chemical optimization program led to the identification of SAR247799. The activity of SAR247799 was compared to S1P, siponimod and FTY-720-P in S1P1-overexpressing cells using G protein–dependent assays [intracellular Ca2+ flux, guanosine 5′-O-(3′-thiotriphosphate) (GTP-y-S) binding, inhibition of forskolin (FSK)–induced cyclic adenosine 3′,5′-monophosphate (cAMP) production, and cell impedance] and G protein–independent assays [β-arrestin recruitment and internalization (receptor trans- location to endosomes)] (Table 1). When visible compound pre- cipitation was noted with FTY-720-P, the data were considered unreliable (noted in Table 1). SAR247799 was a full S1P1 agonist as it had maximal efficacy (Emax) in all assays that was similar to that of S1P and indistinguishable from those of siponimod and FTY-720-P. The median effective concentration (EC50) for SAR247799 was between 12.6 and 493 nM (Table 1). Siponimod and FTY-720-P showed similar potency to each other and were consistently more potent than SAR247799. The potency differential between siponimod and SAR247799 was less in the Ca2+ flux assay (7.5-fold) compared to other assays (28- to 374-fold); however, as Ca2+ flux responses saturate at low receptor occupancy, this differential may be under- estimated in this assay. Overall, the potency of each compound was influenced by the assay type (Table 1).

Because potency is influenced by each compound’s inherent affinity for protein, measurements of inhibition of cAMP production, β-arrestin recruitment, and receptor internalization were adapted to be run in media with identical protein concentrations (0.5% serum) to allow comparison of the potencies and efficacies of SAR247799, siponimod, and S1P (but not FTY-720P due to its limited solubility). Because Emax values were similar for the three compounds in each assay (Table 1 and Fig. 1), the bias ratios for each compound were calculated as a ratio of the EC50 values in the β-arrestin and internal- ization assays (G protein–independent assays) relative to the EC50 for inhibition of cAMP production (Gi signaling) (Fig. 1, A to C). S1P displayed comparable potency in all three assays, whereas SAR247799 was more potent at inhibiting cAMP production than inducing re- ceptor internalization (19×) or β-arrestin recruitment (39×). SAR247799 was a G protein–biased agonist because, compared to the reference molecule S1P, it preferentially activated the Gi pathway. Siponimod was less biased than SAR247799 because it had similar EC50 values in inhibiting cAMP production and in triggering receptor internal- ization and was 14× less potent at β-arrestin assay.

The selectivity of SAR247799 was evaluated against the S1P re- ceptor family (table S1). SAR247799 was selective for S1P1 because it was inactive (or <50% activity at 10 µM) at S1P2, S1P3, and S1P5, and the EC50 value for S1P4 was >100-fold higher than that for S1P1 in inhibiting cAMP production. SAR247799 was also selective in a broad panel of unrelated targets that spanned more than 100 binding and en- zyme assays (inactive or <60% activity at 10 µM in all assays; table S2). To evaluate the ability of SAR247799, siponimod, FTY-720-P, and S1P (Fig. 2A) to desensitize S1P1, the Ca2+ flux assay in S1P1–Chinese hamster ovary (CHO) cells was performed with a preincubation step (to promote potential desensitization), and the S1P1 response was measured following a wash step. Preincubation with 100 nM S1P (>10-fold than the EC50 value; Table 1) or no preincubation produced similar intracellular Ca2+ flux responses following the wash step (Fig. 2B). Therefore, under our experimental conditions, S1P activated S1P1 without desensitization. Although S1P was almost equally potent at inhibiting cAMP production, inducing β-arrestin recruitment, and triggering receptor internalization (Table 1), it did not desensitize S1P1 presumably because of the presence of ligand-degrading en- zymes that spare the receptor from desensitization (25). Similarly, preincubation with SAR247799 did not cause desensitization, whereas FTY-720-P and siponimod caused significant desensitization that did not recover over wash periods of up to 60 min (Fig. 2B).

Efficacy and receptor desensitization in endothelial cells S1P1 activation causes cytoskeletal and morphological changes, which can be measured by cell impedance changes in a Gi-mediated manner (25). Therefore, to evaluate compound activity in a physiologically relevant system without receptor overexpression, forced coupling, or signal amplification, we measured cell impedance changes in human umbilical vein endothelial cells (HUVECs). SAR247799 and siponimod had similar efficacy (Emax), but siponimod was >100-fold more potent than SAR247799 (Fig. 2C and Table 1). The assay was then used to assess the ability of each compound to desensitize S1P1, considering the concentration of each compound in relation to its inherent potency in the assay. Preincubation of cells with a high S1P concentration (10 µM) did not induce S1P1 desensitization because a full response to S1P was evoked after cell washing (Fig. 2D). Sim- ilarly, preincubation of SAR247799, at concentrations up to 10 µM (40-fold higher than the EC50 value), did not induce statistically sig- nificant S1P1 desensitization. Siponimod caused a concentration- dependent desensitization with statistically significant reductions at 10 nM and above (27-fold higher than the EC50 value).

SAR247799 and siponimod induced a concentration-dependent phosphorylation of extracellular-regulated kinase-1/2 (Erk1/2) and protein kinase B (Akt) (Fig. 3, A and B), and these responses were inhibited with a S1P1 antagonist (Fig. 3, C and D, and tables S3 and S4). The EC50 values of SAR247799 for phosphorylation of Erk1/2 and Akt (72.9 to 118 nM) (Fig. 3, E and F, and Table 1) and for cellular impedance in HUVECs (252 nM) (Table 1) were similar. Siponimod showed similar effects in HUVECs on phosphorylated Erk1/2 (pErk1/2), phosphorylated Akt (pAkt), and impedance but, consistent with S1P1-overexpression assays, displayed lower EC50 values than SAR247799 (Table 1). SAR247799 induced phosphoryl- ation of Ser1177 of endothelial nitric oxide synthase (eNOS) in a S1P1-dependent manner, indicating eNOS activation (Fig. 3G).

In HUVECs, SAR247799, siponimod, and FTY-720-P inhibited FSK-induced intracellular cAMP production, and these effects were S1P1 specific (table S5). SAR247799 had a 1.7-fold higher Emax value for inhibiting FSK-induced cAMP than siponimod or FTY-720-P (Fig. 3H and Table 1). In HUVECs, siponimod and SAR247799 had higher EC50 values in inhibiting cAMP production compared to inducing Erk1/2 or Akt phosphorylation or altering cell impedance (Table 1). Furthermore, the EC50 values of siponimod and SAR247799 for inhibiting cAMP production were 100- and 1000-fold higher, respectively, in HUVECs than in S1P1-CHO cells (Table 1). To assess the potential of the compounds to modulate endothelial in- flammation, tumor necrosis factor–a (TNFa)–induced expression of vascular cell adhesion molecule-1 (VCAM-1), a vascular inflam- mation marker, was measured on HUVECs, and this effect was inhibited by SAR247799 [median inhibitory concentration (IC50) 10 to 85% at 1 and 3 mg/kg) (Fig. 5, A and B, respectively). Consistent with these renal effects, pharmacokinetic analysis showed that SAR247799 exposure (Cmax and AUC) increased with dose (table S8). In com- parison, siponimod was not active at 3 mg/kg, and higher doses (10 and 30 mg/kg) showed only partial (34 to 40%) but sig- nificant reductions in serum creatinine and urea (Fig. 5, C and D). At equivalent doses (10 mg/kg), siponimod had about fivefold lower exposure than SAR247799, and both compounds distributed to the kidney with similar kidney-to-plasma ratios (table S9). Thus, the efficacy of the two compounds was compared over a dose range of siponimod (3 to 30 mg/kg) that gave projected plasma and kidney ex- posures at least as high as for the cor- responding dose range of SAR247799 (0.3 to 3 mg/kg).

SAR247799 significantly protected re- nal proximal tubules against necrosis (Fig. 5, E and F) and significantly blunted the development of interstitial hemor- rhage (Fig. 5, G and H). SAR247799 improved indices of early capillary pres- ervation as evidenced by changes in en- dothelial surface markers: There was a significant decrease in endoglin (CD105) and a trend for an increase in platelet endothelial cell adhesion molecule-1 (PECAM-1) (CD31) expression (Fig. 5, I and J). SAR247799 showed a dose- dependent trend for reducing macro- phage (cells positive for macrosialin; also known as CD68) infiltration into the renal parenchyma (Fig. 5K).

I/R was associated with significant drug, also improved the coronary conductance ratio (Fig. 4D). Pharmaco- kinetic analysis demonstrated that SAR247799 exposure (Cmax and AUC) increased with dose (table S6). Hemodynamic parameters [mean arterial pressure (MAP), heart rate (HR), coronary blood flow (CBF), LVdP/dTmax, and double product] were followed throughout the surgical procedure, from baseline con- ditions until the end of the second reperfusion, to assess the stability of each parameter during the procedure and the effect of different con- centrations of SAR247799 (table S7). There were slight, but not statis- tically significant, decreases in MAP, LVdP/dTmax, and CBF over the surgical period in all groups. However, there was no statistically significant difference between vehicle and any of the SAR247799- treated groups, except for MAP at a single dose/time point of SAR247799 (0.3 mg/kg at the end of the second reperfusion) (table S7).

Impact of S1P1-biased signaling on renal protection in rats When given before occlusion, SAR247799 dose dependently reduced the severity of ischemia/reperfusion (I/R)–induced AKI, as reflected by the inhibition of the increase in serum creatinine, a clinically val- idated biomarker (89 and 96% at 1 and 3 mg/kg), and urea (61 and reductions in Pecam1 and Vcam1 mRNAs and an increase in the mRNA for heat shock 70 kDa protein-1 (Hspa1), all of which were counteracted by SAR247799 treatment (Fig. 5, L to N). I/R was associated with an increase in the antiapoptotic protein HSP70, and this was further increased by SAR247799 treatment (Fig. 5O), consistent with its protective role in ischemic renal in- jury (26). Although Hspa1 mRNA was decreased by SAR247799 treatment (which was opposite to the effect of SAR24779 on HSP70 protein), Hspa1 mRNA positively correlated with serum creatinine, a renal injury marker (fig. S1). This discrepancy could be due to differing time courses for mRNA and protein synthesis and/or a negative-feedback loop. These explanations may also apply to Vcam-1: SAR247799 increased Vcam1 mRNA in the AKI model (Fig. 5M) but inhibited TNFa-induced VCAM-1 expres- sion on HUVECs (Fig. 3I). There was a nonsignificant trend for SAR247799 treatment to increase Pecam1 mRNA and PECAM-1– positive cells (as assessed by histology) (Fig. 5, J and L). SAR247799 also prevented vascular leakage, as indicated by the prevention of albumin extravasation into renal tissue (Fig. 5O and fig. S2, A and B).

S1P1 desensitization and lymphocyte reduction in pigs

The S1P1-desensitizing properties of SAR247799 were assessed by measuring lymphocyte reduction in pigs. SAR247799 had minimal to no effect on peripheral blood lymphocytes at 1 and 3 mg/kg (Fig. 6A), the two highest doses tested that were active in the cor- onary microvascular hyperemia study (Fig. 4D). At higher doses (10 and 30 mg/kg), SAR247799 caused a dose-dependent and time-dependent reductions (Fig. 6A). The maximum reductions compared to baseline in blood lymphocytes at 1, 3, 10, and 30 mg/kg were 9, 14, 44, and 74%, respectively (with a nadir between 1.5 and 5 hours), which was associated with corresponding increases in compound exposure (table S3). Blood lymphocytes were not statis- tically significantly reduced from baseline values at 24 hours (and later time points) except at 30 mg/kg, which resulted in a 33% reduction at 24 hours but no statistically significant difference at 48 hours (and later). The half-life of SAR247799 was 5.6 to 7.7 hours, depend- ing on the dose (table S3), and was consistent with the time course of lymphocyte recovery. Fingolimod (0.3 mg/kg) caused a maximum reduction of blood lymphocytes of 79% (at 8 hours), and the reduc- tion persisted at similar levels at 24 and 72 hours, similar to that observed in humans (18).

Lymphocyte-reducing properties in rats

To assess the relationship between renal efficacy and S1P1 desensi- tization, the effect of SAR247799 and siponimod on lymphocytes was evaluated in rats. The profile of blood lymphocytes over 24 hours was similar between vehicle-treated rats and rats treated with SAR247799 (1 or 3 mg/kg) (Fig. 6B), doses that provided significant efficacy in the AKI model (Fig. 5, A and B). Higher doses of SAR247799 (10 and 30 mg/kg) caused a dose-dependent reduction in lymphocytes (at 2 and 6 hours), which returned to vehicle-treated values at 24 hours. Siponimod (1 and 10 mg/kg) caused a marked lymphocyte reduction at 2 and 6 hours that was similar to SAR247799 (30 mg/kg) (~70% reduction; Fig. 6B). However, unlike SAR247799, lymphocytes remained depressed with both doses of siponimod at 24 hours. Thus, the full dose range of siponimod tested in the AKI model (Fig. 5, C and D), whether inactive (3 mg/kg) or partially active (10 and 30 mg/kg), was characterized by marked and sustained S1P1 desensitization.

S1P1 desensitization following repeated administration in rats

The ability of SAR247799, siponimod, and fingolimod to desensitize S1P1 was further assessed after a 7-day oral twice a day administration in rats by assessing both lymphocyte reduction (Fig. 6C) and lung vascular leakage (Fig. 6D). We measured total lymphocytes because the effect of SAR247799 on total lymphocytes paralleled the effect on lymphocyte subpopulations (table S10). Fingolimod reduced blood lymphocytes by 70 to 80% at the three doses evaluated (0.03, 0.3, and 3 mg/kg), but only the higher two doses caused substantial lung vascular leakage. Siponimod at 0.03 and 0.1 mg/kg produced no or minimal lymphocyte reduction, whereas higher doses (0.3, 3, and 30 mg/kg) produced 70 to 80% lymphocyte reduction. The lower doses of siponimod (0.03 and 0.1 mg/kg) produced minimal changes in lung vascular leakage, whereas the higher doses (3 and 30 mg/kg) produced a dose-dependent effect on leakage. SAR247799 at 1 and 3 mg/kg did not reduce blood lymphocytes after 7 days of twice per day administration, similar to the effects of administering a single dose. Higher doses showed a dose-dependent reduction (30% at 10 mg/kg and 50 to 60% at 30 and 100 mg/kg). The three lowest doses of SAR247799 (1, 3, and 10 mg/kg) did not show statistically significant changes in vascular leakage compared to vehicle, and there was a dose-dependent effect at higher doses (30 and 100 mg/kg). Overall, all three compounds showed dose-dependent effects on both lym- phocytes and lung vascular leakage, with lymphocyte reduction being a slightly more sensitive parameter that was detected at lower doses.

DISCUSSION

The biased signaling profile of SAR247799 makes it a functionally selective molecule with therapeutic utility in the setting of endothelial protection, which differs from indications in which S1P1 desensiti- zation and the resulting immunosuppression are desired. Although S1P1 desensitization on lymphocytes has been successfully exploited for the development of lymphopenic S1P1 agonists for MS (fingolimod, siponimod, and ozanimod), retaining intact S1P1 receptors on the surface of endothelial cells is required for realizing the endothelial protective effects mediated through receptor activation. We addressed this challenge by identifying a molecule that can activate S1P1 while escaping β-arrestin–dependent desensitization. A phenotype of β-arrestin-2 knockout mice is the prolongation of opiate-induced analgesia, demonstrating a critical role of β-arrestin-2 in the desen- sitization of opioid receptors (27). These animals also show gastro- intestinal and respiratory side effects to morphine treatment (28), suggesting that circumventing opioid receptor desensitization may lead to longer-lasting and safer analgesics. Subsequently, TRV130, a G protein–biased µ-opioid receptor agonist that demonstrates G protein signaling potency and efficacy similar to morphine but with reduced receptor sequestration, produces comparable analgesia with less gastrointestinal dysfunction and respiratory depression than equianalgesic doses of morphine in mice (29). These preclinical results have been translated to humans, for which TRV130 (trade name: Oliceridine) shows improved analgesia and/or reduced side effects compared to morphine (24). Prompted by the functional selectivity of a G protein–biased agonist in patients, we hypothesized that S1P1 ligands that circumvent receptor desensitization not only would have prolonged endothelial protective effects but also would avoid the undesirable lymphopenia and endothelial-damaging effects characteristic of S1P1 down-regulation. Because multiple clinical S1P1 functional antagonists do not differ in terms of down- stream signaling or in S1P1 down-regulation and degradation (30), we chose for comparison two S1P1 functional antagonists, repre- sentative of the class, that are approved drugs for MS treatment. FTY-720-P, the active metabolite of fingolimod, was chosen as the prototype S1P1 functional antagonist and has demonstrated endothelial-damaging properties in humans, animals, and cells. Siponimod, in contrast to FTY-720-P, is highly selective for S1P1/5. SAR247799 was a G protein–biased agonist with a greater bias ratio compared to S1P or siponimod, which served as reference molecules to confirm that our findings were not an artefact of the varying sensitivities of assay formats. The biased signaling profile of SAR247799 in S1P1-overexpressing cells—inhibition of cAMP production compared to β-arrestin recruitment or receptor internalization—translated to a clear concentration differential between desirable effects (phosphorylation of Erk1/2 and Akt and cellular impedance changes) compared to undesirable desensitiza- tion in endothelial cells. Furthermore, the results in HUVECs and S1P1-overexpressing cells provided independent lines of evidence that SAR247799 was distinct from siponimod with respect to sig- naling and receptor desensitization. Paradoxically, all S1P1 agonists were considerably less potent in HUVECs at inhibiting cAMP pro- duction compared to other end points (phosphorylation of Erk1/2 and Akt and cellular impedance). Potencies for the inhibition of cAMP production also differed considerably between HUVECs and S1P1-CHO cells, suggesting that the coupling of cAMP production to G proteins may differ between cell types. Although inhibition of cAMP production is usually considered to be Gi-mediated, β-arrestins can also lower cAMP concentrations by recruiting phosphodiesterases (31). SAR247799 had an EC50 in HUVECs for inhibiting cAMP production (15.6 µM) that was in the same range as its IC50 for inhibiting TNFa-induced VCAM-1 expression (10 to 30 µM). Fur- thermore, SAR247799 had a higher Emax value than siponimod in HUVECs for inhibiting cAMP production, and SAR247799, but not siponimod, inhibited TNFa-induced VCAM-1 expression (Fig. 3, H and I). These observations suggest that the two activities in HUVECs may be mediated through a common pathway. The mechanism for the compound differences potentially involves β-arrestin signaling, a proposal consistent with the inhibition of TNFa-induced nuclear factor B phosphorylation in HUVECs by high-density lipoprotein (HDL)–bound S1P in a β-arrestin–dependent manner (32).

The endothelial protective effects of SAR247799 were confirmed in pigs using the model of coronary microvascular endothelial dys- function. SAR247799 showed comparable efficacy to L-arginine (Fig. 4D), a molecule reported to improve coronary flow reserve and angina symptoms in patients with microvascular angina (33, 34). SAR247799 caused these microvascular improvements without altering HR, MAP, CBF, and indices of ventricular performance and oxygen consumption (table S7). The absence of blood pressure lowering with SAR247799 is consistent with a lack of functional S1P1 on vascular smooth muscle and S1P2 and S1P3 (for which SAR247799 is inactive) being the primary S1P receptors responsible for blood pressure control (8). Such a profile of an endothelial-protective molecule without vascular smooth muscle effects distinguishes the mechanism of action of SAR247799 from previous vascular agents studied in patients—including ACE inhibitors, nitrates, phosphodiesterase-5 inhibitors, and guanylate cyclase stimulators— and may provide the opportunity for a more specific effect on the microvasculature.

Endothelial damage has been observed in fingolimod-treated patients, in whom macular edema occurs at the approved 0.5-mg dose (which is S1P1 densitizing and results in 70 to 80% lymphocyte reduction), and its incidence increases with higher doses (21). Macular edema is also a side effect in patients treated with siponimod (20). Similarly, dose-dependent pulmonary toxicity has been reported with fingolimod (and ponesimod) at supratherapeutic doses, which manifests as reductions of up to 30% in forced expiratory volume over 1 s, accompanied with dose-limiting chest discomfort and dyspnea (18, 35, 36). The mechanism for these respiratory alterations is consistent with endothelial-damaging effects seen in our rat lung vascular leakage study (Fig. 6D) and by others (14). The three com- pounds tested in the current rat study showed a dose-dependent lung endothelial damage at doses higher than required for lymphocyte reduction. This finding is consistent with the dose-dependent side effects of the aforementioned clinical molecules, mainly at supra- therapeutic doses. Overall, from our preclinical studies and from clinical observations, lymphocyte reduction appears to be a more sensitive marker of S1P1 desensitization than endothelial damage and thus may act as an early safety signal.
Similarly, SAR247799 demonstrated marked renal-protective properties in rats at doses lower than higher S1P1-desensitizing doses. In contrast, siponimod not only showed weaker renal-protective properties (40% reduction in creatinine at 10 and 30 mg/kg and no effect at 3 mg/kg), but the dose-response characterization for lym- phocyte reduction and lung vascular leakage revealed statistically significant effects at even lower doses (1 and 0.3 mg/kg for lympho- cyte reduction after single and repeated dosing, respectively, and 3 mg/kg for lung vascular leakage). Potency and pharmacokinetics could not explain the differences between compound efficacy be- cause the dose range of siponimod produced at least equivalent ex- posure to the dose range of SAR247799, and siponimod was 10- to 100-fold more potent than SAR247799 in most in vitro assays. Al- though selective S1P1 agonists show renal and cerebral protective effects in rodent I/R models, those effects require lymphocyte- reducing doses, similarly to siponimod (37, 38). Lymphocytes ap- pear to play a key role in AKI because lymphocyte-deficient mice are partially protected from the development of AKI (39). Both lymphocyte and nonlymphocyte mechanisms could be responsible for the tissue protective effects of previous S1P1 agonists. S1P1 agonists further protect the kidney in Rag-1 knockout mice, which lack T and B lymphocytes (39, 40). Because the protective effects of SAR247799 were independent of lymphocytes, we propose that the protective effects of previously characterized S1P1 agonists at S1P1-desensitizing doses reflect a lack of G protein–biased receptor signaling (and hence overlapping activation versus desensitization- related effects), rather than lymphocyte pharmacology being a driver of I/R activity. Our proposal is consistent with an engineered S1P chaperone protein that can activate S1P1, S1P2, and S1P3 and that demonstrates protective effects in cardiac and brain I/R models without lymphocyte reduction (41).

Overall, SAR247799 had endothelial protective properties in two distinct animal models with different vascular beds and, in both cases, at doses devoid of lymphocyte reduction. The dose responses of SAR247799 for endothelial protective effects were consistently lower than the dose responses for S1P1-desensitizing effects (lym- phocyte reduction in both species and lung vascular leakage in rats). The animal results were consistent with differential concentration- dependent effects on signaling and desensitization properties in vitro and illustrated a consistent differentiation of SAR247799 versus siponimod. SAR247799 also differed from siponimod with respect to its absolute efficacy on renal function and for inhibition of TNFa-induced VCAM-1 expression on endothelial cells.

S1P1 activation causes HR reduction, and all clinically tested S1P1 functional antagonists display first-dose bradycardia in humans, an effect that is rapidly desensitized. Because SAR247799 was de- signed to be nondesensitizing, it might display sustained S1P1 acti- vation not only at the endothelium but also in the heart causing sustained bradycardia. However, SAR247799 is an acidic molecule (Fig. 2A) and had a low volume of distribution (tables S6 and S8), and concentrations in heart tissue were 20- and 4-fold lower than in kidney tissue or plasma, respectively (table S9). SAR247799 did not reduce HR in pigs (table S7). Whether the concentrations of SAR247799 in heart tissue are sufficiently low to provide a therapeutic index for protective effects versus bradycardia and atrioventricular block will require evaluation in human trials.

The current studies have several limitations. First, the pharma- cological effects on renal function in the rat AKI model may not be solely attributed to endothelial effects because S1P1 activation can also protect renal proximal tubule cells (42). However, we demon- strated the effect of SAR247799 on several vascular parameters (albumin leakage, PECAM-1, and endoglin immunostaining), which were consistent with endothelial-specific effects in pig coronary microvasculature and on endothelial cells. Furthermore, SAR247799 protected the kidney from interstitial hemorrhage, consistent with endothelial-specific S1P1 deletion causing massive vascular hemor- rhage leading to embryonic lethality (11). Inducible deletion of en- dothelial S1P1 in mouse renal I/R models causes marked increases in serum creatinine, vascular inflammation, and renal necrosis (43, 44). The protection conferred by SAR247799 against these effects lends support for an endothelial-mediated effect. Second, the protective properties of SAR247799 in vivo were demonstrated in studies of up to 1-day treatment duration. This was a limitation of the experimental systems chosen and the 3-hour half-life of SAR247799 in rats. The short half-life in rats (although not considered representative of human) presents a technical challenge for a biphasic molecule to achieve steady S1P1-activating concentrations without large peak- to-trough variations that would promote S1P1 desensitization. Never- theless, when pushing the dosing regimen toward S1P1 desensitization (7-day b.i.d. administration), doses demonstrating undesirable effects (lung vascular leakage and lymphocyte reduction) remained 10- to 30-fold higher than renal protective doses for SAR247799 (but not siponimod). This compound-specific differentiation was consistent with our endothelial desensitization studies. However, the effects of long-term SAR247799 treatment on the endothelium need to be evaluated, including in patients.

We showed that high target potency and selectivity are not the only criteria that determine the therapeutic utility of GPCR agonists. SAR247799 is a new approach to S1P1 activation, namely, selective signaling through ligand bias. Although many aspects of S1P1 pharma- cology have been replicated between human and preclinical systems, including lymphopenia, MS efficacy, bradycardia, and endothelial damage, the role of S1P1 in endothelial protection remains to be tested in patients. The endothelial-damaging side effects of fingolimod and siponimod in patients were replicated in the rat lung, and SAR247799 produced these effects only at supra–endothelial protec- tive doses. SAR247799 now provides a unique opportunity to trans- late these findings to patients using a molecule exhibiting an opposite pharmacological profile to fingolimod and siponimod of sustaining S1P1 activation. SAR247799 has features that enable the clinical utility of targeting S1P1 in disorders associated with endothelial dysfunc- tion to be assessed without compromising immune responses.

MATERIALS AND METHODS

Compounds

SAR247799 (4-[5-(3-chloro-phenoxy)-oxazolo [5,4-d]pyrimidin-2-yl]- 2,6-dimethyl-phenoxy}-acetic acid), siponimod, and the S1P1 antagonist were synthesized at Sanofi. Fingolimod and FTY-720-P were purchased from Advanced Technologies and Toronto Research Chemical, respectively. For in vitro assays, compounds were solubilized in a dimethylsulfoxide (DMSO) stock solution, unless otherwise indicated. Siponimod doses for in vivo studies are expressed as milligrams of free base.

Recombinant human S1P1 assays

cAMP, β-arrestin, and total internalization assays were performed (at Sanofi or DiscoverX laboratories) using PathHunter cells (DiscoverX, catalog nos. 95-0142C2, 93-0207C2, and 93-0784C1) using medium with 0.5% fetal bovine serum and all other condi- tions as defined by the manufacturer. For cAMP studies, adenylate cyclase was stimulated with 15 µM FSK, and test compounds were evaluated for cAMP inhibition. FTY-720-P was solubilized in 10% chloroform/90% ethanol to improve solubility.

Calcium flux studies were performed in CHO cells stably over- expressing S1P1 fused to a C-terminal sequence of a modified G protein (Gai4qi4) (Flp-In System, Invitrogen). Cells (40,000 per well) were loaded with Fluo-4 for 60 min and washed, and the fluorescence signal was monitored over 3 min after stimulation with test compounds or vehicle control. In desensitization studies, a first incubation of test com- pound or control was followed by a wash step of defined time intervals (Fig. 2B), and the ability to mount a subsequent calcium response to 100 nM S1P was evaluated.

GTP-y-S assays were performed by preincubating 5 mg of S1P1- CHO cell membrane fractions with GTP-binding buffer [50 mM Hepes buffer (pH 7.5) containing 100 mM NaCl, 5 mM MgCl2, 20 mM guanosine diphosphate, 0.1% bovine serum albumin (BSA), and saponin (20 mg/ml)] containing 10 µM S1P (for maximum re- sponse) or various concentrations of test compound for 30 min at room temperature. The resultant aliquots were further incubated with [35S]GTP-y-S (1200 Ci/mmol; 0.1 nM) for 30 min, and the reaction was terminated by rapid filtration under vacuum. The filter-bound activity was counted with a TopCount instrument (Packard Instruments). Nonspecific binding was determined in the presence of 30 µM GTP-y-S.

Cellular impedance was measured continuously with RTCA SP stations (xCELLigence) essentially as described (25), except for seeding of 3 × 104 S1P1-CHO cells per well in 96-well collagen-coated E-plates and using the maximum response following compound addition (8 min) for EC50 determinations.

Endothelial assays

HUVECs from pooled donors were used (Lonza, CC-2517; Cascade Biologics, C-015-5C; and Promocell, C-12203). For phosphorylation studies, 0.3 × 106 cells per well in six-well plates were serum-starved for 24 hours before 10-min compound treatment. Western blots were performed on cell lysates using the following antibodies from Cell Signalling Technology: anti–phospho-ser473-Akt (pAkt; no. 2965) and anti–phospho-Tyr202/Tyr204-Erk1/2 (pErk1/2; no. 4377), anti– phospho-eNOS-Ser1177 (no. 9570), and anti–a-tubulin (no. 2144). The intensity of phosphorylated bands was normalized to the re- spective total proteins.

Cell surface VCAM-1 was measured in HUVECs (25 × 103 cells per well in 96-well plate) preincubated with test compounds for 3 hours and then stimulated with TNFa (3 ng/ml) for 6 hours. Cells were washed, fixed with RCL2 (Alphelys) and used for enzyme-linked immunosorbent assay (anti–VCAM-1, no. BBA5, R&D Systems). Cell surface VCAM-1 induction was expressed as fold increase over baseline (no TNFa stimulation).

For cAMP measurements, 5 × 103 cells per well were seeded in 96-well plates, grown overnight, and washed in assay buffer [Hanks’ balanced salt solution containing 10 mM Hepes, 0.1% BSA, and 0.5 mM 3-isobutyl-1-methylxanthine] for 15 min, and test compounds were added for 15 min, followed by stimulation with 10 µM FSK for 30 min. cAMP was measured using a cAMP HTRF kit (Cisbio).

Impedance was measured as described for S1P1-CHO cells but using 2 × 104 HUVECs per well. In desensitization studies, after 1 hour incubation with test compounds, measurements were paused, cells were washed twice and allowed to recover for 5.5 hours, and impedance measurements continued for a further 1 hour after re- stimulation with 0.1, 1, or 10 µM S1P. Impedance was expressed in desensitization studies as AUC post-S1P stimulation and normalized to the response of 1 µM S1P (with DMSO preincubation).

Animal studies

All animal studies were performed in accordance with the European Community standard on the care and use of laboratory animals and approved by the Institutional Animal Care and Use Committee of Sanofi R&D. Male animals were used throughout, and the number per group is denoted in the figures and/or legends. Compounds were administered to pigs by 30-min intravenous infusion in 0.9% NaCl and to rats by oral gavage in 0.6% methylcellulose/0.5% Tween 80.

Coronary postocclusive hyperemia in pigs

Anesthetized 10- to 12-week-old farm pigs were prepared by plac- ing a pressure transducer (SPR350, Millar) in the left ventricle for ventricular pressure measurement, an arterial catheter connected to a pressure transducer (P23XL-1, Becton-Dickinson) in the abdom- inal aorta for arterial blood pressure measurement, and a catheter in the right saphenous vein for intravenous infusion of test article. A Doppler flow probe (Transonic) was placed around the left anterior descending coronary artery, and a ligation was loosely placed above the flow sensor for subsequent repeated occlusions.
Animals were randomly assigned to the treatment groups, which differed only with respect to the test article or vehicle administered. Treatment was immediately followed by an 8-min occlusion and a 30-min reperfusion to induce endothelial dysfunction. Endothelial dysfunction was measured by repeating the same sequence of occlusion/reperfusion and measuring the change in coronary con- ductance parameters during the second reperfusion and comparing it relative to the first reperfusion (Fig. 4). Hemodynamic parameters (HR, CBF, MAP, double product, and LVdP/dtmax) were monitored to establish their stability during the surgical procedure and any difference between treatment groups (table S7).

Pig lymphocyte assessment

A separate cohort of 8- to 10-week-old farm pig was used to charac- terize the time response of 30-min intravenous infusions of SAR247799 and fingolimod on peripheral blood lymphocytes. Pharmacokinetic parameters were assessed in SAR247799-treated animals.

Rat AKI

Twelve- to 15-week-old Fischer rats were subjected to a sham surgery or to a 25-min bilateral renal occlusion under pentobarbital (50 mg/kg, i.p.) anesthesia, followed by 24-hour reperfusion. Treat- ment groups differed with respect to random assignment to test article or vehicle, administered 1 hour before renal occlusion. Plasma samples were analyzed for creatinine and urea and kidney tissue for histology, Western blot, and quantitative polymerase chain reaction (PCR). Cortex necrosis was expressed as percentage of tubules that displayed cell necrosis in 12 to 15 fields in the corticomedullary region. CD68, CD105, and CD31 immune-labeling (antibodies: no. BM4000, Acris; no. BAF1320, R&D Systems; and no. SC1506, Santa Cruz Biotechnology) were expressed as percent positively stained pixels (Aperio algorithm) counted on whole sections. Interstitial hemorrhage was percentage of animals with hemorrhage. Albumin and HSP70 (antibodies: nos. SC50536 and SC32239, Santa Cruz Biotechnology) were evaluated qualitatively by Western blotting in renal medulla tissue. mRNA for Vcam1, Pecam1, and Hspa1 were measured by quantitative PCR in renal medulla tissue. Each gene was normalized to the housekeeping gene Rplp1 (CT). TaqMan probes (Applied Biosystems) were Rplp1 (Rn03467157_gH), Pecam1 (Rn01467262_m1), Vcam1 (Rn00563627_m1), and Hspa1 (Rn01525688_s1).

Rat lung vascular leakage and lymphocyte assessment

Eight- to 12-week-old Fischer rats were administered with test article or vehicle twice a day for 7 days, followed by a final treatment on day 8. One hour after the final treatment, Evans Blue dye (30 mg/kg) was injected into the jugular vein, and animals were euthanized 15 min later. Dye was extracted from the left lung using formamide (4 ml per gram of tissue) and quantified at 630 nm. Lymphocytes were assessed by collecting blood before Evans Blue injection. The effect of a single compound administration on lymphocytes in Fischer rats was assessed in a separate study by measuring blood lympho- cytes at baseline (day before treatment) and then subsequently at 2, 6, and 24 hours.

EC50 calculations

Data from each experiment were normalized to the maximal and minimal response observed in the presence of a reference compound (S1P for all assays, except FTY-720-P for internalization and SAR247799 for impedance) and vehicle, respectively, except in the HUVEC cAMP assay (where expressed as percentage inhibition of FSK-induced cAMP) and in phosphorylation assays (where expressed as fold increase over vehicle control).

Data from at least three separate experiments were amalgamated for curve fitting. Effective concentration corresponding to half of the difference between the maximum and minimum effect (EC50) of agonists were determined with Statistical Analysis System procedure NLIN in SAS system release 9.1 under Unix via Biosta@t-SPEED-LTS v2.0 internal software using the four-parameter logistic model.

Statistical analysis

All statistical tests were two sided, are indicated in figure legends, and were performed using SAS version 9.2 for Windows 7. P < 0.05 was considered significant.