Protease-activated receptor-1 signaling by activated protein C in cytokine-perturbed endothelial cells is distinct from thrombin signaling.

Activated protein C (APC) has anti-inflammatory and vascular protective effects independent of anticoagulation. We previously identified the prototypical thrombin receptor, protease-activated receptor-1 (PAR1), as part of a novel APC-endothelial cell protein C receptor (EPCR) signaling pathway in endothelial cells. Experiments in wild-type and PAR1(-/-) mice demonstrated that intravenous injection of APC leads to PAR1-dependent gene induction in the lung. The vascular endothelium undergoes profound changes in severe sepsis, the approved therapeutic indication for APC. Similar to PAR1, APC activated PAR2 through canonical cleavage. Although PAR2 was up-regulated in cytokine-stimulated endothelial cells, APC signaling remained PAR1-dependent. Large scale gene expression profiling documented marked differences in both up- and down-regulated genes between APC and thrombin signaling in cytokine-stimulated cells. APC down-regulated transcripts for proapoptotic proteins including p53 and thrombospondin-1, but p53 was unchanged, and thrombospondin was even up-regulated by thrombin. Concordant PAR1-dependent effects on protein levels were found. Thus, by signaling through the same receptor PAR1, APC, and thrombin can exert distinct biological effects in perturbed endothelium. These data may explain how APC can be therapeutically protective through the EPCR-PAR1 signaling despite ongoing thrombin generation due to disseminated intravascular coagulopathy.

The serine protease thrombin is not only the central procoagulant enzyme in the blood-clotting cascade, but it also activates the anticoagulant protein C pathway. Two receptor proteins on the endothelial cell surface are involved in protein C activation, thrombomodulin and endothelial cell protein C receptor (EPCR). 1 Recruitment of protein C to EPCR enhances its activation by the thrombin-thrombomodulin complex (1,2). Activated protein C (APC) down-regulates thrombin generation in a negative feedback loop. Results from animal models and clinical trials indicate that APC has potent protective effects in systemic inflammation, which are independent from its well established anticoagulant function. Recombinant APC is approved to treat patients with sepsis, but reduction in mortality is most pronounced in severe sepsis (3). The molecular basis for the anti-inflammatory effects of APC is incompletely understood. We have demonstrated recently that APC can induce protective genes in endothelial cells by activating the prototypical thrombin receptor protease-activated receptor-1 (PAR1) dependent on binding to EPCR. Subsequent studies have implicated this APC-EPCR-PAR1 signaling cascade in anti-apoptotic APC signaling in a transformed endothelial cell line (4) and in neuroprotective effects of APC in vivo (5,6). Anti-apoptotic and anticoagulant activities can be dissociated by specific mutations (7) with the potential to apply APC as a specific signaling protease.
The coagulation protease pathway is up-regulated in systemic and local inflammation and contributes to the inflammatory process in complex ways. For example, thrombin-PAR1 signaling has been shown to have proinflammatory effects in a murine model of glomerulonephritis (8), and APC therapy is particularly efficient in severe sepsis where thrombin is formed intravascularly. These findings raise important questions. Is APC signaling contributing to the protective effects in sepsis? If APC signals through PAR1, can the same receptor mediate the proinflammatory effects of thrombin and the protective effects of APC? APC selectively activates PAR1 in quiescent human umbilical vein endothelial cells (HUVECs), but in heterologous expression systems PAR2 supports signaling by APC-EPCR as well (9). Here we show that although PAR2 is up-regulated under inflammatory conditions in endothelial cells, APC-mediated regulation of gene induction remained PAR1-specific. However, PAR1-dependent signaling by thrombin and APC-EPCR produced distinct gene expression profiles in cytokineperturbed cells, and in the regulation of apoptotic genes, APC and thrombin even produced opposite biological effects.
Reporter Gene Assay and Mutagenesis-The spontaneously transformed M6-11 fibroblast cell line from PAR-1-deficient mice was kindly provided by Dr. Patricia Andrade-Gordon. Cells were grown in low glucose Dulbecco's modified Eagle's medium, 2 mM L-glutamine, and 10% fetal calf serum. Cells were transiently transfected using the GenePORTER reagent (Gene Therapy Systems, San Diego, CA) with expression constructs for human PAR2 and EPCR as well as a reporter construct containing 1.2 kb of mouse egr-1 5Ј flanking sequence upstream of the luciferase reporter gene in the promoterless pXP2 as described (9,10,14). 24 h after transfection, cells were serum-starved for 16 h followed by agonist stimulation and a 5-h incubation to allow for the expression of the luciferase protein. Luciferase activity in the cell lysate was determined using the luciferase assay system (Promega, Madison, WI) and a Monolight 2010 luminometer (Analytical Luminescence, San Diego, CA). A mutant PAR2 construct with substitution of Arg 36 with Glu at the P1 position was generated using oligonucleotidedirected mutagenesis (15).
Flow Cytometry and Gene Expression in HUVECs-Primary HU-VECs (Clonetics, Walkersville, MD) were maintained at 37°C in a 5% CO 2 incubator in EGM-2 medium (basal endothelial cell medium supplemented with hydrocortisone, bovine brain extract, human epidermal growth factor, gentamicin/amphotericin, and 10% fetal bovine serum). They were passaged for no more than seven generations. For flow cytometry cells were detached with enzyme-free cell dissociation buffer (Invitrogen) and kept on ice for all subsequent steps. Cells were pelleted in serum-free EGM-2, washed, and resuspended in staining buffer (phosphate-buffered saline, 0.2% bovine serum albumin, pH 7.2). Cells were stained with a 1:300 dilution of the indicated monoclonal antibodies for 30 min followed by washing and a 30-min incubation with a 1:50 dilution of fluorescein isothiocyanate-labeled goat F(abЈ) 2 anti-mouse IgG (Southern Biotechnology Associates, Birmingham, AL). After another washing step, cells were analyzed on a FACScan flow cytometer (BD Biosciences). For TaqMan (Applied Biosystems) real-time reverse transcription-PCR 4 g of total cellular RNA was reverse transcribed followed by RNase treatment using the superscript kit (Invitrogen). TaqMan primers and probes for GAPDH and TR3 were as described (9).
Gene Expression in Murine Lung-PAR1 Ϫ/Ϫ (16) and PAR2 Ϫ/Ϫ (17) mice were provided by Dr. P. Andrade-Gordon, and these mice were backcrossed to Ͼ90% homogeneity with C57Bl/6 mice. Animals used for experiments were between 8 and 12 weeks of age. All studies in mice were approved by The Scripps Research Institute Animal Care and Use Committee and comply with National Institutes of Health guidelines. Carrier control or human APC (2 mg/kg in 100 l) were injected into the retro-orbital sinus in groups of anesthetized wild-type, PAR1-deficient, or PAR2-deficient C57/Bl6 mice. Preparations of activated protein C contain traces of thrombin because thrombin is the only known activator of protein C. Hirudin (100 nM) was added to the injected APC to block contaminating thrombin. The amount of hirudin injected per mouse (70 ng) is not expected to have biological effects because it is more than 100 times lower than a dose that leads to prolongation of the activated partial thromboplastin time in the mouse (18). Hirudin was also included at the same concentration in all control injections. Three h after the bolus injection the mice were sacrificed, and the lower lobe of the left lung was rapidly surgically dissected. Total RNA was extracted in TRIzol reagent using a tissue homogenizer; reverse transcription and real-time PCR were performed as described above. TaqMan primers and probes for murine monocyte chemoattractant protein-1 (MCP-1) and GAPDH were custom-designed.
Microarray Analysis-Confluent HUVEC-derived HUV-EC-C (ATCC CRL 1730) were serum-deprived for 5 h in the presence of 2 nM tumor necrosis factor-␣ (TNF␣) followed by incubation for 90 min with control, 10 nM APC, 10 M PAR1 agonist TFLLRNPNDK (sP1), or 10 nM thrombin (sIIa). Total RNA was isolated and analyzed by hybridization to the HG-U95Av2 array (Affymetrix, Santa Clara, CA). The array contains 12,625 probe sets representing ϳ7000 different human genes. Every gene on the array is represented by a probe set of 16 pairs of oligonucleotide spots, one perfectly matched with the target sequence and the other one with a single mismatch. Based upon the relative hybridization signal intensities for the perfect match versus mismatch pairs, a detection p value (between 0 and 1) and a signal value is calculated for each probe set on the array. A low detection p value indicates that the gene is reliably detected, e.g. p Ͻ 0.05 is the default cut-off for a "present" call. In the comparison analysis (agonist-induced versus buffer control arrays) a change p value (between 0 and 1) and a signal log ratio are calculated from the differences between the perfect match and mismatch intensities comparing the individual pairs in each probe set from the two arrays. A change p value of Ͻ0.05 or Ͼ0.95 indicates genes that are likely expressed at increased or decreased levels in the agonist-treated sample, respectively. The log scale used is base 2; thus a signal log ratio of 1 or Ϫ1 indicates a 2-fold increase or decrease in signal intensity, respectively. Three independent experiments analyzing the effect of stimulation with APC and two experiments with the PAR1 agonist and thrombin analyzing their effect on gene expression levels were performed. Genes were selected based on average change p value and average signal log ratio from the repeat experiments.
Western Blotting for Change in p53 and Thrombospondin Protein Levels-Expression of p53 and thrombospondin-1 were analyzed by Western blotting using anti-p53 DO-7 (Novocastra, Newcastle, UK) and anti-thrombospondin-1 N-20 (sc-12312, Santa Cruz Biotechnology). Staining with anti-actin (Sigma) served as a loading control, and expression levels were quantified using laser densitometry as described (9).
Statistical Analysis-A two-sample two-tailed homoscedastic t test was used to calculate indicated p values.

RESULTS
Canonical Cleavage Is Required for PAR2 Activation by APC-PARs are typically activated by proteolytic cleavage of a specific scissile bond in the N-terminal extracellular domain followed by binding of the newly generated N terminus ("tethered ligand") to a ligand binding site on the receptor body (19). It has been shown previously that APC cleaves the canonical scissile bond in PAR1 (20). To establish that APC activates PAR2 by canonical cleavage we generated a variant of PAR2 with an Arg 36 3 Glu substitution at the P1 position of the scissile bond (PAR2/P1) to block proteolysis by proteases with trypsin-like specificity, including APC and factor Xa. Wild-type PAR2 or PAR2/P1 was co-expressed with EPCR in PAR-deficient fibroblasts. Both human and mouse APC as well as factor Xa led to reporter gene induction in the wild-type PAR2-but not the PAR2/P1-transfected cells (Fig. 1). Cleavage-independent signaling induced by PAR2 agonist peptide confirmed expression of functional receptors. These results demonstrate that APC activates PAR2 through canonical scissile bond cleavage and that PAR2 is also a potential signaling receptor for murine APC.
PAR1 Mediates APC Signaling in the Mouse-To establish an in vivo role for PAR1 or PAR2 in mediating APC signaling, gene expression was analyzed in the lung after intravenous injection of APC in wild-type, PAR1-, or PAR2-deficient mice. As shown in Fig. 2, APC injection up-regulated expression of the transcript for MCP-1 significantly less in PAR1-deficient mice compared with the wild type. The difference between PAR2-deficient mice and wild type was not statistically significant. Because delivery through the retro-orbital sinus likely leads to first passage delivery to the endothelium of the lung, these results are consistent with the in vitro studies, demonstrating that APC in unperturbed endothelial cells specifically activates the prototypical thrombin receptor PAR1.
Enhanced PAR2 Signaling in TNF␣-induced HUVECs-Because endothelial cell PAR2 expression is known to be strongly up-regulated by stimulation with inflammatory cytokines (21), the signaling specificity of APC may change in TNF␣-induced HUVECs. We analyzed whether PAR2 can support APC signaling in cytokine-perturbed cells. Flow cytometry demonstrated that treatment of HUVECs for 5 h with TNF␣ leads to increased PAR2 expression (Table I), whereas the PAR1 level was unchanged. EPCR expression was reduced to 60% of unstimulated control cells, consistent with previous results (22). These results also suggested that these cells might still be responsive to APC signaling. Time course experiments using real-time PCR indeed demonstrated up-regulation of the transcript for the nuclear receptor TR3 by PAR1-and PAR2-specific agonist peptides as well as APC in TNF␣-induced HUVECs with peak levels after 90 min of stimulation (not shown). Compared with quiescent HUVECs, the ratio of TR3 induction by PAR1 and PAR2 was shifted in favor of PAR2, consistent with enhanced expression of functional PAR2 upon TNF␣ stimulation (Fig. 3).
PAR1 Cleavage and Signaling Are Sufficient for TR3 Induction by APC in TNF␣-induced HUVECs-The substantial residual EPCR expression in cytokine-stimulated endothelial cells predicts that these cells are still responsive to APC stimulation. Cleavage-blocking antibodies to PAR1 completely inhibited TR3 induction by APC, whereas signaling by factor Xa was only partially blocked (Fig. 3). These data are in line with previous studies showing that Xa proteolytically activates both PAR1 and PAR2 (10,(23)(24)(25), and they show that APC signaling in cytokine-stimulated endothelial cells requires PAR1 cleavage. Thrombin-cleaved PAR1 has been shown to cross-activate PAR2 in endothelial cells, based on the finding that a small molecule PAR1 antagonist only partially blocks thrombin-mediated signaling (12). To determine whether APC-cleaved PAR1 can cross-activate PAR2 in a similar manner, the effect of the indole-based PAR1 antagonist RWJ-58259 on APC signaling was analyzed. RWJ-58259 binds to the receptor body and inhibits signaling without blocking cleavage of the scissile bond in the N-terminal exodomain (26). As expected, RWJ-58259 acted as a competitive inhibitor of PAR1 agonist peptide induction of TR3. Thrombin signaling was abrogated by cleavage-blocking anti-PAR1 but only partially inhibited by RWJ-58259, consistent with PAR2 cross-activation by thrombincleaved PAR1. APC-mediated signaling was completely inhibited by RWJ-58259, indicating that APC-cleaved PAR1 cannot cross-activate PAR2. Taken together, these results demonstrate that APC leads to gene induction in endothelial cells entirely through PAR1 activation and signaling even when PAR2 is up-regulated upon TNF␣ stimulation.

Distinct Effects of APC and Thrombin on Gene Expression Profile in TNF␣-induced HUVECs-Because
APC is particularly beneficial in severe sepsis, we reasoned that APC signaling may show distinct differences from thrombin signaling in inflammatory cytokine-perturbed endothelial cells. To define the specificity of PAR1 signaling induced by thrombin and APC-EPCR, large scale gene expression profiling was performed in TNF␣-primed HUVECs that were stimulated with thrombin, APC, or a PAR1-specific agonist peptide. Probably because of the greater signaling strength of thrombin, thrombin stimulation caused significant up-or down-regulation of a larger number of genes as compared with APC signaling. Table  II summarizes a manual grouping of 199 or 151 genes selected based on stringency criteria of the average signal log ratios. The complete gene list is provided as supplemental data. Although some genes were induced to a similar extent by thrombin and APC, in general thrombin was a much more potent inducer of transcript changes. Moreover, both thrombin and APC selectively induced transcripts that were not induced by the other protease. Fig. 4 shows hierarchical cluster analysis (27)    stringent subset of genes with signal log ratios of Ͼ0.5 or ϽϪ0.5. Consistent with signaling through the same PAR, thrombin and APC signaling concordantly induced or downregulated several genes (see IIa-induced and -suppressed genes in Fig. 4). Because the gene profiling was performed after 90 min of stimulation, transcription factors and signaling intermediates accounted for a large portion of the changed transcripts. Mediators of potential significance for inflammatory processes, including tissue factor, interleukin 6, or the urokinase receptor, were induced by thrombin as well as APC. However, thrombin was a much more potent inducer of tissue factor in particular. It is therefore likely that thrombin signaling is a major mediator for the coagulant and proinflammatory exacerbation in severe sepsis. In addition, several thrombin-induced or -suppressed genes were little changed by APC, indicating signaling specificity. For example, thrombin but not APC induced interleukin 8 and vascular endothelial cell growth factor, which can increase vascular permeability.
A small group of genes was selectively induced by APC but not by thrombin (see APC-induced in Fig. 4). One of these genes showed concordant up-regulation by APC and PAR1 agonist in our previous gene profiling study on non-stimulated endothelial cells (9). More significantly, a large group of genes was suppressed by APC (see APC-suppressed in Fig. 4), and most strikingly, several of these genes were even up-regulated by thrombin. The APC-suppressed genes include proinflammatory and proapoptotic genes such as members of the NFB family and the proapoptotic p53 (Fig. 4). By suppressing inflammatory signaling intermediates, APC may counteract proinflammatory cytokine signaling and potentiate effects of thrombin signaling that cooperate with these pathways.

PAR1-dependent Down-regulation of p53 and Thrombospondin-1 by APC but Not Thrombin in TNF␣-induced
HUVECs-Because the most striking differences between thrombin and APC were observed in genes down-regulated by APC, we focused on specific examples to confirm concordant changes in the protein levels. To test whether APC regulates p53, cells were treated with doxorubicin to increase p53 protein levels. Western blotting for p53 shows that p53 levels in doxorubicin-treated cells were significantly lower when the cells were pretreated with APC (Fig. 5A). Antibody-blocking experiments demonstrate that this suppressive effect on p53 expression was dependent on APC binding to EPCR as well as PAR1 cleavage (Fig. 5B). APC suppressed thrombospondin-1 mRNA levels, which were increased upon thrombin stimulation. Cytokine-stimulated endothelial cells were treated with either APC or thrombin in the presence or absence of a cleavage-blocking antibody, and thrombospondin-1 protein levels were deter-mined by Western blotting of cell lysates (Fig. 6). APC reduced thrombospondin-1 protein levels in a PAR1 cleavage-dependent manner. In contrast, thrombin induced thrombospondin-1 levels in a PAR1-dependent manner. These findings demonstrate that thrombin and APC by activating the same receptor can mediate opposite biological effects and suggest that EPCR cosignaling may modify PAR1-dependent APC signaling in endothelial cells. DISCUSSION Thrombin signaling in platelets is mediated by PAR1 and PAR4 in humans and PAR3 and PAR4 in mice (19) raising the question whether different PARs are involved in APC signaling in humans and mice. Our finding that gene induction in response to APC injection in pulmonary tissues is significantly diminished in PAR1-deficient mice supports the notion that PAR1 is the major receptor for APC signaling in both humans and mice. This conclusion is consistent with several studies demonstrating that PAR1 but not PAR2 mediates anti-apoptotic signaling (4), calcium flux (28), and microparticle release (29) by APC-EPCR in vitro and the protective effects of APC in mouse models (5,6). A recent study implicates both PAR1 and PAR3 in EPCR-dependent APC signaling in mouse neuronal cells (6). However, it remains to be established through which mechanism PAR3 affects APC signaling in mouse neurons. Available data indicate that mouse PAR3 is not a functional transmembrane signaling receptor but acts as a coreceptor for PAR4 activation by thrombin in mouse platelets (30). Importantly, PAR3 does not seem to be significantly expressed by endothelial cells (31). However, in human HUVECs anti-PAR1 blockade did not prevent late mitogen-activated protein kinase activation and cell proliferation, although anti-PAR1 blockade did suppress early responses (32). Considering the less efficient blocking with a single anti-PAR1 antibody and the very high APC concentration (300 nM) used in this study, the physiological significance of these findings needs to be established. Taken together, emerging evidence adds to our conclusion that PAR1 is the central signaling receptor of the APC-EPCR pathway.
We demonstrated previously that both PAR1 and PAR2 can support APC-EPCR signaling in heterologous expression systems, whereas APC signaling in quiescent human endothelial cells is PAR1-dependent (9). Results reported here show that both human and mouse APC can induce PAR2-dependent signaling in heterologous expression and that canonical PAR2 cleavage is required. However, in endothelial cells APC signaling was strictly PAR1-dependent even when PAR2 expression was strongly up-regulated in cytokine-perturbed cells. Glycosylation of the N-terminal extracellular domain of PAR2 in proximity to the scissile bond has been shown to interfere with PAR2 activation by mast cell tryptase (33). In analogy, posttranslational modification of PAR2 in endothelial cells may preclude proteolytic activation by APC but not factor Xa.
APC apparently not only fails to cleave PAR2 to induce gene expression responses, but APC-cleaved PAR1 also does not induce PAR2 transactivation similar to thrombin-cleaved PAR1 (Fig. 3). Heterodimer formation of PAR1 and PAR2 likely supports transactivation of PAR2 by thrombin-cleaved PAR1 (12), and certain thrombin responses require simultaneous PAR1 and PAR2 signaling (34). Because signaling by APC depends on binding to EPCR, the lack of PAR2 transactivation by APC suggests that EPCR and PAR2 sequester in different membrane compartments. Conversely, only a fraction of PAR1 is predicted to colocalize with EPCR, which also explains why even high concentrations of APC cannot desensitize subsequent thrombin responses (35). Preliminary data from the laboratory of Dr. Esmon (36,37) indicate that EPCR locates to

Distinct Signaling by APC and Thrombin
FIG. 4. Gene expression profiling of APC-and thrombin-induced genes in TNF␣-primed HUVECs. Confluent HUV-EC-Cs were serum-deprived for 5 h in the presence of 2 nM TNF␣ followed by incubation for 90 min with control, 10 nM APC (sAPC), 10 M PAR1 agonist TFLLRNPNDK (sP1), or 10 nM thrombin (sIIa). Gene expression was analyzed with the HG-U95 Av2 array (Affymetrix). Genes were selected based caveolae, and PAR1 activation restricted to caveolar membrane domains may in part account for the signaling specificity of the APC-EPCR-PAR1 pathway.
Although previous studies have shown that APC administration can attenuate the cytokine perturbation of endothelial cells (38,39), APC appears to be particularly effective in severe sepsis where coagulopathy and thrombin generation is frequent. How can it be explained that thrombin-PAR1 signaling can have proinflammatory effects (8), whereas APC signaling is protective in inflammatory conditions if both thrombin and APC signal through the same receptor? Our previous data demonstrate that APC and PAR agonist peptides have very similar effects on gene expression in non-perturbed endothelial cells, including the induction of genes that down-regulate inflammatory pathways or are anti-apoptotic (9). In TNF␣-induced HUVECs, however, thrombin was a more potent inducer of proinflammatory mediators (e.g. interleukin 6 and 8, vascular endothelial growth factor) and tissue factor, whereas APC but not thrombin down-regulated mRNA levels of a number of proinflammatory and pro-apoptotic proteins (Fig. 4), including the tumor suppressor p53 and thrombospondin-1. Down-regulation of these proteins was dependent on EPCR and PAR1; thus activation of the same receptor PAR1 by thrombin and the protein C pathway can lead to different effects on gene expression in TNF␣-primed cells.
Consistent with our findings, Cheng et al. (5) have demonstrated that APC signaling can inhibit the hypoxia-induced tran-sient increase in p53 mRNA in brain endothelial cells. Endothelial cell apoptosis may play an important role in the pathogenesis of sepsis (40). The final decision of a cell to survive or to undergo apoptosis is the result of the integration of a complex and interwoven network of signals. The activity of the pro-apoptotic transcription factor p53 is tightly controlled and fine-tuned through modifications of the protein product. However, there is also evidence that p53 regulation on the mRNA level can be relevant (41,42); e.g. the rise in p53 transcript levels upon serum stimulation (43) may place the cell in a state of anticipation ensuring an effective anti-proliferative or apoptotic response if DNA damage or other stress is encountered (44). On the other hand, downregulation of p53 mRNA levels in endothelial cells by APC signaling may blunt the effects of pro-apoptotic stimuli and protect the endothelium in stressful conditions encountered in inflammation or ischemia/reperfusion. Thrombospondin-1, which was also strongly down-regulated specifically by APC signaling, is a target for positive transcriptional regulation by p53 (45), suggesting that p53 suppression may be an upstream effector mechanism for protective APC signaling. Thrombospondin-1 has been shown to induce apoptosis specifically in cytokine-activated, angiogenic endothelial cells (46). Our data argue that the profound anti-apoptotic properties that distinguish APC from thrombin signaling act specifically in the context of severely cytokine-perturbed endothelial cells. APC has proven therapeutic efficacy in severe sepsis but no statistically significant benefit in less severe sepsis, where inflammatory perturbation of the endothelium is presumed to be partial. The data presented here provide a rationale for the specific benefit of APC in more severe sepsis. on an average change p value of Ͻ0.05 or Ͼ0.95 for either thrombin or APC. Genes with average signal log ratios Ͼ0.5 or ϽϪ0.5 were analyzed by hierarchical clustering and displayed with Treeview; green indicates decreased and red increased expression. For reference, our previously determined (9) transcript changes in response to APC and PAR1 agonist (P1) stimulation on non-TNF␣-primed HUV-EC-C are shown. The groups of genes shown in this figure are based on the tree structure of the clustered genes, and these groups are similar to but not identical with the groups in Table II and in the supplemental data, which were based upon manual data analysis.