Identification of Follistatin-Like 1 by Expression Cloning as an Activator of the Growth Differentiation Factor 15 Gene and a Prognostic Biomarker in Acute Coronary Syndrome

Abstract

BACKGROUND

Growth differentiation factor 15 (GDF15) is a stress-responsive cytokine and biomarker that is produced after myocardial infarction and that is related to prognosis in acute coronary syndrome (ACS). We hypothesized that secreted proteins that activate GDF15 production may represent new ACS biomarkers.

METHODS

We expressed clones from an infarcted mouse heart cDNA library in COS1 cells and assayed for activation of a luciferase reporter gene controlled by a 642-bp fragment of the mouse growth differentiation factor 15 (GDF15) gene promoter. We measured the circulating concentrations of follistatin-like 1 (FSTL1) and GDF15 in 1369 patients with ACS.

RESULTS

One cDNA clone that activated the GDF15 promoter–luciferase reporter encoded the secreted protein FSTL1. Treatment with FSTL1 activated GDF15 production in cultured cardiomyocytes. Transgenic production of FSTL1 stimulated GDF15 production in the murine heart, whereas cardiomyocyte-selective deletion of FSTL1 decreased production of GDF15 in cardiomyocytes, indicating that FSTL1 is sufficient and required for GDF15 production. In ACS, FSTL1 emerged as the strongest independent correlate of GDF15 (partial R² = 0.26). A total of 106 patients died of a cardiovascular cause during a median follow-up of 252 days. Patients with an FSTL1 concentration in the top quartile had a 3.7-fold higher risk of cardiovascular death compared with patients in the first 3 quartiles (P < 0.001). FSTL1 remained associated with cardiovascular death after adjustment for clinical, angiographic, and biochemical variables.

CONCLUSIONS

Our study is the first to use expression cloning for biomarker discovery upstream of a gene of interest and to identify FSTL1 as an independent prognostic biomarker in ACS.

The risk of death and ischemic complications varies considerably in patients with acute coronary syndrome (ACS).⁴ Current guidelines therefore emphasize the importance of risk stratification to enable the making of informed decisions about the tailoring of treatment and allocation of clinical resources (1). Circulating biomarkers may support diagnosis and risk stratification in ACS (2). Accordingly, the search for new cardiovascular biomarkers has emerged as the focus of intense investigation (3). Current approaches for biomarker discovery typically involve large-scale profiling of a disease-related transcriptome, proteome, or metabolome (3– 6). Application of these strategies has led to the identification of new biomarkers, including growth differentiation factor 15 (GDF15) (7–11).

GDF15 is a member of the transforming growth factor β cytokine superfamily that is produced at low levels under healthy conditions (12). Production of GDF15 increases in situations associated with oxidative stress or tissue injury. Indeed, GDF15 has been found, by transcriptional profiling, to be strongly induced in cultured cardiomyocytes subjected to nitrosative or oxidative stress (11, 13) or mechanical stretch (14). Cardiac expression levels of GDF15 increase after myocardial infarction (MI) in mice and patients (11, 15). Increased circulating concentrations of GDF15 are associated with future cardiovascular events in patients with ACS independent of clinical variables and several established biomarkers (16 –18).

Cytokines and growth factors are organized in interactive networks in which they induce or suppress the synthesis of other cytokines, which in turn may trigger the production of further factors (19, 20). We hypothesized that secreted proteins that are upstream of GDF15 may represent new biomarkers relevant to ACS pathophysiology. We therefore devised a eukaryotic expression screen for genes expressed in the infarcted mouse heart that encode activators of the GDF15 promoter. Using this strategy, we identified the secreted protein follistatin-like 1 (FSTL1) as an inducer of GDF15 production and an independent prognostic biomarker in ACS.

Materials and Methods

GDF15 PROMOTER–LUCIFERASE REPORTER PLASMIDS

All animal procedures were performed in accordance with the Committee on Care and Use of Laboratory Animals and were approved by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsi-cherheit. Genomic DNA was isolated from mouse liver. A series of fragments of the 5′-flanking region of the mouse growth differentiation factor 15 (GDF15)⁵ gene was generated by use of the Universal GenomeWalker kit (Clontech) and the Expand Long Template PCR system (Roche). Fragments of the GDF15 5′-flanking region, ranging in size from 248 –1827 bp, were cloned between the KpnI and HindIII restriction sites of the pGL2 basic vector (Promega) to generate the luciferase reporter plasmids, pGL2(-248)GDF15, pGL2(-444)-GDF15, pGL2(-642)GDF15, pGL2(-1027)GDF15, and pGL2(-1827)GDF15. Plasmids were verified by bidirectional DNA sequencing. Ventricular cardiomyocytes were isolated from 1–3-day-old Sprague–Dawley rats and were cultured in 6-well plates (10⁶ cells per well) (21). Cardiomyocytes were transfected with individual reporter plasmids (4 μg per plate) and 80 ng of a β-galactosidase expression plasmid (pCMV-lacZ) to control for transfection efficiency (21). Cardiomyocytes were cultured for 24 h in the absence (control) or presence of 250 μmol/L of the nitric oxide donor S-nitroso-N-acetyl-D,L-penicillamine (SNAP) before luciferase and β-galactosidase activities were determined. The reporter plasmid containing 642 bp of the GDF15 5′-flanking region displayed low basal expression and high inducibility by SNAP and was used in the expression screen (Fig. 1A).

Fig. 1. Identification of FSTL1 by expression screening as an activator of the GDF15 gene.

(A), Data are from 3– 4 independent experiments. (B), Schematic representation of the screen.

EXPRESSION SCREEN

A eukaryotic expression screen for cDNAs encoding activators of the GDF15 promoter was performed by adapting a method described by Olson and coworkers (Fig. 1B) (22). We generated a cDNA expression library from infarcted mouse hearts using the Clone-Miner cDNA library construction kit (Invitrogen). MI was induced by ligation of the left anterior descending coronary artery in 8 –10-week-old male C57Bl/6 mice ventilated with 1% to 3% isoflurane (Baxter) (11). Three days after coronary ligation, mRNA was isolated from the entire left ventricles of 5 mice, pooled, reverse-transcribed into cDNA, cloned into pDONR222 (Invitrogen), and transferred into the pcDNA3.2/V5-DEST Gateway mammalian expression vector (Invitrogen).

The cDNA expression library was separated into pools containing approximately 100 clones each. COS1 cells were plated in 12-well plates (1 × 10⁵ cells per well) in low-glucose DMEM supplemented with 10% fetal calf serum, L-glutamine, penicillin, streptomycin, and 10% Opti-MEM (Invitrogen). The cells were then transfected for 48 h with 0.5 μg of pooled plasmid DNA from the expression library, 0.5 μg of pGL2(-642) GDF15, and 50 ng of internal control pCMV-lacZ using FuGENE 6 (Roche). Cells were then harvested for luciferase and β-galactosidase assays. DH5α Escherichia coli (Invitrogen) were transformed with positive plasmid pools and 12 colonies from each positive pool were picked and combined as a subpool. Plasmids were purified from 16 subpools and used to transfect COS1 cells in the presence of the pGL2(-642)GDF15 reporter and pCMV-lacZ as described above. Positive subpools were transformed into E. coli and 16 individual colonies were picked from each subpool. Plasmids were purified from these individual colonies and were again transfected into COS1 cells in the presence of pGL2 (-642)GDF15 and pCMV-lacZ. DNA from individual positive plasmids was sequenced. DNA sequences were analyzed with Gene Ontology (http://www.geneontology.org). Given our focus on circulating biomarker discovery, genes were filtered to include genes annotated with the term “extracellular” and to exclude genes annotated with the terms “integral to membrane,” “nucleus,” “cytoplasm,” or “intracellular”.

QUANTITATIVE PCR

CGTGGAGCAGAATGAAACAGCCAT and TGAAG GATGGGTTGAGGCACTTGA were used as primers to detect FSTL1. GDF15 primers and quantitative PCR methods have been described (11).

RECOMBINANT FSTL1, FSTL1 ADENOVIRUS, AND FSTL1 TRANSGENIC MICE

Recombinant mouse FSTL1 protein was generated in Sf9 insect cells (23). Mouse FSTL1 cDNA was cloned into a replication-deficient adenovirus using the AdEasy XL Vector System (Stratagene) to generate Ad.FSTL1. An adenovirus encoding β-galactosidase was used as control (Ad.lacZ). Viruses were purified with the Adeno-X Virus Purification Kit (BD Biosciences). Mice with cardiac overexpression of follistatin-like 1 (FSTL1) or cardiomyocyte-restricted deletion of FSTL1 have been described (23). Adult mouse ventricular cardiomyocytes were isolated as described (23).

PATIENT POPULATION

We recruited 1369 consecutive patients with a final diagnosis of ACS [unstable angina, non–ST-elevation MI (NSTEMI), STEMI] who were admitted to Hannover Medical School or to the University Hospital in Heidelberg. Refusal to provide written informed consent was the only exclusion criterion. MI was diagnosed according to the criteria of the Joint European Society of Cardiology/American College of Cardiology/American Heart Association/World Health Federation Task Force (24). Coronary angiography was performed in 1323 patients. The number of vessels showing at least a 50% luminal stenosis was documented. Left main coronary artery stenosis was classified as 2-vessel disease. Left ventricular (LV) systolic function was assessed by angiography or echocardiography in 1346 patients and was graded as normal or mildly, moderately, or severely reduced. Treatment decisions were left to the discretion of the attending cardiologist. Follow-up for cardiovascular mortality and nonfatal MI was accomplished by telephone contact (93% of the patients) or questionnaire (7%) at least 6 months after discharge. Only patients with completed follow-up were included in the analyses. Cardiovascular death was defined as death during hospitalization with MI; death from progressive heart failure; death from documented cardiac arrhythmias, stroke, thromboembolic events, or aneurysms; and sudden or unwitnessed death not related to accidents, suicide, terminal cancer, or other ominous diagnoses. When a patient, spouse, or primary care physician reported another hospital admission for a cardiovascular cause, discharge letters (or final reports on deceased patients) were obtained and searched for a diagnosis of a fatal cardiovascular event or nonfatal MI. All endpoints were adjudicated by C. Widera in Hannover and E. Giannitsis in Heidelberg. During a median follow-up of 252 (range 1–1073) days, 106 patients died from a cardiovascular cause and 47 had a nonfatal MI. The study was approved by the institutional committees on human research at both institutions. All patients provided written informed consent.

LABORATORY ANALYSES

Serum samples were obtained by venipuncture on admission and stored at −70 °C. FSTL1 was measured by immunoluminometric assay (25). GDF15 was measured by IRMA (26). N-terminal pro-B-type natriuretic peptide (NT-proBNP) was measured by a sandwich immunoassay, C-reactive protein (CRP) by a high-sensitivity latex particle-enhanced immunoassay, and cardiac troponin T (cTnT) by a fourth-generation assay (all from Roche Diagnostics). Estimated glomerular filtration rate (eGFR) was calculated with the Modification of Diet in Renal Disease equation.

STATISTICAL ANALYSES

Data are presented as mean with SE, median with 25th and 75th percentiles, or numbers and percentages. We compared continuous variables with the Mann–Whitney test. FSTL1 and GDF15 were related to other baseline characteristics by use of single and multiple variable linear regression analysis. FSTL1, GDF15, NT-proBNP, CRP, and eGFR were not normally distributed and were logarithmically transformed (ln-transformed) before their use in these analyses. Kaplan–Meier survivor curves were examined to assess the relationship between FSTL1 and outcome. To determine whether FSTL1 was independently associated with outcome, Cox regression models were built that included data for variables that were available on admission in patients with ACS (age, sex, diabetes, hypertension, current smoking, previous MI, eGFR, and cTnT; model 1). Information on the number of diseased vessels and LV systolic function was added to these variables to create model 2. NT-proBNP and CRP were added to model 2 to create model 3. Finally, GDF15 was added to model 3. NT-proBNP, CRP, eGFR, and GDF15 were entered as ln-transformed variables into the Cox regression analyses. Because cTnT was undetectable on admission in 40% of the patients, cTnT was treated as a dichotomized variable with 10 ng/L, the detection limit of the assay, used as the cutoff value (27). Previous studies revealed that cTnT concentrations above 10 ng/L on admission are associated with an adverse prognosis in ACS (28 –30). The discriminative strength of FSTL1 concerning 6-month cardiovascular mortality was investigated by ROC-curve analysis, and by calculating the integrated discrimination improvement (IDI) (31). A 2-tailed P value <0.05 was considered to indicate statistical significance. We analyzed data using SPSS 19.0 (SPSS), MedCalc 11.2.1.0 (MedCalc Software), and SAS 9.1 (SAS Institute).

Results

IDENTIFICATION OF FSTL1 BY EXPRESSION SCREENING FOR INDUCERS OF THE GDF15 PROMOTER

We performed a cDNA expression screen by expressing pools of clones from an infarcted mouse heart cDNA expression library in COS1 cells and assaying for activation of a luciferase reporter controlled by a 642-bp fragment of the 5′-flanking region of mouse GDF15. A total of 450 cDNA pools, each containing approximately 100 individual cDNA clones (expression plasmids), were screened. One cDNA expression plasmid that was repetitively found to enhance luciferase expression from the pGL2(-642)GDF15 reporter encoded the full-length 308 amino acid FSTL1 protein (GenBank accession number NM_008047.5).

FSTL1 is a secreted protein sharing a characteristic structural module, the follistatin domain, with follistatin, FSTL3, and members of the SPARC (secreted protein acidic and rich in cysteine) matrix protein family (32). FSTL1 has low protein sequence homology to other follistatin family proteins, and in contrast to follistatin and FSTL3, FSTL1 functions as a ligand for a cell-surface receptor (33, 34).

To confirm FSTL1 as an inducer of the GDF15 promoter, we cotransfected the FSTL1 expression plasmid and the pGL2(-642)GDF15 reporter plasmid into COS1 cells and found that expression of FSTL1 led to a 6.8 (0.9)-fold stimulation of luciferase activity (n = 6; P = 0.002).

FSTL1 INDUCES GDF15 PROMOTER ACTIVITY AND EXPRESSION IN VITRO AND IN VIVO

In cardiomyocytes transfected with pGL2(-642) GDF15, cotransfection of the FSTL1 expression plasmid resulted in a 5.1-fold stronger activation of the GDF15 promoter–luciferase reporter after 24 h compared with cotransfection with a control plasmid encoding DsRed (Fig. 2A). Transfection of the FSTL1 expression plasmid into cardiomyocytes led to a 3.3-fold increase in FSTL1 mRNA production and a 2.7-fold increase in GDF15 mRNA production after 24 h compared with transfection of a control plasmid encoding DsRed (Fig. 2B). Treatment of cardiomyocytes with recombinant mouse FSTL1 for 24 h resulted in a dose-dependent and saturable induction of GDF15 mRNA production (Fig. 2C).

Fig. 2. FSTL1 activation of GDF15 mRNA production in vitro and in vivo.

Data are from 3– 6 independent experiments or 4 –5 mice per group; *P < 0.05; **P < 0.01. Ad., adenovirus; WT, wild type.

To explore the in vivo relevance of these findings, mice were treated with intra–LV cavity injections of adenoviruses encoding FSTL1 or β-galactosidase (1 × 10⁹ plaque-forming units each). After 5 days, mice overexpressing FSTL1 had 3.0-fold higher LV FSTL1 mRNA production levels and 4.1-fold higher LV GDF15 mRNA production levels compared with mice expressing β-galactosidase (Fig. 2D). Similarly, transgenic mice overexpressing FSTL1 in cardiomyocytes (23) had 4.1-fold higher FSTL1 mRNA and 2.9-fold higher GDF15 mRNA production levels in the left ventricle compared with wild-type control mice (Fig. 2E). Conversely, mice with a cardiomyocyte-selective deletion of FSTL1 (23) displayed reduced production levels of FSTL1 mRNA (−88%) and GDF15 mRNA (−61%) in isolated cardiomyocytes (Fig. 2F).

Substantiation of the link between FSTL1 and GDF-15 required the above transgenic models and biologically active recombinant FSTL1, which became available only recently (23). In an earlier investigation, identification of FSTL1 in our screen prompted us to develop an FSTL1 immunoassay (25), which we used to assess the independent prognostic value of FSTL1 in ACS.

CIRCULATING FSTL1 CONCENTRATIONS IN ACS PATIENTS

Baseline characteristics of our patients are summarized in Table 1. The median FSTL1 concentration on admission was 14.1 μg/L (25th–75th percentiles 12.3–16.5 μg/L; range 5.1–50.7 μg/L); 78% of the patients presented with an FSTL1 concentration above 12.0 μg/L, the upper limit of the reference interval in apparently healthy individuals (25). Patients with NSTEMI presented with somewhat higher FSTL1 concentrations compared with patients with unstable angina [median (25th–75th percentiles): 14.7 (12.6 –17.2) μg/L vs 13.4 (11.7–15.5) μg/L; P < 0.001] or STEMI [14.2 (12.2–16.8) μg/L; P = 0.016].

Table 1.

Study population.^a

Age, years	68 (58–76)
Male sex	1011 (74)
Diabetes	324 (24)
Hypertension	1030 (75)
Current smoking	339 (25)
Previous MI	380 (28)
No. diseased vesselsb
0	96 (7)
1	294 (22)
2	295 (22)
3	638 (48)
LV functionc
Normal	424 (32)
Mildly reduced	452 (34)
Moderately reduced	318 (24)
Severely reduced	152 (11)
Index diagnosis
Unstable angina	400 (29)
NSTEMI	597 (44)
STEMI	372 (27)
cTnT >10 ng/L	818 (60)
eGFR, mL · min−1 · (1.73 m2)−1	82 (63–102)
NT-proBNP, ng/L	463 (138–1902)
CRP, mg/L	3.6 (2.0–11.4)
GDF15, ng/L	1961 (1323–3270)
FSTL1, μg/L	14.1 (12.3–16.5)

^aBaseline characteristics of 1369 patients admitted with ACS. Data are n (%) or median (25th–75th percentile).

^bData were available from 1323 patients.

^cData were available from 1346 patients.

ASSOCIATION OF FSTL1 WITH BASELINE VARIABLES IN ACS

In univariate analyses, increasing concentrations of FSTL1 were associated with age (R = 0.10), diabetes (R = 0.16), eGFR (R = −0.22), cTnT (R = 0.18), CRP (R = 0.35), NT-proBNP (R = 0.39), and GDF15 (R = 0.44) (all P < 0.001). FSTL1 concentrations increased in relation to the number of diseased vessels [median (25th–75th percentiles): 0-vessel disease, 13.2 (12.0 –15.9) μg/L; 1-vessel, 13.7 (12.0 –15.9) μg/L; 2-vessel, 14.0 (12.3–16.4) μg/L; and 3 vessel disease, 14.5 (12.4 –17.0) μg/L; P = 0.008] and the degree of LV systolic dysfunction [normal 13.6 (11.8 –15.8) μg/L; mildly reduced 14.0 (12.1–15.9) μg/L; moderately reduced 14.4 (12.5–17.3) μg/L; severely reduced 16.1 (13.3–19.1) μg/L; P < 0.001].

By multiple linear regression analysis, FSTL1 was independently associated with GDF15, NT-proBNP, CRP, age, male sex, eGFR, and diabetes (also see the Supplemental Table 1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol58/issue8). The closest independent association was observed between FSTL1 and GDF15 (partial R² = 0.26; P < 0.001). In an analogous multiple linear regression analysis, GDF15 was independently associated with exactly the same variables and the closest independent association was observed, again, between GDF15 and FSTL1 (also see the online Supplemental Table 1). The relation between FSTL1 and GDF15 in ACS is illustrated in the online Supplemental Fig. 1.

FSTL1 AND THE RISKS OF CARDIOVASCULAR DEATH AND NONFATAL MI

The FSTL1 concentration on admission was associated with cardiovascular mortality during follow-up. The Kaplan–Meier curves indicated that the increase in cardiovascular mortality occurred in the top quartile of FSTL1; patients in the top quartile had a 3.7-fold higher risk of cardiovascular death compared with patients in the first 3 quartiles (95% CI 2.5–5.4; P < 0.001) (Fig. 3). FSTL1 remained associated with cardiovascular death after adjustment for a comprehensive set of clinical variables, eGFR, and cTnT (model 1) (Table 2). Even after further adjustment for the number of diseased vessels and LV systolic function (model 2), and CRP and NT-proBNP (model 3), FSTL1 remained highly significantly associated with cardiovascular death (Table 2). FSTL1 lost its independent association with mortality when GDF15 was added to model 3 (Table 2). Similar results were obtained when index diagnosis, instead of cTnT on admission, was entered into the models (not shown). FSTL1 was not associated with the risk of nonfatal MI.

Fig. 3.

One-year cardiovascular mortality according to quartiles of FSTL1 on admission.

Table 2.

Cardiovascular mortality in relation to FSTL1.^a

Model	Patientsb	Hazard ratio (95% CI)	P
Unadjusted	1369	3.7 (2.5–5.4)	<0.001
Adjusted for
Model 1 variables	1369	2.5 (1.7–3.6)	<0.001
Model 2 variables	1309	2.4 (1.6–3.7)	<0.001
Model 3 variables	1309	2.0 (1.3–3.1)	0.001
Model 3 variables + GDF15	1309	1.5 (0.9–2.3)	0.09

^aHazard ratios (95% CI) associated with an FSTL1 concentration in the fourth quartile compared with an FSTL1 concentration in the first 3 quartiles in the context of model 1, 2, or 3 variables. Model 1 included information on age, sex, diabetes, hypertension, current smoking, previous MI, eGFR, and cTnT on admission. Information on the number of diseased vessels and LV systolic function was added to model 1 to create model 2. NT-proBNP and CRP were added to model 2 to create model 3.

^bNumbers of patients with complete data for all variables.

FSTL1 FOR DISCRIMINATION OF 6-MONTH SURVIVAL STATUS

The c statistic and IDI were calculated to explore whether FSTL1 can enhance the discrimination of patients who die, or don’t die, from a cardiovascular cause during 6 months. The area under the ROC curve (AUC) of FSTL1 was significantly greater than the non-discriminating value of 0.5 (0.70; 95% CI 0.68 – 0.73; P < 0.001). FSTL1 added discriminatory information to model 1 and model 2 variables (Table 3). When added to model 3, FSTL1 led to a further improvement in the c statistic and tended (P = 0.055) to improve also the IDI (Table 3). Similar results were obtained when index diagnosis, instead of cTnT on admission, was included in the models (not shown).

Table 3.

Discrimination of 6-month cardiovascular mortality.^a

	c statistic		IDI
	AUC (95% CI)	P	IDI (95% CI)	P
Model 1	0.807 (0.765–0.848)
Model 1 + FSTL1	0.843 (0.805–0.881)	0.003	0.033 (0.011–0.055)	0.004
Model 2	0.829 (0.787–0.872)
Model 2 + FSTL1	0.858 (0.820–0.896)	0.007	0.029 (0.007–0.050)	0.007
Model 3	0.856 (0.819–0.892)
Model 3 + FSTL1	0.870 (0.834–0.906)	0.020	0.019 (0.000–0.038)	0.055

^ac statistic and IDI for models 1, 2, and 3 without or with inclusion of FSTL1. Models and the number of patients are the same as in Table 2.

Discussion

We performed an expression screen for activators of the GDF15 gene in an effort to discover novel ACS biomarkers. We identified FSTL1 as an activator of GDF15 gene expression in vitro and in vivo. We found the circulating concentrations of FSTL1 to be independently related to GDF15 and to cardiovascular mortality in ACS. Our study is the first to use expression cloning for biomarker discovery upstream of a gene of interest, a strategy that will be applicable outside the specific example provided here.

GDF15 is a stress-responsive cytokine that is induced after injury in the heart and vasculature (11, 15, 35). The circulating concentrations of GDF15 are associated with the risk of fatal and nonfatal cardiovascular events in healthy individuals and patients with cardiovascular disease (16 –18, 36, 37). We thus hypothesized that GDF15 promoter activity can be used as readout in an expression screen for stress-related genes and new prognostic ACS biomarkers. GDF15 production in the heart peaks about 4 days after coronary artery ligation in mice (11, 15). Presuming that activators of the GDF15 gene are expressed before this time point, we established a cDNA expression library from day 3 infarcted mouse hearts. Screening this library, we identified FSTL1 as a potent inducer of a reporter gene driven by a 642-bp fragment of the GDF15 promoter region. Consistent with this finding, a previous study showed FSTL1 to be strongly produced in the infarcted mouse heart on day 3 after coronary artery ligation (38). Validating the screen, we found FSTL1 to be sufficient to activate GDF15 production in cardiomyocytes in vitro and in the murine heart in vivo. In mice with a cardiomyocyte-restricted deletion of FSTL1, we found FSTL1 also to be required for basal GDF15 production.

In ACS, the circulating concentrations of FSTL1 were most closely correlated with the circulating concentrations of GDF15 in a comprehensive multiple linear regression analysis that considered clinical, biochemical, and angiographic variables. In a similar analysis of the independent clinical correlates of GDF15, the correlation between FSTL1 and GDF15 emerged again as the strongest one. In addition, we found FSTL1 and GDF15 to be associated with a remarkably similar set of baseline variables indicating that both proteins reflect overlapping disease pathways in ACS. This hypothesis is supported by data showing that FSTL1 and GDF15 are both induced in the heart in rodent models of MI and aortic stenosis, in which they both promote antiapoptotic and antihypertrophic effects (11, 15, 23, 38, 39).

Despite this increasing evidence for a pathophysiological role of FSTL1 in cardiovascular disease, little is known about the potential role of FSTL1 as a biomarker in patients. In a recent study investigators used immunoblotting to measure FSTL1 in the sera of 21 healthy volunteers and 86 patients with heart failure and found that the circulating concentrations of FSTL1 are increased in heart failure and associated with all-cause mortality (P = 0.09) (40). In our previously reported investigation of the FSTL1 immunoassay, we found FSTL1 to be associated with mortality in 216 patients with ACS (25). Because of the small sample size, we were unable to adjust for confounders of the relationship between FSTL1 and prognosis in that study (25). In this regard, in the this report we present the largest available experience with FSTL1 as a biomarker in patients. FSTL1 was associated with cardiovascular mortality in the context of clinical and biochemical variables that are available on admission (model 1), and after further adjustment for angiographic findings (available in invasively managed patients, model 2), and for NT-proBNP and CRP (model 3). The prognostic strength of FSTL1 was demonstrated by our finding that FSTL1 added comparable discriminatory information to model 2 compared with NT-proBNP plus CRP (AUC 0.858 vs 0.856, see Table 3). These findings should encourage further studies exploring the incremental prognostic value of FSTL1 as a biomarker in cardiovascular disease in the context of established clinical and biochemical risk markers. Future studies must be performed to explore the association of FSTL1 with other cardiovascular endpoints (e.g., heart failure events) and should investigate potential therapeutic implications of high FSTL1 concentrations.

Expression cloning coupled with Gene Ontology analysis to identify secreted proteins as potential biomarkers that are upstream of a gene of interest will be applicable to other disease settings. This new strategy is complementary to the more traditional approach to biomarker discovery downstream of a defined stressful stimulus, such as mechanical strain or nitrosative stress (9, 11). A potential limitation of the use of an existing circulating biomarker as readout in the screen is that the newly identified biomarker(s) will be correlated to some extent with the existing biomarker and may thus represent overlapping pathophysiological dimensions, as we observed here for GDF15 and FSTL1. Although correlated, “orthogonal” biomarkers can underscore the importance of a biological pathway, they might not provide a substantial increase in predictive value (3). Indeed, FSTL1 lost its significant association with cardiovascular mortality when GDF15 was included in the full Cox regression model. Other genes may be used as readout in similar screens in the future, for example intracellular genes involved in specific disease pathways that are not accessible to direct measurement in the circulation.