i2b2: Transla3onal Toolkit

cbmi.med.harvard.edu

i2b2: Transla,onal Toolkit

Isaac S. Kohane, MD, PhD


Where is the missing informa,on?

VP = VG + VE + VGE


Meds Filled

OTC

RxNorm

LOINC, Path/Hist, ECG, Radiology

Pill Dispensers

Lab Values, Vital Signs

Diaries

Fitness Club Memberships

Grocery Store Purchases

Phone GPS

Blogs

Tweets

Facebook Postings

Personal Health

Records

Patients LikeMe

Home Treatments, Monitors and

Tests

Credit Cards

Meds Prescribed Meds Taken Meds Prescribed Meds Taken

Herbals Alt Therapies

CPT, ICD-9

Visit Type/Time

ICD-9, SNOMED Death Records

Climate Public Health Records

HealthMap.org Weather News Feeds GIS Maps EPA

Census

23andMe SNPs, Arrays

Ancestry.com

Zillow LinkedIn

(Indirect from OTC Purchases)

Facebook Friends Twitter Hashtags

Medications

Demographics

Diagnostics (Ordered)

Diagnostics (Results)

Procedures

Lifestyle

Environment

Diagnoses

Family History

Genetics

Social Network

Symptoms

Socioeconomic

Social History Police Records TOB, ETOH Clinical Notes

(Digital)

Reports

Traces, Images

Paper Chart

Physical Exams

Meds Filled NDC, NDF-RT Allergies

Out of Pocket

Differential Dx

Encounters

HL7

HR Sick Days Chief Complaint

Structured Data Un-Structured Data

Link to validate existing data or fill-in missing data

Lin

k to

ob

tain

ne

w ty

pe

s of d

ata

Claims Data, Medicare

Registries, Clinical Trials

Data Outside the Healthcare

System

Pharmacies

Health Care Centers

Probabilistic Linkage

Easier to link to individuals

Harder to link to individuals

Only aggregate data exists

More

Less

Less

Data Quantity


cbmi.med.harvard.edu 03/16/13

Narrative data (NLP text extractions)

Codified data (ICD9 codes, etc)


Accrual Rates


Costs



1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

0

2

4

6

8

10

12

14

16

18

Cusum

of Hospitalizations due to M

yocardial Infarction

Rofecoxib introduced! Rofecoxib withdrawn!

Brownstein, PLoS One, 2007


For weak effects and rare events

Search rou,ne clinical records from 5 major hospitals for: Demographics Diagnosis Medica,ons Lab Results

Reach N Rare Dx Small Effects

10 billion FACTS 6 million pa,ents I2b2

DFCI


Finding the needles ARTICLE

doi:10.1038/nature11040

Cardiac angiogenic imbalance leads toperipartum cardiomyopathyIan S. Patten1,2*, Sarosh Rana3*, Sajid Shahul4, Glenn C. Rowe1, Cholsoon Jang1, Laura Liu1, Michele R. Hacker3, Julie S. Rhee3,John Mitchell4, Feroze Mahmood4, Philip Hess4, Caitlin Farrell1, Nicole Koulisis1, Eliyahu V. Khankin5, Suzanne D. Burke5,8,Igor Tudorache6, Johann Bauersachs7, Federica del Monte1, Denise Hilfiker-Kleiner7, S. Ananth Karumanchi5,8 & Zoltan Arany1

Peripartum cardiomyopathy (PPCM) is an often fatal disease that affects pregnant women who are near delivery, and itoccurs more frequently in women with pre-eclampsia and/or multiple gestation. The aetiology of PPCM, and why it isassociated with pre-eclampsia, remain unknown. Here we show that PPCM is associated with a systemic angiogenicimbalance, accentuated by pre-eclampsia. Mice that lack cardiac PGC-1a, a powerful regulator of angiogenesis, developprofound PPCM. Importantly, the PPCM is entirely rescued by pro-angiogenic therapies. In humans, the placenta in lategestation secretesVEGFinhibitors likesolubleFLT1 (sFLT1), andthis is accentuatedbymultiplegestationandpre-eclampsia.This anti-angiogenic environment is accompanied by subclinical cardiac dysfunction, the extent of which correlates withcirculating levels of sFLT1. Exogenous sFLT1 alone caused diastolic dysfunction in wild-type mice, and profound systolicdysfunction inmice lacking cardiac PGC-1a. Finally, plasma samples fromwomenwith PPCM contained abnormally highlevels of sFLT1. Thesedata indicate thatPPCMismainlyavascular disease, causedbyexcess anti-angiogenic signalling in theperipartumperiod. Thedata also explain how late pregnancyposes a threat to cardiachomeostasis, andwhypre-eclampsiaand multiple gestation are important risk factors for the development of PPCM.

PPCM affects 1 in 300 to 1 in 3,000 pregnancies, with geographic hotspots of high incidence, such as Nigeria and Haiti1,2. The disease ischaracterized by systolic heart failure presenting in the last month ofpregnancy or the first 5months post-partum. Although approximatelyhalf of affected women recover cardiac function post-partum, manypatients progress to chronic heart failure, cardiac transplantation ordeath. Thus, PPCMcandevastate otherwise healthy youngwomenandtheir infants. PPCMremains a disease of unknownaetiology.Theonsetlate in gestation does not coincide with increased haemodynamic loadon the heart, suggesting that othermechanisms are responsible. Recentdatahave suggested that anti-angiogenic prolactin fragmentsmay havean important role in causing the disease in some patients3. Risk factorsfor PPCM also include pre-eclampsia and multiple gestation, suggest-ing potential mechanistic overlap with these processes1,2.PGC-1a is a transcriptional coactivator that drives mitochondrial

biogenesis and other metabolic programs in many tissues, includingthe heart4,5. PGC-1a is highly expressed in the heart, and mice lackingPGC-1a globally have abnormal cardiac energetic reserves and respondpoorly to stressful stimuli such as transverse aortic banding6,7. In additionto its role in mitochondrial homeostasis, PGC-1a also induces theexpression and secretion of pro-angiogenic factors, such as vascularendothelial growth factor (VEGF), which leads to formation of newblood vessels8,9. Although the angiogenic function of PGC-1a has beendescribed in skeletalmuscle, its role in cardiac tissue remains unexplored.

Cardiac-specific PGC-1a deletion leads to PPCMTo study further the role of PGC-1a in the heart, we generatedcardiac-specific PGC-1a knockout (HKO) mice (see Methods). While

studying these mice, we noticed that female HKOmice were fertile, anddelivered normal litter sizes (not shown), but invariably died after one ortwopregnancies (Fig. 1a). Thehearts of thesemicewere large, dilated andfibrotic (Fig. 1b–d), consistent with a dilated cardiomyopathy. Two-dimensional M-mode echocardiography revealed dilated, poorly con-tractile hearts in HKO mice after their second delivery (Fig. 1e). Leftventricular end-diastolic dimensions (LVEDD) and left ventricular end-systolic dimensions (LVESD) were markedly enlarged, and fractionalshortening, a direct measure of cardiac contractile function, was pro-foundly depressed (Fig. 1f–i). Nulliparous mice, as well as post-partumcontrol mice, were not affected. Males were also not affected(Supplementary Fig. 1). Thus, the absence of PGC-1a in cardiomyocytesleads to a profound PPCM in mice.

PGC-1a regulates angiogenesis in cardiac tissueWe have recently shown in skeletal muscle that PGC-1a regulatesangiogenesis by driving the expression of angiogenic factors likeVEGF8,9.Anti-angiogenic therapies, including antibodies that neutralizeVEGF and small-molecule VEGF receptor inhibitors, are increasinglybeing used in oncological and ophthalmological treatments, andcardiomyopathy and heart failure have recently been recognized asimportant side effects10,11, showing that anti-angiogenic therapy canbe harmful to the heart in humans. Impaired VEGF signalling has alsobeen linkedwith cardiac dysfunction inmice12,13. At the same time, latepregnancy is a strong anti-angiogenic environment, partly owing to thesecretion by the placenta of anti-angiogenic factors, like sFLT1, thatbind to and neutralize soluble members of the VEGF family14. Theseobservations led us to postulate that PGC-1a regulates an angiogenic

*These authors contributed equally to this work.

1Cardiovascular Institute, Beth Israel Deaconess Medical Center, Harvard Medical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 2Center for Vascular Biology Research, Beth IsraelDeaconess Medical Center, Harvard Medical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 3Division of Maternal Fetal Medicine/Department of Obstetrics and Gynecology, Beth IsraelDeaconessMedical Center, HarvardMedical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 4Department of Anesthesia and Critical Care, Beth Israel DeaconessMedical Center, HarvardMedicalSchool, 330BrooklineAvenue,Boston,Massachusetts02115,USA. 5DivisionofNephrology/DepartmentofMedicine,Beth IsraelDeaconessMedicalCenter andHarvardMedicalSchool, 330BrooklineAvenue, Boston, Massachusetts 02115, USA. 6Department of Cardiothoracic, Transplantation andVascular Surgery,MedizinischeHochschuleHannover, Carl-NeubergStrasse, D-30625Hannover, Germany.7Department of Cardiology and Angiology, MedizinischeHochschule Hannover, Carl-Neuberg Strasse, D-30625 Hannover, Germany. 8Howard Hughes Medical Institute, 330 Brookline Avenue, Boston,Massachusetts 02115, USA.

1 7 M A Y 2 0 1 2 | V O L 4 8 5 | N A T U R E | 3 3 3

Macmillan Publishers Limited. All rights reserved©2012

11

P5 0.003). Elevated blood pressure in the pre-eclamptic women(Supplementary Table 1) is unlikely to explain the worsening MPI,because MPI is thought to reflect cardiac function independently ofblood pressure22, and pregnant women with similar mild elevationsof blood pressure but without pre-eclampsia have normal cardiac func-tion26. Instead, these data suggest that elevated sFLT1 causes thediastolic dysfunction. To test this idea directly, sFLT1 was deliveredsystemically to pregnant mice by intravenous injection of adenovirusesexpressing sFLT1, and MPI was examined using high-resolutionmurine echocardiography. sFLT1 caused significant increases in MPIin these mice within 10 days (Fig. 5e). These data, taken togetherwith published observations in patients receiving anti-angiogenictherapies10,11, strongly suggest that elevated sFLT1 causes cardiacdysfunction inwomenwith pre-eclampsia. Although the left ventriculardysfunction recovers following delivery in many patients, a secondinsult in some women probably precipitates PPCM.

sFLT1 causes cardiomyopathy and is high in PPCMThe above observations strongly support the idea that PPCM can beinduced by excess anti-angiogenic signalling, including the highexpression of sFLT1 during late gestation seen both in women27 andmice (Supplementary Fig. 5c, P5 0.009). To test this idea directly,sFLT1 was delivered systemically to nulliparous mice, as above. PGC-1aHKOmice that received sFLT1 developed profound cardiac failurewithin 3weeks; these mice had increased cardiac weight and markeddecreases in fractional shortening on echocardiography (Fig. 6a–c).This was accompanied by a marked drop in vascular density (Sup-plementary Fig. 6), although not in larger vessels (SupplementaryFig. 7). Wild-type mice also showed significant, though less extensive,decreases in vascular density and cardiac function after exposure to3 weeks of sFLT1. Thus, sFLT1 alone is sufficient, even in the absenceof pregnancy, to cause dramatic cardiomyopathy in the setting of aheart that is unable to withstand the anti-angiogenic insult.To investigate further whether elevated sFLT1 levels in humans

could be contributing to PPCM, plasma from women with PPCMwas acquired 4–6weeks post-partum and sFLT1 levels were mea-sured. sFLT1 levels usually return to normal within 48–72 h afterdelivery28. sFLT1 levels were elevated in a large subset of thesePPCM patients (P5 0.002), remaining up to five- or tenfold higherthan the levels in control participants (Fig. 6d). Post-partum sFLT1levels can remain slightly higher in subjects with pre-eclampsia29,30,but the levels found here are notably higher. Thus, the findings areconsistent with the idea that a substantial percentage of PPCM sub-jects have been exposed to pre-eclampsia, and that secretion of sFLT1persists inappropriately post-partum. Indeed, in our own institution,33% of the last 75 cases of PPCM were associated with pre-eclampsia(Fig. 6e, f), markedly more than the population rate of 3–5% (ref. 31).The persisting extra-placental source of sFLT1 in the post-partumperiod is not known, and may include placental remnants, circulatingmononuclear cells32 or shed syncytial microparticles33.

DiscussionOur study shows that angiogenic imbalance in the heart during theperi-partum period may lead to PPCM in mice and in humans. Thedata indicate that PPCM is caused by a ‘two-hit’ combination of, first,systemic anti-angiogenic signals during late pregnancy and, second, ahost susceptibility marked by insufficient local pro-angiogenicdefences in the heart. The first hit explains why PPCM is a diseaseof the late gestational period, which is precisely when circulatinganti-angiogenic factors such as sFLT1 peak in pregnancy21,34. Otherpathways, such as prolactin or excess angiotensin II signalling, mayalso be involved3,35. The first hit is also worse in pre-eclampsia, whichis characterized by markedly elevated sFLT1 levels. Associationsbetween pre-eclampsia and PPCM have been well documented inmany populations1,2,36–41(Supplementary Table 3). Interestingly, somestudies involving women of African descent have not found an asso-ciation between hypertensive disorders of pregnancy and PPCM42,suggesting that there is ethnic variability in the pathogenesis ofPPCM. It is also possible that PPCM with and without associatedpre-eclampsia have different pathogeneses43. Overall, our data suggestthat elevated sFLT1 levels in pre-eclampsia contribute to at least thePPCM that is associated with pre-eclampsia. We further propose thatelevated sFLT1 levels in fact present a challenge to the myocardium inall pregnancies, thus explaining why the peri-partum period putswomen at risk of developing heart failure, even in the absence ofpre-eclampsia. Interestingly, other situations of elevated sFLT1(twin pregnancies) and recurrent exposures to sFLT1 (multiplepregnancies) are also strong risk factors for PPCMeven in the absenceof pre-eclampsia2,43,44.Only a minority of women with pre-eclampsia develops PPCM,

consistent with the existence of a second hit. Abnormal PGC-1afunction is such an event in rodents, and it may also be a second hitin the case of humans. A number of previously identified processesmay also constitute this second hit, including myocarditis, immuneactivation, viral infection and/or autoantibodies43. Interestingly,PGC-1a expression is repressed by inflammatory states in the heartand elsewhere45,46, suggesting that many of the above processes thatare implicated in PPCMmay partly converge on PGC-1a. Consistentwith this, we found repressed PGC-1a expression in cardiac samplesfrom women with PPCM (Supplementary Fig. 8). Abnormal STAT3function and ROS production3 and genetic predispositions47 may alsobe contributing factors.In conclusion, the data presented here support the idea that PPCM

is partly a two-hit vascular disease due to imbalances in angiogenicsignalling, and that anti-angiogenic states such as pre-eclampsia ormultiple gestation substantially worsen the severity of the disease. Ourdata may explain why pregnancy triggers PPCM, and also the long-standing epidemiological observation that pre-eclampsia is a riskfactor for developing PPCM. Pro-angiogenic therapies such asexogenous VEGF121, or removal of sFLT1 itself48, may therefore bebeneficial in PPCM.

*

*

0

5

10a

0

0.2

0.4

0.6

0.8

* cm

0

0.1

0.2

0.3 * cm

HW/TL

LVESD FS

CT IgGCT sFLT1HKO sFLT1

100

1,000

10,000

PP CT PPCM sF

LT1

(pg

ml)!

1

b c

PPCM PPCM + pre-eclampsia

All deliveries Pre-eclampsia

e d

f

mg

mm

!1

Figure 6 | sFLT1 is sufficient to inducecardiomyopathy in HKO mice, women withPPCM have elevated sFLT1 levels, and pre-eclampsia as a risk factor for PPCM. a–c, Heartweight:tibial length ratios (a), echocardiographicfractional shortening (b) and LVESD (c) in HKOmice injected with adenovirus expressing sFLT1,versus controls. d, Elevated sFLT1 levels in post-partum women with PPCM. P5 0.002.e, f, Prevalence of pre-eclampsia among alldeliveries (e) and among women with PPCM (f) atHarvard teaching hospitals in the previous 9 years.

ARTICLE RESEARCH

1 7 M A Y 2 0 1 2 | V O L 4 8 5 | N A T U R E | 3 3 7


METHODSAnimal studies.All animal experimentswere performed according to proceduresapproved by the Institutional Animal Care and Use Committee. Mice bearingfloxed alleles of PGC-1a flanking exons 3 and 4, and mice containing thea-MHC::CRE transgene were gifts from B. Spiegelman49 and M. Schneider50,respectively. Mice were maintained on a standard rodent chow diet with 12-hlight and dark cycles. For murine echocardiography, the chest hair was removedwith a topical depilatory agent, and two-dimensional images were visualizedusing a Vivid FiVe echocardiography system (GEMedical Systems) on mice thatwere not anaesthetized. Parasternal short-axis projections were visualized andM-mode recordings at the mid-ventricular level were recorded. Heart rate,LVEDD and LVESD were measured in at least three beats from at least threerecordings and averaged, and left ventricular fractional shortening was thencalculated (fractional shortening5 (LVEDD–LVESD)/LVEDD). For the high-resolutionMPI studies, a VisualSonics 2100 echocardiographymachine was usedon mice anaesthetized with isoflurane, and the MPI was calculated using themanufacturer’s software program. SPECT/CT imaging of mice was performedby the Longwood Area Small Animal Imaging Facility (SAIF). For the VEGFtreatment studies, human VEGF121 (100mg kg21) was injected subcutaneouslydaily, versus saline as the control. For the bromocriptine studies, bromocriptinewas added to the drinking water. Mice were bred starting at the age of 8weekswhile receiving either bromocriptine in the drinking water or daily subcutaneousVEGF121, or both.Cells and reagents. All reagents were from Sigma, unless otherwise indicated.Human soluble VEGF121 was a gift from Scios. Staining of capillaries was per-formed using anti-CD31 antibody (BD Pharmingen) or isolectin B4 (Vector Lab).Quantification of capillaries was performed computationally, using Volocitysoftware (Improvision, PerkinElmer), on three random fields chosen from theseptum of transverse sections from the mid-heart. Staining of arterioles was per-formed using anti-SMA antibody (Santa Cruz) and quantified similarly, usingrandom low-power fields. All quantifications were performed blindly. Isolationand culture of primary NRVMs was performed as described. Cells were infectedwith adenovirus at amultiplicity of infection of 103 to 303, andmRNAexpressionwasmeasured 24 or 48h later. The adenovirus expressing PGC-1a and sFLT1 havebeen described51,52. Prolactin, VEGF and sFLT1 ELISA assays were from R&DSystems. The thiobarbituic acid reactive substances (TBARS) assay was performedon cardiac extracts according to the manufacturer’s instructions (Cayman).Gene expression studies.Total RNAswere isolated frommouse tissue or culturedcells using theTrizolmethod (Invitrogen). Samples for real-timepolymerase chainreaction (PCR) analyses were reverse transcribed (Invitrogen), and quantitativereal-time PCR reactions were performed on the complementary DNAs in thepresence of fluorescent dye (SYBR green) on a BioRad CFX 384 Touch real-timePCRdetection system.DNAproducts of the expected size were confirmed for eachprimer pair.Endothelial migration assay. NRVMs in 24-well plates were infected withadenovirus expressing GFP or PGC-1a for 34 h. bovine serum albumen (BSA)or sFLT1 (100 ngml21) was added to the media for 12 h. Then, 53 104 cells ofHUVECs at 53 104 were put on the upper compartment of transwells (8.0-mmpore size, Corning no. 3422) pre-warmed with EBM2 media overnight at 37 uC.HUVEC migration to the lower compartment of transwells was measured after12 h.MigratedHUVECswere fixed with 4% paraformaldehyde in PBS for 20minat 25 uC, cells remained on the upper compartment were removed with a cottonswab. Cells were blocked with 5% BSA in PBS 0.2% Tween (PBST) and stainedwith phallodin fluorescein isothiocyanate in PBST for 4 h to visualize filamentousactin. Transwell inserts were washed three times in PBST and mounted ontoslides with 49,6-diamidino-2-phenylindole (DAPI) mounting medium.Human studies. The institutional review board of Beth Israel DeaconessMedicalCenter in Boston approved this study. Eligible women were enrolled after pro-viding written informed consent from November 2009 to May 2010. Pregnant

women at least 18 years of age with a singleton pregnancy of at least 24 weeks andless than 41 weeks, and either a diagnosis of pre-eclampsia or without anyhypertensive disorder of pregnancy were eligible. Exclusion criteria includedpre-existing cardiovascular disease, pulmonary disease and non-gestationaldiabetes mellitus. Participants were recruited after admission to labour anddelivery, the ante-partum floor or during a routine prenatal visit. All clinical datawere taken frommedical records. The diagnosis of pre-eclampsia was based on theNationalHighBloodPressure EducationProgramWorkingGroupdefinition, alsoendorsed by the AmericanCongress of Obstetricians and Gynaecologists (ACOG).A maternal–fetal medicine specialist confirmed all diagnoses. Archived plasmasamples from subjects with PPCM have been previously described3. Patients inboth studies were predominantly Caucasian. Retrospective analyses of PPCMand pre-eclampsia in the Harvard teaching hospitals were performed using theHarvard SharedHealthResearch InformationNetwork (SHRINE)53, a de-identifiedrepository of aggregate patient information.Human echocardiography. Bedside transthoracic echocardiograms were per-formed using a Siemens X-300 (Mountainview) machine, by two expert echo-cardiographers using P5-1 Transducer. Images were obtained with the patientlying in the left lateral decubitus position and were reported according to theAmerican Society of Echocardiography guidelines. Images were stored in a cine-loop format with three cardiac cycles of non-compressed data with electrocar-diogram information. The echocardiographers performed a comprehensiveexamination, which included a complete two-dimensional and colour flowDoppler assessment of the left ventricle, right ventricle and intra-cardiac valves.Specifically: ejection fraction with visual quantitative estimation; trans-mitralpulse wave Doppler (E and A waves and deceleration time); Doppler tissue image(both medial and lateral mitral annulus were interrogated, and the final value ofpeak velocity of E’ was calculated as the average of three velocities at eachlocation); MPI, with the calculation performed off-line using a SiemensSyngo DICOM viewing station (Mountainview). The echocardiograms werede-identified before calculating MPI. Ejection fraction, MPI and E/E9 ratios werecalculated. Each image was analysed blind by one of two echocardiographers.Angiogenic factor assays.Women consented to a blood draw at the time of theechocardiogram. All samples for theMPI studywere collected in the ante-partumbefore the delivery, whereas samples in the PPCM study were collected 4–6weekspost-partum. The samples were centrifuged at 1,900g for 8min and plasma wascollected and stored at280uC. Samples were randomly ordered and analysed bya single person in a blind fashion. ELISA assays for sFLT1 were performed withcommercially available kits (R&Dsystems).All assayswere performed in duplicateand values were averaged. If .20% difference was observed between duplicatevalues, the samples were re-analysed.Data and statistical analysis. SAS 9.2 (SAS institute) was used for data analysis.All tests were two sided, and P values of less than 0.05were considered statisticallysignificant. Data are presented as mean 6 standard error, or median and inter-quartile ranges, as indicated. Comparisons were made using the two-tailedStudent’s t-test or the non-parametric Mann–Whitney test, as indicated.

49. Handschin, C. et al. Abnormal glucose homeostasis in skeletal muscle-specificPGC-1a knockoutmice reveals skeletalmuscle-pancreatic b cell crosstalk. J. Clin.Invest. 117, 3463–3474 (2007).

50. Agah, R. et al. Gene recombination in postmitotic cells. Targeted expression ofCre recombinase provokes cardiac-restricted, site-specific rearrangement inadult ventricular muscle in vivo. J. Clin. Invest. 100, 169–179 (1997).

51. Puigserver, P. et al. A cold-inducible coactivator of nuclear receptors linked toadaptive thermogenesis. Cell 92, 829–839 (1998).

52. Kuo, C. J. et al.Comparative evaluation of the antitumor activity of antiangiogenicproteins delivered by gene transfer. Proc. Natl Acad. Sci. USA 98, 4605–4610(2001).

53. Weber, G.M. et al.TheSharedHealth Research InformationNetwork (SHRINE): aprototype federated query tool for clinical data repositories. J. Am. Med. Inform.Assoc. 16, 624–630 (2009).

ARTICLE RESEARCH


ARTICLEdoi:10.1038/nature11040

Cardiac angiogenic imbalance leads toperipartum cardiomyopathyIan S. Patten1,2*, Sarosh Rana3*, Sajid Shahul4, Glenn C. Rowe1, Cholsoon Jang1, Laura Liu1, Michele R. Hacker3, Julie S. Rhee3,John Mitchell4, Feroze Mahmood4, Philip Hess4, Caitlin Farrell1, Nicole Koulisis1, Eliyahu V. Khankin5, Suzanne D. Burke5,8,Igor Tudorache6, Johann Bauersachs7, Federica del Monte1, Denise Hilfiker-Kleiner7, S. Ananth Karumanchi5,8 & Zoltan Arany1

Peripartum cardiomyopathy (PPCM) is an often fatal disease that affects pregnant women who are near delivery, and itoccurs more frequently in women with pre-eclampsia and/or multiple gestation. The aetiology of PPCM, and why it isassociated with pre-eclampsia, remain unknown. Here we show that PPCM is associated with a systemic angiogenicimbalance, accentuated by pre-eclampsia. Mice that lack cardiac PGC-1a, a powerful regulator of angiogenesis, developprofound PPCM. Importantly, the PPCM is entirely rescued by pro-angiogenic therapies. In humans, the placenta in lategestation secretesVEGFinhibitors likesolubleFLT1 (sFLT1), andthis is accentuatedbymultiplegestationandpre-eclampsia.This anti-angiogenic environment is accompanied by subclinical cardiac dysfunction, the extent of which correlates withcirculating levels of sFLT1. Exogenous sFLT1 alone caused diastolic dysfunction in wild-type mice, and profound systolicdysfunction inmice lacking cardiac PGC-1a. Finally, plasma samples fromwomenwith PPCM contained abnormally highlevels of sFLT1. Thesedata indicate thatPPCMismainlyavascular disease, causedbyexcess anti-angiogenic signalling in theperipartumperiod. Thedata also explain how late pregnancyposes a threat to cardiachomeostasis, andwhypre-eclampsiaand multiple gestation are important risk factors for the development of PPCM.

PPCM affects 1 in 300 to 1 in 3,000 pregnancies, with geographic hotspots of high incidence, such as Nigeria and Haiti1,2. The disease ischaracterized by systolic heart failure presenting in the last month ofpregnancy or the first 5months post-partum. Although approximatelyhalf of affected women recover cardiac function post-partum, manypatients progress to chronic heart failure, cardiac transplantation ordeath. Thus, PPCMcandevastate otherwise healthy youngwomenandtheir infants. PPCMremains a disease of unknownaetiology.Theonsetlate in gestation does not coincide with increased haemodynamic loadon the heart, suggesting that othermechanisms are responsible. Recentdatahave suggested that anti-angiogenic prolactin fragmentsmay havean important role in causing the disease in some patients3. Risk factorsfor PPCM also include pre-eclampsia and multiple gestation, suggest-ing potential mechanistic overlap with these processes1,2.PGC-1a is a transcriptional coactivator that drives mitochondrial

biogenesis and other metabolic programs in many tissues, includingthe heart4,5. PGC-1a is highly expressed in the heart, and mice lackingPGC-1a globally have abnormal cardiac energetic reserves and respondpoorly to stressful stimuli such as transverse aortic banding6,7. In additionto its role in mitochondrial homeostasis, PGC-1a also induces theexpression and secretion of pro-angiogenic factors, such as vascularendothelial growth factor (VEGF), which leads to formation of newblood vessels8,9. Although the angiogenic function of PGC-1a has beendescribed in skeletalmuscle, its role in cardiac tissue remains unexplored.

Cardiac-specific PGC-1a deletion leads to PPCMTo study further the role of PGC-1a in the heart, we generatedcardiac-specific PGC-1a knockout (HKO) mice (see Methods). While

studying these mice, we noticed that female HKOmice were fertile, anddelivered normal litter sizes (not shown), but invariably died after one ortwopregnancies (Fig. 1a). Thehearts of thesemicewere large, dilated andfibrotic (Fig. 1b–d), consistent with a dilated cardiomyopathy. Two-dimensional M-mode echocardiography revealed dilated, poorly con-tractile hearts in HKO mice after their second delivery (Fig. 1e). Leftventricular end-diastolic dimensions (LVEDD) and left ventricular end-systolic dimensions (LVESD) were markedly enlarged, and fractionalshortening, a direct measure of cardiac contractile function, was pro-foundly depressed (Fig. 1f–i). Nulliparous mice, as well as post-partumcontrol mice, were not affected. Males were also not affected(Supplementary Fig. 1). Thus, the absence of PGC-1a in cardiomyocytesleads to a profound PPCM in mice.

PGC-1a regulates angiogenesis in cardiac tissueWe have recently shown in skeletal muscle that PGC-1a regulatesangiogenesis by driving the expression of angiogenic factors likeVEGF8,9.Anti-angiogenic therapies, including antibodies that neutralizeVEGF and small-molecule VEGF receptor inhibitors, are increasinglybeing used in oncological and ophthalmological treatments, andcardiomyopathy and heart failure have recently been recognized asimportant side effects10,11, showing that anti-angiogenic therapy canbe harmful to the heart in humans. Impaired VEGF signalling has alsobeen linkedwith cardiac dysfunction inmice12,13. At the same time, latepregnancy is a strong anti-angiogenic environment, partly owing to thesecretion by the placenta of anti-angiogenic factors, like sFLT1, thatbind to and neutralize soluble members of the VEGF family14. Theseobservations led us to postulate that PGC-1a regulates an angiogenic

*These authors contributed equally to this work.

1Cardiovascular Institute, Beth Israel Deaconess Medical Center, Harvard Medical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 2Center for Vascular Biology Research, Beth IsraelDeaconess Medical Center, Harvard Medical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 3Division of Maternal Fetal Medicine/Department of Obstetrics and Gynecology, Beth IsraelDeaconessMedical Center, HarvardMedical School, 330 Brookline Avenue, Boston, Massachusetts 02115, USA. 4Department of Anesthesia and Critical Care, Beth Israel DeaconessMedical Center, HarvardMedicalSchool, 330BrooklineAvenue,Boston,Massachusetts02115,USA. 5DivisionofNephrology/DepartmentofMedicine,Beth IsraelDeaconessMedicalCenter andHarvardMedicalSchool, 330BrooklineAvenue, Boston, Massachusetts 02115, USA. 6Department of Cardiothoracic, Transplantation andVascular Surgery,MedizinischeHochschuleHannover, Carl-NeubergStrasse, D-30625Hannover, Germany.7Department of Cardiology and Angiology, MedizinischeHochschule Hannover, Carl-Neuberg Strasse, D-30625 Hannover, Germany. 8Howard Hughes Medical Institute, 330 Brookline Avenue, Boston,Massachusetts 02115, USA.

1 7 M A Y 2 0 1 2 | V O L 4 8 5 | N A T U R E | 3 3 3


Importance of real-time

exploration

i2b2.org

smartplatforms.org

Systems Informatics Susanne Churchill

Rachel Ramoni Shawn Murphy Griffin Weber

Doug MacFadden Bill Simons

Mike Mendis Nich Wattanasin

Stan Shaw Peter Szolovits

Kat Liao Robert Plenge

Tianxi Cai Finale Doshi Roy Perlis

Jordan Smoller David Kreda Josh Mandel

Daria Prilutsky David Margulies

Bioinformatics Nathan Palmer Patrick Schmid Sek Won Kong

Luke Hutchinson Bonnie Berger

Malcolm Campbell

Genomics Lou Kunkel Alal Eran

Developmental Medicine Lenny Rappaport Chuck Nelson

Stephanie Brewster

Population Science Ken Mandl Ben Reis

John Brownstein

Documents

i2b2: Transla3onal Toolkit