High-Throughput Investigation ofEndothelial-to-Mesenchymal Transformation(EndMT) With Combinatorial Cellular Microarrays

Zongjie Wang,1 Blaise Calpe,2,3,4 Jalil Zerdani,3,4,5 Youngsang Lee,6

Jonghyun Oh,3,7,8 Hojae Bae,3,8,9 Ali Khademhosseini,3,4,8,10 Keekyoung Kim1,3,8

1School of Engineering, University of British Columbia, Kelowna, BC V1V1V7, Canada;

telephone: þ1-250-807-8040; fax: þ1-250-807-9850; e-mail: keekyoung.kim@ubc.ca2Institute of Molecular Health Sciences, ETH Z€urich, Z€urich, Switzerland3Center for Biomedical Engineering, Brigham and Women’s Hospital, Harvard Medical

School, Cambridge, Massachusetts4Wyss Institute for Biologically Inspired Engineering at Harvard University, Boston,

Massachusetts5Institute of Bioengineering, �Ecole Polytechnique F�ed�erale de Lausanne, Lausanne,

Switzerland6Department of Mathematics and Statistics, University of British Columbia, Kelowna, BC,

Canada7Division of Mechanical Design Engineering, Chonbuk National University, Jeonjoo,

Republic of Korea8Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of

Technology, Cambridge, Massachusetts 02139; telephone: þ1-617-768-8395;

fax: þ1-617-768-8477; e-mail: alik@rics.bwh.harvard.edu9Department of Bioindustrial Technologies, College of Animal Bioscience and

Technology, Konkuk University, Hwayang-dong, Kwangjin-gu, Seoul, Republic of Korea10Department of Physics, King Abdulaziz University, Jeddah 21569, Saudi Arabia

ABSTRACT: In the developing heart, a specific subset ofendocardium undergoes an endothelial-to-mesenchymal transfor-mation (EndMT) thus forming nascent valve leaflets. Extracellularmatrix (ECM) proteins and growth factors (GFs) play importantroles in regulating EndMT but the combinatorial effect of GFs withECM proteins is less well understood. Here we use microscaleengineering techniques to create single, binary, and tertiarycomponent microenvironments to investigate the combinatorialeffects of ECM proteins and GFs on the attachment and

transformation of adult ovine mitral valve endothelial cells to amesenchymal phenotype. With the combinatorial microenviron-ment microarrays, we utilized 60 different combinations of ECMproteins (Fibronectin, Collagen I, II, IV, Laminin) and GFs(TGF-b1, bFGF, VEGF) and were able to identify newmicroenvironmental conditions capable of modulating EndMT inMVECs. Experimental results indicated that TGF-b1 significantlyupregulated the EndMTwhile either bFGF or VEGF downregulatedEndMT process markedly. Also, ECM proteins could influence boththe attachment of MVECs and the response of MVECs to GFs. Interms of attachment, fibronectin is significantly better for theadhesion of MVECs among the five tested proteins. Overall collagenIV and fibronectin appeared to play important roles in promotingEndMT process. Great consistency between macroscale andmicroarrayed experiments and present studies demonstrates thathigh-throughput cellular microarrays are a promising approach tostudy the regulation of EndMT in valvular endothelium.Biotechnol. Bioeng. 2015;9999: 1–10.� 2015 Wiley Periodicals, Inc.KEYWORDS: endothelial-to-mesenchymal transformation; heartvalve; high-throughput; microarray

Zongjie Wang and Blaise Calpe contributed equally to this work.

Correspondence to: A. Khademhosseini and K. Kim

Contract grant sponsor: Natural Sciences and Engineering Research Council of

Canada (NSERC) Discovery

Contract grant number: RGPIN-2014-04010

Contract grant sponsor: US Army Engineer Research and Development Center,

Institute for Soldier Nanotechnology, NIH

Contract grant numbers: EB009196; DE019024; EB007249; HL099073; AR057837

Contract grant sponsor: National Science Foundation CAREER award (AK)

Received 23 September 2015; Revision received 3 December 2015; Accepted 7

December 2015

Accepted manuscript online xx Month 2015;

Article first published online in Wiley Online Library

(wileyonlinelibrary.com).

DOI 10.1002/bit.25905

ARTICLE

� 2015 Wiley Periodicals, Inc. Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015 1

Introduction

About 1% of the population suffers from congenital heart valvemalformations. This condition can be life threatening and force thepatient to undergo heart valve replacement with a biological ormechanical valve (Butcher and Markwald, 2007), which account for70,000 surgeries each year in theUnited States (Roberts andKo, 2005;Supino et al., 2004). Unfortunately, both types of prosthetic valveshave drawbacks, as mechanical valves are associated with a risk ofthrombosis and infection (Angell and Angell, 1980; Hoerstrup et al.,2000; Kuntze et al., 1998; Seiler, 2004), while biological valves areprone to degradation and have a poor longevity (Flanagan andPandit, 2003; Paruchuri et al., 2006). Thus, there is an important needfor new therapeutic approaches to treat heart valve diseases. Possibleavenues of research include the development of tissue engineeredheart valves from autologous cells or the modulation of signalingpathways involved in the pathogenesis of the valve (Armstrong andBischoff, 2004; Hinton and Yutzey, 2011). These two approachescould both benefit greatly from a better understanding of theregulation of valve development and remodeling.

In early heart valve development, a subset of endothelial cellsundergoes an endothelial-to-mesenchymal transformation (EndMT)which is triggered by secretions of members of the transforminggrowth factor-beta (TGF-b) family and bone morphogenetic protein(BMP) -2, 4 (Kalluri and Weinberg, 2009; Zeisberg et al., 2007). TheEndMT causes a loss of cell–cell contact adhesion and increases theexpression of a-smooth muscle actin (SMA) and cell migration(Barnett and Desgrosellier, 2003; Srivastava and Olson, 2000).Migratory cells invade and remodel the surrounding extracellularmatrix (ECM) called cardiac cushion and transform to valvularinterstitial cells (VICs), leading to the formation of the nascent valveleaflets (Armstrong and Bischoff, 2004; Butcher andMarkwald, 2007;Paruchuri et al., 2006). However, malfunctioned transition into heartvalve causes congenital valve defects (Hsu and Pearson, 2009). Thesedefects account for cardiovascular diseases such as mitral valvestenosis, mitral valve regurgitation or calcified aortic stenosis(Apostolakis and Baikoussis, 2009; Avierinos et al., 2002; Chua andKalb, 2006; Turi, 2004).

Although EndMT is predominantly observed during embryonicdevelopment, it has been shown that adult ovine mitral valveendothelial cells (MVECs) can undergo EndMT; and the transformedcells have multi-lineage mesenchymal differentiation potential(Wylie-Sears et al., 2011). These results suggest that the adultendothelium contains a reservoir of progenitor cells that couldreplenish VICs in postnatal heart valves. The endothelium of adultvalves is normally quiescent. However, an important population ofcells undergoing EndMT was found in the mitral valve of an ovinemodel ofmitral regurgitation, underlying a possible role of EndMTinthe adaption of the valve to injury (Dal-Bianco et al., 2009). Aprevious study on adult human pulmonary valve endothelial cellsshowed that TGF-b2 inducing EndMT is markedly enhanced whenthese cells are cultured on fibronectin, indicating a regulatory role ofECM proteins (Paruchuri et al., 2006). Even though several growthfactors (GFs) such as TGF-b family members and vascularendothelial growth factor (VEGF) have been shown to regulateEndMT (Riem Vis et al., 2011), their combinatorial effect with ECMproteins is less well understood.

One of the major challenges in analyzing the effects of themicroenvironment on cellular behavior is the experimental diversityresulting from even a small set of independent variables. Moreover,this problem becomes even more complex when interactions amongindependent variables are considered. Because traditional culturemethods are not well suited to investigate a variety of experimentalparameters, many efforts using microscale technologies were madeto address the problem. Themost widely usedmicroscale technologyto study various cell microenvironmental parameters is high-throughput cellular microarray platforms. The microarray uses onlynanoliter volume of samples, and thus reduces the usage of expensivereagents. Therefore, it can facilitate to screen large numberof samplesin a high-throughput way and to investigate the effect betweenexperimental parameters (Fernandes et al., 2009; Howbrook et al.,2003; Nicholson et al., 2007).

Cell-based microarray platforms were employed for drugscreening and toxicology studies (Kwon et al., 2011; Lee et al.,2005, 2008; Wu et al., 2011) and immunofluorescence assay(Fernandes et al., 2008). In addition, the microarray platforms werea great tool to study stem cell differentiation in the combinatorialeffect of microenvironments (Anderson et al., 2004; Brafman et al.,2012; Dolatshahi-Pirouz et al., 2014; Flaim et al., 2005; Gobaa et al.,2011; Klim et al., 2010; Mei et al., 2010; Soen et al., 2006). Theseplatforms enabled the identification of cell regulating microenvi-ronmental parameters that would have been difficult to find withconventional cell culture methods (Charnley et al., 2009) andsignificantly reduce the preparation and optimization process ofnew biomaterials (Dolatshahi-Pirouz et al., 2011).

In this article, we introduced a high-throughput combinatorialmicroarray platform to investigate the microenvironmental regula-tion of EndMT in the heart valve. The combinatorial microarray aimsto uncover the combination of signal regulating EndMTin heart valvethat is necessary for the heart valve development. Here, we used fiveECM proteins (ECMPs) present in the heart valve (Jutta Schaper,1992): Collagen I (C1), II (C2) and IV (C4), Fibronectin (Fn), Laminin(Lm), and three GFs: TGF-b1, VEGF, and basic fibroblast growthfactor (bFGF). These five ECMPs were combined with the fourconditions of the GFs. We seeded mitral valve endothelial cells(MVEC) on these arrays and used immunostaining and highthroughput image analysis to investigate how EndMT can bemodulated by the combinations of ECMPs and GFs.

Materials and Methods

MVECS Culture

Clonally derived ovine MVECs were kindly provided by Bischofflaboratory at Harvard Medical School. Cells were cultured on 1%gelatin coated flasks and fed every 2 days with an endothelialbasal media (Lonza, Basel, Switzerland) supplemented with 10%heat-inactivated fetal bovine serum (GE Healthcare Hyclone,Little Chalfont, United Kingdom), 1% penicillin-streptomycin(Life Technologies, Carlsbad, CA), and 2 ng/mL bFGF (RocheDiagnostics, Indianapolis, IN) as described in (Wylie-Searset al., 2011). To test EndMT process on microarray, cells werecultured in the culturing media with 5 ng/mL of VEGF (R&Dsystems, Minneapolis, MN), bFGF, or TGF-b1 (R&D Systems,

2 Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015

Minneapolis, MN), respectively and without GFs as a controlgroup.

Analysis of Endothelial-to-Mesenchymal Transformationin Macroscale

To verify the concept for the microscale high-throughput experi-ments, we conducted macroscale EndMT experiments with a fewdifferent conditions of the GFs and ECMPs as shown Figure 1. Weseeded MVECs (10,000 cells/cm2) on glass bottom dishes (MatTekCo., Ashland, MA) coated with ECMPs and grew them for 1, 3, and 5days in a typical culturing condition (2 ng/mL bFGF[þ]) as a controlgroup and a culturing condition with 5 ng/mLTGF-b1 and withoutbFGF. Three sets of six dishes were used for the experiment. The firstset of dishes was coated with gelatin from porcine skin (Sigma–Aldrich, St. Louis, MO), the second set with C1 from rat tail (BDBioscience, Franklin Lakes, NJ), and the last set with Fn from bovineplasma (Sigma–Aldrich). After culturing for 1, 3, and 5 days, cellswere fixed in 4% paraformaldehyde (Sigma–Aldrich) for 20min and

washed three times with phosphate buffered saline (PBS) solution.They were then permeabilized and blocked with a solution of 0.5%Triton-X (Sigma–Aldrich) and 1% bovine serum albumin (Sigma–Aldrich). This step was performed for 15min in the case of SMAstaining and only 2min for VE-Cadherin staining. Sampleswere thenwashed three times with PBS, incubated with primary antibodies,such asmouse anti-SMA (dilution 1:500) or rabbit anti-VE-Cadherin(dilution 1:100) both fromAbcam (Cambridge, UnitedKingdom), for1 h, and washed again three times with PBS. Samples were thenincubated with secondary antibodies, such as Alexa Fluor1 goatanti-mouse 546 (dilution 1:200) or Alexa Fluor1 chicken anti-rabbit488 (dilution 1:200) both from Life Technologies (Carlsbad, CA), for40min while being protected from light with an aluminum foil. Afterthree times wash with PBS, phalloidin with 1:40 dilution ratio (AlexaFluor1 647 phalloidin, Life Technologies) was added for 30min foractin staining. Finally, samplesweremountedwith amountingmedia(VECTASHIELD1, Vector Laboratories, Burlingame, CA) beforebeing imaged and analyzed with both an inverted fluorescentmicroscope (Axio Observer, Carl Zeiss, Oberkochen, Germany) and a

Figure 1. Macroscale study of MVEC’s EndMT process. (A) Immunocytochemical staining images of MVECs cultured for 5 days on 1% gelatin in control group (2 ng/mL bFGF)

and TGF-b1(þ) group (5 ng/mL of TGF-b1). Scale bar¼ 100mm. (B) Fluorescent confocal images ofMVECs cultured on Fn, C1, and gelatin for 1, 3, and 5 days (D1, D3, D5) in TGF-b1(þ)

group. Scale bar¼ 100mm. (C) Quantification of SMA(þ) MVECs cultured on Fn, C1, and gelatin for 1, 3, and 5 days in control group and TGF-b1(þ) groups (�P< 0.05). (D) qPCR

quantification of SMA and VE-Cadherin expression of MVECs cultured on Fn, C1, and gelatin for 5 days in control group and TGF-b1(þ).

Wang et al.: High-Throughput Investigation of Endothelial-to-Mesenchymal Transformation (EndMT) 3

Biotechnology and Bioengineering

confocal microscope (SP5 X MP, Leica Microsystems, Wetzlar,Germany) as shown in Figure 1A, B, and C.

For transcription analysis by qPCR, MVECs were seeded on 6-well plates coated with 100mg/mL of C1, Fn or gelatin at a density of10,000 cells/cm2 and grown 5 days in a typical culturing condition(2 ng/mL bFGF [þ]) as a control group and a culturing conditionwith 5 ng/mL TGF-b1 and without bFGF. Total RNA was extractedwith RNeasy Mini Kit (Quiagen, Venlo, Netherlands). cDNA wassynthesized using qScriptTM cDNA Synthesis Kit according to themanufacturer’s instructions and amplified using SMA (forward:tgccatgtatgtggctattca, reverse: accagttgtacgtccagaagc), VE-CAD(forward: acatccgtggttctggactc, reverse: agatggggaagttgtcgttg) andGAPDH (forward: ggcctccaaggagtaaggtc, reverse: cgggagattct-cagtgtggt) specific primers. Finally, gene expression was normalizedto GAPDH expression (Fig. 1D).

Fabrication of Extracellular Matrix Microarrays

Arrays were fabricated as previously described (Bauer et al., 2012;Brafman et al., 2009; Flaim et al., 2005). The schematic of

fabrication process is described in Figure 2A. Briefly, standardmicroscope glass slides (25� 75mm) were cleaned with 10 wt%NaOH, silanized with 3-(trimethoxysilyl) propyl methacrylate(TMSPMA, Sigma–Aldrich), and coated with a thin polyacrylamide(PAA) pad cross-linked by ultraviolet (UV) light. A PAA pre-polymer solution was prepared by mixing 9.5% (w/v) aclylamide(Sigma–Aldrich), 0.5% (w/v) N,N’-Methylenebisacrylamide (bis,Sigma–Aldrich), and 20mg/mL 1-[4-(2-hydroxyethoxy) phenyl]-2-hydroxy-2-methyl-1-propane-1-one (Irgacure 2959, BASF, Lud-wigshafen, Germany) in methanol. Fn, Lm, C1, C2, and C4 dilutedin a printing buffer were prepared in a 384 well plate. Microarrayprinting was performed using a contact microarray printer (SpotBot3, Arrayit, Sunnyvale, CA).

To evaluate the quality of printed proteins, an array of C1:Fncombinations was printed on a PAA coated slide and immunos-tained. Briefly, the slide was washed three times in PBS and blockedfor 1 h in 1% (w/v) BSA. The slide was then incubated with primaryantibodies against C1 (rabbit anti collagen I Ab292, dilution 1:200)and FN (mouse anti fibronectin Ab26245, dilution 1:200) for 1 h,washed three times with PBS, and incubated with secondary

Figure 2. Combinatorial ECM microarray. (A) Schematic of ECM microarray fabrication. (B) Immunostained images of printed spots of C1:Fn binary combination (150mg/mL

each). Scale bar¼ 200mm. (C) Quantification of fluorescent intensity obtained from printed spots of different C1 concentrations.

4 Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015

antibodies (Alexa Fluor chicken anti rabbit 488, dilution 1:200 andAlexa Fluor goat anti mouse 546, dilution 1:200) for 30min. Theslide was then imaged using the inverted fluorescent microscope(Axio Observer, Carl Zeiss, Oberkochen, Germany). In addition,spots of C1 ranging from 50–250mg/mL were printed andimmunostained to examine the effectiveness of printing proteinswith different concentrations. The fluorescence intensity was thenmeasured using CellProfiler (Carpenter et al., 2006).For the investigation of EndMT process, we tested 60 different

combinations of ECMPs (C1, C2, C4, Fn, and Lm) and GFs(TGF-b1, bFGF, and VEGF). We chose 5 ng/mL concentration ofTGF-b1, bFGF, or VEGF and no GFs as a control group in themicroscale EndMT study since 2–10 ng/mL of TGF-b1, bFGF, orVEGF were typically used in the previous macroscale EndMTstudies (Paruchuri et al., 2006; Wylie-Sears et al., 2011). Totalprotein concentration of 300mg/mL (300mg/mL for individualproteins and 150mg/mL each for the binary combinations) wasdiluted in the printing buffer consisted of 100mM acetate, 5 mMEDTA, 20% (v/v) glycerol, and 0.25% (v/v) Triton X-100 to preventprotein aggregation during printing (Flaim et al., 2005). With thefabricated microarrays, we were able to identify new microenvi-ronmental conditions capable of modulating EndMT in MVECs.

Microscopy and Analysis of Fluorescence Images

As mentioned in analysis of endothelial-to-mesenchymal transfor-mation in macroscale Section, we immunostained the SMAs, actinfilaments and nuclei of MVECs cultured on microarrayed 15combination of ECMPs. Series of fluorescent images of immunos-tained samples were acquired using the automated stage of aninverted fluorescent microscope and were analyzed using the cellimage analysis software CellProfiler. Before analysis, the back-ground was removed in the DAPI and the actin channel(sigma¼ 200 pixels) with a script written in ImageJ (NationalInstitutes of Health, Bethesda) and the SMA channel wasnormalized using a linear transformation into an intensity scalefrom 0 to 2000 y ¼ 2000

max�min � ðx� minÞ� �with MATLAB (Math-

works, Natick, MA). The normalization step proved to be a crucialfor the quality of analysis using CellProfiler. The algorithm used todetect the nuclei in CellProfiler was “Background Global” and “OtsuGlobal” to identify the cell bodies. The range of detection for nucleiwas set between 10 and 60 pixels. After a first pass with a thresholdof 0.005 for the classification, the images were separated into fivecategories depending on their percentage of cells expressing SMA:0–19%, 20–39%, 40–59%, 60–79% and 80–100%. Then, a differentthreshold was applied to each of these categories to get the mostaccurate results, typically between 0.004 and 0.009. This methodwas the one that gave the best results.

Statistical Analysis

One way ANOVAwas used to statistically analyze the main effects ofECMPs and GFs for cell attachment. Factorial analysis was used toanalyze the main and interaction effects of two and three factors ofECMPs and GFs for EndMT regulation using scripts written in R3.1.1 (The R Foundation for Statistical Computing, Zurich,Switzerland). Results were shown as average� standard error.

Results and Discussion

Macroscale Study of EndMT

To evaluate the effects of ECMPs on GFs triggered EndMT, weculturedMVECs with two different conditions (bFGF[þ] as a controland TGF-b1[þ]) on dishes coated with Fn, C1 and gelatin.Interestingly, different rates of EndMT could be observed, warrantinga larger scale studywithmore combinations of ECMPs and GFs. After5 days in culture, MVECs in the control group exhibited typicalcobblestone endothelial morphology and VE-Cadherin along cell–cell junctions (Fig. 1A). On the contrary, SMAs were observed in�20% of cells and VE-cadherin was visibly downregulated in thepresence of TGF-b1 (as judged by immunocytochemical analysis;typical images presented in Fig. 1A).Figure 1B shows the SMA expression of MVECs on the Fn, C1 and

gelatin in the presence of TGF-b1 after 1, 3, and 5 days of culture.Based on the quantification analysis presented in Figure 1C, theSMA expression was not significant after 1 day of culture (<1%SMAþ) for both control group and TGF-b1(þ). However, after 3days, the SMA expression was significantly different between TGF-b1(þ) and control group. In the control group, none of protein-coated substrates induced SMA expression greater than 10%. Withthe presence of TGF-b1, regardless of protein substrate modifica-tion (Fn, C1, or gelatin), SMA fibers were observed to have a greatincrement, indicating the growth of overall MVEC population (Fig.1C). After 5 days, the SMA expression was higher than 10% for alltypes of the proteins. The difference between the control group andTGF-b1(þ) indicates that TGF-b1 plays an important role inregulating the SMA expression and EndMT process. UpregulatingSMA expression with TGF-b1 has been reported previously (Dahaland Mahler, 2013; Riem Vis et al., 2011). On the other hand, after 3or 5 days in TGF-b1(þ) group, it turned out that the percentage ofSMAþ cells on Fn and gelatin substrates was significantly higherthan the one cultured on C1 (Fig. 1C). Such difference in SMAexpression remained at 3 and 5 days when comparing Fn, gelatinand C1, indicating that the EndMT process was also regulated byECMPs combined with TGF-b1.The upregulation of SMA expressionwith TGF-b1 was verified by

qPCR analysis (Fig. 1D). The relative expression of SMA and VE-cadherin supports the observation of EndMT after 5 days. Incomparison between the control group and TGF-b1(þ), SMAexpression on all protein substrates was upregulated in the presenceof TGF-b1 and there was no significant difference in the relativeexpression of VE-cadherin. These results are expected as only asubset of the MVEC population undergoes EndMT and expressesSMA and the remainder maintains endothelial phenotype with VE-cadherin expression. Fn or gelatin in the control group alsoupregulated SMA expression and downregulated VE-cadherinexpression in comparison with C1. This phenomenon indicates thatFn and gelatin may upregulate the EndMT process regardless ofTGF-b1. However, from the protein level as shown by theimmunocytochemical analysis (Fig. 1C), Fn or gelatin in the controlgroup did not upregulate SMA expression. This discrepancy can beexplained as mRNA level does not always reflect the protein levelbecause of various post-translational regulation mechanisms(Maier et al., 2009). Taken together, macroscale study revealed

Wang et al.: High-Throughput Investigation of Endothelial-to-Mesenchymal Transformation (EndMT) 5

Biotechnology and Bioengineering

that the GFs, ECMPs, and interaction between ECMPs and GFssignificantly affected EndMT process. Therefore, the furtherinvestigation of different GFs, ECMPs, and the combinatorial effectof ECMPs and GFs for EndMT is required. The high-throughputmicroarray is essential to systematically investigate the effects ofvarious parameters such as main effect and interaction effect of GFsand ECMPs in heart valve development.

Study of EndMT on Microarray

We first examined the quality of protein spots printed with themicroarray printer. Figure 2B shows representative immunostainedimages of printed Fn (red channel) and C1 (green channel) binarycombination spots. In the merged images, Fn was not clear due toits relatively weak fluorescent intensity. However, from the singlechannel images, both Fn and C1 were uniformly distributed over thespots, indicating that C1 and Fn were well mixed in the printedspots. Also, we demonstrated that the microarray printing was ableto effectively control concentrations of C1 greater than 150mg/mLas shown in Figure 2C.

To systematically investigate the effect of the microenvironmenton EndMT, microarrays of single and binary combinations of Fn,Lm, C1, C2, and C4 were produced with a robotic microarraycontact printer. Fifteen combinations of ECMPs (with 10repetitions) were printed per glass slide, for a total of 150 spots,each with a diameter of �300mm. Microarrays were seeded withMVECs and supplementedwith VEGF, bFGF or TGF-b1 dissolved inthe media. A fourth group of microarray was cultured without GFs,for a total of 60 different conditions of microenvironment. Because

of the cell repellant nature of the polyacrylamide pad, MVECsadhered only to the printed ECMPs, forming well defined cellularislands, although small clumps of cells were sometimes observedoutside of the spots. Microarrays were immunostained at 3 dayswith DAPI, anti-SMA and phalloidin and imaged with a bright fieldmicroscope equipped with an automatic stage. A custom imageanalysis routine developed with the software CellProfiler was usedfor nuclei counting and detection of cells expressing SMA. Thisprogram employed a thresholding method (Otsu adaptative) toidentify nuclei and cells body using DAPI and phalloidinfluorescence images, respectively. Detected cells were then classifiedaccording to their SMA fluorescence pixel intensity. Spots with nocells or presenting only a few clumped cells were discarded fromfurther analysis.

Effect of the Microenvironment on Cell Attachment

MVECs were able to adhere to all ECM spots but the number ofattached cells varied significantly from one protein combination tothe other (Fig. 3A). The important standard deviation observed inthe cell count can largely be explained by the unavoidable variabilityof cell-based assays.

MVEC attachment was assessed over 48 h. From days 1 to 3,average numberofMVECs permicroarrayed spot grew from 29 to 71,showing a global increase of the cell population (Fig. 3B). One-wayANOVA analysis was used to determine main effect of ECM on celladhesion (Fig. 3C). These results indicate that Fn significantlyupregulated cell adhesion, resulting in the increase of the cellpopulation. Also, the presence of either VEGF or bFGF is adverse to

Figure 3. MVEC attachment analysis. (A) An array example of a subset block of MVECs without GFs. (B) Quantification of MVECs per microarrayed spot at day 3 after seeding.

Vertical dot lines indicate the average number of MVECs per microarrayed spot at day 1 (white) and day 3 (red). (C) The main effects of cells per microarray spot after 3 days of

culture (�P< 0.05, ��P< 0.005, ���P< 0.0005, n> 20).

6 Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015

MVECs’ proliferation and their influences are stronger than theimpacts ofmost of ECMPs. Therefore, Fn plays a significantly positiverole on MVEC adhesion and growth among all GFs and ECMPs usedin the experiments. Previous studies reported that GFs significantlyinfluencedMVECs’ attachment, proliferation andmigration (Butcheret al., 2007; Riem Vis et al., 2011; Sugi et al., 2003). Theseexperimental results, especially the dominant effects of GFs forMVECs’ attachment, were matched well with previous studies.

Effect of the Microenvironment on EndMT

Images of representative spots on Fn with control, VEGF (þ), andTGF-b1 (þ) conditions are shown (Fig. 4A). The SMA expressionwas more pronounced on the outer edges of the microarray spots.

This may be explained as cells on the outer edge can be moremigratory and lose cell–cell contact to make them more likely toundergo EndMT (Barnett and Desgrosellier, 2003; Srivastava andOlson, 2000). However, cells in the center can receive inhibitorysignals because of relatively tight cell–cell contacts and are lesssusceptible to undergo EndMT. This phenomenonwas not observedin the macroscale study since cells can migrate randomly andundergo EndMT without any patterns.The quantification of SMA expression through immunocyto-

chemical analysis is presented in Figure 4B. Compared to the controlgroup, SMA expression was significantly enhanced with thepresence of TGF-b1 while SMA expression was reduced in thepresence of either VEGF or bFGF. The effects of GFs can be observedfrom another perspective (scatter plot) as well. In Figure 4C, each

Figure 4. Quantification of SMA expression. (A) Images of representative spots on Fn with control, VEGF(þ), and TGF-b1(þ) conditions. Scale bar¼ 100mm. (B) SMA

expression for each protein combination in TGF-b1(þ), VEGF(þ), bFGF(þ), and control group. (C) Scatter diagram of the effect of GFs to cell attachment and SMA expression. Each

dot represents one microarrayed spot. Vertical solid line shows the averaged cell number from control group in day 3 and horizontal solid line the averaged SMA expression

percentage from control group in day 3.

Wang et al.: High-Throughput Investigation of Endothelial-to-Mesenchymal Transformation (EndMT) 7

Biotechnology and Bioengineering

dot represents one microarrayed spot. Most of the dots in bFGF (þ)and VEGF (þ) are in the area below the average of cell number andthe average of SMA expression, indicating that those two GFs playgenerally negative roles in both cell attachment and EndMTregulation. Finally, for TGF-b1, most of the dots are with the SMAexpression value higher than the average, showing its positive effectin stimulating EndMT process.

To investigate the different cell microenvironmental parametersand take advantage of the combinatorial design of the microarray,a factorial analysis was applied to statistically analyze the mainand interaction effects between ECMPs and GFs to SMAexpression (Fig. 5). The main effects (Fig. 5A) illustrates thatthe positive effect of TGF-b1 and the negative effect of VEGF andbFGF to SMA expression are all statistically significant. Theseresults are consistent with the previous discussion and severalreported researches, for example, the study (Dahal and Mahler,2013; Riem Vis et al., 2011) for the effect of TGF-b1 and the study(Riem Vis et al., 2011) for the effect of VEGF. Compared to themain effects of ECMPs, the influence of GFs dominates the SMAexpression. Among ECMPs, C4 induced more SMA expressionthan other ECMPs. In addition, the interaction effect of C4 and Fnsignificantly upregulated SMA expression (Fig. 5B). On the other

hand, the interactions between individual protein and GF areminor so that their contributions to SMA expression are notsignificant (Fig. 5C). However, when considering the three factorinteractions between protein, protein and GF (Fig. 5D), some ofthe combinations show a significantly positive effect to SMAexpression (e.g., C1:Fn:TGF-b1, C1:C2:TGF-b1, and C4:Fn:TGF-b1). Taken together, the presence of TGF-b1 significantlypromoted the SMA expression and thus stimulates the EndMTprocess. Furthermore, C4 and Fn over all ECMPs plays the majorroles in SMA expression. Those results are well matched with theresult from the macroscale experiment. In contrast, the presenceof VEGF or bFGF apparently inhibited the EndMT process.Considering the perspective of regulating EndMT process, wesuggest that two different strategies: First, the combinations ofECMPs (i.e., C1:Fn, C1:C2, and C4:Fn) with TGF-b1 is used toaccelerate the EndMT Process because most of the interactioneffects between ECMPs are positive. Second, single ECMP is usedwith either FGF or VEGF to suppress EndMT process.

It is notable that although the inhibitory effect of bFGF is muchmore significant than VEGF, unlike for VEGF, there is nosystematical study on the impact of bFGF on EndMT process atthe cellular level. bFGF may cause other unclear effects which

Figure 5. Statistical analysis for the combinatorial effect of ECMPs and GFs on SMA expression. (A) Main effect of ECMPs and GFs. (B) Interaction effect of binary combination

ECMPs. (C) Interaction effect of individual ECMPs and GFs. (D) Interaction effects of binary combination ECMPs and GFs (���P< 0.005, ��P< 0.01, �P< 0.05, n>20).

8 Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015

require further study. Also, it has been reported that a mechanicalstimulus can also induce EndMT as well as valvular morphogen-esis and stratification (Biechler et al., 2010; Combs and Yutzey,2009; Riem Vis et al., 2011). In order to systematically studyparameters, as well as complex interaction effects regulatingEndMT process, recently developed microarrays with controllablestretch function can be utilized (Moraes et al., 2013; Winkleret al., 2014). The data presented in this study will be useful toguide further investigations of EndMT process for heart valvedevelopment.In summary, this microarray platform provided a high-

throughput method to screen and study complex cell–ECMP-GFinteractions and was well-suited for the investigation of EndMTprocess in heart valve development. By taking into considerationmore parameters regulating the EndMT, further studies usingmicroarray platforms will help elucidating heart valve developmentmechanisms which can potentially help generating tissueengineered heart valves for transplantation.

Conclusion

In this article, a single, binary, and tertiary componentmicroarrays platform is presented to investigate the combinato-rial effects of ECMPs and GF on the attachment and EndMT ofadult ovine MVECs. Compared to traditional macroscalemethods, this microarray has the advantages of high-throughput, low cost, and minimum chemical usage. Thismicroarray is also highly compatible with cell screening andanalysis, which allows the study of complex combinations ofECMPs and GFs to EndMT regulation. The experimental resultsindicate that VEGF and bFGF reduced the attachment of MVECsin microarray environments. Among five types of proteins, Fnsignificantly upregulated MVEC adhesion and growth. Inaddition, the presence of VEGF and bFGF significantlynegatively influenced the EndMT while TGF-b1 promotedEndMT process markedly. The consistency of our results withprevious results and macroscale study proves the feasibility ofemploying this microarray as a high-throughput, low-costplatform for screening ECMPs and GFs influencing MVECsdifferentiation and attachment. Also, it is notable that some ofthe binary interaction effects between ECMPs and the tertiaryinteraction effects between ECMPs and GFs further promotedthe SMA expression. Those combinations include C4:Fn, C1:Fn:TGF-b1, C2:C4:TGF-b1, and C4:Fn:TGF-b1. The positive effectsof such combinations also play important roles in upregulatingEndMT process, which can be further employed to manipulatethe EndMT process more effectively.

This work was supported by the Natural Sciences and Engineering ResearchCouncil of Canada (NSERC) Discovery Grant (RGPIN-2014-04010). Thisresearch was funded by the US Army Engineer Research and DevelopmentCenter, the Institute for Soldier Nanotechnology, the NIH (EB009196;DE019024; EB007249; HL099073; AR057837), and the National ScienceFoundation CAREER award (AK). Authors wish to thank Prof. JoyceBischoff ’s group at Boston Children’s Hospital, Harvard Medical School forproviding MVEC cells and helping qPCR analysis. Finally, authors thankProf. Ian Wheeldon at UC Riverside for his valuable discussion and ideaabout this research.

References

Anderson DG, Levenberg S, Langer R. 2004. Nanoliter-scale synthesisof arrayed biomaterials and application to human embryonic stemcells. Nat Biotechnol 22:863–866. http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd¼Retrieve&db¼PubMed&dopt¼Citation&list_uids¼15195101

Angell WW, Angell JD. 1980. Porcine valves. Prog Cardiovasc Dis 23:141–166. http://www.ncbi.nlm.nih.gov/pubmed/6774381

Apostolakis EE, Baikoussis NG. 2009. Methods of estimation of mitral valveregurgitation for the cardiac surgeon. J Cardiothorac Surg 4:34. http://www.ncbi.nlm.nih.gov/pubmed/19604402

Armstrong EJ, Bischoff J. 2004. Heart valve development: Endothelial cell signalingand differentiation. Circ Res 95:459–470. http://www.ncbi.nlm.nih.gov/pubmed/15345668

Avierinos JF, Gersh BJ, Melton LJ, 3rd, Bailey KR, Shub C, Nishimura RA, Tajik AJ,Enriquez-Sarano M. 2002. Natural history of asymptomatic mitral valveprolapse in the community. Circulation 106:1355–1361. http://www.ncbi.nlm.nih.gov/pubmed/12221052

Barnett JV, Desgrosellier JS. 2003. Early events in valvulogenesis: A signalingperspective. Birth Defects Res C Embryo Today 69:58–72. http://www.ncbi.nlm.nih.gov/pubmed/12768658

Bauer M, Kim K, Qiu Y, Calpe B, Khademhosseini A, Liao R, Wheeldon I. 2012. Spotidentification and quality control in cell-based microarrays. ACS Comb Sci14:471–477. http://pubs.acs.org/doi/abs/10.1021/co300039w

Biechler SV, Potts JD, Yost MJ, Junor L, Goodwin RL,Weidner JW. 2010. Mathematicalmodeling of flow-generated forces in an in vitro system of cardiac valvedevelopment. Ann Biomed Eng 38:109–117.

Brafman DA, Chien S, Willert K. 2012. Arrayed cellular microenvironments foridentifying culture and differentiation conditions for stem, primary and rare cellpopulations. Nat Protoc 7:703–717. http://www.ncbi.nlm.nih.gov/pubmed/22422316

Brafman DA, de Minicis S, Seki E, Shah KD, Teng D, Brenner D, Willert K, Chien S.2009. Investigating the role of the extracellular environment in modulatinghepatic stellate cell biology with arrayed combinatorial microenvironments.Integr Biol (Camb) 1:513–524. http://www.ncbi.nlm.nih.gov/pubmed/20023766

Butcher JT, Markwald RR. 2007. Valvulogenesis: The moving target. Philos Trans RSoc Lond B Biol Sci 362:1489–1503. http://www.ncbi.nlm.nih.gov/pubmed/17569640

Butcher JT, Norris RA, Hoffman S, Mjaatvedt CH, Markwald RR. 2007. Periostinpromotes atrioventricular mesenchyme matrix invasion and remodelingmediated by integrin signaling through Rho/PI 3-kinase. Dev Biol 302:256–266.

Carpenter AE, Jones TR, LamprechtMR, Clarke C, Kang IH, FrimanO, GuertinDA, ChangJH, Lindquist RA,Moffat J, GollandP, SabatiniDM. 2006. CellProfiler: Image analysissoftware for identifying and quantifying cell phenotypes. Genome Biol 7:R100.

Charnley M, Textor M, Khademhosseini A, Lutolf MP. 2009. Integration column:Microwell arrays for mammalian cell culture. Integr Biol (Camb) 1:625–634.

Chua D, Kalb K. 2006. Statins and progression of calcified aortic stenosis. AnnPharmacother 40:2195–2199. http://www.ncbi.nlm.nih.gov/pubmed/17132803

Combs MD, Yutzey KE. 2009. Heart valve development: Regulatory networks indevelopment and disease. Circ Res 105:408–421.

Dahal S, Mahler G. Role of Biochemical and Biophysical Factors on Endothelial-to-Mesenchymal Transformation. 2013 39thAnnu. Northeast Bioeng Conf 2013:125–126.http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber¼6574389

Dal-Bianco JP, Aikawa E, Bischoff J, Guerrero JL, Handschumacher MD, Sullivan S,Johnson B, Titus JS, Iwamoto Y, Wylie-Sears J, Levine RA, Carpentier A. 2009.Active adaptation of the tethered mitral valve: Insights into a compensatorymechanism for functional mitral regurgitation. Circulation 120:334–342. http://www.ncbi.nlm.nih.gov/pubmed/19597052

Dolatshahi-Pirouz A, Nikkhah M, Gaharwar AK, Hashmi B, Guermani E, Aliabadi H,Camci-Unal G, Ferrante T, Foss M, Ingber DE, Khademhosseini A. 2014.A combinatorial cell-laden gel microarray for inducing osteogenic differentiation ofhuman mesenchymal stem cells. Sci Rep 4:3896. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid¼3905276&tool¼pmcentrez&rendertype¼abstract

Dolatshahi-Pirouz A, Nikkhah M, Kolind K, Dokmeci MR, Khademhosseini A. 2011.Micro- and nanoengineering approaches to control stem cell-biomaterialinteractions. J Funct Biomater 2:88–106.

Fernandes TG, Kwon SJ, Lee MY, Clark DS, Cabral JM, Dordick JS. 2008. On-chip,cell-based microarray immunofluorescence assay for high-throughput analysis

Wang et al.: High-Throughput Investigation of Endothelial-to-Mesenchymal Transformation (EndMT) 9

Biotechnology and Bioengineering

of target proteins. Anal Chem 80:6633–6639. http://pubs.acs.org/doi/abs/10.1021/ac800848j

Fernandes TG, Diogo MM, Clark DS, Dordick JS, Cabral JMS. 2009. High-throughputcellular microarray platforms: Applications in drug discovery, toxicology andstem cell research. Trends Biotechnol 27:342–349. http://www.ncbi.nlm.nih.gov/pubmed/19398140

Flaim CJ, Chien S, Bhatia SN. 2005. An extracellular matrix microarray for probingcellular differentiation. Nat Methods 2:119–125. http://www.ncbi.nlm.nih.gov/pubmed/15782209

Flanagan TC, Pandit A. 2003. Living artificial heart valve alternatives: A review. EurCell Mater 6:28–45; discussion 45 http://www.ncbi.nlm.nih.gov/pubmed/14639553

Gobaa S, Hoehnel S, Roccio M, Negro A, Kobel S, Lutolf MP. 2011. Artificial nichemicroarrays for probing single stem cell fate in high throughput. Nat Methods8:949–955.

Hinton RB, Yutzey KE. 2011. Heart valve structure and function in development anddisease. Annu Rev Physiol 73:29–46. http://www.ncbi.nlm.nih.gov/pubmed/20809794

Hoerstrup SP, Sodian R, Daebritz S, Wang J, Bacha EA, Martin DP, Moran AM,Guleserian KJ, Sperling JS, Kaushal S, Vacanti JP, Schoen FJ, Mayer JE, Jr. 2000.Functional living trileaflet heart valves grown in vitro. Circulation 102:III44–III49. http://www.ncbi.nlm.nih.gov/pubmed/11082361

Howbrook DN, Van Der Valk AM, O’Shaughnessy MC, Sarker DK, Baker SC, Lloyd AW.2003. Developments in microarray technologies. Drug Discov Today 8:642–651.http://www.sciencedirect.com/science/article/pii/S1359644603027739

Hsu DT, Pearson GD. 2009. Heart failure in children: Part I: History, etiology, andpathophysiology. Circ Heart Fail 2:63–70. http://www.ncbi.nlm.nih.gov/pubmed/19808316

Jutta Schaper BS. 1992. Cellular and molecular alterations in the failing human heart.Heidelberg, Germany: Springer International Publishing. http://www.springer.com/us/book/9783642724763?wt_mc¼ThirdParty.SpringerLink.3.EPR653.About_eBook.p 303–309. http://link.springer.com/chapter/10.1007/978-3-642-72474-9_26

Kalluri R, Weinberg RA. 2009. The basics of epithelial-mesenchymal transition.J Clin Invest 119:1420–1428. http://www.ncbi.nlm.nih.gov/pubmed/19487818

Klim JR, Li L, Wrighton PJ, Piekarczyk MS, Kiessling LL. 2010. A definedglycosaminoglycan-binding substratum for human pluripotent stem cells. NatMethods 7:989–994.

Kuntze CE, Blackstone EH, Ebels T. 1998. Thromboembolism and mechanical heartvalves: A randomized study revisited. Ann Thorac Surg 66:101–107. http://www.ncbi.nlm.nih.gov/pubmed/9692447

Kwon CH, Wheeldon I, Kachouie NN, Lee SH, Bae H, Sant S, Fukuda J, Kang JW,Khademhosseini A. 2011. Drug-eluting microarrays for cell-based screening ofchemical-induced apoptosis. Anal Chem 83:4118–4125.

Lee M-Y, Kumar RA, Sukumaran SM, Hogg MG, Clark DS, Dordick JS. 2008. Three-dimensional cellular microarray for high-throughput toxicology assays. ProcNatl Acad Sci USA 105:59–63. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid¼2224231&tool¼pmcentrez&rendertype¼abstract

Lee M-Y, Park CB, Dordick JS, Clark DS. 2005. Metabolizing enzyme toxicology assaychip (MetaChip) for high-throughput microscale toxicity analyses. Proc NatlAcad Sci USA 102:983–987. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid¼545843&tool¼pmcentrez&rendertype¼abstract

Maier T, G€uell M, Serrano L. 2009. Correlation of mRNA and protein in complexbiological samples. FEBS Lett 583:3966–3973. http://dx.doi.org/10.1016/j.febslet.2009.10.036

Mei Y, Saha K, Bogatyrev SR, Yang J, Hook AL, Kalcioglu ZI, Cho S-W, Mitalipova M,Pyzocha N, Rojas F, Van Vliet KJ, Davies MC, Alexander MR, Langer R, JaenischR, Anderson DG. 2010. Combinatorial development of biomaterials for clonalgrowth of human pluripotent stem cells. Nat Mater 9:768–778. http://dx.doi.org/10.1038/nmat2812

Moraes C, Likhitpanichkul M, Lam CJ, Beca BM, Sun Y, Simmons CA. 2013.Microdevice array-based identification of distinct mechanobiological responseprofiles in layer-specific valve interstitial cells. Integr Biol (Camb) 5:673–680.http://www.ncbi.nlm.nih.gov/pubmed/23403640

Nicholson RL, Welch M, Ladlow M, Spring DR. 2007. Small-molecule screening:Advances in microarraying and cell-imaging technologies. ACS Chem Biol2:24–30.

Paruchuri S, Yang JH, Aikawa E, Melero-Martin JM, Khan ZA, Loukogeorgakis S,Schoen FJ, Bischoff J. 2006. Human pulmonary valve progenitor cells exhibitendothelial/mesenchymal plasticity in response to vascular endothelial growthfactor-A and transforming growth factor-beta2. Circ Res 99:861–869. http://www.ncbi.nlm.nih.gov/pubmed/16973908

Riem Vis PW, Kluin J, Sluijter JP, van Herwerden LA, Bouten CV. 2011.Environmental regulation of valvulogenesis: Implications for tissue engineer-ing. Eur J Cardiothorac Surg 39:8–17. http://www.ncbi.nlm.nih.gov/pubmed/20637649

Roberts WC, Ko JM. 2005. Frequency by decades of unicuspid, bicuspid, andtricuspid aortic valves in adults having isolated aortic valve replacement foraortic stenosis, with or without associated aortic regurgitation. Circulation111:920–925. http://www.ncbi.nlm.nih.gov/pubmed/15710758

Seiler C. 2004. Management and follow up of prosthetic heart valves. Heart90:818–824. http://www.ncbi.nlm.nih.gov/pubmed/15201262

Soen Y, Mori A, Palmer TD, Brown PO. 2006. Exploring the regulation of humanneural precursor cell differentiation using arrays of signaling microenviron-ments. Mol Syst Biol 2:37.

Srivastava D, Olson EN. 2000. A genetic blueprint for cardiac development. Nature407:221–226. http://www.ncbi.nlm.nih.gov/pubmed/11001064

Sugi Y, Ito N, Szebenyi G, Myers K, Fallon JF, Mikawa T, Markwald RR. 2003.Fibroblast growth factor (FGF)-4 can induce proliferation of cardiac cushionmesenchymal cells during early valve leaflet formation. Dev Biol 258:252–263.

Supino PG, Borer JS, Yin A, Dillingham E, McClymont W. 2004. The epidemiology ofvalvular heart diseases: The problem is growing. Adv Cardiol 41:9–15. http://www.ncbi.nlm.nih.gov/pubmed/15285213

Turi ZG. 2004. Cardiology patient page. Mitral valve disease. Circulation 109:e38–e41. http://www.ncbi.nlm.nih.gov/pubmed/14970121

Winkler M, Simon MG, Vu T, Gartner TL, Jester JV, Lee AP, Brown DJ. 2014.A microfabricated, optically accessible device to study the effects of mechanicalcues on collagen fiber organization. Biomed Microdevices 1–13.

Wu J, Wheeldon I, Guo Y, Lu T, Du Y, Wang B, He J, Hu Y, Khademhosseini A.2011. A sandwiched microarray platform for benchtop cell-based highthroughput screening. Biomaterials 32:841–848. http://dx.doi.org/10.1016/j.biomaterials.2010.09.026

Wylie-Sears J, Aikawa E, Levine RA, Yang JH, Bischoff J. 2011. Mitral valveendothelial cells with osteogenic differentiation potential. Arterioscler ThrombVasc Biol 31:598–607. http://www.ncbi.nlm.nih.gov/pubmed/21164078

Zeisberg EM, Tarnavski O, Zeisberg M, Dorfman AL, McMullen JR, Gustafsson E,Chandraker A, Yuan X, Pu WT, Roberts AB, Neilson EG, Sayegh MH, Izumo S,Kalluri R. 2007. Endothelial-to-mesenchymal transition contributes to cardiacfibrosis. Nat Med 13:952–961. http://www.ncbi.nlm.nih.gov/pubmed/17660828

10 Biotechnology and Bioengineering, Vol. 9999, No. xxx, 2015