•  Scott MacDonnell, Jake Megna, Qin Ruan, Olivia Zhu, Gabor Halasz, Dan Jasewicz, Kristi Powers, Hock E, Maria del Pilar Molina-Portela, Ximei Jin, Dongqin Zhang, Justin Torello, Nicole T. Feric, Michael P. Graziano, Akshay Shekhar, Michael E. Dunn, David Glass and Lori Morton

Activin A directly impairs human cardiomyocyte contractile function indicating a potential role in heart failure development

Published in Frontiers in Cardiovascular Medicine, Volume 9 - 2022

Activin A has been linked to car­diac dys­func­tion in aging and dis­ease, with ele­vat­ed cir­cu­lat­ing lev­els found in patients with hyper­ten­sion, ath­er­o­scle­ro­sis, and heart fail­ure. Here, we inves­ti­gat­ed whether Activin A direct­ly impairs car­diomy­ocyte (CM) con­trac­tile func­tion and kinet­ics uti­liz­ing cell, tis­sue, and ani­mal mod­els. Hydro­dy­nam­ic gene deliv­ery-medi­at­ed over­ex­pres­sion of Activin A in wild-type mice was suf­fi­cient to impair car­diac func­tion, and result­ed in increased car­diac stress mark­ers (N‑terminal pro-atri­al natri­uret­ic pep­tide) and car­diac atro­phy. In human-induced pluripo­tent stem cell-derived (hiP­SC) CMs, Activin A caused increased phos­pho­ry­la­tion of SMAD2/3 and sig­nif­i­cant­ly upreg­u­lat­ed SERPINE1 and FSTL3 (mark­ers of SMAD2/3 acti­va­tion and activin sig­nal­ing, respec­tive­ly). Activin A sig­nal­ing in hiP­SC-CMs result­ed in impaired con­trac­til­i­ty, pro­longed relax­ation kinet­ics, and spon­ta­neous beat­ing in a dose-depen­dent man­ner. To iden­ti­fy the car­diac cel­lu­lar source of Activin A, inflam­ma­to­ry cytokines were applied to human car­diac fibrob­lasts. Inter­leukin ‑1β induced a strong upreg­u­la­tion of Activin A. Mech­a­nis­ti­cal­ly, we observed that Activin A‑treated hiP­SC-CMs exhib­it­ed impaired dias­tolic cal­ci­um han­dling with reduced expres­sion of cal­ci­um reg­u­la­to­ry genes (SERCA2, RYR2, CACNB2). Impor­tant­ly, when Activin A was inhib­it­ed with an anti-Activin A anti­body, mal­adap­tive cal­ci­um han­dling and CM con­trac­tile dys­func­tion were abro­gat­ed. There­fore, inflam­ma­to­ry cytokines may play a key role by act­ing on car­diac fibrob­lasts, caus­ing local upreg­u­la­tion of Activin A that direct­ly acts on CMs to impair con­trac­til­i­ty. These find­ings demon­strate that Activin A acts direct­ly on CMs, which may con­tribute to the car­diac dys­func­tion seen in aging pop­u­la­tions and in patients with heart failure.


Activin A is a mem­ber of the trans­form­ing growth factor‑β (TGF‑β) super­fam­i­ly and con­sists of a homod­imer of two βA sub­units (1). It was ini­tial­ly char­ac­ter­ized as an induc­er of fol­li­cle-stim­u­lat­ing hor­mone secre­tion but is now known to be involved in many crit­i­cal bio­log­i­cal process­es, includ­ing embry­on­ic devel­op­ment, cel­lu­lar dif­fer­en­ti­a­tion, hematopoiesis, tis­sue repair, and fibro­sis (26). Func­tions of Activin A are medi­at­ed by acti­va­tion of the SMAD2/3 path­way by bind­ing to type I and type II activin recep­tors (7). Type I recep­tors include activin recep­tor type 1A (ALK2), activin recep­tor type 1B (ALK4), and activin recep­tor type 1C (ALK7), in humans. Activin recep­tor type 2A and activin recep­tor type 2B are type II recep­tors (8).

Emerg­ing evi­dence sug­gests that Activin A also has an effect on the car­dio­vas­cu­lar sys­tem, with ele­vat­ed lev­els found in patients with pul­monary hyper­ten­sion, ath­er­o­scle­ro­sis, and more recent­ly coro­n­avirus dis­ease 2019 (COVID-19) (913). In heart fail­ure, serum lev­els of Activin A are ele­vat­ed and pos­i­tive­ly cor­re­late both with sever­i­ty of dis­ease and age-depen­dent car­diac dys­func­tion (14, 15). Based on these find­ings, sev­er­al labs have worked to explore poten­tial mechanism(s) by which Activin A con­tributes to the patho­gen­e­sis of heart fail­ure. Using cul­tured neona­tal rat car­diomy­ocytes (CMs), Activin A caused a marked increase in expres­sion of car­diac injury mark­ers (atri­al natri­uret­ic pep­tide, and brain natri­uret­ic pep­tide), mark­ers of extra­cel­lu­lar matrix remod­el­ing (14), and reduced expres­sion of sarco/​endoplasmic retic­u­lum Ca2+-ATPase 2a (SERCA2a) RNA and pro­tein (15). Extend­ing these find­ings in vivo, over­ex­pres­sion of Activin A in mice trig­gered car­diac activin recep­tor type II-medi­at­ed sig­nal­ing in the heart, caus­ing impaired sys­tolic and dias­tolic car­diac func­tion (15). Impor­tant­ly, Roh et al. demon­strat­ed that block­ing activin recep­tors or activin recep­tor lig­ands pre­vent­ed pres­sure over­load-induced car­diac dys­func­tion, sug­gest­ing that Activin was a cen­tral con­trib­u­tor to the car­diac dys­func­tion observed in that mod­el (15). Mech­a­nis­ti­cal­ly, deple­tion of SERCA2a was observed in both neona­tal rat CMs and in mice exposed to high lev­els of Activin A. These reduc­tions in SERCA2a are plau­si­ble expla­na­tions for the reduced con­trac­tile func­tion observed (15, 16). Despite these impor­tant advances in the under­stand­ing of how Activin A con­tributes to car­diac dys­func­tion, there are no func­tion­al data to pro­vide clear evi­dence that Activin A has a direct impact on human CMs.

Human-induced pluripo­tent stem cells (hiP­SCs) are a key tool for study­ing phys­i­ol­o­gy and dis­ease at the cel­lu­lar lev­el. The pluripo­tent nature of these cells allows for dif­fer­en­ti­a­tion into spe­cif­ic cell types of inter­est that may not be read­i­ly avail­able for research. Dif­fer­en­ti­a­tion into human CMs pro­vides researchers the oppor­tu­ni­ty to cul­ture, treat, chal­lenge, and mon­i­tor cell func­tion over time, char­ac­ter­is­tics that are not fea­si­ble with adult rodent or human CMs (17). While hiP­SC-CMs pro­vide a mod­el sys­tem for explor­ing dis­ease biol­o­gy, in 2D cul­tures they dis­play dis­or­ga­nized sar­com­ere struc­ture, express devel­op­men­tal genes, and are not cul­tured under a defined load. There­fore, when using hiP­SC-CMs to define a phe­no­type, it is impor­tant not to rely sole­ly on 2D cul­tures (17, 18). In this study, we char­ac­ter­ize the direct impact of Activin A on murine ven­tric­u­lar func­tion, hiP­SC-CMs, and human engi­neered car­diac tis­sues (HECTs). We pro­pose a mech­a­nism by which inflam­ma­to­ry sig­nal­ing dri­ves car­diac Activin A expres­sion and con­tributes to CM dys­func­tion in patients with heart failure.

Mate­ri­als and methods

Tis­sue culture

For the cul­ture of hiP­SC-CMs, tis­sue cul­ture ves­sels were pre-coat­ed with 10 μg/​mL fibronectin (Ther­mo Fish­er Sci­en­tif­ic, Waltham, MA, USA) for 1 h at 37°C. hiP­SC-CMs (iCell Car­diomy­ocytes2; Fuji­film Cel­lu­lar Dynam­ics, Madi­son, WI, USA) were stored, thawed, and plat­ed accord­ing to the manufacturer’s instruc­tions. Briefly, cells were flash thawed (37°C, 3 min) and slow­ly dilut­ed in plat­ing medi­um. Day 0 is the time­point of dif­fer­en­ti­at­ed hiP­SC-CM plat­ing. Dif­fer­en­ti­a­tion is com­plete as the iCell2 car­diomy­ocytes fol­low a 17 – 20-day dif­fer­en­ti­a­tion pro­to­col per­formed pri­or to cell ship­ment. The iCell2 car­diomy­ocytes arrived cry­op­re­served, and are genet­i­cal­ly puri­fied to approx­i­mate­ly 98% cTnT + cells. For gene expres­sion analy­sis and phos­pho­ry­la­tion assays, 5 × 105 cells were plat­ed per well of a 12-well plate. For imped­ance, elec­tro­phys­i­ol­o­gy, and cal­ci­um tran­sient assays, cells were plat­ed in 96-well plates at a den­si­ty of 5 × 104 cells per well. Cells were main­tained in a humid­i­fied 37°C incu­ba­tor with 5% CO2, with media changed every 48 h. Cells were main­tained in cul­ture until a syn­chro­nous, beat­ing mono­lay­er of cells formed (∼10 – 14 days).

Pri­ma­ry human car­diac fibrob­lasts (HCFs; Pro­mo­Cell, Hei­del­berg, Ger­many) were plat­ed in 24-well plates and grown (37°C, 5% CO2) in medi­um con­tain­ing 10% serum, until 80% con­flu­ent. Cells were then incu­bat­ed with or with­out 1 nM recom­bi­nant human inter­leukin (IL)-1β, onco­statin M, or IL6 (all pur­chased from R&D Sys­tems, Min­neapo­lis, MD, USA) in Dulbecco’s Mod­i­fied Eagle Medi­um + 0.1% bovine serum albu­min for 24 h.

West­ern blot analysis

Cells were exposed to 1 nM Activin A (R&D Sys­tems) for 30 min in the pres­ence or absence of anti – Activin A (10 nM REGN2477) or iso­type con­trol (10 nM REGN1945) mon­o­clon­al anti­bod­ies (Regen­eron Phar­ma­ceu­ti­cals, Tar­ry­town, NY, USA) (19, 20). Cells were washed twice with cold phos­phate-buffered saline and lysed using RIPA Lysis and Extrac­tion Buffer (Ther­mo Fish­er Sci­en­tif­ic) sup­ple­ment­ed with Halt™ Pro­tease and Phos­phatase Inhibitor Cock­tail (Ther­mo Fish­er Sci­en­tif­ic). Lysates were cen­trifuged (14,000 × g, 15 min), and the total pro­tein was quan­ti­fied using the Pierce BCA Pro­tein Quan­ti­ta­tion Kit (Ther­mo Fish­er Sci­en­tif­ic). Pro­tein detec­tion in cell lysates was per­formed under reduc­ing con­di­tions using a 12 – 230 kDa Sep­a­ra­tion Mod­ule for the cap­il­lary elec­trophore­sis™ Wes sys­tem (Pro­tein­Sim­ple, San Jose, CA, USA), accord­ing to the manufacturer’s instruc­tions. Pro­tein sam­ples were dilut­ed with 5 times reduc­ing buffer to a final con­cen­tra­tion of 0.5 mg/​mL, dena­tured (5 min, 95°C), and placed on ice. Car­tridge plates were assem­bled, spun (1,000 × g, 5 min), and placed into the Wes™ instru­ment. Pri­ma­ry anti­bod­ies were obtained from Cell Sig­nal­ing Tech­nolo­gies (Dan­vers, MA, USA). Phospho-SMAD2(Ser465/467)/SMAD3(Ser423/425) was dilut­ed 1:50, SMAD2/3 was dilut­ed 1:50, and GAPDH was dilut­ed 1:100. The Anti-Rab­bit Detec­tion Mod­ule (Pro­tein­Sim­ple) con­sist­ed of anti-rab­bit sec­ondary anti­body, a strep­ta­vidin-horse­rad­ish per­ox­i­dase con­ju­gate, and chemi­lu­mi­nes­cent detec­tion reagents. Pro­tein detec­tion was ana­lyzed using Com­pass soft­ware (Pro­tein­Sim­ple), which quan­ti­fied areas under the curves and height for peak chemi­lu­mi­nes­cent sig­nals from the pro­teins of interest.

RNA iso­la­tion and cDNA synthesis

hiP­SC-CMs were exposed to Activin A (R&D Sys­tems) acute­ly (1 nM for 24 h) or chron­i­cal­ly (1 nM every 48 h for 6 dos­es), and gene expres­sion was ana­lyzed. RNA was iso­lat­ed using an RNeasy Mini Kit (Qia­gen, Ger­man­town, MD, USA) accord­ing to the manufacturer’s instruc­tions. RNA con­cen­tra­tion and qual­i­ty were assessed using a Nan­oDrop Spec­tropho­tome­ter (Ther­mo Fish­er Sci­en­tif­ic). cDNA was syn­the­sized using the Max­i­ma™ H Minus cDNA Syn­the­sis Mas­ter Mix (Ther­mo Fish­er Scientific).

For mouse gene expres­sion analy­ses, tis­sues were homog­e­nized in TRI­zol (Ther­mo Fish­er Sci­en­tif­ic) with chlo­ro­form for phase sep­a­ra­tion. Total RNA was then puri­fied from the aque­ous phase using a MagMAX™-96 for Microar­rays Total RNA Iso­la­tion Kit (Ther­mo Fish­er Sci­en­tif­ic) accord­ing to the manufacturer’s instruc­tions. Genom­ic DNA was removed using an RNase-Free DNase Set (Qia­gen, Ger­man­town, MD, USA). cDNA syn­the­sis was per­formed using the Super­Script® VILO™ Mas­ter Mix (Ther­mo Fish­er Sci­en­tif­ic) and quan­ti­fied using Nan­oDrop (Ther­mo Fish­er Scientific).

Bulk RNA sequencing

Strand-spe­cif­ic RNA-seq libraries were pre­pared from 500 ng RNA using a KAPA strand­ed mRNA-Seq Kit (KAPA Biosys­tems, Wilm­ing­ton, MA, USA). Twelve-cycle poly­merase chain reac­tion (PCR) was per­formed to ampli­fy libraries. Sequenc­ing was per­formed on an Illu­mi­na HiSeq® 2500 (Illu­mi­na, San Diego, CA, USA), using a mul­ti­plexed paired-read run with 33 cycles. Raw sequence data were con­vert­ed to FASTQ for­mat via Illu­mi­na Casa­va 1.8.2. Reads were decod­ed based on their bar­codes, and qual­i­ty was eval­u­at­ed with FastQC. Reads were mapped to the human genome (NCBI B37.3) and a Uni­ver­si­ty of Cal­i­for­nia, San­ta Cruz gene mod­el using ArrayStu­dio® soft­ware (Omic­Soft, Cary, NC, USA), allow­ing 2 mis­match­es. Reads mapped to the sense-strand exons of a gene were summed at the gene level.

Bulk RNA-seq dif­fer­en­tial analysis

Dif­fer­en­tial gene expres­sion analy­sis was car­ried out with DESeq2 v1.30.0 (Bio­con­duc­tor soft­ware) (21). Counts at the gene lev­el as sum­ma­rized by ArrayStu­dio® soft­ware were used as input after round­ing. Genes were pre-fil­tered for a min­i­mum of 10 reads. Size fac­tors were esti­mat­ed per sam­ple using a medi­an-of-ratios method to account for inter-sam­ple sequenc­ing depth vari­a­tions. Hypoth­e­sis test­ing was car­ried out using the Wald test, with mul­ti­ple test­ing cor­rec­tion by the Ben­jami­ni and Hochberg method. Sig­nif­i­cance was defined as adjust­ed P < 0.05 and fold change > 1.5.

Gene set enrich­ment analysis

All genes were sort­ed using –log10(pval)*FC from DESeq2 analy­sis (regard­less of sig­nif­i­cance). The pre-sort­ed gene lists for each com­par­i­son were used as input into Gene Set Enrich­ment Analy­sis (GSEA) using the fast gene set enrich­ment analy­sis R pack­age v1.16.0 (Bio­con­duc­tor soft­ware) (22). Gene sets test­ed were obtained from the Mol­e­c­u­lar Sig­na­ture Data­base c2.cp.reactome.v7.1 col­lec­tion (23).

Gene expres­sion analy­sis by RT-qPCR

Poly­merase chain reac­tions (20 μL total) con­tained 10 μL of 2 × Taq­Man Gene Expres­sion Mas­ter Mix (Ther­mo Fish­er Sci­en­tif­ic), 1 μL of a 20 × Taq­Man probe (Sup­ple­men­tary Table 1; Ther­mo Fish­er Sci­en­tif­ic), 5 μL (10 ng) of cDNA, and 4 μL of water, and were run on a QuantStu­dio™ 3 Real-Time PCR Sys­tem (Ther­mo Fish­er Sci­en­tif­ic). Ther­mo­cy­cler set­tings were as fol­lows: 95°C for 15 min, then 40 cycles of 95°C, 15 s fol­lowed by 60°C, 60 s. Ampli­fi­ca­tion plots were gen­er­at­ed by QuantStu­dio 3 instru­ment soft­ware, and result­ing cycle thresh­old (Ct) val­ues were derived. GAPDH was used as an endoge­nous con­trol. Analy­sis of gene expres­sion in mouse tis­sues was per­formed using a Sen­si­FAST Probe Lo-ROX Kit (Merid­i­an Bio­science, Mem­phis, TN, USA) with the QuantStu­dio 12K Flex Real-Time PCR Sys­tem (Ther­mo Fish­er Sci­en­tif­ic). GAPDH was used as an endoge­nous con­trol gene to nor­mal­ize any cDNA input dif­fer­ences. The delta-delta Ct (2–ΔΔCt) method (24) was used to cal­cu­late rel­a­tive fold change in gene expres­sion for all quan­ti­ta­tive reverse tran­scrip­tion PCR (RT-qPCR) analyses.

Imped­ance and elec­tro­phys­i­o­log­i­cal char­ac­ter­i­za­tion of hiPSC-CMs

Fol­low­ing an ini­tial dose-response assess­ment of cells exposed to titrat­ed Activin A (R&D Sys­tems), a 1 nM dose was cho­sen for sub­se­quent exper­i­ments unless oth­er­wise stat­ed. For imped­ance, elec­tro­phys­i­ol­o­gy, and cal­ci­um assays, chron­ic expo­sure to Activin A was per­formed as shown in the pres­ence or absence of an anti – Activin A (REGN2476) or an iso­type con­trol anti­body (REGN1945) (Regen­eron Phar­ma­ceu­ti­cals Inc.) (25). Con­trac­til­i­ty and elec­tro­phys­i­ol­o­gy of hiP­SC-CMs were char­ac­ter­ized using Car­dioEx­cyte 96 (Nan­ion Tech­nolo­gies, Munich, Ger­many), a hybrid sys­tem that simul­ta­ne­ous­ly records the imped­ance and extra­cel­lu­lar field poten­tial (EFP) of a beat­ing mono­lay­er of CMs in a label-free envi­ron­ment under phys­i­o­log­i­cal cul­ture con­di­tions. In this study, hiP­SC-CMs were plat­ed on elec­trode-con­tain­ing 96-well plates (NSP-96; Nan­ion Tech­nolo­gies) and record­ed for 30 s every 4 h. Imped­ance and EFP data were ana­lyzed using Dat­a­Con­trol 96 soft­ware (Nan­ion Tech­nolo­gies). All beat­ing inflec­tions record­ed dur­ing the 30‑s inter­val were sec­tioned and aver­aged to derive a mean beat.” The fol­low­ing elec­tri­cal pac­ing para­me­ters were used dur­ing imped­ance record­ing: pace fre­quen­cy, 1 Hz; burst ampli­tude, 1 volt; burst fre­quen­cy, 1 kHz; burst length, 2 ms. The ampli­tude (peak-to-trough sig­nal), fall/​rise time (time from 90 to 10% ampli­tude and vice ver­sa), upstroke/​relaxation veloc­i­ty (max­i­mal pos­i­tive and neg­a­tive slopes), and over­all cov­er­age of the elec­trode through the base imped­ance (steady com­po­nent of the imped­ance mag­ni­tude) were char­ac­ter­ized from the pro­file of each mean beat (26, 27). For EFP record­ing, tran­sient elec­tri­cal activ­i­ty out­side of the cell was mea­sured, and mean beats were sub­se­quent­ly inversed to resem­ble the action poten­tial of CMs. Ampli­tude, down­stroke veloc­i­ty (max­i­mal slope dur­ing depo­lar­iza­tion), and field poten­tial dura­tion (the time between the first deflec­tion for depo­lar­iza­tion and the max­i­mum of the repo­lar­iza­tion curve) were char­ac­ter­ized for each mean beat.

Cal­ci­um tran­sient mea­sure­ments in hiPSC-CMs

Cal­ci­um tran­sients were assessed using an Ear­ly­Tox Car­diotox­i­c­i­ty Kit (Mol­e­c­u­lar Devices, San Jose, CA, USA) (28). Cal­ci­um dye load­ing was per­formed accord­ing to the manufacturer’s instruc­tions. Ear­ly­Tox cal­ci­um dye was resus­pend­ed in the sup­plied buffer and added to the cells in a 1:1 ratio with the CM main­te­nance media. The plate was incu­bat­ed for 2 h (37°C, 5% CO2) before record­ing cal­ci­um tran­sients for 2 min at 37°C on the FLIPR Tetra Sys­tem (Mol­e­c­u­lar Devices), using the fol­low­ing para­me­ters: exci­ta­tion, 470 – 495 nM; emis­sion, 515 – 575 nM; expo­sure time, 50 ms; light-emit­ting diode inten­si­ty, 50%; and inter­val time, 100 ms. Cal­ci­um traces were pro­duced and ana­lyzed using Soft­Max Pro Soft­ware (Mol­e­c­u­lar Devices).

Gen­er­a­tion of HECTs and image-based con­trac­til­i­ty measurements

Three-dimen­sion­al HECTs sus­pend­ed between poly[octamethylene maleate (anhy­dride) cit­rate] wires were pre­pared from hiP­SC-CMs and human ven­tric­u­lar car­diac fibrob­lasts (Lon­za, Allen­dale, NJ, USA), and image-based con­trac­til­i­ty mea­sure­ments were gen­er­at­ed using the Biowire II plat­form (TARA Biosys­tems, Inc.), as described pre­vi­ous­ly (29, 30). HECTs were assessed for auto­matic­i­ty (i. e., spon­ta­neous beat rate), force-fre­quen­cy rela­tion­ship (FFR; active force at 1 – 4 Hz), and post-rest poten­ti­a­tion. HECTs with min­i­mal spon­ta­neous activ­i­ty that exhib­it­ed a pos­i­tive FFR and post-rest poten­ti­a­tion were used for com­pound test­ing. HECTs were incu­bat­ed for 30 min in an envi­ron­men­tal cham­ber (37°C, 5% CO2) before video record­ing under field stim­u­la­tion at 1 Hz for 30 s fol­lowed by 30 s with­out field stim­u­la­tion. Cul­ture media were then removed and replaced with fresh media con­tain­ing the test com­pound or vehi­cle. Arrhyth­mic activ­i­ty was deter­mined from con­trac­tile mea­sure­ments acquired every 48 h fol­lowed by reap­pli­ca­tion of the test com­pound or vehi­cle. Con­trac­til­i­ty videos were ana­lyzed using cus­tom analy­sis soft­ware. Twitch ampli­tude (active force), log2 trans­for­ma­tion of twitch ampli­tude, twitch dura­tion (twitch width at half ampli­tude), time to twitch ampli­tude (10% twitch height to ampli­tude), time from twitch ampli­tude (ampli­tude to 10% twitch height), max­i­mum con­trac­tion and relax­ation slopes, and per­cent arrhyth­mic beats (dou­blet beats/​total beats x 100) were char­ac­ter­ized for each 30‑s video acquired with field stim­u­la­tion. The spon­ta­neous beat rate was cal­cu­lat­ed from each of the 30‑s videos acquired with­out field stim­u­la­tion (total beats/​30 s).


Male C57BL/6N mice aged ∼12 weeks and weigh­ing 22 – 28 g (Tacon­ic Bio­sciences, Rens­se­laer, NY, USA) were accli­mat­ed for a min­i­mum of 7 days pri­or to exper­i­men­ta­tion. Mice were co-housed in poly­car­bon­ate, sol­id-bot­tom cages in a tem­per­a­ture-con­trolled envi­ron­ment (22 ± 2°C) with an approx­i­mate 12‑h light-dark cycle and with access to research diets of stan­dard pel­let chow and reverse osmo­sis – fil­tered water. Mice were anes­thetized with a mix­ture of 2% isoflu­rane in 100% oxy­gen before sur­gi­cal pro­ce­dures. All ani­mal pro­ce­dures and pro­to­cols described in this work were approved by Regen­eron Phar­ma­ceu­ti­cals, Inc., are in accor­dance with state and fed­er­al guide­lines, and are aligned with reg­u­la­tions set forth by Regen­eron Phar­ma­ceu­ti­cals, Inc.’s Insti­tu­tion­al Ani­mal Care and Use Committee.

Hydro­dy­nam­ic deliv­ery of an Activin A expres­sion construct

cDNA-encod­ing Activin A was gen­er­at­ed by PCR from a plas­mid tem­plate har­bor­ing untagged Activin A (Ref­Seq acces­sion num­ber NM_002192.4) using the fol­low­ing primers: for­ward, 5′-TCCCCCACCCCAGGATCCGAGGG GCCA3′ and reverse, 5′-GCACCTCCTCACACCCACGAGTAT CCGCCGGCGCCTAGGATCTAG3′. The result­ing cDNA was cloned into a mam­malian expres­sion vec­tor pRG977 and con­firmed by DNA sequenc­ing. On study day 0, mice (n = 20) were strat­i­fied to groups based on body weight. Mice were inject­ed via tail vein with 2.5 μg plas­mid in ster­ile saline at 10% of body weight. Serum was col­lect­ed to assess Activin A expres­sion and bio­mark­ers. At study ter­mi­na­tion (14 days post-hydro­dy­nam­ic deliv­ery [HDD]), ter­mi­nal body weight was acquired and imme­di­ate­ly fol­lowed by admin­is­tra­tion of a ter­mi­nal dose of ketamine/​xylazine and exsan­guina­tion via the abdom­i­nal aor­ta. The heart was resect­ed and weighed, and the right and left atria and ven­tri­cles were dis­sect­ed and placed imme­di­ate­ly into indi­vid­ual tubes con­tain­ing RNAlater (Ther­mo Fish­er Sci­en­tif­ic). The tib­ia was removed and mea­sured using a hand caliper.

Pres­sure – vol­ume loop acquisition

Mice (n = 26) that under­went hydro­dy­nam­ic deliv­ery of Activin A for 14 days were placed in a supine posi­tion on a warm sur­face with exter­nal heat­ing lamps to main­tain nor­mal body tem­per­a­ture (35.5 – 37.5°C). Endo­tra­cheal intu­ba­tion was per­formed, and mice were ven­ti­lat­ed (Har­vard MiniVent with PEEP attach­ment) at 135 breaths/​minute and tidal vol­ume 0.2 mL. An inci­sion was made in the upper abdomen above the xiphoid toward the ante­ri­or. The skin was then sep­a­rat­ed from the chest wall via blunt dis­sec­tion, and a small inci­sion was made above the apex of the heart between the third and fourth inter­costal space. A small saline-dipped swab was placed through the inci­sion. Using the swab to pro­tect the heart and lungs, a cau­ter­ized cut was made through the ribs and ster­num to the oppo­site side of the chest to expose the apex of the heart and the infe­ri­or vena cava (IVC). For­ceps were used to remove the peri­cardi­um. The apex of the left ven­tri­cle was punc­tured with a 25-gauge nee­dle, and a 1.2 F tetrap­o­lar admit­tance catheter (Tran­son­ic Sys­tems, Itha­ca, MY, USA) attached to an ADV500 pres­sure – vol­ume mea­sure­ment sys­tem (Tran­son­ic Sys­tems) and Pow­er­Lab 1635 inter­face (ADIn­stru­ments, Col­orado Springs, CO, USA) was insert­ed. Base­line record­ing was con­duct­ed for 5 min before mak­ing 3 IVC occlu­sions in suc­ces­sion. Data were ana­lyzed using Lab­Scribe (iWorx, Dover NHUSA).


Activin A lev­els were assessed using the Activin A Quan­tikine enzyme-linked immunosor­bet assay (ELISA) kit (R&D Sys­tems), and N‑terminal pro-atri­al natri­uret­ic pep­tide (NT-proANP) lev­els were assessed using an NT-proANP ELISA (Bio­med­ica Medi­z­in­pro­duk­te, Vien­na, Aus­tria) accord­ing to the manufacturer’s instructions.

Sta­tis­ti­cal analyses

For in vit­ro para­me­ters, data are expressed as mean ± stan­dard devi­a­tion. One-way analy­sis of vari­ance with Tukey’s post hoc analy­sis was used to deter­mine sig­nif­i­cance. For all in vivo para­me­ters eval­u­at­ed, sta­tis­ti­cal analy­ses were per­formed using Graph­Pad Prism (Graph­Pad Soft­ware, San Diego, CA, USA). Data are expressed as mean ± stan­dard error of the mean. Sta­tis­ti­cal analy­ses were per­formed using an unpaired 2‑tailed t‑test. P val­ues ≤ 0.05 were con­sid­ered indica­tive of sta­tis­ti­cal significance.


Ele­vat­ed cir­cu­lat­ing Activin A results in car­diac dys­func­tion in the murine heart

To inves­ti­gate whether Activin A is suf­fi­cient to alter glob­al heart func­tion, we chal­lenged wild-type mice with a mam­malian expres­sion vec­tor encod­ing Activin A (Activin A HDD) and assessed car­diac func­tion using pres­sure – vol­ume loop analy­sis. Lev­els of cir­cu­lat­ing Activin A were ∼80 times high­er in mice inject­ed with Activin A HDD com­pared with mice inject­ed with con­trol vec­tor (P < 0.01; Fig­ure 1A). In mice exposed to high lev­els of Activin A for 14 days, the heart-to-body weight ratio was ∼9% low­er (P < 0.01), and the heart weight-to-tib­ia length ratio was ∼15% low­er (P < 0.01) (Fig­ure 1B). Lev­els of cir­cu­lat­ing NT-proANP, a bio­mark­er of car­diac wall stress, were > 1.7 times high­er (P < 0.01) in mice inject­ed with Activin A HDD (Fig­ure 1C).

Figure 1. Activin A overexpression in mice produced cardiac dysfunction. Comparison of physical and molecular characteristics in control and Activin A–treated mice. (A) In vivo circulating levels of Activin A. (B) Heart-to-body weight ratio. (C) Circulating NT-proANP. (D) Gene expression in LV + S tissue. (E) Pressure-volume loop analyses. Statistical analyses were performed using an unpaired two-tailed t-test. *P < 0.05, **P < 0.01. ActA, Activin A; ESPVR, end systolic pressure–volume relationship; HDD, hydrodynamic delivery; LV + S, left ventricle and septum; NT-proANP, N-terminal pro-atrial natriuretic peptide.

Sig­nif­i­cant increas­es in expres­sion of Fstl3 (∼1.5‑fold [P < 0.05]), Nppa (∼1.8‑fold [P < 0.05]), and Myh7 (∼1.4‑fold [P < 0.05]) in left ven­tri­cle and sep­tum tis­sue com­pared with con­trol were seen with Activin A, while no sig­nif­i­cant changes were observed in Activin A‑induced expres­sion of Nppb, Myh6, Tgf1, and Gdf15 (Fig­ure 1D).

Pres­sure – vol­ume loop analy­sis revealed that Activin A caused sig­nif­i­cant decreas­es in end-sys­tolic pres­sure (∼15%, P < 0.01) and ejec­tion frac­tion (∼19%, P < 0.05), and caused sig­nif­i­cant increas­es in end-sys­tolic and end-dias­tolic vol­umes (∼53%, P < 0.01 and ∼26%, P < 0.05, respec­tive­ly) (Fig­ure 1E). Although Activin A HDD result­ed in cir­cu­lat­ing Activin A con­cen­tra­tions (mean val­ue: 4.7 ng/​mL) that were high­er than those observed in human heart fail­ure (<1 ng/​mL observed in human heart fail­ure patients) (14), strik­ing­ly, Activin A was suf­fi­cient to induce impaired sys­tolic and dias­tolic car­diac func­tion in oth­er­wise healthy young mice.

Activin A increas­es SMAD phos­pho­ry­la­tion and expres­sion of SMAD2/3 tar­get genes in hiPSC-CMs

We next explored whether Activin A could direct­ly dri­ve down­stream sig­nal­ing in human CMs. SMAD2/3 phos­pho­ry­la­tion was sig­nif­i­cant­ly increased by 83% in hiP­SC-CMs exposed to 1 nM Activin A (P < 0.01) for 30 min com­pared with con­trol (Fig­ure 2A). This increase in SMAD phos­pho­ry­la­tion was blocked by the anti – Activin A anti­body but not by an iso­type con­trol (Fig­ure 2A).

Figure 2. Activin A induces SMAD3 phosphorylation and downstream signaling in hiPSC-CMs. (A) Automated western blot data for the effect of Activin A plus anti–Activin A antibody or an isotype control on SMAD2/3 phosphorylation. (B) RNA sequencing data for expression of FSTL3, and SERPINE1 (PAI1) expression following exposure to 1 nM Activin A. One-way ANOVA was used to determine significance. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ActA, Activin A; hiPSC-CM, human-induced pluripotent stem cell-derived cardiomyocyte; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; TPM, transcripts per million.

RNA sequenc­ing was used to deter­mine the effect of acute (24 h) and chron­ic (6 repeat­ed dos­es) Activin A expo­sure on expres­sion of genes direct­ly involved in activin sig­nal­ing, includ­ing recep­tors, lig­ands, and SMAD pro­teins. With chron­ic expo­sure, expres­sion of the genes encod­ing ALK2 (ACVR1) and bone mor­phogenic pro­tein recep­tor type 2 (BMPR2) increased by 46% and 28%, respec­tive­ly, from base­line fol­low­ing chron­ic expo­sure to Activin A (Table 1). Sim­i­lar­ly, expres­sion of SMAD3 and TGF1 increased by 22% and 40%, respec­tive­ly. Expres­sion of down­stream SMAD2/3 tar­get genes was also high­er com­pared with con­trol cells fol­low­ing chron­ic Activin A expo­sure, includ­ing FSTL3 (∼83%, P < 0.0001) and SERPINE1 (∼27%, P < 0.001), indi­cat­ing that activin-depen­dent SMAD sig­nal­ing was intact in these hiP­SC-CMs (Fig­ure 2B). Inter­est­ing­ly, acute Activin A expo­sure caused a ∼33% increase in expres­sion of FSTL3 (P < 0.01), but had no sig­nif­i­cant effect on expres­sion of SERPINE1, while chron­ic treat­ment induced upreg­u­la­tion of both FSTL3 and SERPINE1 (Fig­ure 2B).

Table 1. Effect of Activin A on expres­sion of genes involved in activin signaling.

Activin A impairs hiP­SC-CM con­trac­tile function

To study the direct con­se­quence of engaged Activin A sig­nal­ing on hiP­SC-CM con­trac­tile func­tion, we con­duct­ed Activin A dose-esca­la­tion stud­ies and con­tin­u­ous­ly mea­sured con­trac­tile dynam­ics for 21 days post-expo­sure. A dose-depen­dent decrease in con­trac­tile imped­ance ampli­tude was observed in hiP­SC-CMs after expo­sure to Activin A (Fig­ure 3A). From day 10 (base­line) to day 20, con­trac­tile ampli­tude decreased by 63% with 100 nM Activin A (11.6 ± 0.264 ohms to 4.33 ± 0.726 ohms), 62% with 10 nM Activin A (12.6 ± 0.334 ohms to 4.76 ± 0.117 ohms), 53% with 1 nM Activin A (11.7 ± 0.142 ohms to 5.46 ± 0.126 ohms), and 32% with 0.1 nM Activin A (11.9 ± 0.177 ohms to 8.06 ± 0.336 ohms), com­pared with a 21% reduc­tion in con­trol media (12.2 ± 0.175 ohms to 9.65 ± 0.222 ohms). The Activin A – induced decrease in con­trac­tile imped­ance ampli­tude was pre­vent­ed by 25 nM anti – Activin A mon­o­clon­al anti­body (10.50 ± 0.87 vs 6.97 ± 0.43 ohms in cells exposed to Activin A plus iso­type con­trol anti­body; P < 0.0001). The inhibito­ry effect of the anti – Activin A anti­body decreased upon anti­body dilu­tion (Fig­ure 3B).

Figure 3. Activin A exposure time course on hiPSC-CM function. (A) The effect of different concentrations of Activin A on hiPSC-CM contractile impedance amplitude. (B,C) The impact of Activin A inhibition on Activin A–induced contractile dysfunction in hiPSC-CMs: dose-dependent effect on contractile impedance amplitude (B) and contractile parameters at both intrinsic and normalized beat rate (C). (D) Representative traces at both intrinsic beat rate and paced at 1 Hz hiPSC-CM, induced pluripotent stem cell-derived CM. Vertical lines indicate media changes in panels (A,B). One-way ANOVA was used to determine significance. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ActA, Activin A; CM, cardiomyocyte; hiPSC-CM, human-induced pluripotent stem cell-derived cardiomyocyte.

At day 20, hiP­SC-CMs were assessed at their intrin­sic beat rate or paced at 1 Hz. When paced at 1 Hz, 1 nM Activin A reduced the imped­ance con­trac­tile ampli­tude (6.46 ± 0.23 vs 10.27 ± 0.36 ohms, P < 0.0001) and relax­ation veloc­i­ty kinet­ics (35.0 ± 1.1 vs 48.8 ± 2.7 ohms/​s, P < 0.0001) com­pared with media con­trol (Fig­ure 3C). Anti – Activin A anti­body (≥5 nM) pre­vent­ed the Activin A – medi­at­ed reduc­tion in imped­ance ampli­tude com­pared with cells exposed to Activin A plus con­trol anti­body (anti-Activin A 25 nM: 10.8 ± 0.99 vs 6.72 ± 0.38 ohms/​s, P < 0.0001). Sim­i­lar­ly, cells exposed to Activin A plus ≥ 5 nM anti – Activin A anti­body had quick­er relax­ation veloc­i­ty kinet­ics com­pared with cells exposed to Activin A plus con­trol anti­body (anti – Activin A 25 nM: 49.6 ± 4.0 vs 35.6 ± 2.7 ohms/​s; P < 0.0001). Rep­re­sen­ta­tive imped­ance traces are pre­sent­ed in Fig­ure 3D.

IL-1β induces Activin A expres­sion in HCFs

Giv­en our evi­dence that Activin A was detri­men­tal for CM func­tion, we test­ed whether pro-inflam­ma­to­ry cytokines present in fail­ing human hearts con­tributes to local Activin A secre­tion in non-CMs. HCFs were treat­ed with IL-1β, IL6, onco­statin M, and Activin A. Expres­sion of genes asso­ci­at­ed with activin and TGF‑β sig­nal­ing was then quan­ti­fied using RT-qPCR (Fig­ure 4A). Lev­els of INH­BA mRNA were almost 4 times high­er in cells treat­ed with IL-1β com­pared with con­trol cells (P < 0.0001). Treat­ment of HCFs with Activin A result­ed in an approx­i­mate­ly 2.5 times high­er expres­sion of SERPINE1 and FSTL3 (P < 0.0001), con­firm­ing the SMAD2/3 depen­dance of these genes in car­diac fibrob­lasts. Although Activin A had no effect on MSTN expres­sion, it was reduced by ∼80% in cells treat­ed with IL-1β or onco­statin M (P < 0.0001) and by ∼50% in cells treat­ed with IL6 (P < 0.0001). Sig­nif­i­cant reduc­tions in GDF11 expres­sion were observed in cells exposed to onco­statin M (P < 0.05) and Activin A (P < 0.05), although expres­sion was reduced (non-sig­nif­i­cant­ly) by all treatments.

Figure 4. IL-1β induces Activin A gene expression in cardiac fibroblasts. (A) The effect of inflammatory cytokines, IL-1β, IL-6, and oncostatin M on gene expression in cardiac fibroblasts. (B) The concentration of Activin A in control- and IL-1β-treated cells. (C) The effect of IL-1β on induced pluripotent stem cell-derived CM contractility (vertical lines indicate media changes). One-way ANOVA was used to determine significance. ****P < 0.0001. ActA, Activin A; CM, cardiomyocyte.

Con­sis­tent with IL-1β – induced increase in expres­sion of INH­BA mRNA, the con­cen­tra­tion of Activin A was approx­i­mate­ly 9 times high­er in the super­natant of HCFs treat­ed with IL-1β than in that of con­trol cells (Fig­ure 4B). To inves­ti­gate whether IL-1β is suf­fi­cient to reca­pit­u­late the impaired con­trac­til­i­ty induced by Activin A, con­trac­tile imped­ance ampli­tude was assessed in hiP­SC-CMs exposed to IL-1β. IL-1β was found to have no direct effect on con­trac­tile imped­ance ampli­tude (Fig­ure 4C). Sim­i­lar­ly, IL6 has no direct impact on CM func­tion (data not shown). As TGF-β1 gene expres­sion was increased with chron­ic treat­ment, we test­ed the impact of a pan-TGF‑β inhibito­ry anti­body on activin A‑induced CM dys­func­tion. No direct impact on CM func­tion was observed with anti-TGF‑β (up to 30 nM) sug­gest­ing that TGF‑β is not involved in the observed activin A‑induced CM dys­func­tion (Sup­ple­men­tary Fig­ure 1).

Engaged Activin A sig­nal­ing in hiP­SC-CMs down­reg­u­lates cal­ci­um cycling genes while pro­mot­ing fibrot­ic gene expression

To elu­ci­date the tran­scrip­tion­al net­works that lead to Activin A – medi­at­ed CM dys­func­tion, RNA sequenc­ing was per­formed after acute (24 h) and chron­ic (6 repeat­ed dos­es) Activin A expo­sure. Analy­sis of dif­fer­en­tial­ly expressed genes iden­ti­fied sev­er­al path­ways that were altered by expo­sure of hiP­SC-CMs to Activin A (Fig­ure 5A). Of the path­ways that were sig­nif­i­cant­ly enriched fol­low­ing treat­ment with acute or chron­ic Activin A, all were upreg­u­lat­ed with a pos­i­tive nor­mal­ized enrich­ment score, except for glyco­gen metab­o­lism and car­diac con­duc­tion. Com­pared with con­trol, sig­nal­ing by TGF‑β path­ways was sig­nif­i­cant­ly (P < 0.05) upreg­u­lat­ed with acute Activin A treat­ment. Fig­ure 5B shows lead­ing-edge genes extract­ed from GSEA. The 10 most up-reg­u­lat­ed TGF‑β sig­nal­ing genes and the 21 most down-reg­u­lat­ed genes involved in cal­ci­um cycling, elec­tro­phys­i­ol­o­gy, and con­trac­tion after chron­ic admin­is­tra­tion of Activin A are plot­ted. Note that, although the genes were select­ed based on per­tur­ba­tion in the chron­ic con­di­tion, these genes show con­sis­tent reg­u­la­tion in the acute con­di­tion as well.

Figure 5. Activin A–dependent gene signature in hiPSC-CMs. (A) Significantly enriched pathways in hiPSC-CMs treated with acute (24 h) and chronic Activin A. All pathways were upregulated with positive normalized enrichment scores except glycogen metabolism and cardiac conduction. Cardiac conduction and signaling by TGF-β family members were only significant (P < 0.05) at 24 h. (B) Heatmap of leading-edge gene expression in cardiac conduction and signaling by TGF-β family members. Although both pathways were only significant after acute Activin A exposure (24 h), a similar pattern of gene expression was observed in chronic exposure. (C) TPM values for a subset of cardiac genes highlighted in the heat map. One-way ANOVA was used to determine significance. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ActA, Activin A; hiPSC-CM, human-induced pluripotent stem cell-derived cardiomyocyte; ns, not significant; TGF-β, transforming growth factor-β; TPM, transcripts per million.

Tak­ing a can­di­date gene approach, dif­fer­ences for indi­vid­ual genes were present after both acute and chron­ic Activin A treat­ment. Chron­ic Activin A expo­sure caused upreg­u­la­tion of TGF1 (∼39% high­er com­pared with con­trol, P < 0.001; Fig­ure 5C). Chron­ic Activin A expo­sure caused sig­nif­i­cant down­reg­u­la­tion, com­pared with con­trol, of the crit­i­cal cal­ci­um-han­dling genes ATP2A2 (∼22%, P < 0.0001) and RYR2 (∼35%, P < 0.0001), encod­ing sarcoplasmic/​endoplasmic retic­u­lum cal­ci­um ATPase 2 and ryan­odine recep­tor 2, respec­tive­ly. Addi­tion­al­ly, expres­sion of CACNB2, cod­ing for the volt­age-depen­dent L‑type cal­ci­um chan­nel sub­unit beta‑2, and CORIN, cod­ing for atri­al natri­uret­ic pep­tide-con­vert­ing enzyme, were reduced by ∼33% (P < 0.0001) and ∼55% (P < 0.0001), respec­tive­ly (Fig­ure 5C). These obser­va­tions sug­gest that Activin A – medi­at­ed sig­nal­ing direct­ly impacts genes impor­tant to coor­di­nate nor­mal CM con­duc­tion and contraction.

Effect of Activin A on hiP­SC-CM elec­tro­phys­i­ol­o­gy and cal­ci­um transients

To deter­mine the effect of Activin A on elec­tro­phys­i­ol­o­gy and cal­ci­um han­dling, hiP­SC-CMs were assessed at their intrin­sic beat rate after 6 Activin A dos­es at day 20. Elon­gat­ed field poten­tial dura­tion was observed in hiP­SC-CMs treat­ed with chron­ic Activin A com­pared with con­trol (0.56 ± 0.01 vs 0.49 ± 0.02 s, P < 0.01), which was pre­vent­ed by 5 nM and 25 nM anti – Activin A anti­body (0.51 ± 0.05 vs 0.56 ± 0.02 s with 25 nM con­trol anti­body; Fig­ures 6A,B). Chron­ic Activin A expo­sure also result­ed in a reduc­tion in field poten­tial ampli­tude and down­stroke veloc­i­ty com­pared with con­trol (48.58 ± 6.52 vs 74.52 ± 11.66 μV, P < 0.01 and 0.018 ± 0.006 vs 0.034 ± 0.005 V/​s, P < 0.001), respec­tive­ly. Both of these were pre­vent­ed by 25 nM anti – Activin A anti­body (85.55 ± 19.82 vs 42.87 ± 2.00 μV with con­trol anti­body and 0.039 ± 0.009 vs 0.019 ± 0.002 V/​s with con­trol anti­body), respec­tive­ly (Fig­ures 6A,B).

Figure 6. Effect of Activin A on myocardial electrophysiology and calcium transients. The effect of Activin A or Activin A plus inhibitory antibody on (A,B) electrophysiology and (C,D) calcium transients in induced pluripotent stem cell–derived cardiomyocytes. One-way ANOVA was used to determine significance. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ActA, Activin A; BPM, beats per minute; FLIPR, fluorescent imaging plate reader; FPD, field potential duration; IPSC, induced pluripotent stem cells; RFU, relative fluorescence units.

Cal­ci­um han­dling was ana­lyzed to explore whether the mech­a­nism of altered Activin A – induced elec­tro­phys­i­o­log­i­cal changes in hiP­SC-CMs was a result of impaired cal­ci­um flux (rep­re­sen­ta­tive traces are dis­played in Fig­ure 6C). Chron­ic expo­sure with 1 nM Activin A reduced peak (max-min rel­a­tive flu­o­res­cence units [RFU]) cal­ci­um ampli­tude (447 ± 33 vs 609 ± 99 RFU, P < 0.02 with con­trol media), increased cal­ci­um falling time (0.70 ± 0.04 vs 0.52 ± 0.05 s, P < 0.0001), and increased cal­ci­um ris­ing time (0.41 ± 0.06 vs 0.24 ± 0.04 s, P < 0.01; Fig­ure 6D). Peak cal­ci­um han­dling ampli­tude was 751 ± 129 RFU in cells treat­ed with Activin A plus anti – Activin A anti­body (25 nM) com­pared with 484 ± 37 RFU in cells treat­ed with Activin A plus con­trol anti­body. Slow­er cal­ci­um tran­sient fall and rise times were also observed fol­low­ing chron­ic expo­sure to Activin A com­pared with Activin A plus inhibito­ry anti­body (0.75 ± 0.08 vs 0.58 ± 0.04 s and 0.36 ± 0.05 vs 0.27 ± 0.1 s, respec­tive­ly; Fig­ure 6D).

Effect of Activin A on arrhyth­mia and con­trac­tile func­tion in HECTs

To bet­ter under­stand the con­trac­tile force and kinet­ics in a more mature mod­el of human myocardi­um, a third mod­el was stud­ied using HECTs derived from hiP­SC-CMs and HCFs. From a base­line lev­el of 0% arrhyth­mic activ­i­ty, chron­ic Activin A treat­ment caused a dose-depen­dent increase in arrhyth­mic activ­i­ty over time, ris­ing to 80% on day 8 and 67% on day 34 (24 dou­blet beats/​30 total beats and 20 dou­blet beats/​30 total beats, respec­tive­ly; rep­re­sen­ta­tive con­trac­tion curves in Sup­ple­men­tary Fig­ure 2). Con­cur­rent with (and like­ly under­ly­ing) the increased arrhyth­mic activ­i­ty, there was an observed increase in spon­ta­neous beat rate activ­i­ty. This increase in spon­ta­neous activ­i­ty induced by 1 nM Activin A was pre­vent­ed (treat­ment start­ing on day 1, Fig­ures 7A,B) and reversed (treat­ment start­ing on day 20; Fig­ure 7B) by 10 nM anti – Activin A anti­body. At 8 days post-treat­ment (peak arrhyth­mic activ­i­ty in this mod­el), spon­ta­neous beat rate was 0 Hz in HECTs treat­ed with 1 nM Activin A and 10 nM anti – Activin A anti­body, mir­ror­ing the con­trol (Fig­ures 7A,B). Inter­est­ing­ly, alter­nate TGF‑β fam­i­ly mem­bers, myo­statin (MSTN) (growth dif­fer­en­ti­a­tion fac­tor [GDF] 8) and GDF11 dosed at 1 nM failed to induce an increase in spon­ta­neous activ­i­ty (Fig­ure 7B). Spon­ta­neous beat rate was sig­nif­i­cant­ly faster in HECTs treat­ed with 1 nM Activin A and 10 nM iso­type con­trol anti­body com­pared with con­trol (1.3 Hz vs 0.2 Hz; P < 0.0001) (Fig­ure 7B).

Figure 7. Effect of Activin A on contractile function and arrhythmia in HECTs. (A) Effect of Activin A on cardiac rhythm in HECTs for spontaneous beat rate (beating in the absence of pacing) during the full treatment period and at 8 days post treatment. (B) Effect of Activin A, MSTN (GDF8), and GDF11 on cardiac rhythm in HECTs for spontaneous beat rate during treatment and at 8 days post treatment. (C,D) impact of Activin A and preventive Activin A inhibition on HECT contraction kinetics on day 20 and the reversal by applying anti–Activin A starting on day 20 to day 34. (E,F) Impact of Activin A, Activin A inhibition, MSTN (GDF8), and GDF11 on HECT contraction kinetics on day 20. Shading on D denotes the 95% confidence interval. One-way ANOVA was used to determine significance. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ActA, Activin A; GDF, growth differentiation factor; HECT, human engineered cardiac tissue; MSTN, myostatin.

There was no sig­nif­i­cant dif­fer­ence in twitch ampli­tude, max­i­mum con­cen­tra­tion slope, or max­i­mum relax­ation slope between all treat­ment con­di­tions and media-only con­trol on day 20 (Fig­ure 7C). Con­sis­tent with our hiP­SC-CMs, time to and from twitch ampli­tude and twitch dura­tion increased in a dose-depen­dent man­ner fol­low­ing chron­ic treat­ment with Activin A on day 20 (Fig­ure 7D). Mean time to twitch ampli­tude was sig­nif­i­cant­ly longer fol­low­ing treat­ment with 1 nM (P < 0.01) or 10 nM Activin A (P < 0.001) com­pared with con­trol con­di­tions. Sim­i­lar­ly, mean time from twitch ampli­tude was sig­nif­i­cant­ly longer fol­low­ing treat­ment with 0.1 nM (P < 0.01), 1 nM (P < 0.0001), or 10 nM (P < 0.0001) Activin A, and mean twitch dura­tion was sig­nif­i­cant­ly pro­longed fol­low­ing treat­ment with 1 nM (P < 0.001) or 10 nM Activin A (P < 0.0001). Treat­ment with anti – Activin A, start­ed at day 0, nor­mal­ized twitch para­me­ters, while the anti­body iso­type con­trol had no effect (Fig­ure 7D).

The effects of treat­ment with 1 nM Activin A on HECT spon­ta­neous beat rate and twitch dura­tion per­sist­ed for 34 days in the pres­ence of 10 nM con­trol anti­body (Fig­ures 7B,E). Impor­tant­ly, these effects were sup­pressed in the pres­ence of anti – Activin A and reversed when treat­ment was ini­ti­at­ed on day 20 (Fig­ure 7D). It is inter­est­ing to note that, at day 20, 1 nM MSTN (GDF8) and 1 nM GDF11 had inter­me­di­ate effects when com­pared with Activin A on con­trac­tile kinet­ics com­pared with con­trol (Fig­ure 7F). It may be sim­ply that Activin A is a more potent lig­and than either GDF11 or MSTN; how­ev­er, these find­ings sup­port the con­tention that engaged Activin A sig­nal­ing, and to a less­er extent GDF8 or GDF11, in CMs impairs car­diomy­ocyte con­trac­tile dynam­ics to pro­mote dys­func­tion­al myocar­dial performance.


The grow­ing preva­lence of heart fail­ure among aging pop­u­la­tions is a major con­cern (31) that has recent­ly been exac­er­bat­ed by the fre­quen­cy of myocar­dial dys­func­tion in patients with COVID-19 (32, 33). IL sig­nal­ing, SMAD2/3 phos­pho­ry­la­tion, and Activin A sig­nal­ing path­ways have all been impli­cat­ed in the devel­op­ment of car­diac dys­func­tion and age-relat­ed car­dio­vas­cu­lar dis­ease (15, 3436). Recent­ly, high lev­els of Activin A have also been found in patients with COVID-19, and are asso­ci­at­ed with poor out­comes and mor­tal­i­ty (10, 11). Activin A may dri­ve eti­ol­o­gy-spe­cif­ic heart fail­ure phe­no­types, includ­ing car­dio­pro­tec­tion from ischemic injury (37), while ele­vat­ed cir­cu­lat­ing lev­els of Activin A are asso­ci­at­ed with car­diac dys­func­tion in non-ischemic heart fail­ure mouse mod­els as well as in humans (12, 15). The cell-type spe­cif­ic role of Activin A in the human heart is yet to be ful­ly defined.

Here, we demon­strate that engage­ment of Activin A sig­nal­ing is respon­si­ble for damp­ing CM gene net­works required to main­tain opti­mal con­trac­tile func­tion in the murine heart, hiP­SC-CMs, and HECTs. In our Activin A over-expres­sion mouse mod­el (HDD), ele­vat­ed cir­cu­lat­ing Activin A was suf­fi­cient to impair car­diac func­tion in oth­er­wise healthy young mice, con­firm­ing pre­vi­ous reports (15). Activin A caused increased phos­pho­ry­la­tion of SMAD3 in CMs as well as sig­nif­i­cant upreg­u­la­tion of SERPINE1 and FSTL3, mark­ers of SMAD2/3 acti­va­tion and activin sig­nal­ing. Expres­sion of genes asso­ci­at­ed with car­diac func­tion, ATP2A2, RYR2, and CACNB2 (asso­ci­at­ed with cal­ci­um dynam­ics); blood pres­sure and flu­id bal­ance (CORIN – process­es natri­uret­ic pep­tides); ion chan­nel sta­bil­i­ty and func­tion (KCNJ2, SCN9A, SCN8A, SCN2B); and pro­tec­tion from car­diac remod­el­ing and heart fail­ure (KCNJ11) were down­reg­u­lat­ed in hiP­SC-CMs fol­low­ing Activin A expo­sure. Impor­tant­ly, these same genes are known to be down-reg­u­lat­ed in patients with heart fail­ure and in exper­i­men­tal heart fail­ure mod­els (3846).

We also observed pro­longed cal­ci­um fall times fol­low­ing expo­sure to Activin A, which can be indica­tive of impaired CM dias­tolic func­tion (47). The impair­ment of cal­ci­um chan­nel kinet­ics was pre­vent­ed by an anti – Activin A anti­body. Down­reg­u­la­tion of key cal­ci­um-han­dling genes RYR2 and ATP2A2, induced by Activin A expo­sure, may con­tribute to the reduced peak cal­ci­um ampli­tude and increased cal­ci­um rise and fall times (48). Activin A – medi­at­ed upreg­u­la­tion of FSTL3 and SERPINE1, down­stream SMAD2/3 tar­get genes, in our hiP­SC-CMs showed that the activin recep­tor and its asso­ci­at­ed sig­nal­ing are func­tion­al­ly intact in our mod­el, and that Activin A can tar­get hiP­SC-CMs direct­ly. These data lend sup­port for Activin A in the upreg­u­la­tion of activin recep­tors fol­low­ing myocar­dial infarc­tion, and is in line with pre­vi­ous results in rat mod­els (14).

Inhi­bi­tion of Activin A pre­vent­ed and even reversed Activin A – induced CM dys­func­tion, sug­gest­ing that Activin A plays an impor­tant role in the observed patho­log­ic effects. At the con­cen­tra­tion test­ed (1 nM), nei­ther GDF11 nor MSTN (GDF8) induced as robust a response as Activin A in the HECT mod­el. These find­ings are con­sis­tent with oth­er pre­clin­i­cal data sup­port­ing Activin A as the pri­ma­ry lig­and dri­ving car­diac dys­func­tion (15), as well as evi­dence show­ing that inhi­bi­tion or dis­rup­tion of the activin recep­tor type 2 sig­nal­ing path­way improves car­diac func­tion fol­low­ing exper­i­men­tal heart fail­ure or myocar­dial infarc­tion (15, 49). Inhibit­ing activin recep­tor type 2 and/​or TGF‑β recep­tor sig­nal­ing fol­low­ing myocar­dial infarc­tion may also improve cal­ci­um han­dling (49).

In both mice and humans, IL-1β and IL6 are sig­nif­i­cant­ly upreg­u­lat­ed fol­low­ing myocar­dial infarc­tion and with advanced heart fail­ure (5053). In human pri­ma­ry car­diac fibrob­lasts, we found that IL-1β induced a strong upreg­u­la­tion of INH­BA, the gene encod­ing the homod­imer­ic sub­units of Activin A. How­ev­er, in con­trast to pub­lished data col­lect­ed in neona­tal CMs (54), IL-1β had no direct effect on con­trac­til­i­ty in our hiP­SC-CM assay. These data sug­gest that ele­vat­ed lev­els of IL-1β in the serum of patients with heart fail­ure or COVID-19 may act on car­diac fibrob­lasts to upreg­u­late Activin A, which then direct­ly impairs car­diac con­trac­til­i­ty. While COVID-19 was the not the ini­tial focus of our work, recent clin­i­cal stud­ies high­light­ed cor­re­la­tions between ele­vat­ed Activin A lev­els and dis­ease sever­i­ty in COVID-19 (10, 11). In one pre­clin­i­cal study, hiP­SC-CM treat­ed with serum from patients with COVID-19 dis­played marked arrhyth­mia (55). How­ev­er, the arrhyth­mia was not abol­ished by adding the IL-1β inhibitor canakinum­ab to the cul­ture. These data from Dimai et al. sug­gest that, while IL-1β is ele­vat­ed in COVID-19, addi­tion­al factor(s) are respon­si­ble for induc­ing arrhyth­mia (56). While based on our data this may be Activin A, it is impor­tant to note that lev­els of cir­cu­lat­ing TGF‑β and Activin A have both been shown to cor­re­late with dis­ease sever­i­ty in COVID-19 (15, 57). Since TGF‑β and Activin A both induce SMAD3 phos­pho­ry­la­tion and sim­i­lar down­stream sig­nal­ing, the rel­a­tive impor­tance of Activin A– vs TGF‑β – induced car­diac dys­func­tion in patients with COVID-19 requires fur­ther elu­ci­da­tion. Our data sug­gest that, in our hiP­SC-CM mod­el, activin induces CM dys­func­tion inde­pen­dent of TGF‑β.

The results of this study show that Activin A direct­ly impairs CM func­tion and sug­gest that ele­vat­ed lev­els of Activin A with aging, post COVID-19 infec­tion, or after car­diac injury may direct­ly con­tribute to car­diac dys­func­tion. Fur­ther­more, we have iden­ti­fied a poten­tial mech­a­nism through which IL-1β may indi­rect­ly impair CM con­trac­til­i­ty through its upreg­u­la­tion of Activin A in car­diac fibrob­lasts (sum­ma­rized in Sup­ple­men­tary Fig­ure 3). This is the first study to demon­strate that Activin A acts direct­ly on CMs, which may con­tribute to the car­diac dys­func­tion seen in aging pop­u­la­tions, in patients with heart fail­ure, and in patients with COVID-19 – relat­ed morbidities.

Data avail­abil­i­ty statement

Qual­i­fied researchers may request access to study doc­u­ments and gene expres­sion data that sup­port the meth­ods and find­ings report­ed in this man­u­script. The RNAseq data (Fig­ure 5) has been uploaded to NCBI. Sub­mis­sion ID: SUB12161861 and Bio­Pro­ject ID: PRJNA891053.

Ethics state­ment

All ani­mals received humane care in com­pli­ance with IACUC and the Prin­ci­ples of Lab­o­ra­to­ry Ani­mal Care for­mu­lat­ed by the Nation­al Soci­ety for Med­ical Research and the Guide for the Care and Use of Lab­o­ra­to­ry Ani­mals pre­pared by the Insti­tute of Lab­o­ra­to­ry Ani­mal Resources and pub­lished by the Nation­al Insti­tutes of Health (NIH pub­li­ca­tion 85 – 23, revised 1985).

Author con­tri­bu­tions

SM, JM, OZ, GH, MPM‑P, MPG, and MED con­ceived and designed the research. JM, QR, DJ, KP, HE, XJ, DZ, JT, NTF, and AS per­formed the exper­i­ments. SM, JM, and QR ana­lyzed the data. SM, JM, QR, OZ, GH, DJ, KP, NTF, MPG, MED, DG, and LM inter­pret­ed the results of exper­i­ments. SM, JM, OZ, and DJ pre­pared the fig­ures. SM, JM, OZ, and GH draft­ed the man­u­script. All authors edit­ed and revised the man­u­script and approved the final ver­sion of the manuscript.


This pre­clin­i­cal analy­sis was fund­ed by the Regen­eron Phar­ma­ceu­ti­cals, Inc. Med­ical writ­ing sup­port was pro­vid­ed by Prime, Knutsford, UK, sup­port­ed by the Regen­eron Phar­ma­ceu­ti­cals, Inc., accord­ing to Good Pub­li­ca­tion Prac­tice guide­lines (https://​www​.acpjour​nals​.org/​d​o​i​/​10​.​7326​/​M​15​-0288). The spon­sor was involved in the study design and col­lec­tion, analy­sis and inter­pre­ta­tion of data, as well as data check­ing of infor­ma­tion pro­vid­ed in the manuscript.

Con­flict of interest

SM, JM, QR, YZ, GH, DJ, KP, HE, MPM‑P, XJ, DZ, JT, AS, MED, and LM were employ­ees and stock­hold­ers of Regen­eron Phar­ma­ceu­ti­cals, Inc. NTF and MPG were employ­ees and stock­hold­ers of TARA Biosys­tems. SM, JM, and LM had a patent pend­ing (10771US01).

The remain­ing authors declare that the research was con­duct­ed in the absence of any com­mer­cial or finan­cial rela­tion­ships that could be con­strued as a poten­tial con­flict of interest.