AMPK b 1 and AMPK b 2 define an isoform-specific gene signature in human pluripotent stem cells, differentially mediating cardiac lineage specification

AMP-activated protein kinase (AMPK) is a key regulator of energy metabolism that phosphorylates a wide range of proteins to maintain cellular homeostasis. AMPK consists of three subunits: a , b , and g . AMPK a and b are encoded by two genes, the g subunit by three genes, all of which are expressed in a tis-sue-specific manner. It is not fully understood, whether individual isoforms have different functions. Using RNA-Seq technology, we provide evidence that the loss of AMPK b 1 and AMPK b 2 lead to different gene expression profiles in human induced pluripotent stem cells (hiPSCs), indicating isoform-specific function. The knockout of AMPK b 2 was associated with a higher number of differentially regulated genes than the deletion of AMPK b 1, suggesting that AMPK b 2 has a more comprehensive impact on the transcriptome. Bioinformatics analysis identified cell differentiation as one biological function being specifically associated with AMPK b 2. Correspond-ingly, the two isoforms differentially affected lineage decision toward a cardiac cell fate. Although the lack of PRKAB1 impacted differentiation into cardiomyocytes only at late stages of cardiac maturation, the availability of PRKAB2 was indispensable for mesoderm specification Technologies) was used following the manufacturer ’ s protocol, starting with 1 3 10 5 cells/cm 2 . For long-term maintenance of CMs, cells were selected in DMEM without glucose supple-mented with 1% sodium lactate solution and 25 m g/ml genta-micin (72) for 48 h. Afterward, cells were reseeded on Matri-gel-coated cultureware using STEMdiff TM Cardiomyocyte Dissociation Kit (STEMCELL Technologies). Fat oxidation was measured as described (73).

AMP-activated protein kinase (AMPK) is a key regulator of energy metabolism that phosphorylates a wide range of proteins to maintain cellular homeostasis. AMPK consists of three subunits: a, b, and g. AMPKa and b are encoded by two genes, the g subunit by three genes, all of which are expressed in a tissue-specific manner. It is not fully understood, whether individual isoforms have different functions. Using RNA-Seq technology, we provide evidence that the loss of AMPKb1 and AMPKb2 lead to different gene expression profiles in human induced pluripotent stem cells (hiPSCs), indicating isoformspecific function. The knockout of AMPKb2 was associated with a higher number of differentially regulated genes than the deletion of AMPKb1, suggesting that AMPKb2 has a more comprehensive impact on the transcriptome. Bioinformatics analysis identified cell differentiation as one biological function being specifically associated with AMPKb2. Correspondingly, the two isoforms differentially affected lineage decision toward a cardiac cell fate. Although the lack of PRKAB1 impacted differentiation into cardiomyocytes only at late stages of cardiac maturation, the availability of PRKAB2 was indispensable for mesoderm specification as shown by gene expression analysis and histochemical staining for cardiac lineage markers such as cTnT, GATA4, and NKX2.5. Ultimately, the lack of AMPKb1 impairs, whereas deficiency of AMPKb2 abrogates differentiation into cardiomyocytes. Finally, we demonstrate that AMPK affects cellular physiology by engaging in the regulation of hiPSC transcription in an isoform-specific manner, providing the basis for further investigations elucidating the role of dedicated AMPK subunits in the modulation of gene expression.
The serine/threonine kinase 59-adenosine monophosphate (AMP)-activated protein kinase (AMPK) is a central player in cellular energy homeostasis and stress response (1,2). Upon a reduction of intracellular energy levels that is reflected by an increased AMP/ATP ratio, AMPK becomes activated and induces the phosphorylation of substrate proteins. Consequently, catabolic pathways are promoted while anabolism is inhibited to restore energy balance. Because of its fundamental role in various metabolic processes, AMPK is considered a promising target for the treatment of several pathologies, most prominently metabolic disorders, but also cancer and inflammatory diseases (3,4).
AMPK is a heterotrimeric protein complex that is composed of a catalytic a subunit and two regulatory subunits b and g. The a subunits contain the kinase domain and a phosphorylation site that is required for activation of the protein. AMPK activity is supported by myristoylation occurring at the b subunits (5), which additionally comprise a carbohydrate-binding module. The g subunits are responsible for nucleotide binding and thereby allow AMPK to adapt to changes in the energetic requirement of the cell. Although two genes each encode the a and b subunits, three g isoforms exist. Consequently, 12 different heterotrimeric combinations can be formed, which are expressed in a tissue-and species-specific manner. Whereas a1, b1, and g1 seem to be ubiquitously expressed, availability of the other isoforms is more differential (6), whereby different AMPK trimers vary in their physiological properties. Along this line, AMPK protein complexes containing the b2 instead of the b1 isoform possess a greater binding affinity to glycogen (7) and an increased activity following sumoylation (8). Although the specificity of individual AMPK isoforms and distinct trimer combinations is gaining more and more attention, the isoformspecific functions of AMPK are not completely understood.
In the heart, AMPK plays an important role in the adaption of cellular energy balance by modulating glucose and fatty acid metabolism. It has been extensively studied in the context of cardiac dysfunction, particularly in ischemia-reperfusion injury and hypertrophy (23,24). Although a high prevalence of a2and b2-containing AMPK trimers is described for human heart tissue, the only subunit that could clearly be associated with human pathologies is g2 (6). The metabolic characteristics of selective AMPK complexes and single isoforms in different organs are gaining more and more attention, especially with regard to the development of pharmacological modulators of AMPK activity. In contrast to pan-AMPK activators or inhibitors, isoform-selective compounds take advantage of the specificity of AMPK complexes. They feature high kinase selectivity and reduced off-target effects. However, the physiological consequences resulting from the tissue-selective composition of AMPK complexes are not fully understood so far.
In the present study, we investigated the functional differences between b1and b2-containing AMPK trimers in human induced pluripotent stem cells (hiPSCs), providing evidence that the lack of either subunit differently alters the gene expression profile. As a consequence, the differentiation toward the cardiac lineage is affected in an isoform-specific manner, indicating different physiological functions of AMPKb1 and b2 isoforms.

Human induced pluripotent stem cells as a model to investigate isoform-specific functions of AMPKb1 and AMPKb2
To investigate cellular effects that are differentially mediated by AMPKb1 and AMPKb2, we used hiPSCs harboring a gene knockout for one or the other isoform. The WT cell line was selected based on its approximately equal expression of the two AMPKb subunits (Fig. 1A) and KO cell lines were created by CRISPR/Cas9-mediated gene modification (Fig.  1B). PRKAB12/2 cells were generated by insertion of one base pair within exon 1, resulting in an in-frame stop codon. For the deletion of PRKAB2, a fragment of exon 2 was excised, generating a stop codon in exon 2. The growth behavior as well as the morphology of the cell colonies did not differ between PRKAB12/2 or PRKAB22/2 hiPSCs and WT cells (Fig. 1C). Importantly, deficiency of neither AMPKb1 nor AMPKb2 resulted in a compensatory up-regulation of the respective other isoform (Fig. 1D). Of note, protein levels of AMPKa1 and AMPKa2 were unchanged (Fig.  1E). We further confirmed that basal activity of AMPK was not affected by the knockout of either AMPKb1 or AMPKb2. All cell lines tested showed levels of phosphorylated AMPK and of acetyl-CoA carboxylase (ACC) similar to those in WT hiPSCs (Fig. 1, F and G).

AMPKb1and AMPKb2-deficient undifferentiated hiPSCs show isoform-specific gene expression
The transcriptomes of WT hiPSCs as well as two clones of each KO cell line were analyzed by RNA-Seq. For that, as for the experiments described before, two knockout cell lines for each genotype were used. Cell lines were generated from single clones after gene knockout of either PRKAB1 or PRKAB2 to ensure the cultivation of individual cell populations. RNA-Seq was furthermore performed in quadruplicates to allow for solid statistical evaluation of the data. Although 404 genes were regulated with a -fold change .1.3 (p-value ,0.05) relative to WT in PRKAB22/2 hiPSCs, this applied to only 132 genes in PRKAB12/2 hiPSCs ( Fig. 2A). To identify the cellular effects that might arise from these differences, major downstream pathways of AMPK were investigated. However, analysis of the phosphorylation status of regulatory associated protein of mTOR (Raptor) (Fig. S1A) and tuberous sclerosis complex 2 (TSC2) (Fig. S1B) revealed no significant changes. Genes being associated with glucose and lipid metabolism did not vary between PRKAB12/2 or PRKAB22/2 and WT hiPSCs (data not shown). Oxidative metabolism also appeared to be unchanged under basal conditions, as expression levels of PPARGC1A and, consistently, visualization of mitochondria by MitoTracker staining (Fig. S1C) did not reveal any differences between the genotypes. In line with this, b-oxidation of fatty acids in the absence of glucose did not differ between WT and AMPKb1or AMPKb2-deficient cells (Fig. S1D). Corresponding to our previous data, comparative pathway analysis of the gene sets did not show major differences between WT and any of the KO hiPSC lines (data not shown). Although, because of the low number of regulated genes, no solid GO enrichment analysis could be performed in PRKAB12/2 cells, we could identify several biological processes that might be affected by the lack of AMPKb2. Thereby, a large data set of the regulated genes in PRKAB22/2 hiPSCs could be assigned to cell differentiation (Fig. 2B). As we did not detect differences between the two clones of identical genotypes, further analyses were carried out with one clone of each genotype. Gene expression of the pluripotency markers POU Class 5 Homeobox 1 (POU5F1/ OCT4), SRY-box transcription factor 2 (SOX2), and NANOG was analyzed in undifferentiated hiPSCs and was not altered between the three cell lines (Fig. 2C). However, the independent differentiation into the three germ layers revealed lower expression of the mesoderm and endoderm markers T-box transcription factor T (T/Brachyury) and SOX17 in PRKAB22/2 relative to WT cells (Fig. 2D). In contrast, mRNA levels of the ectoderm marker PAX6, although lower in undifferentiated cells, was significantly higher in cells derived from PRKAB22/2 than in those from WT hiPSCs. PRKAB12/2 hiPSCs consistently showed a differentiation potential similar to WT cells. Together, our data indicate that AMPKb1and AMPKb2-containing AMPK trimers attain unique physiological effects that, in hiPSCs, affect differentiation potential. Specifically, the availability of AMPKb2 seems to be required for the development toward the mesodermal and endodermal lineage.

Deficiency of PRKAB1 impairs, whereas loss of PRKAB2 abrogates, differentiation into cardiomyocytes (CMs)
To further evaluate the impact of either AMPKb1 or AMPKb2 deficiency on mesoderm differentiation, we applied all three cell lines to a standardized cardiomyocyte differentiation protocol (Fig. 3A). Protein expression of AMPKb1 and of AMPKb2 increased during differentiation in WT as well as in both knockout cell lines, whereas no continuous differences in protein levels were observed between WT and PRKAB12/2 or PRKAB22/2 hiPSCs (Fig. S2A). Thereby, the appearance of lineage-specific marker genes was monitored (Fig. 3B). SOX2 and OCT4 as well as early differentiation markers including the mesendoderm-associated gene T and the homeobox protein MIXL became massively down-regulated over time in WT and PRKAB12/2 hiPSCs. Strikingly, in PRKAB22/2 cells, expression of these marker genes remained rather high, whereas genes that indicate early cardiac cell fate such as islet-1 (ISL1) and myocyte enhancer factor 2C (MEF2C) showed much lower induction than in WT. The same was true for TNNT2, encoding for cardiac troponin T2 (cTnT) as well as for signal regulatory protein alpha (SIRPA), two markers commonly used to identify cardiac cell populations. Unlike PRKAB22/2 cells, WT and PRKAB12/2 hiPSCs strongly increased TNNT2 transcription starting from around day 6 of differentiation. The differences in gene expression were equally translated to con-siderably lower protein content (Fig. 3, C-E and Fig. S2B). To evaluate differentiation efficiency, we quantified the number of cells co-expressing cTnT and a-actinin, one feature of mature and functional CMs. On average 30, 18, and 1.4% cTnT-and a-actinin-positive cells were obtained after metabolic selection and 13 days of maturation in WT, PRKAB12/2, and PRKAB22/2cells, respectively (Fig. 3E). Consistently, expression of genes encoding for sarcomere and calciumhandling proteins, a prerequisite for cardiac contractility, was lower in PRKAB12/2 cells relative to WT and only minimal in PRKAB22/2 cells (Fig. S3A). Interestingly, phosphorylation of ACC was similar in all three cell lines throughout differentiation (Fig. S2C), suggesting that AMPK activation did not significantly change. In search for regulatory mediators of cardiac development that might be affected by the deficiency of AMPKb1 and b2, we found that the up-regulation of the key cardiac transcription factors including GATA4 and Figure 1. Characterization of the cellular model used to study isoform-specific effects of AMPK. A, amount of AMPKb1/b2 protein in WT hiPSCs determined by Western blotting and calculated by means of a standard curve of recombinant AMPK (n = 10 independent experiments, *, p , 0.05). B, KO strategy used to generate PRKAB12/2 and PRKAB22/2 hiPSCs. Gray boxes frame gRNA sequences, arrows indicate cutting sites. One base insertion is in bold. C, representative phase-contrast images of cell colonies (scale bars, 50 mM). D, Western blots and quantification of total AMPKb1 in WT and PRKAB22/2 hiPSCs (left) and AMPKb2 in WT and PRKAB12/2 hiPSCs (right). Two KO clones per genotype were analyzed (n = 8 independent experiments; n.d., not detectable). E, Western blots and quantification of AMPKa1 (left) and AMPKa2 (right) in WT, PRKAB12/2, and PRKAB22/2 hiPSCs. Signal intensity is shown relative to WT (n = 4/ 6 independent experiments, two clones per KO cell line were analyzed and plotted as individual bars). F and G, representative Western blotting and quantification for phosphorylated AMPK in WT, PRKAB12/2, and PRKAB22/2 hiPSCs. Signal intensity is shown relative to WT (n = 6/5 independent experiments). Dotted lines in (F) indicate blot splicing, to delete irrelevant sample lanes.
The AMPK target PGC-1a, encoded by PPARGC1A, is essential for proper development of CMs from embryonic stem cells (25) and was shown to be part of a co-expression network defining cardiac identity (53). PPARGC1A expression was significantly lower in PRKAB22/2 cells starting from day 8 onward (Fig. 3G). Furthermore, a number of miRNAs being required for the activation of CM-specific genes and thus for fine-tuning lineage specification were analyzed (26)(27)(28). Their expression pattern mirrored that of the cardiomyocytespecific genes (Fig. S4). Taken together, these data indicate that AMPKb1 and AMPKb2 differentially affect the differentiation potential of hiPSCs toward the cardiac lineage, involving various regulatory mediators. The lack of AMPKb1 affects cardiac differentiation at the progenitor stage, impairing maturation into fully functional cardiomyocytes. In contrast, PRKAB2-deficient hiPSCs hardly give rise to cTnT-positive cells, suggesting that AMPKb2-containing isoforms are required for early lineage specification.

Lack of AMPKb2 impairs mesendoderm lineage decision of hiPSCs
The loss of pluripotency is an essential prerequisite for the differentiation of stem cells into any lineage. Following up on our previous experiments that suggest deficiencies in the development of PRKAB22/2 hiPSCs already at early time points, we observed a higher number of Oct-3/4-positive cells at day 8 of differentiation in PRKAB22/2 relative to WT and PRKAB12/2 cells (Fig. 4, A and B). Similar results were obtained by qRT-PCR for OCT4 (Fig. 4C) and SOX2 (Fig. 4D). Based on the fact that AMPKb2-deficient hiPSCs seem to be specifically limited in their differentiation potential toward mesoderm and endoderm (Fig. 2H), we speculated that significant impairments might manifest already before the time point when both lineages separate from a common precursor. Consistently, the mesoderm/mesendoderm marker Brachyury/T showed markedly lower expression at early stages of differentiation, whereas levels appeared to be slightly higher at later time points compared with WT as well as PRKAB12/2 hiPSCs (Fig. 4E). Expression of MESP1, however, a marker for cardiac mesoderm, was barely detectable and below levels of WT and PRKAB12/2 cells throughout differentiation (Fig. 4F). The process of stem cell development toward mesendoderm is mediated by a coordinated crosstalk of different signaling pathways including transforming growth factor-b, Wnt, and fibroblast growth factor signaling (29)(30)(31)(32). Consequently, the AMPKb1 and b2 have different functions in hiPSCs expression of some key mediators responsible for the coordinated translation of external signals into a transcriptional response (33)(34)(35) was analyzed in PRKAB22/2 and WT hiPSCs at day 2 of differentiation, the time point at which the cells pass the mesendoderm stage (Fig. 3B). Expression of SMAD2, SMAD4, and FOXH1, which modulate BMP3 and activin signaling, as well as of CTNNB1 encoding for the Wnt mediator b-catenin, were repressed in PRKAB22/2 relative to WT hiPSCs (Fig. 4G). Altogether, these data predict that lack of AMPKb2 impairs already early stages of mesendoderm differentiation.
AMPKb1 and AMPKb2 are required for the expression of pluripotency and cardiac lineage genes To confirm that the alterations in cardiac gene expression can be attributed to the lack of either AMPKb1 or AMPKb2, we re-expressed the missing isoform in the respective KO cells. For that, PRKAB1 and PRKAB2 cDNA was cloned into a lentiviral vector under the control of the EF-1a promoter (Fig. 5A). Successful integration and expression of the protein was confirmed (Fig. 5B). Although the rescue of AMPKb1 and AMPKb2 protein expression was only around 10 and 5% of the WT levels, the partial reconstitution of AMPKb2 was sufficient to significantly decrease OCT4 as well as SOX2 expression relative to PRKAB22/2 cells starting from day 4 of differentiation (Fig. 5C). This suggests that even a low amount of AMPKb2 is sufficient to initiate cell differentiation. Consistently, expression levels of GATA4 and NKX2.5 were increased in cells reexpressing AMPKb1 and AMPKb2 (Fig. 5D). Together, these experiments illustrate the isoform-specific function of AMPK on the expression of dedicated genes, thereby modulating cell physiology in different ways.

Discussion
In the present study, we demonstrate that AMPK heterotrimers containing either the b1 or the b2 subunit in the protein complex account for distinct gene expression profiles in human iPSCs and, as a consequence, differentially affect downstream physiological processes. Specifically, we show that AMPKb1 and b2 influence cardiac differentiation potential in an isoform-specific manner. Lack of PRKAB1 impairs, but does not prevent, the development of CMs. In contrast, PRKAB2 deficiency nearly abrogates cardiac differentiation most likely by prohibiting lineage decision at the mesendoderm stage.
The hiPSC line used in our study was chosen because it possesses similar amounts of both AMPKb isoforms. Although we failed to generate iPSCs lacking both subunits, single knockout of either AMPKb1 or AMPKb2 was successful. The knockout of neither PRKAB1 nor PRKAB2 caused a significant counter-regulation of other AMPK subunits, excluding compensatory mechanisms masking AMPKb-specific effects. RNA-Seq analysis of WT, PRKAB12/2, and PRKAB22/2 hiPSCs revealed major differences in their transcriptomes and thereby for the first time provides evidence for an AMPKb isoform-specific gene expression profile. The WT cells used in our study were not transfected with a control gRNA. However, as differences in gene expression were observed in two clones of each genotype relative to WT and also between cells lacking one or the other AMPKb isoform, off-target effects caused by the knockout procedure could be excluded. Furthermore, reconstitution of the missing AMPK subunit restored gene expression in the respective knockout cell line, additionally supporting our conclusion that the differences between WT and KO hiPSCs are because of isoform-specific effects on gene expression. Basal phosphorylation of AMPK was the same in WT as in both KO cell lines. With that, the changes in gene expression were most likely not because of variations in overall cellular AMPK activity, but rather caused by the downstream effects that selectively depend on the presence or selective activation of AMPKb1 and b2. The two isoforms possess only 71% sequence homology (36), so that b1and b2-containing AMPK trimers undergo different posttranslational modifications and vary in their affinity to carbohydrates. This might result in different substrate affinities and subcellular localization of the protein (7,8,18), which could provide the molecular basis for the functional differences between AMPK isoforms containing either the b1 or the b2 subunit observed in our study. Another possibility is that b1and b2-containing trimers might be differentially regulated. Evidence for an isoform-selective activation of AMPK has been obtained in skeletal muscle (37). AMPK can directly affect gene expression via several mechanisms and by the interaction with different factors. For example, it could be shown that AMPK is able to phosphorylate a number of transcription factors and transcriptional coactivators including PGC-1a, FOXO3, CREB proteins and others, thereby impacting gene expression profiles in different tissues (9)(10)(11). Furthermore, AMPK was proven to have functions in histone modification and epigenetic mechanisms as histone H2B and p300 were identified as substrates of AMPK (15,16). At this point, we can only speculate that the selection of AMPKb1 or AMPKb2 in the protein complex intervenes with those diverse and multifactorial mechanisms and, especially in stem cells, directs cell fate decision. However, further studies will be needed to decipher the role of dedicated AMPK subunits on stem cell physiology and function.
Analysis of the differentially expressed genes identified by RNA-Seq did not allow an assignment of any pathway to either b1or b2-containing trimers, suggesting that these discrepancies do not significantly impact downstream effects of AMPK in the pluripotent state. This corresponds to the suggestion of

AMPKb1 and b2 have different functions in hiPSCs
Young et al. (38) that AMPK plays a minor role in undifferentiated stem cells under basal conditions. However, several developmental processes and cell differentiation GO terms could specifically be associated with AMPKb2, indicating that the consequences induced by differential gene expression predominantly emerge under differentiating conditions. Trilineage differentiation experiments confirmed this finding. Although cell fate decision into mesoderm and endoderm was reduced in AMPKb2-deficient hiPSCs relative to WT, ectoderm development was facilitated. These findings substantiate previous results, indicating that AMPK activity is not required for the lineage decision toward ectoderm (38).
However, AMPK has been shown to be involved in the differentiation of stem and progenitor cells (39)(40)(41) but, unlike for the two isoforms of AMPKa, any specificity for either b1or b2-containing trimers in respect to cell fate determination was not described before. Our three stem cell genotypes varied in their cardiac differentiation potential, which was mirrored by major discrepancies in the expression of cardiac lineage markers as well as genes being associated with sarcomere function. Expression levels of AMPKb1 as well as of AMPKb2 continuously increased between day 0 and day 15 of differentiation. Significant differences between WT and either of the knockout hiPSCs could only be detected at one time point of differentia-tion in PRKAB22/2 cells, suggesting that the lack of one b isoform does not induce a general mechanism leading to a compensatory up-regulation of the respective other b isoform. Literature data illustrating the expression pattern of AMPKb during cardiac development in vivo are rare. Kim et al. (42) investigated this aspect in mice and observed a drop of the expression of both AMPKb1 and b2 from E15 to day 1 after birth as well as to adulthood. However, those studies are hardly comparable to human in vitro models as species-specific differences are likely. Furthermore, in vitro stem cell models poorly represent the in vivo cardiovascular environment (43)(44)(45). To our knowledge, our study for the first time investigates the dedicated impact of AMPKb isoforms on human cardiac development in a stem cell-based model. Isoform-specific functions of AMPK have been reported for cardiac physiology as AMPK activation modulates contraction of CMs by phosphorylating cTnT, leading to an increase in calcium sensitivity and contractility. This is mainly attributed to AMPKa1 (46,47). Moreover, AMPK is involved in the O-GlcNAcylation of certain proteins including troponin (48), which in turn has an impact on cTnT phosphorylation and cardiac function (49,50), underlining the important role of AMPK in heart physiology. The switch from glycolysis to fatty acid oxidation as the main energy source is a prerequisite for proper differentiation and maturation of CMs  and was shown to be indispensable for cell cycle exit and cardiac progenitor cell maturation of hPSC-derived cardiac organoids (51). The shift to oxidative phosphorylation requires mitochondrial network formation that is mediated by PGC-1a, a well-described target of AMPK. PGC-1a becomes up-regulated during cardiac development that is tightly associated and temporally dependent on the expression of the transcription factor NKX2.5 in hESC-derived cardiac progenitor cells (25). NKX2.5, together with other highly conserved transcription factors including GATA4, TBX5 and ZEB1, belongs to a regulatory network controlling heart development in vivo and cardiogenesis in vitro (52,53). Lack of NKX2.5 in human ESC-derived CMs causes decreased expression of cardiac differentiation markers as well as impaired electrophysiology and contractile function (54). Overall, these reports correspond to our data. In WT cells, massive induction of NKX2.5 and TBX5 transcription and a simultaneous up-regulation of PPARGC1A were observed at day 8 of differentiation that exactly coincides with an increase in the expression of TNNT2 and other sarcomere genes. Strikingly, lack of AMPKb1 as well as AMPKb2 showed major disturbances in this gene regulatory network, although to different extents. This is additionally confirmed by changes in the expression of specific miRNAs that are equally important for cardiac differentiation and physiology (26,(55)(56)(57), among others, altering the expression of cardiac transcription factors (58)(59)(60).
Although AMPKb1-deficient hiPSCs maintain the ability to differentiate into CMs, lack of AMPKb2 seems to abrogate it by attenuating the exit from pluripotency and the progression of differentiation. As the expression of effectors mediating activin, BMP and Wnt signaling, which coordinately promote the exit from pluripotency and subsequent mesoderm induction, was significantly decreased in cells lacking AMPKb2, isoforms containing this subunit are obviously critical at very early stages of differentiation and required for cells to pass the mesendoderm stage. Wnt and activin signaling are modulated by YAP (61,62), which can directly be phosphorylated by AMPK (63) and is under certain circumstances a target of O-GlcNAcylation (64). In addition, metabolite availability in hESCs can affect lineage decision via activation of AMPK (65).
Taken together, the mechanism underlying the phenotype observed in our stem cell model is most likely multifaceted and might include direct phosphorylation and posttranslational modifications of effector proteins at different levels of regulation. Besides an altered miRNA expression pattern, one can expect epigenetic modulations to interfere with an AMPKb-dependent regulatory network that finally defines the transcriptional program and metabolic adaptions of the cells. Moreover, subcellular localization of AMPK might be different (66). However, further investigations are needed to identify the mechanism(s) underlying the isoform-specific mRNA profiles in the present model.
In an apparent contradiction to our data, a major heart phenotype in whole body AMPKb2 knockout animals has not been described (67,68). Nevertheless, mice with a muscle-specific deletion of AMPKb1/b2 showed cardiac dysfunction going along with decreased phosphorylation of troponin (69). We speculate that, in vivo, mechanisms compensating for the lack of AMPKb2 might exist that cannot equally take place during an in vitro differentiation protocol. Furthermore, a time-dependent regulation of AMPK activity that might take place during embryonic development is probably not equally adapted in a cell culture model. Along this line, it was shown recently that pharmacological activation of AMPK in cardiac progenitor cells under differentiation is neither sufficient to alter gene expression of differentiation and CM markers, nor promote PPARGC1A transcription and mitochondrial biogenesis (70). A recent publication suggests an increase in AMPK activation during cardiac differentiation of hiPSCs, reaching highest levels of phosphorylation at days 9 and 11 followed by a drop of protein activation at day 14 (71). Our data do not confirm this finding, which might be because of differences in the differentiation protocol. Furthermore, analyzing the phosphorylation status of ACC to monitor AMPK activation during differentiation, we could not detect significant differences between WT and either of the KO cell lines. However, we cannot preclude that those differences might still exist but are not detectable because of insufficient sensitivity of Western blot analysis.
Although further investigations are required to precisely decipher the molecular mechanisms underlying the isoform-specific effects of AMPKb1/b2 during differentiation, our data provide evidence that the AMPK isoforms exhibit different biological functions in hiPSCs. Whether those differences can also be observed in other cellular systems will be the substance of future studies.

Experimental procedures
Generation of hiPSC KO cell lines PRKAB1 and PRKAB2 KO hiPSC lines were purchased as customized products from Applied StemCell. In brief, KO cell lines were generated from a WT hiPSC line (no. ASE-9202; ASC) by CRISPR/Cas9-mediated gene modification. For this, one (CGCGCCGCGCTGGAGCGGCANGG) or two (ACCA-CCAGCGACCGGGTGTCNGG, GGGACGCTTACCTTGG-AGTCNGG) gRNAs were utilized to create a frameshift KO mutation of the PRKAB1 or PRKAB2 gene, respectively. Individual iPSC colonies were expanded and KO was verified by sequence analysis. Positive staining for Oct-4, Sox-2, SSEA-4, TRA-1-60, and TRA-1-81 and for alkaline phosphatase as well as normal karyotype was approved by Applied StemCell.

Maintenance of human iPS cells
Undifferentiated hiPSCs were cultured on Matrigel-coated plates seeded with mTeSR TM Plus medium; passaging of the cells was performed using ReLeSR TM (STEMCELL Technologies). Single cell suspensions were prepared using Accutase. After cell seeding, medium was substituted with ROCK inhibitor (BioVision) for 24 h.

Differentiation of hiPSCs and maintenance of CMs
Differentiation into mesoderm, ectoderm, and endoderm was performed using STEMdiff TM Trilineage Differentiation Kit (STEMCELL Technologies). For cardiac differentiation, the STEMdiff TM Cardiomyocyte Differentiation Kit (STEMCELL AMPKb1 and b2 have different functions in hiPSCs Technologies) was used following the manufacturer's protocol, starting with 1 3 10 5 cells/cm 2 . For long-term maintenance of CMs, cells were selected in DMEM without glucose supplemented with 1% sodium lactate solution and 25 mg/ml gentamicin (72) for 48 h. Afterward, cells were reseeded on Matrigel-coated cultureware using STEMdiff TM Cardiomyocyte Dissociation Kit (STEMCELL Technologies). Fat oxidation was measured as described (73).

Reconstitution of hiPSC KO cell lines
The coding sequences of PRKAB1, PRKAB2, and T2A-Puro were ordered as gBlocks from Integrated DNA Technologies. The respective PRKAB1/2-T2A-Puro in-frame fusion constructs were assembled via Golden Gate cloning technique. The sequence integrity was verified by Sanger sequencing. PRKAB1/2-T2A-Puro elements were excised with EcoRI/XbaI and cloned into a lentiviral vector under the EF1a promoter. VSV-G pseudotyped lentiviruses were produced with LV-MAX lentiviral production system according to standard protocols (Gibco).

RNA-Seq and bioinformatics analysis
Total RNA was purified using RNeasy kit (Qiagen) and sequencing was performed at ATLAS Biolabs GmbH. Bioinformatics analyses of RNA-Seq raw data were performed using the Array Studio software package version 10.1.2.9 (OmicSoft, Qiagen). Mapping of RNA-Seq reads (fastq files) to the human genome was done using Omicsoft gene model 20130723 with human B37.3 as reference genome and OSA4 as aligner. Genes with FPKM ,1 were filtered out according to 50% fraction. Principal component analysis has been applied to all samples as a quality control assessment. Differentially expressed genes were detected by applying the DESeq2 statistical test to the comparisons of PRKAB12/2 and PRKAB22/2 clones versus WT. P-values were adjusted by the Benjamini-Hochberg procedure to control FDR. Changes in expression levels .1.3fold and an adjusted p-value ,0.05 were considered significant. Pathway and mechanistic analyses of differentially expressed genes were performed using Ingenuity Pathway Analysis (Qiagen).
Gene Ontology (GO) enrichment analysis GO enrichment analysis was performed using an online tool (RRID:SCR_002811) being based on the PANTHER classification system. Datasets containing genes being differentially expressed between WT and PRKAB12/2 or PRKAB22/2 hiPSCs were subjected to the PANTHER Overrepresentation test for GO biological process. Release version of the GO database was 2019-10-08. Fisher's exact test with FDR correction was calculated automatically. Only results with FDR ,0.05 were displayed.

qRT-PCR
RNA isolation was performed in a QIACube device using an RNeasy kit (Qiagen). High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) was used for reverse transcrip-tion and qRT-PCR was performed with TaqMan ® Assays (Life Technologies) in a LightCycler 480 instrument (Roche). Raw values were normalized to GAPDH. Assay IDs are listed in Table S1.
Western blotting SDS-Page followed by Western blotting was performed by standard procedures. Antibodies were from Cell Signaling Technology. Recombinant AMPK protein (A2/B1/G2, A2/B2/ G2; SignalChem) was plotted as a concentration curve and used to calculate the actual amount of AMPKb1 and AMPKb2 protein from total cell lysates.

Immunocytochemistry
For immunofluorescence staining, cells were detached and reseeded on laminin-coated 96-well plates (PerkinElmer). After attachment, cells were fixed in 4% PFA and staining was performed following a standard protocol. Primary antibodies were rabbit anti-cTnT (Abcam), mouse anti-a-actinin (Sigma), goat anti-Oct3/4 (R&D Systems). Alexa Fluor 488 or 555-conjugated secondary antibodies were used (Invitrogen) and nuclei were stained with Hoechst 33342 (Roche). Images were taken and analyzed using a spinning disc confocal microscope (Perki-nElmer) and the corresponding imaging software Harmony V4.8.

Statistical analysis
Data are shown as mean 6 S.E. Statistical significance was analyzed by two-tailed Student's t test (GraphPad Prism software version 8.0.2 for Windows). p values ,0.05 were considered significant. Maximal three significant digits are shown.

Data availability
Raw data of RNA-sequencing is available at NCBI-GEO under accession number GSE144043. GO, Gene Ontology; hESC, human embryonic stem cell; CM, cardiomyocyte.