Activation of Nkx2.5 transcriptional program is required for adult myocardial repair

The cardiac developmental network has been associated with myocardial regenerative potential. However, the embryonic signals triggered following injury have yet to be fully elucidated. Nkx2.5 is a key causative transcription factor associated with human congenital heart disease and one of the earliest markers of cardiac progenitors, thus it serves as a promising candidate. Here, we show that cardiac-specific RNA-sequencing studies reveal a disrupted embryonic transcriptional profile in the adult Nkx2.5 loss-of-function myocardium. nkx2.5−/− fish exhibit an impaired ability to recover following ventricular apex amputation with diminished dedifferentiation and proliferation. Complex network analyses illuminate that Nkx2.5 is required to provoke proteolytic pathways necessary for sarcomere disassembly and to mount a proliferative response for cardiomyocyte renewal. Moreover, Nkx2.5 targets embedded in these distinct gene regulatory modules coordinate appropriate, multi-faceted injury responses. Altogether, our findings support a previously unrecognized, Nkx2.5-dependent regenerative circuit that invokes myocardial cell cycle re-entry, proteolysis, and mitochondrial metabolism to ensure effective regeneration in the teleost heart.

D ue to the limited responsiveness of endogenous cellular and molecular repair mechanisms in injured myocardium, cardiac dysfunction remains a key cause of morbidity and mortality in patients. Coronary artery disease leads to myocardial infarction which results in fibrotic scarring, adverse remodeling, and heart failure 1 . Furthermore, in patients with congenital heart disease, cardiac ischemia yields cardiomyocyte (CM) loss and the induction of inappropriate healing responses 2 . Despite these pathogenic mechanisms that ensue following injury, recent reports demonstrate that adult CMs preserve the capacity to replicate in both mice [3][4][5][6][7][8] and humans [9][10][11] . Therefore, identifying key factors that enhance CM proliferation offers opportunities to stimulate myocardial repair in patients suffering from cardiac dysfunction or hypoplasia.
Interestingly, in contrast to adult mice and humans, neonatal mice and adult zebrafish retain the ability to undergo CM dedifferentiation and proliferation, optimizing cardiac regenerative potential [12][13][14][15] . Specifically, new CMs emerge following division of pre-existing CMs to repopulate the injury site 12,13 . Furthermore, despite the propensity of most mammalian CMs to undergo polyploidization 16 , a positive correlation between the maintenance of a mononuclear, diploid state, and retention of cardiac regeneration capacity in both mice and zebrafish has been established 17,18 . Taken together, exploiting the pro-regenerative mechanisms innate in zebrafish and applying them to diseased human hearts would provide an arsenal of tools to awaken constructive healing in both adult and pediatric patients 19,20 .
Parallel transcriptional networks have been shown to be essential for both development and regeneration of several tissues including the heart 13,21,22 . For example, reactivation of genes essential during cardiogenesis is critical to stimulate CM production in damaged myocardium 23 . Thus, we took advantage of our prior investigation of the Nkx2.5 transcriptional regulatory pathways in the zebrafish embryo to provide insights into the developmental mechanisms that are harnessed for activation of cardiac regenerative potential. The homeodomain transcription factor, NKX2-5, is one of the most commonly mutated genes associated with human congenital heart disease (CHD) and is a master regulator of cardiac development 24 . Studies in model organisms highlight crucial roles of vertebrate and invertebrate homologs of NKX2-5 in cardiac specification and morphogenesis [25][26][27][28] . While the functions of Nkx genes in early development have been examined, it remains unclear whether Nkx2-5 is necessary to mount an efficient response for CM renewal. In this study, we inspect the distinct cellular and molecular processes regulated by Nkx2-5 in injured myocardium to elucidate effective therapeutic strategies for myocardial regeneration.
Our previous work illuminates that temporally controlled expression of nkx2.5 is required to preserve chamber-specific identity maintenance in both the first and second heart fields during discrete developmental time windows [29][30][31] . Moreover, in the context of otherwise embryonic lethal nkx2.5 −/− embryos, early overexpression (21 somites) of nkx2.5 is sufficient to sustain cardiac function and ensure embryonic viability into adulthood 29 . These results regarding the temporal regulation of nkx2.5 broaden our appreciation of the essential functions of early cardiac transcriptional regulation in establishing long-standing myocardial health. Thus, we exploited this unique model to study the role of nkx2.5 in the adult heart following injury. Here, we show that rescued nkx2.5 −/− fish demonstrate a diminished regenerative response following amputation of the ventricular apex. Compromised sarcomere dedifferentiation and decreased proliferation dampen the requisite regrowth of new CMs from pre-existing CMs in the nkx2.5 −/− fish. Moreover, the epicardium fails to penetrate the regenerate to establish the tissue environment necessary to induce a cascade of signals required for healing. Altogether, our data illustrate the critical function of Nkx2.5 in the stressed adult myocardium to invoke regenerative cues for CM dedifferentiation, renewal, and effective patterning of a supportive microenvironment. Understanding the mechanisms guiding Nkx2.5-dependent cardiac regeneration will have an impact on the management of patients with CHD or cardiomyopathies associated with NKX2-5 mutations.
To substantiate our findings with quantitative strategies, we applied previously established morphometrics to assess for normal postembryonic cardiac growth. Precisely, we observed no statistically significant difference in the percentage of trabeculation between non-transgenic, transgenic wild-type, and nkx2.5 −/− hearts ( Fig. 1Q) 32 . However, ventricular volume is slightly increased in the nkx2.5 −/− fish when compared to nontransgenic and transgenic wild-type fish, suggesting mild ventricular expansion (Fig. 1R). To assess the significance of this subtle variability in chamber size, we tested for possible signs of cardiac failure by subjecting the fish to an exercise routine 33 . In a swim tunnel, we measured endurance, the average time during which animals could maintain orientation facing a high velocity current, and found no statistically significant differences between all genotypes assessed (Fig. 1S). Together, we conclude that, although future investigation may help to unravel these slight differences in ventricular size, Nkx2.5 is not necessary following embryogenesis to ensure long-term viability and fitness into adulthood.
Molecular signatures of adult nkx2.5 −/− hearts are perturbed. Despite intact cardiac formation in the nkx2.5 −/− compared to wild-type fish, we posited that there may be alterations in the underlying molecular signatures given the absence of this critical cardiac transcription factor. Thus, we performed RNA sequencing (RNA-seq) analysis of ventricular apices extracted from nontransgenic wild-type, transgenic wild-type, and nkx2.5 −/− hearts.
We investigated the fundamental molecular similarities and differences of these three genotypes by retrieving the subsets of statistically significant, differentially expressed genes (DEGs). To isolate the set of DEGs that are misregulated in the nkx2.5 −/− fish (MT0) independent of the control sample used (WT0 or TG0), we established a subgroup of 1073 genes (hereafter referred to as group C or the 'overlap' group, Fig. S1A). This intersection category reflects the critical genes that are differentially expressed in the nkx2.5 −/− heart when compared to both wild-type conditions while excluding those genes that are misregulated between the transgenic and non-transgenic wild-type samples.
We then explored the Gene Ontology (GO) categories most highly enriched in each of the comparisons. Although several GO terms were similarly enriched in the MT0 vs WT0 (group A) and   Fig. 2A), we refer to the 'overlap' group as the most accurate depiction of the DEGs between the adult nkx2.5 −/− and wild-type hearts. Muscle structure development, actin filament-based processes, heart development, and circulatory system processes are the enhanced GO terms ( Fig. 2A, 'overlap' group), underlining our findings that the developmental pathways regulated by Nkx2.5 in the embryo are analogously disrupted in the adult loss-of-function model. We also generated a heat map from the 'overlap' group that displays the z-score for genes with relatively higher or lower expression levels between samples (Fig. 2F). This representation of the 'overlap' dataset underscores the consistency between the replicates within each genotype. Moreover, these data emphasize that approximately 2/3 of the DEGs in the nkx2.5 −/− fish are upregulated and 1/3 is downregulated, consistent with reported repressive functions of Nkx2.5 [34][35][36] .
While this methodical strategy of defining the 'overlap' category captures variations in the molecular profile unique to the loss of nkx2.5 gene function, we questioned whether hsp70l promotor leakiness could explain the observed differences between TG0 and WT0. Our prior data illustrates absence of Tg(hsp70l:nkx2.5-EGFP) expression during embryogenesis 29 . We thus inspected the potential for overexpression in the adult heart. Interestingly, qPCR analysis of nkx2.5 expression illuminates no statistically significant difference between levels detected in TG0 and WT0 hearts (Fig. S1B). Given that the presence of Tg(hsp70l:nkx2.5-EGFP) does not explain the differences observed between TG0 and WT0 datasets, we dissected the most highly enriched GO terms in group D ( Fig. 2A) and found autophagy, mitochondrion organization, and ribosome biogenesis. Thus, we conclude that aberrations in these 'housekeeping' functions constitute the primary differences in this comparison and are distinct from the molecular changes observed in the MT0 comparisons. Indeed, while 'housekeeping' functions represent >75% of the DEGs in the TG0 vs WT0 comparison, these GO terms compose only~5% of the DEGs in the MT0 comparisons, reinforcing the benefits of utilizing the 'overlap' dataset to isolate genetic differences attributed to the loss of nkx2.5 gene function in the adult heart.
In this 'overlap' set, we investigated the genetic interaction network pertaining to the heart development GO category. Our studies identify bmp4, a vital signal in cardiac development and regeneration, as a core node (Fig. 2B). bmp4 is upregulated in the nkx2.5 −/− fish when compared to both non-transgenic and transgenic wild-type fish. Our results are closely aligned with previous murine studies in which a negative feedback loop is identified between Nkx2-5 and Bmp2 25 , the functional equivalent of zebrafish Bmp4 37 . Given that bmp4 serves as a central connector in our RNA-seq dataset, we propose that modulation of myriad Nkx2.5 effectors, both direct and indirect, function to disrupt the cardiac development regulatory program in the nkx2.5 −/− fish. As verification of the biological significance of this gene regulatory network (GRN), we detected statistically significant upregulation of versican (vcana) in nkx2.5 −/− compared to the wild-type myocardium employing RNAscope strategy ( Fig. 2C-E). Taken together, our results illuminate that overexpression of Nkx2.5 at 21 somites rescues nkx2.5 −/− -associated embryonic defects but results in enduring perturbation of the essential developmental pathways throughout adulthood.
Activation of nkx2.5 occurs following myocardial injury. Given that the Nkx2.5 downstream network is modified despite normal myocardial infrastructure in nkx2.5 −/− hearts, we hypothesized that the integrity of the Nkx2.5 transcriptional profile is robust enough to obfuscate gross structural abnormalities. We speculated that stress or injury would expose a more dramatic disruption in the molecular program transcriptionally regulated by Nkx2.5 in the adult heart. Yet, the specific nature of nkx2.5 expression in the mature wild-type zebrafish heart upon injury has yet to be fully elucidated. Previous studies have reported expression of nkx2.5 along the resection plane 13,23 , while others have been unable to detect nkx2.5 transcripts through in situ hybridization (ISH) 12 . Thus, we performed ventricular resections in wild-type fish, as previously described 38 . Using RNAscope, our data reveal expression of nkx2.5 throughout the ventricular myocardium in uninjured and at 7 days and 14 days post amputation (dpa) (Fig. 3A-C). Moreover, upon examination of the injured area, it is evident that a few cells are also expressing nkx2.5 inside the regenerate (Fig. 3B, C). Intriguingly, at 7 dpa, minimal nkx2.5 expression is observed in the wound area. However, the abundance of transcripts increases progressively over time, a pattern that reflects the process of redifferentiation of newly generated CMs. We repeated the injury in wild-type fish carrying Tg(nkx2.5:ZsYellow) and detected both endogenous ZsYellow and Tropomyosin through immunostaining. Our RNAscope findings were reinforced by co-expression of ZsYellow and Tropomyosin in both uninjured and injured hearts ( Fig. 3D-F'). Together, these data substantiate our conclusion that Nkx2.5 is expressed in the adult zebrafish myocardium and in the regenerate following ventricular amputation.
Nkx2.5 is required for myocardial regeneration. Confirmation of nkx2.5 expression in the amputation plane of the adult zebrafish myocardium provided further ammunition to our hypothesis that loss of Nkx2.5 transcriptional regulation in the context of injury would intensify the disruption in the downstream molecular program. To address this question, we performed ventricular resections in genotyped nkx2.5 −/− fish. We selected this injury tactic to circumvent inadvertent activation of the hsp70l transgene given previous studies documenting transgenic expression for as long as six days following cryoinjury 41 . To verify that the physical trauma of amputation does not yield a similar effect, we measured the magnitude and duration of nkx2.5 transcriptional activation in Tg(hsp70l:nkx2.5-EGFP) fish following ventricular resection, but in the absence of heat shock. Executing qPCR, we observed peak expression 1hour post amputation and normalization to endogenous levels by 2 dpa (Fig. S2F). Moreover, employing an anti-GFP antibody, we implemented immunostaining to detect the spatiotemporal expression of Nkx2.5-EGFP following ventricular amputation. Our results uncover parallel, albeit slightly delayed in comparison to qPCR evaluation of nkx2.5-EGFP transcripts (Fig. S2F), expression of Nkx2.5-EGFP with diminution of the protein by 2 dpa (Fig. S2A-E). Nkx2.5-EGFP is primarily evident within ventricular myocardium and not specifically at the injury border. Moreover, Nkx2.5-EGFP expression is cytoplasmic initially Color and size of circles correspond to log (q-value) and enrichment, respectively. B Network analysis of genes associated with the enriched GO term, heart development, from the 'overlap' set (group C). Color represents the fold change comparisons between nkx2.5 −/− and wild-type hearts (MT0 vs WT0). Red is upregulated and blue is downregulated, as shown in the key. C, D Representative images of RNAscope in situ hybridization analysis on sections from uninjured hearts employing a probe directed against vcana highlight upregulation in nkx2.5 −/− compared to wild-type myocardium. Scale bar, 50 μm. E Quantification of integrated signal density of vcana is depicted in wild-type (n = 3) and nkx2.5 −/− (n = 2) hearts. Mean and standard error of each data set are shown. Unpaired, twotailed t-test yields a statistically significant difference between WT0 and MT0 (p = 0.0031). F Hierarchical clustering heatmap analysis of DEGs (n = 1071) noted in the 'overlap' set (group C). The Z-scores of gene expression measurements are displayed as colors ranging from blue to red. The rows are clustered using Euclidean distance measures and average linkage. Source data are provided as a Source Data file. Taken together, these data suggest that, while Tg(hsp70l:nkx2.5-EGFP) is activated in response to ventricular resection, nkx2.5 expression is limited in duration and location, thus, minimizing the likelihood that these secondary effects modulate the response to injury.
To test our hypothesis that Nkx2.5 is required for adult myocardial repair, we exploited this ventricular apex amputation approach to injure wild-type and nkx2.5 −/− fish (Fig. 4A). We were fascinated to detect diminished regenerative potential with increased fibrin retention and deposition of collagen at the injury site at 30 dpa in nkx2.5 −/− compared to non-transgenic and transgenic wild-type hearts (Figs. 4B-D′, S3). Quantification of the percentage of scar tissue in relation to each ventricle (Fig. 4E) and the degree of residual scar (Fig. 4F) at 30 dpa substantiates these qualitative findings. Moreover, at 50 dpa, the pronounced defect in muscularization is still evident in the nkx2.5 −/− compared to the non-transgenic and transgenic wild-type hearts ( Fig. S4A-L′) with statistically significant differences manifest by residual scar tissue and severity analyses (Fig. S4M, N). In addition to evaluating the repair mechanisms in the nkx2.5 −/− fish, we also examined their chamber identity maintenance. Although the ventricular myocardium in nkx2.5 −/− fish exhibits areas of ectopic amhc expression (Fig. 1M, S5C), the healing injury site preserves its chamber-specific characteristics, similar to non-transgenic and transgenic wild-type hearts (Fig. S5A, B). Our findings suggest that, while nkx2.5 is required for embryonic 31 and adult ventricular identity maintenance (Fig. 1M, S5C), other transcriptional regulatory mechanisms are essential to prevent transdifferentiation of ventricular myocardium in the context of CM renewal. Altogether, these experiments reveal an essential requirement for Nkx2.5 to induce effective myocardial regeneration in the adult teleost heart. Given our findings that the molecular signature of adult nkx2.5 loss-of-function myocardium exhibits disruption in the normal developmental networks without evidence of detectable cardiac structural abnormalities (Figs. 1B-G and 2B), we questioned whether these disrupted pathways are required for successful stress response or injury repair. Thus, we performed a transcriptomic profiling from non-transgenic wild-type (WT7), transgenic wild-type (TG7), and nkx2.5 −/− (MT7) fish at 7 days dpa. First, we investigated the intersection of the regenerative responses in both control systems, TG7 vs TG0 and WT7 vs WT0. From this approach, we identified 882 and 408 DEGs that are commonly upregulated and downregulated, respectively ( Fig. S6A, B, green). Applying GO enrichment for these shared DEGs, we affirmed that the normal cardiac regenerative transcriptional profile is recruited as illustrated by highly enriched categories such as extracellular matrix organization, actin cytoskeleton organization, and mitotic sister chromatid segregation (Fig. S6C). Furthermore, we validated the RNA-seq expression profile of essential genes known to be imperative for repair mechanisms using qPCR. For the WT7 vs WT0 comparison, we focused on 14 genes derived from the overlapping dataset representing the shared DEGs in the RNA-seq  control systems (Fig. S6A, B, green). While only two genes reached statistical significance globally, we ascertained that 11 of the 14 transcripts were similarly up or downregulated by qPCR, corroborating the accuracy of our wild-type sequencing data (Fig. S6D, E). Finally, PCA comparing all RNA-seq samples shows that duplicates for each condition cluster together, indicating that the DEGs are not due to noise (Fig. S6F). Second, the first dimension tends to separate injured versus uninjured samples, while the second dimension tends to separate samples by genotype (WT, TG, and MT). In summary, these data indicate robust, global transcriptional differences in MT versus WT or TG samples. However, we observed a larger separation between the WT0 and TG0 samples, which we have resolved to be secondary to differences in 'housekeeping' functions (Fig. S1C), and account for this variation in our analysis by removing the contribution of these DEGs to the WT0 and TG0 comparison by generating the 'overlap' set ( Fig. 2A, group C). Together, we show that the intersection of the regenerative responses in both control systems reflects previously identified cellular repair processes and that our RNA-seq transcriptional profiles echo in vivo alterations in transcriptional regulation.
Nkx2.5 deficiency disrupts proliferation and proteolysis. Following validation of the endogenous reparative mechanisms in our control fish after amputation (Fig. S6), we aimed to probe the underlying causes of the diminished regenerative potential in the nkx2.5 −/− fish at 7 dpa. To accomplish this goal, we generated an UpSet plot to extract the specific gene sets that authentically represent the injury response and are not confounded by differences in genotypes or controls (Fig. 5A). In this analysis strategy, the rows represent all appropriate two-way comparisons between the six individual samples (WT0, TG0, MT0, WT7, TG7, and MT7), summarizing the dataset size and delineating which comparisons are incorporated into the column sets. After contemplating appropriate inclusion and exclusion criteria, we narrowed in on three essential columns to scrutinize the reparative processes that are and are not initiated in the nkx2.5 −/− fish independent of GRN aberrations from the loss of nkx2.5 function. We identified set A (Fig. 5A, turquoise column) as those genes that are deployed in wild-type hearts following amputation but fail to be recruited in the absence of nkx2.5 gene function. Set B (Fig. 5A, magenta column) denotes DEGs that are utilized in the nkx2.5 loss-of-function model but are not activated in control scenarios. Finally, set C (Fig. 5A, black column) embodies the DEGs that are appropriately enlisted in the nkx2.5 −/− fish, consistent with wild-type regenerative mechanisms. The aforementioned set A encompasses 277 genes that fail to be activated in the nkx2.5 −/− fish during cardiac regeneration (Fig. 5A, set A, turquoise column). This group is functionally enriched for several cellular processes including cell cycle, translation, integrin-mediated signaling pathway, and ATP synthesis coupled proton transport (Fig. 5B, set A). Further network analysis employing set A illuminates upregulation of the cell cycle and integrin-mediated signaling pathways, pinpointing critical, regenerative mechanisms that are boosted in the wildtype but not in the nkx2.5 −/− fish following amputation (Figs. 5C and S7). Specifically, while wild-type fish mount a proliferative response and mediate extracellular matrix (ECM) production following amputation 19,40,42,43 , nkx2.5 loss-of-function model fails to mirror these cellular reactions. Moreover, downregulation of the tricarboxylic acid cycle and the ATP synthesis coupled proton transport modules and similar dampening of the mitochondrial translation pathways in injured wild-type (but not nkx2.5-deficient) myocardium corroborate previous results emphasizing the necessary depression in metabolism during myocardial repair 13,[44][45][46] . Further aligned with our findings that nkx2.5 −/− fish demonstrate abnormal metabolism (Fig. 5C), recent murine studies describe an adult Nkx2-5 mutant model with significant abnormalities in mitochondrial respiratory pathways 47 . Altogether, analysis of set A accentuates that failure of the appropriate injury response in two essential cellular processes, CM proliferation and mitochondrial OXPHOS activity, is due to the loss of Nkx2.5 function.
Next, we queried the molecular pathways that are inappropriately regulated in the nkx2.5 −/− fish and are not altered in the wild-type conditions (Fig. 5A, set B, magenta column). Set B encompasses 165 genes that are functionally enriched in proteasome-mediated ubiquitin-dependent protein and proteolysis involved in cellular protein catabolic processes (Fig. 5B, set  B). Applying these GO terms, we produced a network depicting downregulation of several, closely linked members of the psmd gene family, proteosome 26S subunit, non-ATPase (Figs. 5D and S8). Proteasome-mediated ubiquitin-dependent protein catabolic processes degrade proteins in a precise and fastidious fashion employing the 19S regulatory subunit of the 26S proteasome 48 . More specifically, in the context of zebrafish cardiac regeneration, expression of psmb1 and psmb3 have been discovered in woundedge CMs 49 , evoking a functional role in sarcomere disassembly 12,49,50 . Thus, failure to stimulate expression of these proteolytic drivers may impair the ability of nkx2.5-deficient hearts to undergo dedifferentiation 12,13 , an indispensable step in CM replenishment.
Finally, we investigated those genes which are differentially expressed in the same direction in non-transgenic wild-type, transgenic wild-type, and nkx2.5 −/− fish in the uninjured and injured states at 7 dpa (Fig. 5A, set C, black column). Set C contains 205 genes appropriately engaged in the regenerative response in the nkx2.5 loss-of-function model. In this GO category, we found enrichment in the regulation of heart contraction and morphogenesis and in the innate immune response (Fig. 5B, set C; Figs. 5E and S9). These data emphasize that normal immunological pathways (Lai et al. 51 ) are adequately stimulated in the nkx2.5 −/− fish following amputation. Furthermore, while heart morphogenesis pathways downstream of Nkx2.5 are misregulated in the adult myocardium (Fig. 2D), these networks are not responsible for CM renewal in the injured heart. Taken together, our transcriptomic results indicate that reduction in regenerative potential in the nkx2.5 −/− fish is due to the breakdown in three mechanistic injury responses vital for effective cardiac regeneration: cell cycle, mitochondrial metabolism, and proteolysis.
Nkx2.5 targets identify nodes that link regenerative GRNs. To determine the key effectors mediating these specific injury responses, we harnessed data from our previous studies to uncover a consolidated set of Nkx2-5 direct targets identified in mouse HL-1 cells by DamID 52 . We then intersected our RNA-seq results with these DamID targets to identify the key GRNs nodes that function immediately downstream of Nkx2-5 to promote regeneration. To perform this overlap, we ascertained the zebrafish orthologues of the EnsEMBL IDs of the mouse DamID targets using BioMart (release 100) 53 . Overlapping these datasets with the entire gene lists in sets A and B from our RNA-seq results ( Fig. 5A; independent of GO annotation) highlights interconnecting nodes that are under the control of Nkx2-5 and link the enriched modules for cell cycle, mitochondrial metabolism, and proteolysis (Fig. 6). When comparing uninjured and injured non-transgenic wild-type hearts (Fig. 6A), our analysis supports prior data by exhibiting appropriately diminished ATPdependent and translational pathways and stable protein catabolic processes to support CM renewal 46,54 . Moreover, proliferative genes are triggered following injury to simulate repair (Fig. 6A) 17 . Importantly, Nkx2.5 effectors coordinate these GRNs through critical hubs, such as ect2, psmb3, and psmd7, which assemble a consolidated response to injury. Alternatively, when evaluating the uninjured and injured nkx2.5 −/− fish (Fig. 6B), these crucial modules either fail to be inhibited, in the case of mitochondrial translation, or fail to be activated, in the case of proliferation. Further, inappropriate downregulation of the genes required for sarcomere disassembly exacerbates the collapse in reparative mechanisms (Fig. 6B). Altogether, integration of our new RNA-seq dataset and our existing DamID target list uncovers strategic nodes by which Nkx2.5 synchronizes the adult myocardial injury response.
To inspect the biological significance of these GRNs, we implemented in vivo analyses to detect expression of crucial nodes in non-transgenic wild-type and nkx2.5 −/− fish at 7 dpa. ect2, a putative direct target of Nkx2.5 from our DamID study and confirmed to be bound by NKX2-5 in human pluripotent cell-derived cardiomyocytes 55 , is well positioned to promote cell cycle re-entry and sarcomere disassembly during cardiac regeneration given the communication that it establishes between these two modules (Fig. 6A, B). Previous studies have elucidated the function of ect2 in contractile ring assembly and initiation of cytokinesis; Ect2 activity converts RhoA-GDP (guanosine diphosphate) to RhoA-GTP (guanosine triphosphate) at the spindle 56 . Employing transient dominant negative expression in zebrafish, recent reports have illuminated the role of ect2 loss-of-function in inhibiting cytokinesis, inducing polyploidization, and ultimately, in yielding a loss of cardiac regenerative potential 17 . Applying RNAscope technology, we documented upregulation of ect2 expression in the injury zone and wound edge of injured, nontransgenic wild-type apices (Fig. 7A), validating the importance of Ect2 function in CM cytokinesis. Moreover, ect2 expression is significantly depleted in injured nkx2.5 −/− hearts which supports our hypothesis that Nkx2.5 is required upstream of ect2 in promoting cell cycle reentry and proliferation in regenerating myocardium (Fig. 7B). Finally, our conclusion is further reinforced by evidence that nkx2.5 and ect2 transcripts colocalize in individual nuclei at the injury border (Fig. 7C, D), suggesting cell autonomous interactions.
Two additional Nkx2.5 effectors, psmb3 and psmd7, sit high in the Nkx2.5 transcriptional hierarchy linking the mitochondrial translation and the proteolysis modules (Fig. 6A, B). Given data depicting psmb3 and psmb1 transcripts in the wound edge following cardiac injury in wild-type zebrafish 49 , downregulation of these proteasome gene family members has the potential to prevent proteolytic mechanisms necessary to disassemble mature CM architecture. We postulate that Nkx2.5 regulates psmb3 and psmd7 expression following ventricular injury and, in the nkx2.5 loss-of-function model, depleted psmb3 and psmd7 expression restrains dedifferentiation of pre-existing CMs. Employing ISH, we document robust expression of psmb3 (Fig. 7E) and psmd7  (Fig. 7F, H). Moreover, pixel intensity quantification corroborates these histological findings; both psmb3 and psmd7 exhibit a statistically significant decrease in expression in the wound (Fig. 7I, J). Altogether, we conclude that Nkx2.5 operates via key effectors, ect2, psmb3, and psmd7, to orchestrate regenerative mechanisms involving mitochondrial metabolism, dedifferentiation, and cytokinesis to repopulate CMs following injury.
Nkx2.5 promotes proliferation and dedifferentiation. Considering this defective transcriptional response in nkx2.5 −/− fish, we examined the source of the impaired healing in vivo by evaluating the CM proliferation indices at 7 dpa and 14 dpa in wild-type and nkx2.5 −/− fish. Given CM-specific nkx2.5 expression (Fig. 3), we hypothesized that Nkx2.5 is required cell autonomously to induce dedifferentiation, cell division, and redifferentiation of new CMs. Our data reveal a statistically significant decrease in the CM proliferation index in animals with reduced nkx2.5 transcriptional activity versus animals with intact nkx2.5 function at 7 dpa (Fig. 8A, B, E). In contrast, there is no statistically significant discrepancy at 14 dpa in the proliferative rates between wild-type and nkx2.5 −/− fish (Fig. 8C-E). These findings illuminate an essential role for Nkx2.5 in promoting CM proliferation at 7 dpa when an escalation in cellular renewal is required for myocardial regeneration, validating the GRNs created from the RNA-seq data (Figs. 5C and 6).
We next examined the injured nkx2.5 −/− fish for evidence of impaired sarcomere disassembly given downregulation of the proteolysis module of the Nkx2.5-dependent GRN (Figs. 5D and 6). Prior work has detected reactivation of embryonic cardiac sarcomeric proteins at the wound edge, defining embryonic cardiac myosin heavy chain (embCMHC) as a marker of undifferentiated CMs 57 . To assess whether nkx2.5 regulates dedifferentiation and, therefore, re-expression of immature myosins, we performed immunostaining with embCMHC and MF20 at 7 dpa (Fig. 8F, G). While wild-type hearts demonstrate co-localization of embCMHC and MF20 along the border of the regenerate (Fig. 8F), nkx2.5 −/− myocardium shows minimal expression of embCMHC at the site of injury (Fig. 8G). These findings are further corroborated with immunostaining directed against activated leukocyte adhesion molecular a (Alcam), another marker of dedifferentiation 58,59 , demonstrating diminished Alcam expression in the nkx2.5 −/− compared to the wildtype regenerates (Fig. S10). From these data, we conclude that adult CMs fail to dedifferentiate in the absence of nkx2.5 gene function. Taken together, our results validate GRNs underscoring that Nkx2.5 is required cell autonomously to instruct proteolytic pathways for dedifferentiation of mature CMs to an immature myocardial fate (Figs. 5D and 6).

Nkx2.5 non-cell autonomously limits epicardial integration.
Although our data establish that CM-specific nkx2.5 expression directly regulates CM proliferation and dedifferentiation via Nkx2.5-dependent cell cycle and proteolysis modules, we sought to evaluate whether Nkx2.5 also mediates non-cell autonomous regenerative functions. Both the cardiac endocardial and epicardial lineages exhibit sentinel roles in heart regeneration and interaction among these specific cell types allows for orchestration of the complex healing process. Coronaries arteries are required to regenerate in order to perfuse new myocardium 23 and revascularization has been shown to occur within hours of cryoinjury to stimulate CM proliferation 41 . Thus, using nkx2.5 −/− fish carrying Tg(kdrl:EGFP) la116 , we investigated the process of angiogenesis in wild-type and nkx2.5 loss-of-function animals. Interestingly, we observed normal EGFP + endothelial tubes in the wound area establishing that nkx2.5 expression is not responsible for coronary angiogenesis during cardiac regeneration (Fig. S11).
Next, we compared the regenerative response in the epicardium, a thin mesothelial cell layer overlying the atrial and ventricular chambers, between wild-type and nkx2.5 −/− fish. Previous studies have elucidated the essential roles of epicardial activation, proliferation, and colonization of the regenerate 23 . Specifically, epicardial cells undergo an epithelial-tomesenchymal (EMT) transition to yield epicardial derived cells (EPDCs). These EPDCs invade the injury site between 7 dpa and 30 dpa to provide a cellular scaffold and produce perivascular cells and myofibroblasts, ultimately in support of CM proliferation and angiogenesis 39,60 . As the wound heals, the epicardium undergoes EMT and establishes a thickened epicardial layer covering the injured myocardium [60][61][62][63][64] . Employing immunostaining with Tg(tcf21:DsRed2) to label the epicardium in wild-type and nkx2.5 −/− fish, we discerned expression at the injury site in wild-type and nkx2.5 −/− fish at 14 dpa (Fig. 9A-D). Yet, impaired integration of tcf21 + cells in the nkx2.5 −/− compared to wild-type fish is evident; tcf21 + epicardial cells fail to penetrate the regenerate and remain exterior to the CM layer (Fig. 9B, D). Quantification of EPDC migration reveals a statistically significant decrement in the distance measured from the apex of the heart to the maximal interior position of the Tcf21:DsRed2 + cells when comparing wild-type and nkx2.5 −/− hearts (Fig. 9E). Taken together, these data indicate that, while the epicardium mounts a suitable response to tissue damage in the loss-of-function nkx2.5 model, compromised epicardial penetration into the wound weakens the healing potential in the nkx2.5 −/− compared to wildtype fish.
We sought to clarify the underlying cellular mechanism responsible for this deficient epicardial infiltration into the regenerate. Specifically, we tested our hypothesis that Nkx2.5 mediates non-cell autonomous functions in addition to the CMintrinsic roles of dedifferentiation and proliferation. Thus, we performed ex vivo assays using non-transgenic wild-type and nkx2.5 −/− fish carrying Tg(tcf21:dsRed), as previously described 65 . At 3 dpa, we resected the ventricular apices, cultured them in fibrin-coated plates, and quantified the outgrowth of epicardial cells after four days. There were no significance differences identified following quantification of epicardial cell expansion between the wild-type and the nkx2.5 −/− cultured apices (Fig. 9F-H). Moreover, employing EdU incorporation, we established that the proliferation rate in the wild-type epicardial cultured cells is the same as that in the nkx2.5 −/− epicardium (Fig. 9I-K). These findings support our conclusion that the defect in epicardial infiltration of the regenerate is due to mechanisms extrinsic to the epicardial cells. This conclusion aligns elegantly with our network analyses which invoke a small, integrinmediated signaling pathway module (itgb2, itba6a, and itgbl1) that fails to be stimulated in the nkx2.5 loss-of-function model (Fig. 5C). Given the importance of ECM remodeling in cardiac regeneration 43,45,66 , we expect that future investigation of targets of Nkx2.5 will elucidate novel ECM regulatory networks exemplifying non-cell autonomous functions of this cardiacspecific transcription factor.

Discussion
These data offer a unique perspective on mechanisms regulated by Nkx2.5 that are responsible for building new muscle following injury in the mature zebrafish heart. In particular, we show that CM-specific expression of nkx2.5 is required for cardiac regeneration following amputation of the ventricular apex in an innovative, adult Nkx2.5 loss-of-function zebrafish model. Using this genetic tool, we harnessed both new RNA-seq and our prior DamID data to reveal that the developmental molecular profile regulated by Nkx2.5 is perturbed in the adult nkx2.5-deficient myocardium. Moreover, our multi-variate network analysis underlines Nkx2.5 as a vital regulator of CM renewal, functioning to promote cell cycle re-entry and to maintain proteolysis genetic modules. Direct targets, ect2, psmb3, and psmd7 provide examples of essential hubs in the Nkx2.5-dependent GRNs guiding proliferation, sarcomere disassembly 67,68 , and mitochondrial metabolism 47 . We observe cellular evidence of these molecular mechanisms in the diminished number of PCNA + CMs and the decreased expression of immature cardiac myosin and Alcam in the injury border of nkx2.5 −/− hearts. Yet, additional non-cell autonomous functions of Nkx2.5 are necessary for epicardial penetration into the wound to support the reparative microenvironment niche. Moreover, ECM modules fail to be upregulated in the absence of nkx2.5 gene function, suggesting a paradigm whereby Nkx2.5 mediates crucial components in ECM patterning. Taken together, these studies underscore the importance of activation of embryonic Nkx2.5 for CM dedifferentiation and proliferation during cardiac regeneration and suggest opportunities for therapeutic intervention in patients suffering from myocardial infarction and adult CHD.
While it is clear that evolutionarily conserved developmental genetic signatures are critical to facilitate regenerative responses from cuttlefish to mouse [69][70][71][72][73] , the toolkits required to investigate the underlying essential mechanisms remain underdeveloped. There is a scarcity of genetic lines with loxP sites flanking genes of interest that limits our ability to institute conditional deletions efficiently. Furthermore, while some pivotal cardiac transcription factors have been examined functionally 21,22,74 , many members of the cardiac developmental transcriptional hierarchy have yet to be investigated. Although murine studies have benefitted from conditional deletion of Nkx2-5 to examine its roles in the adult cardiac conduction system and trabecular myocardium 75,76 , the Nkx2-5 requirement in myocardial repair was not addressed due to the limited regenerative potential in these models. Here, we describe the application of an innovative technique expediting rescue of otherwise lethal nkx2.5 −/− embryos using heat shock overexpression without confounding evidence of non-specific defects 29 . We have taken advantage of our nkx2.5-deficient fish to dissect the cell type-specific functional requisites, uncovering inappropriate responses in proliferation, sarcomere disassembly, and mitochondrial metabolic pathways in the CMs and cell extrinsic abnormalities in ECM deposition of the regenerative niche. This unique model provides a substrate to study the distinct molecular profiles that are altered in the Nkx2-5 loss-offunction myocardium, mimicking late-stage challenges faced by adult patients with Nkx2-5-associated CHD. Our data also illustrate insights generated from a comprehensive assessment of the injured Nkx2.5 loss-of-function adult model and multi-variate network interpretation of transcriptional profile alterations in various genetic and temporal conditions. By probing each comparison, we identified distinct responsesboth appropriate and inappropriatethat are regulated by Nkx2.5. These findings suggest that therapeutic measures need to be tailored to the individual clinical scenarios in order to ameliorate pathological mechanisms and redirect the interlocked chain of events that ensue following myocardial damage 73 . Specifically, in the context of adult CHD patients carrying NKX2-5 mutations, we envisage that discrete and directed remedies, such as stimulation of cell cycle re-entry and suppression of mitochondrial translation, would augment successful regenerative outcomes. Moreover, the proteasome-mediated ubiquitin-dependent protein catabolic processes necessitate preservation to ensure proper dedifferentiation of Nkx2.5-deficient CMs. Our data suggest that modulation of the rapid inflammatory response following injury occurs aptly and, therefore, does not need remediation to support effective repair. Taken together, careful delineation of the molecular and cellular processes that go awry following injury in both wild-type and genetically altered myocardium will enhance our understanding of regenerative strategies and permit recruitment of explicit endogenous mechanisms deficient in each patientspecific context.
While evidence of CM-specific expression of nkx2.5 in the adult zebrafish heart, diminished proliferation, and arrested dedifferentiation illuminate cell intrinsic functions of Nkx2.5 following injury, impaired epicardial migration in vivo points to the contribution of CM extrinsic roles to the decreased reparative capability in nkx2.5 −/− fish. In light of our ex vivo studies depicting normal epicardial migration and proliferation and our network analyses illustrating failure of nkx2.5 −/− fish to upregulate an integrin-mediated ECM module, we speculate that Nkx2.5 directly or indirectly regulates crucial constituents of the cardiac matrix. Further investigation is required to uncover the key players acting downstream of Nkx factors. Yet, recent reports indicate that distinct embryonic ECM cargo can augment cytokinesis in postnatal rat CMs in vitro and in vivo 77 , providing a glimpse of therapeutic potential. Moreover, zebrafish studies have advanced our appreciation of the mechanisms responsible for the cardiac valve regeneration 78,79 , further emphasizing the importance of studying ECM regulation in adult Nkx2.5 loss-offunction models given the incidence of valvular defects in these patients 80,81 . To address questions on the impact of Nkx2.5regulated ECM on cardiac and valvular repair, future inquires of the regenerative proteome with emerging techniques is necessary to understand global protein-protein interaction dynamics 82,83 .
Altogether, our studies underscore that activation of the embryonic Nkx2.5 transcriptional network is critical in triggering productive intrinsic and extrinsic cellular responses to myocardial damage in the adult zebrafish heart. Exploiting insights into the specific mechanistic processes that are hampered in the nkx2.5 −/− fish such as CM proliferation and dedifferentiation, our work open doors for the cultivation of targeted therapies in Nkx2-5-deficient myocardium. Studies demonstrating opportunities for myriad factors to instruct CM regeneration indicate that therapeutic intervention is feasible [84][85][86] as long as the coordination of systematic approaches is realized 73 . The identification of previously unrecognized Nkx2.5 targets, ect2, psmb3, and psmd7, as critical nodes in orchestrating cell cycle re-entry, proteolysis, and mitochondrial metabolism in response to injured myocardium offers initial steps to achieve this mission.
Heat shock conditions. Embryos from outcrosses of fish carrying Tg(hsp70l:nkx2.5-EGFP) were maintained at 28.5°C and exposed to heat shock at 37°C at 21 somites. To implement heat shock, 50 embryos were placed in 2.5 mL of preheated embryo medium in a 35 mm Petri dish on top of a covered heat block for one hour. Three hours following initiation of heat shock, transgenic embryos were identified by visualization of ubiquitous EGFP expression. Non-transgenic sibling embryos exposed to heat shock served as controls.
Histological analysis. Zebrafish were anesthetized with tricaine and then fixed in 4% paraformaldehyde overnight at 4°C. Subsequently, the dissected hearts were washed several times with PBS and cryopreserved with 30% sucrose before immersion in O.C.T. (Tissue-Tek). The blocks were stored immediately at −80°C. They were sectioned at 10 µm with a cryostat (Leica CM3050 S) and collected for Acid Fuchsin Orange G (AFOG), RNA in situ hybridization (ISH), RNAscope, and immunostaining.
AFOG staining was performed as previously described on 6-10 animals per time point 38 . Briefly, slides were dried for one hour at room temperature and refixed with 4% PFA for five minutes. Subsequently, they were incubated in a preheated Bouin's solution for two hours at 60°C followed by one hour at room temperature. After extensive washes with H 2 O, the slides were rinsed in 1% phosphomolybdic acid for five minutes. Following a five-minute wash with H 2 O, they were incubated for 10 min in AFOG staining solution (0.5% Aniline Blue, 1% Orange G, 1.5% Acid Fuchsin, pH 1.09). Finally, the slides were rinsed in H 2 O, dehydrated, cleared in xylene, and mounted in Cytoseal 60 (Thermo Fisher Scientific).
Immunostaining was performed as previously described 92 . Briefly, the hearts were collected in cold PBS and brought to room temperature. Following washes with PBS plus 0.1% Tween20 (PBT), sections were permeabilized with methanol for one hour at room temperature, blocked in blocking solution (PBS containing 2% sheep serum, 0.2% Triton X-100, 1% DMSO) for 30 min at room temperature, and incubated with primary antibodies in blocking solution at 4°C overnight. Sections were then incubated with secondary antibodies between one and three hours at room temperature. Finally, the slides were mounted with Vectashield medium (Vector Laboratories) and DAPI prior to imaging.
Mef2/PCNA staining was performed as previously described 93 . Briefly, the slides were equilibrated to room temperature prior to performing antigen retrieval in 10 mM citrate buffer at 98°C for 20 min. Then, they were cooled for 20 min, rinsed in H 2 O twice, and immersed in pre-cooled acetone (−20°C) for 10 min. Subsequently, acetone was allowed to evaporate for 20 min at room temperature. The slides were then washed once in H 2 O, washed twice in PBA, and incubated in blocking buffer (5% normal goat serum, 0.3% Triton X-100 in PBS) for one hour at room temperature. Incubation with primary antibody was performed in antibody dilution buffer (1% BSA, 0.3% Triton X-100 in PBS) overnight at 4°C, and secondary antibody for two hours at room temperature.
Primary RNAscope staining was executed following the manufacturer's instructions for the RNAscope multiplex fluorescent assay v2 (Advanced Cell Diagnostics or ACD). RNAscope probes for vcana (C3), nkx2.5 (C2), and ect2 (C1) were synthesized by ACD and nuclei were stained with DAPI prior to analysis.
Images for AFOG, ISH, and RNAscope were acquired with a Zeiss Axio Imager.D2m and a Hamamatsu Nanozoomer SQ. Following immunostaining protocols, confocal imaging was performed with a Nikon Ti Eclipse inverted confocal microscope and a Zeiss LSM 880 with z-stacks analyzed using ImageJ (version 1.53k, NIH).
Quantification and statistical analysis. All quantifications were performed blinded. Unless otherwise indicated, experiments were implemented with at least four biological replicates and at least three samples per replicate. For quantification strategies, three non-consecutive sections with the largest area were measured to assign an index for each heart. To calculate the proportion of scar tissue, the section with the largest injury site was selected and normalized to the ventricular area. Furthermore, the severity of the scar was classified as previously described 94 . Ventricular volume was calculated on 10 µm sections by totaling the ventricular area in each section and multiplying by the distance between each section 32 . The percentage of trabeculation was calculated as the percentage of muscular tissue within the ventricular area using the threshold tool and manually excluding the internal cavity area 32 . For quantification of the ISH signals, previously reported protocols were executed 95 . For fluorescent intensity calculations, embCMHC + and Alcam + areas were quantified in pixels employing ImageJ in the region of the regenerate and normalized to background. To measure proliferation, Mef2 + /PCNA + cells in a 200 µm area from the border were counted manually 96 . Individual CMs were categorized by the presence of Mef2 and by a size greater than 5 µm diameter. Of these CMs, those cells also positive for PCNA signal were counted and averaged to calculate a proliferative index for each heart. In vivo migration of epicardial derived cells (EPDCs) was quantified by measuring the distance between the ventricular apex and the furthest EPDC inside the myocardium. For statistical analyses, two-tailed Student's t-tests were used when normality test was passed. Statistical values are displayed as mean ± standard error of the mean (SEM). The following nomenclature was employed to present results: ns not significant, *p < 0.05, **p < 0.01; ***p < 0.001, ****p < 0.0001.
Quantitative PCR and RNA sequencing. For qPCR, total RNA was extracted from six pooled ventricles for each condition and timepoint analyzed. The purified RNAs were reverse transcribed with the iScript TM cDNA Synthesis Kit (Bio-Rad). qPCR was performed with iQ TM SYBR Green Supermix (Bio-Rad) and a CFX96 Touch TM Real-Time PCR Detection System (Bio-Rad). All experiments were completed using biological triplicates. Primer sequences are noted in Supplemental Table 1.
For RNA Sequencing, total RNA from six pooled ventricular apices of two biological replicates was extracted using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Next, the quality was assessed using a Bioanalyzer. The cDNA library was prepared using TruSeq RNA Prep Kit v2 (Illumina) and the samples were submitted to the JP Sulzberger Columbia Genome Center at Columbia University Medical Center. A HiSeq 2500 System was used to obtain single-end 100 bp reads for each sample.
RNA-sequencing data processing and analysis. Reads were aligned to the GRCz10-91 zebrafish reference genome using STAR (version 2.5.2) 97 and were quantified using featureCounts (version 1.5.0-p3) 98 . Differential gene expression between conditions was performed using DEseq2 (version 3.14) (Anders and Huber, 2012). Statistically significant differentially expressed genes (DEGs) were retrieved by filtering for genes with absolute log2 fold-change (FC) > 0.5 and false discovery rate (FDR) < 0.5 in R (version 3.6.0).
Venn overlaps of DEGs were generated with the VennDiagram package (version 1.7.1) 99 . Hierarchical clustering was performed using the gplots::heatmap.2() function which employs Euclidean distance measures with the average linkage method (version 3.1.1) 100 . The UpSet plot was generated with the UpSetR package (version1.4.0) 101 . Functional enrichment was performed using Metascape 3.5 102 and validated by orthogonal approaches using DAVID (version 6.8) 103 and Panther (version 14.1) 104 , employing default parameters. Gene Ontology (GO) plots were created using the ggplot2 package (version 3.3.2).
Overlap with DamID targets of Nkx2-5 in mouse HL-1 cells (GSE44902) 52 was performed by first retrieving the zebrafish orthologues of the EnsEMBL IDs of the mouse DamID targets using BioMart (release 100) 53 . Then, the intersection of the EnsEMBL IDs for the zebrafish DamID targets and zebrafish DEGs were obtained with BioVenn (version 1.1.3) 105 . Network reconstructions were performed by retrieving protein-protein interacting relationships using STRING (version 11.0, default parameters, https://string-db.org) and visualized with Cytoscape (version 3.8.0) 106 . For networks associated with GO terms, interactions only between genes associated with specified GO terms were examined. However, the networks highlighting DamID candidates as hubs (Fig. 6) were reconstructed based on an overlap of all targets and sets A and B (Fig. 5). The first neighbors of the DamID targets and modules with more than seven nodes were included for analysis.
Epicardial explant culture. Heart explant cultures were performed as previously described using Tg(tcf21:DsRed2) pd37 fish 65 . Briefly, hearts were extracted at three days post-amputation (dpa) from wild-type and nkx2.5 −/− ;Tg(hsp70l:nkx2.5-EGFP) fish. The ventricular apices were cultured in 24-well plates, pre-coated with fibrin, and incubated for four days at 28°C, 5% C0 2 . Cell migration was assessed using an EVOS Digital Microscope with ImageJ to measure the maximal radius between the edge of the ventricular apex and the edge of the monolayer. For EdU incorporation experiments, 25 mM EdU was added to the culture medium and the explants were fixed one hour later 107 . The Click-iT EdU Imaging Kit (Invitrogen) was employed following the manufacturer's instructions.
Swimming endurance assay. Adult 4-month-old zebrafish were kept unfed for 24 hours prior to initiation of the challenge in the swim tunnel respirometer device (Mini Swim-170, Loligo Systems). After a minute of acclimation inside the tunnel with no current, water (26°C) flow speed was initiated at 10 cm/s and increased by 5 cm/s increments every minute. Once the maximum speed was attained, fish were allowed to swim until exhaustion or the end of the test (maximum duration of 35 min) 33 .