RNA SEQ Analysis Indicates that the AE3 Cl−/HCO3− Exchanger Contributes to Active Transport-Mediated CO2 Disposal in Heart

Loss of the AE3 Cl−/HCO3 − exchanger (Slc4a3) in mice causes an impaired cardiac force-frequency response and heart failure under some conditions but the mechanisms are not known. To better understand the functions of AE3, we performed RNA Seq analysis of AE3-null and wild-type mouse hearts and evaluated the data with respect to three hypotheses (CO2 disposal, facilitation of Na+-loading, and recovery from an alkaline load) that have been proposed for its physiological functions. Gene Ontology and PubMatrix analyses of differentially expressed genes revealed a hypoxia response and changes in vasodilation and angiogenesis genes that strongly support the CO2 disposal hypothesis. Differential expression of energy metabolism genes, which indicated increased glucose utilization and decreased fatty acid utilization, were consistent with adaptive responses to perturbations of O2/CO2 balance in AE3-null myocytes. Given that the myocardium is an obligate aerobic tissue and consumes large amounts of O2, the data suggest that loss of AE3, which has the potential to extrude CO2 in the form of HCO3 −, impairs O2/CO2 balance in cardiac myocytes. These results support a model in which the AE3 Cl−/HCO3 − exchanger, coupled with parallel Cl− and H+-extrusion mechanisms and extracellular carbonic anhydrase, is responsible for active transport-mediated disposal of CO2.

. Significantly enriched Gene Ontology (GO) categories. GO categories were identified using the GOrilla program 15 and grouped by related functions as described in Methods and Supplementary Information. The degree of enrichment = (b/n)/(B/N), with (N, B, n, b) defined as follows: N is the total number of genes, B is the total number of genes associated with a specific GO term, n is the number of genes in the target set, b is the number of genes in the intersection. For more complete information on GO categories and how to access specific gene lists in Excel Files, see legends for Supplementary  reduction in the use of ATP and substrates for biosynthesis would allow greater utilization of ATP for contraction. Expression changes in energy metabolism GO categories were highly significant (Table 1), and when the directions of changes were examined, they indicated increased glucose metabolism, reduced fatty acid metabolism, and reduced biosynthesis (Fig. 3).
Changes in genes encoding major regulators of energy metabolism (Fig. 3A) included down-regulation of the transcription factors Ppard (a member of the hypoxia response GO category) and Pparg. The expression of both Ppard and Pparg are reduced by hypoxia 66,67 , consistent with a role for AE3 in maintenance of O 2 /CO 2 balance. Prkag2 and Prkab1, regulatory subunits of AMPK, which plays a major role in energy sensing, energy metabolism 68 , and response to hypoxia 69 , were up-regulated. Ppip5k2 (diphosphoinositol pentakisphosphate kinase 2, 1.38-fold increase, not shown) regulates inositol pyrophosphate metabolism and is an AMPK-independent energy sensor and regulator 70 . The hypoxia-responsive ATP-sensitive K + channels, Abcc8 (Sur1) and Abcc9 (Sur2), were up-regulated. Both channels interact with Kcnj11 (Kir6.2, not changed), with Abcc8 the predominant form in atria and Abcc9 the predominant form in ventricles 71 . Abcc9 protects the heart against ischemia and both channels serve as metabolic sensors, couple energy metabolism and membrane excitability, play major roles in carbohydrate metabolism, and are induced by hypoxia 72,73 .
Expression of genes for proteins that stimulate glycolysis, glucose uptake, and glucose oxidation were increased. These include Hif1a and Thra (Fig. 1), which play major roles in glucose metabolism 74,75 . Changes in additional genes involved in glucose metabolism are shown in Fig. 3B. The α1-adrenergic receptor (Adra1b) is cardioprotective during myocardial infarction and ischemia in part because of enhanced glucose metabolism 74,76 . Many of the encoded proteins are affected by activation of Akt (indicated in Fig. 3B). This occurs via phosphorylation of Ser473, which was significantly increased in AE3-null hearts subjected to atrial pacing 4 . For example, C2cd5 contributes to insertion of GLUT4 into the plasma membrane 77 , Cd28 stimulates glucose uptake and glycolysis 78 , and Ehbp1 is involved in insulin-regulated GLUT4 recycling and glucose transport 79 , all in response to Akt activation. Smarcd3 is a transcriptional cofactor that drives glycolytic metabolism through a mechanism involving Akt 80 . Entpd5, a UDPase, is involved in Akt responses and has been shown to increase the catabolic efficiency of aerobic glycolysis in tumor cells 81 . Ptk2b (Pyk2 tyrosine kinase, focal adhesion kinase 2) mediates insulin-independent insertion of GLUT4 into the plasma membrane 82 and contributes to α1-adrenergic receptor-mediated activation of Akt 83 . Tpk1 (thiamine pyrophosphokinase), generates thiamine pyrophosphate, a  Supplementary Table S2 and in a subset  of 158 signaling genes in Supplementary Table S3; values are means ± SE; n = 4 for each genotype; *p ≤ 0.01 vs WT controls.
cofactor needed for oxidative decarboxylation of pyruvate and other substrates in mitochondria 84 . Up-regulation of Pdp1 (pyruvate dehydrogenase phosphatase 1) would favor dephosphorylation and activation of pyruvate dehydrogenase, which would enhance pyruvate (glucose) oxidation 85 . Activation of Crhr2, the urocortin 2 receptor, causes increased AMPK activation, glucose uptake, and phosphorylation of acetyl-CoA carboxylase (which inhibits fatty acid biosynthesis) in cardiomyocytes 86 ; cardioprotective effects of Crhr2 activation during ischemic injury include activation of phosphatidylinositol 3-kinase/Akt signaling 87 . Gene knockout studies showed that loss of Tbc1d1 impairs insulin-stimulated glucose uptake and increases fatty acid oxidation 88 , so up-regulation of Tbc1d1 should favor a switch to glucose metabolism.
Expression of the leptin receptor (Lepr), which is induced by hypoxia 89 and affects both glucose and lipid metabolism, was up-regulated (Fig. 3C). In isolated hearts, treatment with leptin stimulates fatty acid oxidation, reduces cardiac efficiency, and has no effect on glucose oxidation 90 so one could question a role for its receptor in glucose metabolism; however, hearts of diabetic mice lacking a functional long form of Lepr (ObRb), which might be expected to show the opposite effect, exhibit an increase in palmitate oxidation and a decrease in glycolysis and glucose oxidation 91 . Also, after myocardial infarction, Lepr was upregulated and contributed to a shift from fatty acid to glucose metabolism in a process involving phosphatidylinositol 3-kinase/Akt signaling 92 . The latter study raises the possibility that up-regulation of Lepr mRNA in AE3-null hearts, which includes increased expression of the long form of Lepr ( Supplementary Fig. S1), may affect glucose metabolism.
Down-regulation of Pparg and Ppard (Fig. 3A) is consistent with reduced fatty acid oxidation, as it has been shown that cardiac-specific ablation of Pparg 93 or Ppard 94 causes a reduction in fatty acid oxidation. Also, hypoxia-induced microRNA-mediated repression of Ppard facilitates a reduction in fatty acid metabolism and an increase in glucose metabolism in the failing heart 66 . Reduced utilization of fatty acids for energy metabolism is also suggested by reduced expression of the following genes ( Fig. 3C): Cd36, which mediates fatty acid uptake across the plasma membrane 95 ; Tlr4, which interacts with Cd36 and stimulates fatty acid uptake 96 ; Iqgap2, which also interacts with Cd36 and serves a signaling pathway that stimulates fatty acid uptake and processing 97 ; Mgll, which hydrolyzes monoglycerides to produce fatty acids and glycerol for energy metabolism and biosynthetic processes 98 ; and Acadl (long-chain acyl-CoA dehydrogenase), which is expressed at high levels and catalyzes the initial step of fatty acid β-oxidation. Acadl expression has been shown to be down-regulated by Hif1α in tumor cells 99 . Mid1ip1 (Mig12, 1.22-fold) by itself or in a complex with Thrsp (Spot14, 0.77-fold) interacts with acetyl-CoA carboxylase 100 ; increased Mid1ip1 and decreased Thrsp expression leads to higher acetyl-CoA carboxylase activity, which reduces β-oxidation of fatty acids 100 . Scd1 and Scd2 (stearoyl-CoA desaturase 1 and 2) were  Supplementary  Table S2; values are means ± SE; n = 4 for each genotype; *p ≤ 0.01 vs WT controls.
both up-regulated, which would be expected to reduce fatty acid β-oxidation and improve glucose oxidation 101,102 . Although an increase in stearoyl-CoA desaturase activity might be expected to increase lipid accumulation 102 , reduced expression of Plin3 (perilipin 3) and Plin4, which coat cytosolic lipid droplets 103 , may reflect a reduction  Supplementary Table S4; values are means ± SE; n = 4 for each genotype; *p ≤ 0.01 vs WT controls.
in storage of cytosolic lipids in cardiomyocytes due to reduced fatty acid uptake. Abcd2 (1.61-fold increase) transports very long chain acyl-CoA into peroxisomes and contributes to fatty acid degradation and to synthesis of docosahexaenoic acid (DHA) 104 . Hypoxia has been shown to increase the DHA content of lipid membranes in heart 105 and, along with eicosapentaenoic acid (EPA), DHA is protective in hypoxia-reoxygenation injury in cardiomyocytes 106 .
Expression changes indicate reduced use of ATP and substrates for biosynthesis. A number of changes would be expected to reduce the use of ATP and substrates for biosynthesis (Fig. 3D). Reduced expression of Pcx (pyruvate carboxylase), which exhibits reduced activity in Drosophila flies adapted to hypoxia 107 , would reduce the conversion of pyruvate to oxaloacetate 108 . The reduction in Ucp2 (uncoupling protein 2), a member of the hypoxia response GO category that is down-regulated by hypoxia via repression of Pparg 109 , would reduce the transport of oxaloacetate and other 4-carbon intermediates out of mitochondria 110 . Thus, both changes would increase the use of pyruvate and other intermediates for oxidative phosphorylation and reduce their use in ATP-utilizing biosynthetic processes. Reduced expression of Mlxipl (ChREBP), which regulates glycolysis and fatty acid synthesis, would also be expected to reduce the use of glucose metabolites for biosynthesis 111 . Two genes involved in glycolysis, Pgam1 (phosphoglycerate mutase 1) and Pfkfb2 (6-phosphofructo-2-kinase/ fructose-2,6-bisphosphatase 2; 0.77-fold, not shown), were down-regulated. The substrates and products (2-phosphoglycerate, 3-phosphoglycerate, and fructose 2,6-bisphosphate) of the enzymes encoded by these genes serve regulatory functions in glycolysis 112,113 . Pgam1 is up-regulated in cancer cells and stimulates the use of glycolytic intermediates for biosynthesis 113 , so down-regulation could have the opposite effect. Increased expression of Eef2k (eukaryotic elongation factor-2 kinase, 1.31-fold), which reduces consumption of energy by inhibiting protein synthesis during O 2 deficiency 114 , would also conserve ATP for contraction. Hspb2, encoding a heat shock protein that dramatically enhances the efficiency of coupling between ATP hydrolysis and contractile work 115 , was up-regulated ( Fig. 3D), suggesting an increase in cardiac efficiency.
Expression changes indicate altered membrane excitability and contractile function. Some particularly prominent groups of differentially expressed genes encoded: (i) proteins that regulate heart rate, membrane excitability, and cardiac conduction ( Table 1, Fig. 4), and (ii) myofibrillar proteins localized to the  Supplementary Table S5 and 84 transporter, pump, and channel genes in Supplementary Table S6; values are means ± SE; n = 4 for each genotype; *p ≤ 0.01 vs WT controls, except Kcne1 (p = 0.015).
SCiEntifiC REpoRtS | 7: 7264 | DOI:10.1038/s41598-017-07585-y sarcomere, M-band, Z-discs, t-tubules, and intercalated discs ( Table 1, Fig. 5). These changes indicate major remodeling of electrophysiological and contractile functions in AE3-null hearts. Although some of these changes may have the potential to provide compensation for deficits resulting from impaired CO 2 disposal, they do not provide strong evidence for or against any of the three major hypotheses being considered. See Supplementary Results and Discussion for explanations of the functions of specific genes in Figs 4 and 5, and evidence that they represent adaptive rather than pathological changes.
Expression changes relevant to the Na + -loading and pH i regulation hypotheses. Genes for several proteins that affect contractility by increasing Na + -loading and Ca 2+ -loading were sharply up-regulated (Fig. 6). These include Agtr1a (angiotensin receptor 1a), Nr3c2 (mineralocorticoid receptor), and Egf (epidermal growth factor), which are involved in a pathway that affects Na + -and Ca 2+ -loading 116 . Additional changes consistent with the Na + -loading hypothesis were reduced expression of Atp1a2 and Atp1a4 (α2 and α4 isoforms of the Na + , K + -ATPase).
A number of changes appear related to impaired HCO 3 − and pH i homeostasis. Myoc (myocilin, 1.52-fold) associates with syntrophins in the dystrophin complex 117 , a component of costameres. Increased myocilin, a member of GO:0014066 (Regulation of Phosphatidylinositol 3-Kinase Signaling), stimulates Akt signaling 117 . Processing and secretion of myocilin is altered by extracellular pH and HCO 3 − concentrations 118 and could therefore be affected by loss of AE3, which we have shown to cause an increase in Akt signaling in response to elevated heart rates 4 . Thus, changes in myocilin could be a response to localized changes in pH i or HCO 3 − in AE3-null myocytes and may contribute to up-regulation of Akt signaling during pacing.
Slc4a4 (NBCe1 Na + /HCO 3 − cotransporter), the major HCO 3 − uptake mechanism in cardiomyocytes 1, 119 was down-regulated, which should also reduce Na + -uptake, and Slc26a6, which mediates both Cl − /HCO 3 − exchange and Cl − /formate exchange 120 , was up-regulated (Fig. 6). These changes could, in principle, provide compensation for alkalinization resulting from HCO 3 − overload; however, NHE1 protein 4 and Carns1 (carnosine synthase) mRNA (Fig. 6) were increased. NHE1 is a powerful H + -extrusion mechanism and Carns1 is involved in the synthesis of histidyl dipeptides 121 , which are present at very high levels in cardiac muscle and serve as a major intracellular buffer for H + .
Car14, which encodes CA XIV, the most abundant carbonic anhydrase in mouse heart, was up-regulated ( Fig. 6, see Supplementary Table S8 for expression levels of all CA isoforms). CA XIV is associated with the  Supplementary Table S7.
Values are means ± SE; n = 4 for each genotype; *p ≤ 0.01 vs WT controls.
SCiEntifiC REpoRtS | 7: 7264 | DOI:10.1038/s41598-017-07585-y sarcoplasmic reticulum and colocalizes with mitochondria, where it facilitates CO 2 venting 13 . CA XIV also binds to AE3 on the extracellular surface of mouse cardiomyocytes, where it catalyzes the conversion of HCO 3 − extruded by AE3 to CO 2 14 , which would require H + extruded by some other mechanism. Although mitochondrial CA is expressed at very low levels in heart 13 , Car5b, a mitochondrial CA isoform, was significantly increased. Potential Cl − and H + extrusion mechanisms to function in AE3-mediated CO 2 disposal. It has been noted that AE3-mediated HCO 3 − extrusion has the potential to contribute to CO 2 disposal 10-12 ; however, Cl − recycling and parallel H + -extrusion would also be needed. In heart, there are many Cl − channels that could mediate Cl − recycling (Supplementary Table S9). The negative membrane potential during most of the excitation-relaxation cycle would allow efficient Cl − recycling, although this would not provide charge balance, and extrusion of a net negative charge would require additional ion transport processes to maintain the resting membrane potential. H + -extrusion is particularly problematic because if Na + /H + exchange were responsible, the process would not be electrically balanced and extrusion of Na + would require additional expenditure of energy, with an increase in both ATP production and CO 2 disposal.
The simplest and most energetically efficient H + -extrusion mechanism would be the HVCN1 voltage-sensitive proton channel 122 . Although HVCN1 has not been reported previously in cardiac myocytes, data in the EMBL-EBI Expression Atlas (see Methods) indicates that it is expressed in all mammalian tissues. This includes mouse heart and the heart of other mammalian species, including human (Supplementary Table S10). The available expression data indicate that HVCN1 mRNA is expressed in heart at levels comparable to those of NHE1 and also show that AE3 is the most abundant Cl − /HCO 3 − exchanger in mammalian hearts.

Discussion
Although AE3 was identified almost 30 years ago and shown to be expressed at high levels in heart 123, 124 , its physiological functions are not yet established. Here we used RNA Seq analysis to assess the major hypotheses about its functions in cardiac muscle. The differential expression data provide strong support for the CO 2 disposal hypothesis and suggest major avenues of investigation that could be used to further test this hypothesis. This is potentially important as CO 2 disposal is generally thought to occur entirely by diffusion, either directly across the plasma membrane or through gas channels 125,126 . However, transporters involved in transepithelial ion transport processes are able to extrude large quantities of HCO 3 − and H + that are derived from CO 2 (see Supplementary Results and Discussion), so transport-mediated CO 2 disposal is a reasonable mechanism. Also, the prior demonstrations of carbonic anhydrase-mediated CO 2 hydration in the venting of CO 2 from cardiomyocyte mitochondria 13 , which is needed to avoid inhibition of oxidative phosphorylation, and the association of CA XIV with the extracellular domains of AE3 12, 14 are consistent with transport -mediated CO 2 disposal in heart.
The differential expression patterns most strongly indicating impaired CO 2 disposal in AE3-null cardiomyocytes were the changes in genes mediating hypoxia responses, coupled with changes that would likely increase blood flow, and reduced expression of angiogenesis genes, all of which indicated impaired O 2 /CO 2 balance in heart. The changes in angiogenesis genes that play direct roles in vascular tissues, while modest, were unambiguously consistent with a reduction in angiogenesis in AE3-null hearts, which initially seemed inconsistent with the apparent hypoxia response. However, the hypoxia response is not due to systemic hypoxia, as global loss of AE3 does not affect respiratory function, systemic acid-base homeostasis, or blood gasses (O 2 , CO 2 ) 127 . Also, the level of blood lactate, which is utilized as an energy source by cardiac myocytes, is slightly reduced 127 rather than increased as in systemic hypoxia. Furthermore, systemic hypoxia should increase, not decrease, angiogenesis. The data are consistent with a mild hypoxia response in AE3-null myocytes, which in turn secrete vasodilators in order to increase blood flow. Increased oxygenation of the stromal tissue would lessen the hypoxia in myocytes but would also reduce the stimulus for angiogenesis. These results are consistent with impaired O 2 /CO 2 balance occurring specifically in cardiomyocytes, as would be expected if AE3-mediated extrusion of HCO 3 − plays a major role in CO 2 disposal.
The RNA Seq data provide limited support for the Na + -loading hypothesis. Activation of Agtr1a 128 and treatment with Egf 129 stimulate contractility in cardiac myocytes and isolated hearts. Increased contractility in response to myocardial stretch requires angiotensin, mineralocorticoid, and Egf receptor activities 130 , along with the downstream effector NHE1, which was up-regulated at the protein level in AE3-null hearts 4 . Activation of this pathway is known to increase Na + -and Ca 2+ -loading 116 and, as proposed previously 2, 6-8 , when NHE1 activity is activated to increase Na + -loading, H + -extrusion could be balanced by AE3-mediated HCO 3 − extrusion. This function is compatible with a CO 2 disposal function, since AE3 could be involved in maintaining pH i balance in response to H + -extrusion by either NHE1 or HVCN1. Nevertheless, in previous studies 3, 4 we observed no changes in Ca 2+ -handling that might explain the reduction in force-frequency response in AE3-null mice, and the expression changes observed here provide only limited support for the Na + -loading hypothesis.
The current data and previous studies do not support the hypothesis that the major function of AE3 is to mediate recovery from an alkaline load in vivo, even though AE3-mediated recovery from an alkaline load can be demonstrated following experimental manipulations in vitro 2 . NHE1 protein was increased in AE3-null hearts 4 and NHE1 mRNA was increased in isolated AE3-null cardiac myocytes 2 . Also, mRNA encoding Carns1, involved in histidyl dipeptide synthesis, was increased. Histidyl dipeptides would likely be important for efficient venting of CO 2 from mitochondria 13 as they would buffer H + produced by CO 2 hydration, thus facilitating the reaction that converts waste CO 2 to H + and HCO 3 − . Increased expression of NHE1 and Carns1 mRNA suggest that the loss of AE3 leads to an increased need for NHE1-mediated H + -extrusion and H + buffering capacity. Given that AE3 extrudes HCO 3 − , this may seem paradoxical; however, CO 2 hydration generates equimolar amounts of HCO 3 − and H + and a reduction in HCO 3 − extrusion could affect not only the rate of CO 2 hydration, but also a parallel mechanism of H + extrusion that is affected by extracellular carbonic anhydrase associated with AE3 (discussed below).
The up-regulation of Car14 is potentially relevant, as CA XIV has been proposed to play an important role in both CO 2 venting from mitochondria 13 and in AE3-mediated HCO 3 − extrusion from cardiac myocytes, where it is associated with the extracellular domains of AE3 14 . CA XIV also associates with AE3 in retina and brain 12,14 . Also, its mRNA was increased in neurons of AE3-null mice 14,131 , consistent with a deficit in CO 2 disposal in AE3-null neurons. This could be responsible for the epilepsy phenotype in AE3-null mice 14,132 and in humans with a heterozygous AE3 mutation 133 , as hypoxia can contribute to epilepsy 134 . Interestingly, expression of Car5b, a mitochondrial CA isoform, was increased in AE3-null hearts, suggesting a perturbation of CO 2 venting from mitochondria. This is consistent with the reduction in AE3-mediated extrusion of HCO 3 − from the cell, which would shift the equilibrium toward a reduction in CO 2 hydration. These changes support the CO 2 disposal hypothesis.
Because a reduction in carbonic anhydrase-mediated hydration of waste CO 2 as it exits the mitochondria causes an inhibition of oxidative phosphorylation 13 , it is reasonable to expect that a reduction in the ability to dispose of HCO 3 − would elicit adaptative changes in energy metabolism. The changes in metabolic genes, while modest, suggest an increase in glucose metabolism and a reduction in fatty acid uptake and metabolism, which would provide a more favorable ATP/O 2 ratio during mild hypoxia. Interestingly, a number of the proteins involved in glucose metabolism are affected by activation of Akt (indicated in Fig. 3), which is known to have a major effect on glucose metabolism in heart 135 . The increased Akt phosphorylation observed when AE3-null mice were subjected to atrial pacing 4 and upregulation of glucose metabolism genes that respond to Akt suggest that glucose metabolilsm may be stimulated in AE3-null hearts during acute biomechanical stress, when O 2 utilization would be increased. In addition to an improved ATP/O 2 ratio, a shift in the relative balance between glucose and fatty acid metabolism would also be expected to improve cardiac function due to a reduction in the negative effects of fatty acid metabolism on cardiac efficiency 136 . An increase in cardiac efficiency is supported by the up-regulation of Hspb2 (Fig. 3D). When challenged with β-adrenergic stimulation, mice lacking Hspb2 hydrolyzed more ATP but performed less work 115 . Thus, an increase in Hspb2 should provide better protection of energy reserves and improved contractility in response to β-adrenergic stimulation and other stress conditions. In addition, there were a number of expression changes that would be expected to reduce the use of substrates and ATP for biosynthesis, which would conserve energy for muscle contraction. A reduction in use of ATP and substrates for biosynthesis is consistent with the smaller hearts in AE3-null mice 2, 4 .
The major difficulty in proposing a role for AE3 in transport-mediated CO 2 disposal in cardiac myocytes is that it would require the parallel operation of an energetically-efficient H + disposal mechanism. Of the known mechanisms of H + extrusion, only the HVCN1 H + channel, which is expressed in heart (Supplementary Table S8) and seems to be ubiquitous in mammalian tissues, would appear to have the necessary properties. It is perfectly selective for H + , mediates outward transport only, and is strongly activated by intracellular acidity and a positive membrane potential 122 . As illustrated in Fig. 7, HVCN1 and AE3 have the potential to form an efficient mechanism for transport-mediated CO 2 disposal. Because AE3 is electroneutral and unaffected by changes in membrane potential, its HCO 3 − extrusion activity is driven by the inwardly directed Cl − gradient. With recycling of Cl − through sarcolemmal Cl − channels while the cell is in the polarized resting phase, efficient export of HCO 3 − being produced via CO 2 hydration 13 would be maintained. H + passing through HVCN1 during each action potential would be catalytically combined with HCO 3 − present in the unstirred layer via extracellular CA XIV that is associated with AE3 12, 14 , thereby preventing a buildup of acid on the cell surface. It should be noted that generation of CO 2 by this mechanism would not require tight coupling in which the HCO 3 − being extruded by AE3 is directly combined with H + being extruded by HVCN1. With HCO 3 − serving as the major extracellular buffer and AE3 continuously replenishing HCO 3 − being consumed by CO 2 production via extracellular CA activity, the HCO 3 − concentration in the unstirred layer would be maintained and the CO 2 generated would be washed away in the blood.
Although additional studies will be needed to test various aspects of the proposed CO 2 disposal mechanism and to assess the metabolic, electrical, and contractile consequences of its perturbation, the expression data described here suggest that AE3 plays a central role in transport-mediated CO 2 disposal. Transport-mediated CO 2 disposal occurs in epithelial tissues as a side effect of acid-base and electrolyte transport; however, this is the first description of an energetically-efficient system for transport-mediated CO 2 disposal. In addition, the data set (included in its entirety in Supplementary Files) provides both a framework and a rich source of additional information for further investigations of the cardiac functions of AE3.

RNA Seq Analysis.
Total RNA was isolated from whole hearts of 4-month-old FVB/N WT and AE3-null male mice (n = 4 of each genotype). All procedures using animals conformed to guidelines published by the National Institutes of Health (Guide for the Care and Use of Laboratory Animals) and were approved by the Institutional Animal Care and Use Committee at the University of Cincinnati. Samples were subjected to RNA Seq analysis in the University of Cincinnati Genomics and Sequencing Core using the Illumina HiSeq. 1000 platform. Sequence reads were aligned to the reference mouse genome using TopHat aligner 137 . The full data set was deposited in the Gene Expression Omnibus (GEO accession number GSE70471) and is also provided as an Excel File in Supplementary  Table S1. Additional Excel Files with subsets of expression data are provided in Supplementary Tables S2-S7. Foldchange expression data in the Excel files relate expression in AE3-null hearts relative to that observed in WT hearts.
Statistical analysis to identify differentially expressed genes was performed using the negative-binomial model of read counts as implemented in DESeq Biocondoctor package 138 . Most of the genes (80%) included in the figures had an FDR (False Discovery Rate) < 0.05, which is a measure of the probability of a false positive given the inclusion of over 23,000 genes in the analysis. Among the 536 genes with FDR < 0.05, 284 were up-regulated and 252 were down-regulated. Some genes with FDR > 0.05 were also included because they were in significantly enriched GO categories that were a major part of the phenotype (see Table 1 and legends for Supplementary  Tables S2-S7). Inclusion of genes with a significant P value but an FDR > 0.05 that were also present in significant GO categories, which were derived from a small subset of genes, protected against exclusion of false negatives. mRNA expression is presented as Reads Per Kilobase per Million mapped reads (RPKM), which normalizes for the size of the mRNA and provides a measure of the relative abundance of specific transcripts. Approximately 25 million reads were achieved per sample.
Cardiac RNA Seq expression data for mammalian hearts (Supplementary Table S10) and other tissues was obtained from the European Bioinformatics Institute (EBI) Expression Atlas (http://www.ebi.ac.uk/gxa/home). Heart data was for mRNA expression in hearts of both male and female Fisher 344 rats at 4 developmental stages 139 and in hearts of C57Bl6 mice, opossum, rhesus monkey, and human 140 . The latter study used a normalization procedure that allowed cross-species comparisons of specific genes.
Gene Ontology Analysis. The online GOrilla program 15 (Gene Ontology enRIchment anaLysis and visu-aLizAtion tool; http://cbl-gorilla.cs.technion.ac.il/) was used for Gene Ontology analyses. This program can be accessed and used without registering and is self-explanatory and easy to use. Two analysis options were used: (1) Single Rank List in which the entire gene set was ranked according to P values and (2) Two Unranked Lists, in which a target list of genes with a P value or FDR (false discovery rate) value within a specified range was compared against the background list of all genes (over 23,000 total, with 18,892 expressed genes) to which reads were mapped. Neither analysis option alone was completely satisfactory for identifying relevant GO categories and genes. Thus, we used both the Single Rank option, which considers only the relative significance ranking of all genes and does not depend on an arbitrary cutoff and the Two Unranked lists option, in which various cutoffs, Figure 7. Model for the role of the AE3 Cl − /HCO 3 − exchanger in transport-mediated CO 2 disposal. Oxygen entering the myocyte is rapidly converted to CO 2 in mitochondria. CO 2 venting from mitochondria 13 is facilitated by CA-mediated conversion of CO 2 to HCO 3 − and H + , with H + buffered by histidyl dipeptides (HDP) and other components, thereby effectively blocking the back reaction by keeping the concentration of free H + low. CO 2 disposal is proposed to be mediated by a combination of HCO 3 − extrusion by AE3, Cl − recycling via Cl − channel activity, H + -extrusion via HVCN1 during each depolarization, and extracellular carbonic anhydrase (CA) activity to generate CO 2 .
ranging from very stringent (FDR < 0.01-0.05) to less stringent (P < 0.01) were used. The Single Rank option identified some functions that were highly enriched at the top end of the significance range and other functions that were significantly enriched over a very large range, but with lower enrichment scores. Both approaches are needed to minimize inclusion of false positives or exclusion of false negatives.
Enrichment analysis for both options uses the hypergeometric distribution to calculate statistical probabilities 15 . The Single Rank option avoids setting an arbitrary cutoff of P values to be considered and was particularly useful for identifying GO categories that were highly enriched at the top end of the significance range for individual genes. When using the Single Rank analysis, all genes were ranked by P values and the distribution of genes in each GO category was used to calculate significance scores and enrichment of specific GO categories. When using the Two List analysis, significance and enrichment in each GO category is calculated based on the number of genes for specific GO categories appearing in the target list and background list. Enrichment (N, B, n, b) is defined as follows: N is the total number of genes, B is the total number of genes associated with a specific GO term, n is the number of genes in the top of the input list or in the target set when appropriate, b is the number of genes in the intersection; Enrichment = (b/n)/(B/N). As shown in Supplementary Fig. S2, the program provides a visual display of hierarchically arranged GO categories, with large broad categories displayed at the top and smaller, more specific categories nested within one or more of the broad categories. In addition, the program provides a list of GO categories ranked by significance and a list of the genes in each GO category, which can be shown or hidden. When performing GOrilla analysis, the genes were not separated into up-regulated or down-regulated genes since a given process might be affected positively (or negatively) by up-regulation of some genes and down-regulation of others. Significant GO categories were inspected, grouped by related functions, and a list of non-redundant genes for each broad category was prepared as described in the legends for Supplementary Tables S2-S7. In some cases, particularly (1) Hypoxia, Vasodilation, Angiogenesis (Supplementary Table S2) and (2) Energy Metabolism (Supplementary  Table S4), additional genes were included in the lists based on PubMed literature searches, which relied heavily on PubMatrix. Those searches can be publicly accessed as described below.
PubMatrix Analyses. Pubmatrix (https://pubmatrix.irp.nia.nih.gov/), an online literature search tool 16 , was used to search PubMed to identify relevant publications describing the functions of specific genes in various biological processes ( Supplementary Fig. S3). In each run, PubMatrix performs pairwise literature searches of up to 100 search terms and up to 10 modifier terms. The search terms corresponded to the 536 genes with an FDR ≤ 0.05; for each gene, the gene symbol was grouped with common names for the gene. For example, search terms for the angiotensin II receptor, type 1 was: (Agtr1a or AT1a or Agtr1). The 536 genes with FDR < 0.05 were grouped into 4 sets of up-regulated genes (Sets 1-4) and 4 sets of down-regulated genes (Sets 5-8) and each set was searched against modifier terms for 1) Hypoxia and related processes [Modifier terms: (hypoxia or hypoxic), (HIF1 or Hif1alpha or Hif1a or Hif), (Egln3 or PHD3), (Epas or Hif2a), (Vegf or Vegfa), Angiogenesis, Vasodilation, Vasoconstriction, (ischemia or ischemic), (heart or cardiac)] and 2) Energy Metabolism [Modifier terms: "energy metabolism", "lipid metabolism", "beta oxidation", (mitochondria or mitochondrial), "glucose metabolism", "glucose oxidation", Glucose, Glycolysis, Ampk, AKT]. As illustrated in Supplementary Fig. S3, each set of searches is displayed on a grid that allows any pairwise search to be opened; the searches are stored and can be reopened at any time. The reader can access the searches performed here for Hypoxia and Metabolism genes (under Public Results, username Vair&Shull) by registering on PubMatrix with a username (typically email address) and a simple password. Quantitative Real-time PCR (RT-PCR) analysis. Total RNA was isolated from hearts of 4-month-old FVB/N male mice (N = 4 of each genotype) using Tri-reagent (Molecular Research Center, Cincinnati, OH). cDNA was prepared by random priming using Superscript III First-strand synthesis kit from Life Technologies.