Iron-dependent post transcriptional control of mitochondrial aconitase expression

Abstract

Iron regulatory proteins (IRPs) control the translation of animal cell mRNAs encoding proteins with diverse roles. This includes the iron storage protein ferritin and the tricarboxylic cycle (TCA) enzyme mitochondrial aconitase (ACO2) through iron-dependent binding of IRP to the iron responsive element (IRE) in the 5′ untranslated region (UTR). To further elucidate the mechanisms allowing IRPs to control translation of 5′ IRE-containing mRNA differentially, we focused on Aco2 mRNA, which is weakly controlled versus the ferritins. Rat liver contains two classes of Aco2 mRNAs, with and without an IRE, due to alterations in the transcription start site. Structural analysis showed that the Aco2 IRE adopts the canonical IRE structure but lacks the dynamic internal loop/bulge five base pairs 5′ of the CAGUG(^U/C) terminal loop in the ferritin IREs. Unlike ferritin mRNAs, the Aco2 IRE lacks an extensive base-paired flanking region. Using a full-length Aco2 mRNA expression construct, iron controlled ACO2 expression in an IRE-dependent and IRE-independent manner, the latter of which was eliminated with the ACO2^3C3S mutant that cannot bind the FeS cluster. Iron regulation of ACO2^3C3S encoded by the full-length mRNA was completely IRE-dependent. Replacement of the Aco2^3C3S 5′ UTR with the Fth1 IRE with base-paired flanking sequences substantially improved iron responsiveness, as did fusing of the Fth1 base-paired flanking sequences to the native IRE in the Aco^3C3S construct. Our studies further define the mechanisms underlying the IRP-dependent translational regulatory hierarchy and reveal that Aco2 mRNA species lacking the IRE contribute to the expression of this TCA cycle enzyme.

Abbreviations

ACO2:

mitochondrial aconitase

ALAS2:

erythroid isoform of 5-aminolevulinate synthase

C:

control

D:

desferal

DMT1:

divalent metal transporter 1

eGFP:

enhanced green fluorescent protein

FTH1:

H-ferritin

FTL:

L-ferritin

H:

hemin

IRE:

iron responsive element

IRP:

iron regulatory protein

ORF:

open reading frame

SD:

standard deviation

SLC11A2:

DMT1

SLC40A1:

ferroportin

TCA:

tricarboxylic acid cycle

TFRC:

transferrin receptor 1

TISU:

translation initiator for the short 5′ UTR

UTR:

untranslated region

Introduction

RNA-based regulatory networks are central to maintaining organismal homeostasis in prokaryotes and eukaryotes. RNA regulons composed of RNA binding proteins, non-coding RNAs, and mRNA targets form networks that control intermediary metabolism, growth control, and other critical cellular processes.^1–5 Many such networks rely on hierarchical regulation of multiple RNA targets.^4,5 Dysregulation of the activity of RNA binding proteins or mutation of their binding sites leads to human diseases characterized by, for instance, oncogenesis, intellectual disability, or disrupted metal metabolism.³,^6–8 Iron regulatory proteins (IRPs) are cytosolic iron-regulated RNA binding proteins activated by iron deficiency in animal cells. IRPs classically bind to short stem-loop structures, iron responsive elements (IREs), in the 5′ or 3′ untranslated region (UTR) of mRNA encoding proteins that control cellular iron homeostasis and compensatory actions within the adaptive response to iron deficiency.

The expanding scope of cellular processes influenced by IRPs extends beyond those involving proteins with direct roles in iron metabolism and includes the tricarboxylic acid cycle (TCA), adaptive responses to hypoxia, possibly cell cycle regulation, and others.^9–16 The initial mRNAs found to be targets of IRPs are the mRNAs encoding the L (Ftl) and H (Fth1) subunits of the iron storage protein ferritin. IRPs bind to a single IRE in the 5′ UTR of these transcripts to block initiation.¹⁷  Tfrc mRNA contains five IREs in the 3′ UTR, which in collaboration with IRPs and a nuclease mechanism, control mRNA stability and cellular uptake of transferrin iron.^18–20 To date, functional canonical IREs have been identified in 10 mRNAs, although recent work suggests IRP may control a much larger regulon, including RNAs with non-canonical IREs.^10,16 The functional diversity of the proteins encoded by IRE-containing mRNA suggests IRPs may control mRNA fate in a hierarchical manner.^21,22 Studies on mRNA with 5′ IREs showed that the ferritin subunits were much more strongly regulated than mitochondrial aconitase (ACO2), succinate dehydrogenase B (SDHB), or the erythroid isoform of 5-aminolevulinate synthase (ALAS2).^23–28 Furthermore, the synthesis of proteins required for iron storage (ferritins, FTH1, and FTL) and export (ferroportin, SLC40A1) is strongly repressed in iron-replete cells and thus respond most to iron excess. In contrast, weaker targets like Alas2, Aco2, and Sdhb mRNA respond most to iron deficiency or genetic activation of IRP RNA binding activity, suggesting there are two classes of IRP-dependent translational control.^23,27,28 Further evidence of selective action of IRPs was noted by the finding that the binding affinity of IRP1 for six 5′ IRE extended over a 9-fold range delimited by Ftl and Fth1 IREs having the highest and Aco2 the lowest affinity.²¹ The relevance of this RNA binding hierarchy is illustrated in hereditary hyperferritinemia cataract syndrome, where the most significant activation of ferritin expression due to mutations in and flanking the IRE occurred over the first 10-fold drop in the RNA binding affinity.⁶ Understanding the mechanistic basis responsible for the selective targeting of mRNA by the IRP–IRE system is needed to fully understand the dysregulation of iron metabolism in disease and develop approaches to ameliorate it.

Canonical IRE families [e.g. ferritins (Fth1, Ftl) vs. Aco2] can differ in the sequence and structure of their RNA helix and the extent and structure of flanking sequences.^29–32 Of the six 5′ IRE-containing mRNAs described to date, Fth1 and Ftl are the most extensively studied. Canonical IREs contain an essential unpaired C (C11 in the nomenclature used herein) five basepairs 5′ of the terminal loop, but in ferritin IRE, a conserved unpaired U, two nucleotides further 5′, which may exist as a single unpaired nucleotide or as part of a larger UGC internal loop, is also present.^29–31,^33,34 Deleting this unpaired U reduces IRP binding and enhances ferritin mRNA translation in cell-free systems.^21,31,35 Furthermore, Fth1, Ftl, and Slc40a1 IRE are part of a larger RNA helix due to more extensive base-paired flanking sequences compared to Aco2 IRE.^21,29,30,36 Deleting the flanking sequences significantly lowers IRP1 RNA binding affinity for these RNAs.²¹ In the case of ferritin, structural integration of the IRE with flanking sequences is a vital part of its translational repression by IRP in vitro.^30,36 While these studies have begun to define the basis for the strong regulation of ferritin mRNA translation by IRPs, cellular studies have not examined the mechanistic basis for weaker regulation of 5′ IRE-containing mRNAs such as Aco2.

In this study, we show that multiple species of Aco2 mRNA exist in rat liver and, because of differences in 5′ UTR length, fall into two classes depending on whether an IRE is present or not. We structurally defined the IRE in Aco2 mRNA using nuclease mapping and chemical probing and assessed it functionally in transfected HEK 293 cells. We demonstrate that the Aco2 IRE secondary structure aligns with the canonical IRE structure but differs from the IRE in ferritin mRNAs in two ways. First, it cannot form a three-nucleotide internal loop containing the critical unpaired C five base pairs from the terminal loop. Second, it lacks the extensive base-paired flanking sequence seen in other IRP-targeted mRNAs because of its proximity to the mRNA 5′ end. Functional analysis in HEK 293 cells using a full-length IRE-containing Aco2 mRNA encoding myc-tagged ACO2 revealed roles for translational regulation and protein stability dependent on the [4Fe–4S] of ACO2 in response to iron. In contrast, an IRE-containing Aco2 mRNA encoding the protein with each of its Fe–S Cys ligands mutated to Ser (ACO2^3C3S) shows complete dependence on the IRE for iron responsiveness of protein expression. Substitution of the Fth1 IRE with its base-paired flanking sequence in place of the Aco2 IRE significantly improved iron responsiveness, as did insertion of Fth1 base-paired flanking sequences into the Aco2 IRE. Our studies demonstrate the competency of the Aco2 IRE for regulation by IRP in vivo. Because the Aco2 IRE species lack an extended RNA helix, ACO2 synthesis is less sensitive to iron scarcity. Conversely, lengthening the Aco2 IRE helix through joining it to the sequence flanking the Fth1 IRE improved iron sensitivity, demonstrating a role for extended IRE structure. Taken together, our studies support the concept that 5′ IREs within an extended RNA helix confer stronger iron-dependent translational regulation in part because of increased thermodynamic stability and hence ‘lifetime’ of the IRE structure and, as implicated by the crystal structure of IRP1 with the ferritin IRE, the ability to form additional protein-RNA contacts with the base-paired flanking region.³⁴

Methods

Cell culture

HEK 293 cells were grown in high glucose DMEM with 10% fetal bovine serum (Hyclone). Cells were transfected with cesium chloride-purified plasmid DNA using lipofectamine (Thermo Fisher Scientific). HEK 293 cells were treated with hemin (Frontier Scientific) to induce iron overload or deferoxamine (Millipore Sigma) to cause iron deficiency.

5′ RLM RACE

The 5′ ends of capped Aco2 mRNA were cloned using the RLM-RACE Kit (Invitrogen). Total rat liver RNA was used for cDNA synthesis. Aco2 cDNA was made using the gene-specific primer CCTCAATCAGATGATCACAGTGGATGG. In the second PCR reaction, the primers CATAGCCATCTGGGCTGTTG and CTTATCGATGCTGATGGCGATGAATGAACACTG were used to amplify Aco2 cDNA further. The cDNA products (11 obtained; 9 unique sequences) were cloned and sequenced.

Human ACO2 expression plasmids and cell culture

To construct a mammalian expression plasmid that encoded the full-length Aco2 mRNA, a human Aco2 cDNA plasmid along with the expression vector GR4 was used.³⁷ GR4 has the advantage that the primary transcription start site has been mapped, facilitating the expression of Aco2 mRNA with the proper 5′ UTR length. GR4 also encodes mCherry in a second transcription unit, thereby allowing correction for transfection efficiency. The expression of ACO2 and mCherry is driven by separate CMV promoters. Human Aco2 cDNA was purchased from Open Biosystems (clone ID: 5264179; accession number: BC026196) and modified to contain the complete 5′ and 3′ UTRs of the mRNA and a C-terminal myc tag before insertion into GR4. The full-length ACO2 cDNA was inserted in place of enhanced green fluorescent protein (eGFP) at the primary transcription start site for the CMV promoter driving its transcription, as determined previously³⁷ (Table S1). Our approach allowed for relatively easy insertion of mutant or chimeric IRE sequences in place of the wild-type ACO2 5′ UTR. Lastly, the three Cys (C385, C448, and C451) required for ligation of the [4Fe–4S] cluster in ACO2 were mutated to Ser to uncouple ACO2 expression from potential differences in protein stability between apo- and holo-ACO2.³⁸ The Quikchange Multi Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA, USA) was used. The accuracy of mutagenesis was confirmed by DNA sequencing. Additional details concerning the construction of the Aco2 expression plasmids are provided in the supplement (Table S1). We determined ACO2 expression by immunoblotting for the myc tag and normalized by mCherry and α-tubulin protein level.

All cell culture experiments used HEK 293 cells grown in DMEM high glucose (Gibco) with 10% fetal bovine serum (Hyclone). Cells were transfected with lipofectamine (Invitrogen).

Quantifying ACO2 expression

ACO2 level was determined by immunoblotting with an anti-C-myc monoclonal antibody (9E10). The results were normalized for transfection efficiency using m-Cherry expression and protein load using α-tubulin.

Antibodies and immunoblotting

The mouse 9E10 antibody was produced from a hybridoma (ATCC CRL-1729) through ascites production in mice (ENVIGO). The rabbit polyclonal antibody against α-tubulin was from Abcam (#15246). The secondary antibodies used were goat anti-mouse (LiCor IR 680) and goat anti-rabbit (LiCor IR 800). Immunoblots were quantified on a LiCor Odyssey imaging system. Results were compared using the Student's t-test.

Nuclease mapping and chemical modification of Aco2

The supplement provides these details.

In silico prediction of RNA secondary structure

The secondary structure of RNA was predicted using the Mfold web server (http://mfold.rna.albany.edu/?q=mfold) applying default settings.^39–41 Mfold is now available through UNAFold (http://www.Mfold.org/mfold/applications/rna-folding-form.php). We considered the structures with the most stable secondary structure (lowest ΔG). Default settings were used to predict the Aco2 IRE structure. After mapping results were obtained, nucleotides U12–14 and A25–27 in the upper stem of the IRE were forced to base-pair. Terminal loop nucleotides G19 and C22 were forced to be single-stranded.

Results

Aco2 mRNA 5′ end and IRE secondary structure analysis

As a foundation for our secondary structure and translational regulation analysis of Aco2 mRNA, we performed a 5′ RLM RACE analysis using rat liver RNA as the substrate to detect 5′ capped Aco2 RNAs. Sequencing of the cDNAs revealed nine unique 5′ ends for Aco2 mRNA (Fig. 1A). Because the mRNA cap is added co-transcriptionally and the sequences we obtained reflected the genomic sequence at this locus, the 5′ end heterogeneity for Aco2 mRNA appears to reflect different transcription start sites and does not arise from alternative splicing. Relative to the AUG thought to initiate synthesis of the ACO2 preprotein,⁴² the 5′ UTR ranged in length from 4 to 29 nucleotides. The 27 and 29 nucleotide species we observed in rat liver are similar or identical to the 29 nucleotide 5′UTR identified by primer extension of porcine heart Aco2 mRNA.⁴² Of these RNA species identified, only two of the nine identified (22%) were long enough to be able to form an IRE, as predicted by Mfold⁴⁰ although we cannot exclude the possibility that these Aco2 mRNAs with the longer 5′ UTR are underrepresented in our analysis due to an inability of reverse transcriptase to reach the mRNA 5′ end. Four of the nine Aco2 mRNA species identified had 5′ UTRs that were <10 nucleotides in length and are likely poorly translated by the canonical initiation mechanism. However, the sequence surrounding the Aco2 initiation codon (CAAAATGGCGCCU) has similarities to the translation initiator for the short 5′ UTR (TISU) family of elements (SAASATGGCGGC where S is G or C). TISU elements promote CAP-dependent, scanning-independent mRNA translation with very short 5′ UTRs encoding proteins required during cellular energy stress, including those involved in mitochondrial ATP generation.^43,44 Our results indicate that a small fraction of Aco2 mRNA is likely subject to IRP action, but a more significant fraction of Aco2 mRNA in the liver contains a very short 5′UTR and may be subject to other mechanisms of translational control.

Fig. 1.

RNA sequence and secondary structures: (A) 5′ RLM RACE analysis of 5′ capped Aco2 mRNA species in rat liver; (B) RNA secondary structure analysis of Aco2 IRE as determined by nuclease mapping of (³²P) 5′-end-labeled RNA. A, T refers to single strand specific ribonucleases, and V1 is the double-strand specific ribonuclease; (C) RNA secondary structure analysis of Aco2 IRE as determined by nuclease mapping and chemical modification of (³²P) 3′-end labeled RNA; (D) RNA secondary structure of bullfrog H-ferritin IRE as determined by Theil and associates using chemical probing and nuclease mapping³⁰; (E) RNA secondary structure of human Fth1 IRE and flanking regions as predicted by Mfold; (F) RNA secondary structure of Fth1-Aco2 hybrid composed of the Aco2 IRE and Fth1 flanking regions as predicted by Mfold. The region marked by the bracket is from Fth1. The arrow points to the adenosine at nucleotide 44, added to increase the stem stability more in line with the Fth1 IRE plus flanking sequences. The 3′ terminal guanosine was added to keep the Aco2 coding region in frame for the production of ACO2. In panels (B) and (C) the AGU of the terminal loop and unpaired C at position 11 make the majority of contacts with IRP1,³⁴ which are shown in white with black background. In panels (C), (D) and (E) the base paired region flanking these IRE are highlighted in yellow. For panels (B) through (F) the dotted line between the first C and second G of the terminal loop is to indicate that these nucleotides base pair in the IRP1 bound RNA allowing the formation of a pseudotriloop that displays the A, G and U in optimal conformation for binding the protein.³⁴

We used nuclease mapping and chemical modification of the Aco2 IRE to map the secondary structure of the RNA directly. The 5′ UTR plus six nucleotides of the coding region of Aco2 mRNA were synthesized, [³²P]-labeled at the 5′ or 3′ end, and subjected to mapping or modification. The summation of this work is provided in Fig. 1B and C, while the original data are provided in the Supplement (Figs. S1–S9). We used the 27 nucleotide Aco2 5′ UTR that includes five nucleotides of the coding region for nuclease mapping because MFOLD predicts the same structure as for the 29 nucleotide species, and the 27 nucleotide species initiates with a G, which allows for much higher RNA yields using T7 RNA polymerase.⁴⁵

The structure of the Aco2 IRE as determined by nuclease and chemical sensitivity is shown (Fig. 1B and C). Nucleotides highlighted in black with white lettering represent those making the majority of contacts with IRP1.³⁴ Secondary structure mapping confirmed the presence of the terminal loop, as noted by the susceptibility of this region to single-strand specific nucleases A, S, and T1 (Fig. 1B and C; Figs. S1–S5). The first G of the terminal loop displayed enhanced cleavage by T1 and S1 relative to other loop nucleotides, as observed in the ferritin IRE.^29,30,36 Susceptibility of the single adenosine in the terminal loop (A18) to diethylpyrocarbonate (DEPC) followed by chemical cleavage indicates it is not involved in base pairing (Fig. S6). DEPC susceptibility of A18 was blocked in the presence of saturating levels of IRP1 (Figs. S7 and S8).

The upper stem (between the unpaired C11 and the terminal loop) of the Aco2 IRE displayed significant sensitivity on both the 5′ and 3′ sides to double-strand specific nuclease V1 (Fig. 1B and C; Figs. S5 and S9). The upper stem region of the IRE closest to the terminal loop exhibited limited sensitivity to single-strand specific nucleases A and T1 (Figs. S1, S2, and S4), perhaps suggesting ‘breathing’ of this region as has been previously noted for the ferritin IRE.^29,30,36 Similarly, the lower stem exhibited significant cleavage by RNase V1 on both the 5′ and 3′ sides. Cleavage at C8 by both RNases V1 and A suggests the presence of an open versus closed internal loop, perhaps enlarging the internal loop involving A4–C6 in the lower stem. The structures for the Aco2 IRE shown in Fig. 1B and C, which are compatible with the canonical IRE secondary structure, arise from Mfold after incorporation of the mapping results. The Aco2 IRE is closer to the 5′ end of Aco2 mRNA and consequently cannot form the extensive base-paired region flanking the core IRE predicted for the ferritin IREs and which was observed in the Fth1 IRE structure determined by Theil and associates (Fig. 1D, base-paired flanking region in yellow).^29,30,36 We also note that the Aco2 IRE is much less thermodynamically stable with a ΔG of –5.8 kcal/mol compared to, for instance, Fth1, where the ΔG is –22.3 kcal/mol.

An ACO2 FeS cluster mutant as a reporter of IRE-dependent regulation

To determine the regulatory elements involved in post-transcriptional control of ACO2 expression, we inserted a full-length ACO2 cDNA, with a single C-terminal myc tag, in place of eGFP in the GR4 expression plasmid.³⁷ The ACO2 cDNA and its derivatives used here were inserted at the primary transcription start site for the CMV promoter allowing us to control the 5′ UTR length³⁷ (Table S1). Our initial studies revealed iron regulation of the abundance of the myc-tagged protein after a 24-h treatment with hemin or desferal (Fig. S10). However, the effect of iron chelation with desferal was not ablated by the deletion of the critical unpaired C (C11 in our constructs), five basepairs 5′ of the terminal loop of the IRE, a mutation known to strongly reduce IRP binding.^34,46,47 This suggests that both IRE-dependent and IRE-independent factors were driving ACO2 expression in relation to iron status.

A second construct, replacing the wildtype Aco2 open reading frame (ORF) with Aco2^{C385S, C448S, C451S} (abbreviated as Aco2^3C3S) where all FeS cluster ligating Cys were mutated to Ser, was used (Fig. 2A). This eliminated FeS cluster occupancy as a potential variable in iron regulation of the ACO2 protein level because ACO2^3C3S can only exist as the apoprotein (Fig. 2C). Previous studies of iron regulation of the expression of other FeS proteins revealed that they are frequently more unstable in the apo form.^38,48,49 The 5′ UTR in the Aco2^3C3S construct was modified by deletion or insertion mutagenesis to determine the role of the IRE in the iron-dependent control of ACO2^3C3S expression in cells and to compare its regulatory strength to that of other IREs (Fig. 2B). We found that the expression of ACO2^3C3S encoded by the 5′ UTR with an intact IRE was lower than that of wildtype ACO2^3C3S with the ΔC11 mutant IRE, but it retained a strong response to desferal and a ratio of expression in hemin versus desferal of 1.9-fold (Fig. S11). In the seven independent experiments reported in Fig. 3 through Fig. 5, the ratio of expression of the ACO2^3C3S construct in hemin-treated relative to desferal-treated cells was highly reproducible, with a value of 2.2 ± 0.2 (mean ± SD). Most importantly, the iron-regulation response of the ACO2^3C3S construct was eliminated when its mRNA contained the ΔC11 IRE mutant 5′ UTR (Fig. 3A). To further evaluate the functionality of the Aco2 IRE, we replaced the 27 nucleotide 5′ UTR of Aco2^3C3S with the 36 nucleotide unstructured 5′ leader of the alfalfa mosaic virus (AMV) mRNA (Fig. 2B, AMV).⁵⁰ Expression of ACO2^3C3S driven by the AMV 5′ leader was lower than ACO2^3C3S containing the Aco2 5′ UTR and failed to respond to changes in iron status (Fig. 3B). Thus, neither the 5′ UTR with the ΔC11 mutant of the Aco2^3C3S IRE nor the AMV 5′ leader on its own could confer iron-dependent control on ACO2^3C3S production. We conclude that the native 27 nucleotide Aco2 5′ UTR with the intact IRE is required for iron-dependent control of ACO2^3C3S synthesis.

Fig. 2.

ACO2 expression constructs. (A) Schematic of a portion of the modified GR4 dual expression plasmid.³⁷ In this modified construct, we replaced the eGFP gene in the original plasmid with the 5′ UTR, ORF, and 3′ UTR of Aco2 The unique Sac 1, along with other restriction sites, facilitates insertion of wildtype or mutant Aco2 5′ UTR and additional 5′ UTR at the primary transcription start site for the CMV promoter enhancer.³⁷ (B) The following 5′ UTR constructs used in this study are described; see also Table S1. Aco2: The intact 27 nucleotide 5′ UTR is shown in Fig. 1A. Aco2^ΔC11: This deletes the critical unpaired C in the stem of the IRE yielding a 26 nucleotide 5′ UTR. AMV: The 36 nucleotide 5′ UTR of alfalfa mosaic virus (AMV) RNA 4⁵⁰ replaced the wildtype 5′ UTR of Aco2 in the AMV construct. Fth1: The Aco2 5′ UTR is replaced with 70 nucleotides from the human Fth1 5′ UTR, including the IRE and its flanking sequences. Alas2: The Aco2 5′ UTR is replaced with 30 nucleotides of the human Alas2 IRE.^52,53  Amv-Aco2: The 5′ most eighteen nucleotides of the AMV RNA 4 5′ UTR was fused to the 5′ end of wildtype Aco2 5′ UTR. Fth1-Aco2: The base paired helical region flanking the Fth1 IRE was fused to a 24 nucleotide portion (from C9 to G36) of Aco2 5′ UTR plus the AUGG of that RNA. An additional adenosine (A44) was inserted on the 3’ side of the flanking sequence, allowing the production of an IRE + flanking sequence with a similar ΔG as the wildtype Fth1 region (see Figs. 1E and 1F). Because the Aco2 segment retained the AUGG the 3′ flanking sequence (not shown) from Fth1 plus additional of 2 nts to maintain the reading frame resulted in 4 amino acids added to the N-terminus or ACO2. This construct maintained the reading frame for the Aco2 coding region. (-GU)Fth1-Aco2: The GU bulge on the 5′ side of the Fth1 flanking sequence was deleted. (C) ACO2 expression constructs encoded either wildtype protein or a mutant version in which all three Fe-S cluster ligating Cys had been mutated to Ser. Each construct included a c-myc epitope tag at the carboxyl terminus.

Fig. 3.

IRE-dependent control of ACO2^3C3S expression. (A) The expression of the apoprotein mutant ACO2^3C3S in transiently transfected HEK 293 cells that were untreated (control, C) or treated with 100 μM desferal (D) or 100 μM hemin (H) for 12 hrs. ACO2 level was determined by immunoblotting with an anti-C-myc monoclonal antibody (9E10), and the results were normalized as described in Methods. Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars) or the ACO2^3C3S protein under the control of ΔC11 Aco2 IRE (Aco2^ΔC11, black bars). These results are representative of n = 5 experiments. (B) The expression of the apoprotein mutant ACO2^3C3S in transiently transfected HEK 293 cells that were untreated (control, C) or treated with 100 μM desferal (D) or 100 μM of hemin (H) for 12 hrs. ACO2^3C3S level was determined by western blotting with an anti-C-myc monoclonal antibody (9E10), and the results were normalized as described in Methods. Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars) or the ACO2^3C3S protein under the control of alfalfa mosaic virus 5′ UTR (AMV, black bars). These results are representative of n = 4 experiments. Differences were determined using Student's T-test.

Fig. 5.

Impact of Sequences of the IRE on Regulation of ACO2^3C3S. The expression of the apoprotein mutant ACO2^3C3S in transiently transfected HEK 293 cells that were untreated (control, C) or treated with 100 μM desferal (D) or 100 μM hemin (H) for 12 hrs. ACO2^3C3S level was determined by western blotting as described in Methods. Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars), the ACO2^3C3S protein under the control of the AMV-Aco2 chimeric 5' UTR (AMV-Aco2, black bars) or the ACO2^3C3S protein under the control of the Aco2 IRE with Fth1 flanking sequences (Fth1-Aco2, striped bars). The predicted secondary structure of the Fth1-Aco2 5′ UTR is shown in Fig. 1F. These results represent n = 4 experiments. (B) Role of a conserved GU bulge in Fth1 flanking sequence: Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars) or the ACO2^3C3S protein under the control of the Aco2 IRE with Fth1 flanking sequences in which the two nucleotide GU bulge unique to human H-ferritin was deleted ((-GU) Fth1-Aco2, black bars). These results are representative of n = 4 experiments. Differences were determined using Student’s T-test.

IREs and associated base-paired flanking region confer differential regulation in cells

We next compared the ability of the Fth1 IRE and its base-paired flanking region to control ACO2^3C3S expression relative to Aco2^3C3S mRNA with the intact native Aco2 5′ UTR. We replaced the 27 nucleotide Aco2 5′ UTR with a 70-nucleotide segment of the Fth1 5′ UTR containing the IRE and flanking region that we fused with the AUG of Aco2^3C3S (Fig. 2B, Fth1). Expression of ACO2^3C3S controlled by the Fth1 sequences trended lower in untreated control (C) cells (P = 0.07) and was ∼40% lower in desferal-treated cells compared to the ACO2^3CS3 controlled by the native Aco2 5′ UTR (Fig. 4A). The ratio of expression in hemin compared to desferal was 2.1 for ACO2^3C3S controlled by the native Aco2 5′ UTR and 7 for ACO2^3C3S containing the Fth1 5′ UTR. Thus, the Fth1 IRE with base-paired flanking sequences confers a more robust response of ACO2^3C3S to changes in iron status than is the case for the native 27 nucleotide Aco2 5′ UTR.

Fig. 4.

Functional analysis of the Fth1 and Alas2 IRE with flanking sequence. The expression of the apoprotein mutant ACO2^3C3S in transiently transfected HEK 293 cells that were untreated (control, C) or treated with 100 μM of the iron chelator desferal (D) or 100 μM of the iron source hemin (H) for 12 hrs. ACO2^3C3S level was determined by western blotting as described in Methods. (A) Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars) or the ACO2^3C3S protein under the control of Fth1 IRE plus flanking sequences (Fth1, black bars). These results are representative of n = 4 experiments. (B) Comparison of the expression level of ACO2^3C3S controlled by the WT Aco2 IRE (Aco2, open bars) or the ACO2^3C3S protein under the control of the wildtype human erythroid 5- aminolevulinate synthase (Alas2) IRE (Alas2, black bars). These results are representative of n = 4 experiments. Differences were determined using Student's T-test.

Synthesis of Alas2, a critical regulator of erythroid heme synthesis, is controlled by IRP.^17,27,^51–54 In an in vitro model of erythroid differentiation, Alas2 mRNA was more weakly controlled in response to changes in cellular iron status than was Ftl mRNA.²⁷ When linked to the Aco2^3C3S ORF (Fig. 2B, Alas2), the Alas2 5′ UTR, the expression of ACO2^3C3S was 2.1-fold higher in hemin than in desferal, which was essentially identical to the 2.2-fold range of regulation observed with the native Aco2 5′ UTR (Fig. 4B).

Fth1 flanking sequences improve ACO2 regulation

Notable differences between the Aco2 and Fth1 IREs include the closer proximity to the mRNA 5′ end and a related lack of an extensive structured base-paired flanking region in the Aco2 5′ UTR (compare Fig. 1B with 1D and 1E).⁵⁵ Starting two nucleotides 5′ of the unpaired C (C11, as noted in Fig. 1C), the Aco2 IRE is 10 nucleotides or less from the 5′ end, while Fth1 or Ftl is generally between 31 and 39 nucleotides from the 5′ end (Fig. 1A).^24,56,57 In addition to stabilizing the IRE structure, the sequences flanking the ferritin IRE have been shown to improve IRP binding and translational regulation.^21,36,58 Two constructs were used to determine if adding flanking sequences could improve the iron regulation of ACO2^3C3S. First, the Aco2 5′UTR was lengthened by adding the first 18 nucleotides of AMV RNA 4 5′ UTR, producing a 45 nucleotide AMV–Aco2 5′UTR (Fig. 2B, AMV–Aco2). The AMV RNA 4 5′ UTR is unstructured, particularly in the U-rich 5′ end used here.^50,59 Second, using the 23 nucleotide core of the Aco2 IRE, we extended the helical stem by inserting 21 nucleotides (12 on the 5′ side and 9 on the 3′ side of the IRE) from the H-ferritin flanking region, producing the 41 nucleotide Fth1–Aco2 5′ UTR (Fig. 1F). We used somewhat less than the entire base-paired flanking region of Fth1 to create a stem-loop with similar thermodynamic stability (ΔG –20.5 kcal/mol) as the intact Fth1 IRE plus flanking region (ΔG –22.3 kcal/mol) (Figs. 1F and 2B, Fth1–Aco2). The native Aco2 IRE has a ΔG of –5.8 kcal/mol. Lengthening the Aco2 5′ UTR by adding the AMV sequence at the 5′ end reduced overall translatability under all conditions tested while leaving the hemin to desferal ratio at 2.6, which is similar to the native Aco2 5′ UTR ratio of 2.4 in this experiment (Fig. 5A). In contrast, the addition of the Fth1 flanking sequences resulted in reduced expression in control and desferal-treated cells but a similar level of expression in hemin relative to the native Aco2 5′ UTR. The hemin to desferal ratio increased relative to the native Aco2 5′ UTR from 2.4 to 5.9 for the Fth1–Aco2 5′ UTR construct.

An additional feature of the region flanking the human Fth IRE (Fig. 1E) is a GU bulge of eight nucleotides 5′ of the essential unpaired C that is not predicted in bullfrog (Fig. 1D), rat, mouse, chicken, or zebrafish Fth mRNA. To address the possible role of the GU bulge in IRP-dependent translational control, it was deleted from the human Fth1–Aco2 5′ UTR producing the (-GU)Fth1–Aco2 5′ UTR [Fig. 2B, (-GU)Fth1–Aco2]. Under control conditions, the (-GU)Fth1–Aco2^3C3S construct showed a level of expression that was not different from the Aco2^3C3S construct (Fig. 5B). This contrasted with the Fth1–Aco2^3C3S construct, which was expressed at a lower level than Aco2^3C3S construct under control conditions (Fig. 5A). In the presence of desferal, the (-GU)Fth1–Aco2^3C3S construct was more strongly repressed than was the Aco2^3C3S construct (Fig. 5B), as was the case for the Fth1–Aco2^3C3S construct (Fig. 5A). However, the ratio of expression in hemin versus desferal was 5 for (-GU)Fth1–Aco2^3C3S construct compared to 5.9 for the Fth1–Aco2^3C3S, suggesting that the GU bulge in human Fth1 IRE stem is a minor contributor to IRP-dependent regulation.

Discussion

Our analysis of the iron-dependent mechanisms controlling Aco2 expression supports several novel conclusions. First, Aco2 mRNA species exist in the liver with differing lengths of their 5′ UTR; the majority are too short to form a functional IRE. Second, Aco2 mRNA with the two longest 5′ UTRs is predicted to form a classical IRE structure but lacks UGC internal loop/bulge and the extended base-paired flanking sequences of the ferritin IRE. The ability to form the IRE structure was confirmed by nuclease mapping in vitro. Third, in transiently transfected HEK 293 cells, iron-dependent control of the synthesis of the ACO2 apoprotein, as modeled by Aco2^3C3S encoded by the full-length mRNA, is entirely dependent on the IRE. Fourth, substituting the H-ferritin IRE plus flanking sequences in place of the native Aco2 5′ UTR substantially improves iron regulation of ACO2 synthesis. Fifth, the addition of the base-paired flanking sequence from the H-ferritin IRE to the Aco2 IRE enhanced the response to iron deficiency but not iron overload. Sixth, ACO2 protein expression can be iron-regulated in a manner that is independent of the IRE and likely involves stabilization of the protein when the iron–sulfur cluster is present, as has been noted for other Fe–S proteins.^38,48,49 Our study demonstrates that ACO2 is regulated in a complex manner involving IRP-dependent and independent mechanisms of iron regulation.

Previous studies demonstrated that ACO2 synthesis and abundance depend on iron status over a more limited range than observed for ferritin.^24,26,60,61 Furthermore, studies of the Aco2 mRNA translation state in the liver showed strong translational activation by iron and significant, albeit not complete, repression in response to iron deficiency.^23,61,62 Given that ACO2 function is essential in animal cells,⁶³ near complete repression of the synthesis of this TCA cycle enzyme, as can be the case for the ferritins,²³ is predicted to be deleterious. Our current results suggest that the inability to completely repress ACO2 synthesis in iron deficiency in the liver appears to be determined by two mechanisms. First, most Aco2 mRNA species in adult rat liver lack the IRE. However, in contrast to the role of alternative splicing in producing IRE-deficient forms of Slc40a1 (ferroportin) and Slc11a2 (DMT1) mRNA,^64,65 the 5′ end heterogeneity of Aco2 mRNA appears to be due to differences in transcription start site as the sequence of these species are found in genomic DNA without the presence of introns. Alternations in the transcription start site are common, and associated differences in the 5′ UTR sequence can confer distinct differences in translation rate.^66–68 Second, 5′ end analysis and nuclease mapping demonstrate that while the longer species of Aco2 mRNA can form a canonical IRE structure, it lacks substantial base-paired flanking sequences as found in the ferritin IRE.^29,30 The addition of base-paired sequences flanking the IRE of Fth mRNA to Aco2 mRNA substantially improved iron regulation by enhancing translational repression in response to iron deficiency without impacting translation in response to iron excess. Earlier studies showed that conserved base-paired sequences flanking the ferritin IRE improved its function in a cell-free translation system.³⁶ Notably, IRE mutations giving rise to hyperferritinemia cataract syndrome include those in the base-paired flanking region of the human L-ferritin IRE.⁸ Furthermore, we formerly demonstrated a significant reduction in the RNA binding affinity of IRP1 for the Ftl, Slc40a1, and Alas2 IREs when the sequences flanking these IRE are deleted.²¹ Previous studies indicated that on binding IRP1, structural changes in the base-paired flanking region of the H-ferritin enhance translational repression.^30,36,69 While the mechanism of how base-paired flanking sequences improve IRP-dependent translational regulation remains to be fully elucidated, we conclude that the absence of this structural feature in the Aco2 5′ UTR weakens the function of the associated IRE.

Our nuclease mapping and chemical modification results show that Aco2 mRNAs with the 27 nucleotide 5′ UTR and presumably the longer 29 nucleotide species can form the canonical IRE secondary structure. This includes substantial accessibility of the AGU in the terminal loop CAGUGC sequence to single-strand specific nucleases and, in the case of the adenosine, modification by DEPC in the absence of IRP1 indicating it is not involved in base pairing. Although the first, fifth, and sixth nucleotides of the terminal loop are also sensitive to single-strand specific nucleases, the increased reactivity of G19 compared to G21 to RNases S1 and T1 as well as of the AGU sequence to S1 nuclease is reflective of the accessibility of these nucleotides to IRP1.³⁴ This likely reflects the formation of the RNA pseudo-triloop on base pairing of C17 and G21, which allows the AGU nucleotides to become accessible for multiple interactions with IRP1.³⁴,^70–72 The extensive accessibility of nucleotides on the 5′ and 3′ sides of the IRE stem to double-strand specific nuclease V1 is consistent with the formation of an RNA helix that allows display of the unpaired C five basepairs 5′ of the terminal loop and that makes multiple contacts with IRP1.³⁴ Although the Aco2 IRE forms a significant 12 base pair stem, the proximity of the IRE to the mRNA 5′ end prevents it from forming a more extensive RNA helix, as is the case for the 20 base pair helix observed with the Fth1 IRE.

Our work reveals that most Aco2 mRNA species in rat liver lack an IRE and are thus not subject to IRP action. Other extensively characterized examples of known mRNA isoforms with and without an IRE include those encoding DMT1 and ferroportin.^65,73 In each case, alternative promoters coupled with alternative splicing generate mRNA isoforms that contain or lack the IRE. The 3′ IRE controls the stability of Dmt1 mRNA postnatally in mice, while in adults, the evidence suggests it is a positive effector of Dmt1 mRNA translation.^64,73 In contrast, the ferroportin transcript lacking the IRE is believed to promote dietary iron absorption when duodenal mucosal cells are iron deficient. However, in erythroblasts, it is hypothesized to sensitize the cells to iron restriction.⁶⁵ For ACO2 regulation, it remains to be determined if the relative abundance of Aco2 mRNA species with or without an IRE varies in a tissue or cell-type specific manner or if their quantity can be regulated. Furthermore, our finding that some Aco2 mRNAs lacking the IRE have very short 5′ UTRs that may share properties of TISU elements raises the question of whether or not this provides a means to enhance the production of ACO2 in energy stress scenarios possibly including iron deficiency.^43,44

Conclusions

Hierarchical translation of 5′ IRE-containing mRNA allows IRPs to selectively dictate the activity of pathways necessary for controlling iron metabolism and other metabolic responses to iron deficiency. Differences in 5′ UTR length result in two classes of Aco2 mRNA, which contain or lack the IRE. Structural studies show that the Aco2 IRE fits the canonical structure for these regulatory elements but cannot form a larger three nucleotide internal loop containing the required unpaired C five base pairs 5′ of the terminal loop or extensive base-paired flanking region observed in ferritin IRE. Expression constructs for wildtype ACO2 responded to iron in IRP-dependent and independent manners. In contrast, the ACO2^3C3S mutant, which cannot incorporate a FeS cluster, exhibited complete dependence on the IRE for iron responsiveness. Replacement of the native IRE in Aco2^3C3S with the Fth1 IRE and its flanking sequences significantly enhanced iron-responsiveness, as did insertion of only Fth1 base-paired flanking sequences into the native Aco2 IRE. We conclude that the lack of base-paired flanking sequences and the previously established lower binding affinity of IRP1 to the Aco2 core IRE allow active translation of IRE-containing Aco2 mRNA in iron-replete cells and significant, but not complete, translational repression in iron deficiency. Aco2 mRNA lacking an IRE provides further protection from the likely lethal consequences of a total loss of expression of this TCA enzyme.

Conflict of interest

The authors declare no conflicts of interest.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.