Electron transfer flavoprotein and its role in mitochondrial energy metabolism in health and disease

Abstract

Electron transfer flavoprotein (ETF) is an enzyme with orthologs from bacteria to humans. Human ETF is nuclear encoded by two separate genes, ETFA and ETFB, respectively. After translation, the two subunits are imported to the mitochondrial matrix space and assemble into a heterodimer containing one FAD and one AMP as cofactors. ETF functions as a hub taking up electrons from at least 14 flavoenzymes, feeding them into the respiratory chain. This represents a major source of reducing power for the electron transport chain from fatty acid oxidation and amino acid degradation. Transfer of electrons from the donor enzymes to ETF occurs by direct transfer between the enzyme bound flavins, a process that is tightly regulated by the polypeptide chain and by protein: protein interactions. ETF, in turn relays electrons to the iron sulfur cluster of the inner membrane protein ETF: QO, from where they travel via the FAD in ETF:QO to ubiquinone, entering the respiratory chain at the level of complex III. ETF recognizes its dehydrogenase partners via a recognition loop that anchors the protein on its partner followed by dynamic movements of the ETF flavin domain that bring redox cofactors in close proximity, thus promoting electron transfer. Genetic mutations in the ETFA or ETFB genes cause the Mendelian disorder multiple acyl-CoA dehydrogenase deficiency (MADD; OMIM #231680). We here review the knowledge on human ETF and investigations of the effects of disease-associated missense mutations in this protein that have promoted the understanding of the essential role that ETF plays in cellular metabolism and human disease.

1. Introduction

Electron transfer flavoprotein (ETF) is an enzyme in the mitochondrial matrix space that mediates transfer of electrons from a series of mitochondrial flavoenzymes to the respiratory chain. It was first described in 1954, and named dye/cyt reducing factor (Crane and Beinert, 1954). In 1956 its name was changed to electron-transferring flavoprotein, acronym ETF (Crane and Beinert, 1956), that more precisely describes its activity. ETF is a dimeric enzyme composed of two different subunits, ETF-α and ETF-β, encoded by two different genes, ETFA and ETFB, respectively, and one molecule each of FAD and AMP. Two decades later, the Beinert lab again discovered the enzyme that links ETF to the respiratory chain: Electron transfer flavoprotein-ubiquinone oxidoreductase (ETF:QO; gene name ETFDH) (Ruzicka and Beinert, 1977). Sequential transfer of electrons from a series of dehydrogenation reactions via these enzymes thus channels electron flow to the respiratory chain for ATP production. At least 14 human enzymes transfer electrons to ETF posing the startling challenge to bind and specifically interact with all these enzymes. A personal record of the discovery story can be read in Helmut Beinert’s article in the Proceedings of the 1988 Fatty Acid Oxidation meeting at Philadelphia (Beinert, 1990).

In 1976 the first patient with a metabolic disorder characterized by urinary excretion of glutaric acid but also a series of other metabolites was described (Przyrembel et al., 1976) and in 1985 this biochemical phenotype was shown to be caused by deficient synthesis of the ETF or ETF:QO proteins (Frerman and Goodman, 1985a). Deficiency of ETF or ETF:QO results in metabolic blocks of the upstream dehydrogenation reactions, therefore the disorder is called multiple acyl-CoA dehydrogenase deficiency (MADD). With the elucidation of the cDNA sequences for ETFA and ETFB (Finocchiaro et al., 1988, 1993) the genetics for the subgroup of patients with mutations in these two genes were elucidated. We will in the following discuss the accumulated knowledge on the structure, function, integration in metabolic pathways, and genetic deficiency of human ETF, a key enzyme in mitochondrial energy metabolism.

2. Evolutionary origins

ETFs orthologues are found in all kingdoms of life, and belong to a family of α/β-heterodimeric FAD-containing proteins, which participate in electron transfer reactions in a variety of metabolic pathways (Finocchiaro et al., 1988; Tsai and Saier, 1995; O’Neill et al., 1998). Based on sequence homology and functional types, the members of this family can be clustered in three different groups, Group I which comprises mammalian and also a few bacterial ETFs, group II that contains proteins from nitrogen-fixing and diazotrophic bacteria and group III, which includes two putative proteins, YaaQ and YaaR, in Escherichia coli (Toogood et al., 2007). Group I ETFs are usually physiological electron acceptors of several dehydrogenases, and electrons flow to a membrane-bound ETF:ubiquinone oxidoreductase (ETF:QO). Besides mammalian ETF, ETF from the bacterium Paracoccus denitrificans has also been extensively studied (Husain and Steenkamp, 1985; Bedzyk et al., 1993; Roberts et al., 1999). In group II, the ETF-α and ETF-β are homologs of FixB and FixA proteins, respectively (Scott and Ludwig, 2004), and have the ability to donate electrons to dehydrogenases such as butyryl-CoA dehydrogenase or trimethylamine dehydrogenase, and can also act as electron acceptors from ferredoxin and NADH (Pace and Stankovich, 1987). It is postulated that the electron pathway to dinitrogen occurs via ETF:ferredoxin oxidoreductase, ferredoxin, nitrogenase reductase and nitrogenase (Scott and Ludwig, 2004). The best studied ETF of this group is the one from the bacterium Methylophilus methylotrophus strain W3A1. The crystal structure of W3A1 ETF has been reported, and although the overall fold is similar to group I ETFs, the FAD binding domain is rotated by about 40° relative to that of group I (Leys et al., 2003). Another well studied ETF is the one from the anaerobe Megasphaera elsdenii, which is atypical, as it contains two FAD-binding sites and no AMP cofactor (O’Neill et al., 1998; Sato et al., 2003). Altogether this diversity illustrates a relative flexibility of the ETF fold and provides clues into how this enzyme has evolved into a mitochondrial protein in eukaryotes. In the following we will focus on the human ETF protein and discuss knowledge from its homologs where it adds extra insights into ETF structure and function.

3. Pathways using ETF as electron transporter to the respiratory chain

Human ETF receives electrons from at least 14 FAD-linked dehydrogenation reactions and hands them over to electron transfer flavoprotein-ubiquinone oxidoreductase (ETF:QO), which in turn transfers them to Coenzyme Q₁₀ (CoQ10) in the inner mitochondrial membrane (Fig. 1; Table 1). This feeds the CoQ10 pool of the electron transport chain (ETC), similarly to complexes I and II.

Fig. 1.

ETF is a hub in mitochondrial redox metabolism. Amino acid catabolism, fatty acid oxidation, and choline metabolism are among the metabolic routes in mitochondria in which ETF intervenes as an electron acceptor, oxidizing multiple other flavoproteins (for acronyms see Table 1). The cartoon illustrates the central position of ETF in mitochondrial bioenergetics, as this matrix protein collects reducing power from multiple metabolic pathways, funneling electrons into the respiratory chain.

Table 1.

Enzymes transferring electrons to human ETF.

Acronym	Protein name	Gene name	UNIPROT	Metabolism/ function	Primary substrates
VLCAD	Very long-chain specific acyl-CoA dehydrogenase	ACADVL	P49748	Fatty acid oxidation	C12- to C24-CoA
LCAD	Long-chain specific acyl-CoA dehydrogenase	ACADL	P28330	Fatty acid oxidation	C6- to C24-CoA
MCAD	Medium-chain specific acyl-CoA dehydrogenase	ACADM	P11310	Fatty acid oxidation	C4- to C16-CoA
SCAD	Short-chain specific acyl-CoA dehydrogenase	ACADS	P16219	Fatty acid oxidation	C4- to C6-CoA
ACAD9	Complex I assembly factor ACAD9	ACAD9	Q9H845	Complex I assembly, fatty acid oxidation	long-chain unsaturated acyl-CoAs
ACAD10	Acyl-CoA dehydrogenase family member 10	ACAD10	Q6JQN1	n.d.	2 methyl-C15-CoA
ACAD11	Acyl-CoA dehydrogenase family member 11	ACAD11	Q709F0	n.d.	C22-CoA
SBCAD	Short/branched chain specific acyl-CoA dehydrogenase	ACADSB	P45954	Amino acid catabolism, fatty acid oxidation	2-methylbutyryl-CoA, isobutyryl-CoA, 2-methylhexanoyl-CoA, C4-CoA and C8-CoA
GCDH	Glutaryl-CoA dehydrogenase	GCDH	Q92947	Amino acid catabolism	glutaryl-CoA
IBDH	Isobutyryl-CoA dehydrogenase	ACAD8	Q9UKU7	Amino acid catabolism	2-methylbutanoyl-CoA
IVD	Isovaleryl-CoA dehydrogenase	IVD	P26440	Amino acid catabolism	isovaleryl-CoA and 3-methylbutanoyl-CoA
SARDH	Sarcosine dehydrogenase	SARDH	Q9UL12	Choline metabolism	sarcosine
DMGDH	Dimethylglycine dehydrogenase	DMGDH	Q9UI17	Choline metabolism	N,N-dimethylglycine
D2HGDH	D-2-hydroxyglutarate dehydrogenase	D2HGDH	Q8N465	n.d.	D-2-hydroxyglutarate

n.d. - not determined.

One group of enzymes transferring electrons to ETF are acyl-CoA dehydrogenases (ACDHs) involved in β-oxidation of straight-chain fatty acids (VLCAD, LCAD, MCAD, and SCAD) and three other ACDHs (ACAD9, ACAD10, and ACAD11) that dehydrogenate acyl-chain substrates (see detailed gene names and function in Table 1). ACAD9 is essential for assembly of complex I and its dehydrogenation activity towards long-chain acyl-CoA’s appears to be associated with its assembly function (Nouws et al., 2010). ACAD10 and ACAD11 have been shown to dehydrogenate certain particularly long-chain and branched-chain acyl-CoA’s (Ensenauer et al., 2005; He et al., 2011). A second group consists of ACDHs handling intermediates in the degradation of amino acids (IBDH, SBCAD, IVD, GCDH, details in Table 1). The proteins of both groups are orthologues that have originated from a common ancestor (Swigonova et al., 2009). The enzymes involved in fatty acid and amino acid degradation are homotetramers or homodimers with one FAD molecule bound per subunit.

A third group is composed of dehydrogenases involved in choline metabolism (SARDH, and DMGDH, details in Table 1), that are evolutionary orthologues but share no significant sequence similarity with the ACDHs. Recently, D-2-hydroxyglutarate dehydrogenase (D2HGDH), mutations in which are associated with the neurometabolic disorder D-2-hydroxyglutaric aciduria, has been shown to use ETF as electron acceptor (Toplak et al., 2019). This protein has no clear evolutionary relationship to either of the three groups. There are possibly additional human enzymes that transfer electrons to ETF, however, no published evidence is available.

Taken together, this means that ETF must recognize, bind, and interact with multiple evolutionarily distinct groups of enzymes. The involvement in a multiplicity of metabolic processes in the mitochondrial matrix indicates that ETF and ETF:QO represent a regulatory hub that controls the electron flow to the respiratory chain from a series of flavoenzymes. As further discussed below, inappropriate expression of these enzymes or genetic deficiencies in their encoding genes will impair upstream metabolic pathways, resulting in the accumulation of intermediate metabolites, and in a decrease in overall ATP production.

4. Gene structure and biosynthesis

The human ETF subunits are nuclear encoded. The α-subunit is synthesized as a precursor protein of 35 kDa, while the β-subunit is synthesized as a 27 kDa protein in the cytosol in a form that is indistinguishable from the mitochondrial form (Ikeda et al., 1986). After import into mitochondria the α-subunit precursor sequence is cleaved off yielding a mature form with 32 kDa (Ikeda et al., 1986). The genes for α and β subunits mapped to chromosomes 15q23-q25 and 19q13.3, respectively (Barton et al., 1987; Antonacci et al., 1994).

ETFA is transcribed as a 96117 bp pre-mRNA, which is spliced into a 2289 bp mRNA consisting of 12 protein encoding exons. The start codon is located in exon 1 (NM_000126.4). ETFB is transcribed as a 21220 bp pre-mRNA, which is processed into a 872 bp mRNA and 6 exons with the start codon being localized in exon 1. These are the major ETFA (NM_000126.4) and ETFB (NM_001985.3) transcripts (Olsen et al., 2003) (Table 2; see also Section 9). Alternative transcript variants of both genes have been identified. The ETFA transcript variant (NM_001127716.1) lacks an in-frame exon 2 (49 amino acids). The encoded protein lacks 12 of the 25 amino acids making up the mitochondrial targeting sequence of the precursor protein (Olsen et al., 2003). The ETFB transcript variant (NM_001014763.1) uses a downstream transcript start site in intron 1. The encoded protein has a longer and distinct N-terminus compared to the major ETFB transcript variant. The biological functions of the two transcript variants are not known.

Table 2.

Characteristics of the human genes and transcripts encoding ETF subunits^a.

Gene	Chro.	CCDS code	RefSeq Trascript ID	Ensemble Transcript ID	Pre-spliced (nt)	Post-spliced (nt)	Exons	Coding exons	Uniprot name	Length (aa)	Mass (kDa)
ETFA	15	32299.1	NM_000126.4	ENST00000557943.6	96117	2289	12	12	P13804	333	35
	15	45311.1	NM_001127716.1	ENST00000433983.6	96117	1346	11	11	P13804	284	30
ETFB	19	12828.1	NM_001985.3	ENST00000309244.8	21220	872	6	6	P38117	255	28
	19	33085.1	NM_001014763.1	ENST00000354232.8	21220	3551	5	5	P38117	346	37

^aBased on the NCBI, Ensemble and Uniprot databases.

5. Biochemical and structural properties

In 1975 Hall and Kamin reported the purification, biochemical and spectroscopic properties of ETF and several dehydrogenases from pig liver mitochondria (Hall and Kamin, 1975). ETF was firstly described as containing two subunits and two FAD molecules, with a global molecular weight of 58 kDa (Hall and Kamin, 1975). However, seven years later Gorelick et al showed, using ETF purified from pig kidney, that it binds only one FAD per dimer (Gorelick et al., 1982). A subsequent report on the purification of pig liver ETF corroborated this finding, establishing that in fact mammalian ETF has only one FAD per dimer (Husain and Steenkamp, 1983). Later, in 1993, it was uncovered that ETF has an additional AMP-binding site (Sato et al., 1993). The AMP cofactor, has no influence on the activity of the protein but seems to be important for the assembly of the holo structure (Sato et al., 1996).

In 1996, Roberts and colleagues solved the crystal structure of human ETF to 2.1 Å resolution, revealing that the protein has three distinct domains: domain I is composed of the N-terminal portion of ETF-α; domain II consists of C-terminal portion of ETF-α and a small C-terminal portion of ETF-β, and; domain III is made up mostly from ETF-β (Fig. 2, inset) (Roberts et al., 1996). Analysis of the human ETF structure reveals that the flavin cofactor is located adjacent to the Rossmann fold (Hanukoglu, 2015) in domain II, and it occupies a cleft between the two subunits, interacting mainly with the C-terminal portion of the α-subunit. Moreover, the C7 and C8 of the methyl groups in the dimethylbenzene ring make van der Waals contact with Tyr16 and Phe41 from the ETF-β (Fig. 2, bottom right). The AMP cofactor is buried deeply within domain III making mostly backbone interactions. The phosphate moiety of the cofactor forms hydrogen bonds with residues Ala126, Asp129, Asn132, Gln133 and Thr134 from the β-subunit (Fig. 2, top right). This region is moderately conserved in all ETF proteins, and it has been suggested that the AMP binding site in human ETF could be a remnant of a NADH binding site, similar to the one present in the physiological electron donor for ETF from M. elsdenni (Roberts et al., 1996). Later, Toogood and colleagues suggested that this could also be a remnant of a second binding site for FAD that would receive electrons from the first FAD site as in the ETF:MCAD complex structure (see below) and because some ETF proteins have two FAD cofactors (Toogood et al., 2005).

Fig. 2.

Human ETF structure and cofactor contacts. The central panel represents the ETF structure color-coded by subunit (ETF-α, green; ETF-β, blue) depicting the AMP and FAD cofactors, whose contacts are detailed in ligplots (right panels). The inset (top left) depicts a cartoon of ETF color-coded by its composing structural domains. Figures prepared using Pymol and LigPlot+ (Wallace et al., 1995) using the human ETF crystal structure (PDB code: 1efv). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

The flavin cofactor in oxidized ETF proteins has a particular fingerprint in the absorption spectra with the oxidized form presenting maxima at 373 nm and 436 nm, with a shoulder around 470 nm. Reduction of ETF by dehydrogenases results in a rapid formation of an anionic semiquinone, with a characteristic spectrum of an anionic red semiquinone. Its further reduction occurs slower (Hall and Lambeth, 1980; Beckmann et al., 1981; Steenkamp and Husain, 1982). Reoxidation of ETF is catalyzed by the membrane-bound iron-sulfur flavoprotein ETF: QO (Beckmann and Frerman, 1985). This partner enzyme is localized in the inner mitochondrial membrane in the matrix side, and it has two redox-active cofactors, one [4Fe-4S] cluster and one FAD (Alves et al., 2012). It has been shown that the iron-sulfur cluster is the electron acceptor from ETF, and the interaction between the two enzymes is made through ETF-β (Swanson et al., 2008).

Another important characteristic is the ETF greenish fluorescence, with an emission peak around 490 nm. Depending on the ETF source, flavin intensity could be at least 3.5 times higher than that of free FAD. Most flavoproteins, such as the ACDHs, usually present flavin emission less than 1% of the corresponding free FAD (Crane and Beinert, 1956; Hall and Kamin, 1975; Herrick et al., 1994). The FAD from porcine ETF has a particularly high fluorescence, which is quenched upon reduction. Taking advantage of this characteristic, a fluorometric assay for ACDHs activity has been set up and used over the years (Frerman and Goodman, 1985b; Gorelick et al., 1985). Recently Zhang and colleagues reported an improved dehydrogenase microplate activity assay using the fluorescence of recombinant porcine ETF, which facilitates its application for basic research and also clinical diagnostics (Zhang et al., 2019). As in all redox assays, it is however critical to correctly consider non-enzymatic reduction events, which can yield false positives in this type of assays. Interestingly, the oxidation of the FAD cofactor in human ETF to its 8-formyl-derivative has been recently described (Augustin et al., 2018). This modification is irreversible and pH dependent, involving several conserved amino acid residues within the cofactor binding site resulting in a highly stable flavin semiquinone (see below). ETF with 8-formyl-FAD presented an enhanced affinity to dimethylglycine dehydrogenase, suggesting that it could be an important modulator of ETF interaction with its redox partners (Augustin et al., 2018).

6. Functional interactions – structure of the ETF:MCAD complex

To better understand the interaction between human ETF and the dehydrogenases the crystal structure of ETF in complex with MCAD was solved (Toogood et al., 2004). The crystal structure revealed a recognition loop in the β-subunit of ETF that acts as an anchor. It is centered on the conserved residue Leu195, and docked to a hydrophobic patch on the dehydrogenase, and a highly dynamic region on the ETF FAD domain, allowing fast interprotein electron transfer (Toogood et al., 2004) (Fig. 3). Further studies led the authors to present a dynamic multistate model for the ETF:MCAD complex that involves random motion between three distinct positions for the flavin cofactor domain (Toogood et al., 2005).

Fig. 3.

Cartoon depicting the interaction between human ETF and MCAD. The ETF structure is color-coded by subunit (ETF-α, green; ETF-β, blue) highlighting the recognition loop in domain III of the β-subunit (red) and two MCAD subunits from the MCAD tetramer. While one of the MCAD subunits (light grey) interacts with the recognition loop in ETF, another contiguous subunit (dark grey) is brought to the vicinity of the ETF flavin and electron transfer is further facilitated by a swinging motion and conformational adjustments within domain II. These conformational adjustments not only approach the FAD from both proteins to proximity but also promote changes in the cofactor coordination sphere that tune the FAD redox potential of ETF favouring two-electron transfer, and mitigating the formation of stable one-electron semiquinone species (see text for details). The figure is made using Pymol and based on the MCAD:ETFβ-Glu165Ala complex crystal structure (PDB code: 2A1T (Toogood et al., 2005)). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Interestingly, from different crystal structures of ETF:Dehydrogenase partner complexes it was observed that ETF presents always a recognition loop that interacts with sequence-wise distinct, but structurally equivalent hydrophobic patches (Toogood et al., 2007). Also, in the different crystal structures the flavin domain in the complex has low electron density map, indicating that the flavin moiety is highly dynamic and capable of sampling different positions. After recognition of the hydrophobic patch, the flavin domain is promiscuous in searching for transfer-competent states that place the redox cofactors in the most adequate position for the different substrates (Toogood et al., 2007). These characteristics associated with the protein:protein interactions are critical to understand why ETF can serve as electron acceptor for diverse dehydrogenases.

Furthermore, to improve electron flow within the different pathways it has been suggested that ETF is present in some specific functional assemblies, composed of dehydrogenases from mitochondrial fatty acid β-oxidation, ETF:QO, CoQ10 and complex III (Parker and Engel, 2000). This result comes in line with the previous suggestion that some form of functional organization between dehydrogenases and the inner mitochondrial membrane enzymes must occur, as it has been observed that intact mitochondria oxidize fatty acids faster than disrupted mitochondria (Kispal et al., 1986). Analysis of extracts from isolated rat liver mitochondria using blue native polyacrylamide gel electrophoresis suggested the presence of a multifunctional fatty acid β oxidation complex, that congregates with respiratory chain enzymes (Wang et al., 2010). Also, a recent report provides further evidence for the formation of a respiratory super-complex composed of multiple ACDHs, ETF, ETF: QO, and trifunctional protein (Wang et al., 2019).

7. Electron transfer properties and redox tuning by interactions

ETF is a two-electron and two-proton transporter as its FAD undergoes successive reduction via two-consecutive one-electron transfer steps, with the formation of an intermediate one-electron red flavin semiquinone species (FAD^•−), which is then fully reduced to FADH₂ with the uptake of one additional electron and two protons (Fig. 4a). Protonation of the red (anionic) semiquinone form, usually at pKa > 8, yields a blue (neutral) semiquinone. The redox potentials of the FAD cofactor in ETF have been determined (Salazar et al., 1997) in anaerobic potentiometric titrations followed by visible spectroscopy. This takes advantage of the spectral difference between oxidized and fully reduced FAD: while the former exhibits the absorption maxima at 373 nm and 436 nm, the latter is colorless (Salazar et al., 1997; Rodrigues and Gomes, 2012). Therefore, the relative fraction of each of these species (FAD, FAD^•− and FADH₂) can be accurately determined from multi component analysis of the spectra taken at each titration point, as a function of a poised redox potential, allowing the determination of the oxidation reduction potential of each step (E₁ and E₂) from fitting to the Nernst equation (Rodrigues and Gomes, 2012). A titration of purified recombinant human ETF at pH 8.0 yields redox potentials of −75 mV and +15 mV for the oxidized/semiquinone and semiquinone/hydroquinone couple, respectively (Fig. 4b). In a protein, the FAD redox potentials are determined by the nature of the side-chain interactions from nearby and FAD bonding residues, which thus regulate the electron transfer process and the possibility to stabilize one-electron species (Entsch and Ballou, 2013).

Fig. 4.

ETF electron transfer properties and redox tuning by protein interactions. Reaction scheme (a) depicting the different FAD species formed upon reduction (FAD, oxidized; FAD^●-, semiquinone; FADH₂, reduced). Titration curves of ETF in the absence (b) and in the presence of MCAD (c) in 100 mM Tris-HCl pH 8.0. The represented data points were recorded at wavelengths where fully oxidized flavin (filled circles) and semiquinone species (open circles) have major contributions, 437 nm and 478 nm, respectively. Solid lines correspond to the best fit of Nernst equations for two consecutive one-electron reductions steps (FAD → FAD^•− → FADH₂) (Rodrigues and Gomes, 2012). Diagram representing the reduction potentials of FAD from ETF and in a complex with MCAD, also for the ETFβ-Arg191Cys mutant (d) and representation of the amount of semiquinone radical (FAD^•−) formed for the different conditions depicted in quantitative plots and qualitatively by the yellow glow behind cartoon structures (e). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

The determination of the crystal structure of the ETF:MCAD complex, and the substantial motions observed, suggested that such dynamic conformational changes might alter the FAD redox potential due to rearrangements of its bonding network residues. This would imply that the efficiency of electron transfer between MCAD and ETF might be dictated by the interaction between the two proteins. A study by Rodrigues et al showed that the redox potential of ETF, when determined in the presence of its partner MCAD, is indeed changed: the redox potential of the oxidized/semiquinone pair was substantially raised from −75 to –30 mV, while that of the semiquinone/hydroquinone couple was lowered from +15 to −5 mV (Fig. 4c and d) (Rodrigues and Gomes, 2012). This result showed that the two-electron transfer process was substantially favored upon formation of the MCAD:ETF complex. To demonstrate that this effect was a consequence of the protein:protein interaction, the same experiments were performed under high salt conditions to perturb electrostatic interactions, and using the ETFβ-p. Arg191Cys variant, which comprises a mutation in the recognition loop (see Fig. 3). Indeed, perturbing the interaction surface between the two proteins diminishes the redox tuning of the FAD cofactor in ETF, and the redox potentials determined for the mutant were in fact closer to those measured for ETF alone (Fig. 4d and e). Interestingly, ETFβ-p.Arg191-Cys is a disease variant (see following section) and this effect explains why it generates functional deficiency in MADD patients.

This effect of redox tuning of FAD in ETF by protein interactions also provides a conceptual framework to rationalize the production of reactive oxygen species (ROS). It is well established that mitochondrial enzymes are a source of ROS generation, and ETF has been shown to be one of the proteins involved in the deleterious generation of these species. Seifert et al. in a cellular study of ROS production during long-chain fatty acid oxidation identified ETF as a likely site for ROS production (Seifert et al., 2010). Subsequently, thorough mechanistic explanation was provided by Rodrigues and Gomes, 2012), who, using an enhanced superoxide and hydrogen peroxide detection assay, demonstrated that during turnover conditions ETF generates both superoxide and hydrogen peroxide in the presence of octanoyl-CoA and catalytic amounts of MCAD (Rodrigues and Gomes, 2012). Indeed, the basis for this effect lies in thermodynamics: in the absence of its substrate (MCAD) the split in partial redox potentials leads to the formation of 74% of one-electron semiquinone (FAD^•−), whose high reactivity with O₂ promotes the formation of toxic ROS. Indeed, molecular dynamics studies elicited several O₂ binding sites near the ETF flavin cofactor (Husen et al., 2019). Formation of the complex with MCAD decreases semiquinone levels down to 45%, and a disease mutation impairing complex formation raises this figure up to 59%, which suggests a potential origin for ROS increase in MADD (discussed below). These results showed that, in physiological conditions, a decrease in electron transfer efficiency between ETF and its partner reductants is an important and frequently overlooked site of ROS generation in mitochondrial bioenergetics. As higher ratios of MCAD:ETF decrease ROS production by ETF, it can be suggested that in vivo this effect might be minimized by the formation of super-complexes comprising ‘docked’ ETF to its partners and/or by higher expression levels of its different electron donors, thus maximize encounters of ETF with its substrate enzymes.

8. Regulation and modifications

Inspection of human transcriptome and proteome expression databases like ProteomicsDB (https://www.proteomicsdb.org; (Schmidt et al., 2018)) indicates that ETFA and ETFB are stably expressed at relatively high levels in most tissues, both at transcript and protein level. The highest expression levels are observed in liver, heart, kidney, skeletal muscle and fat tissue. As the two genes are localized on different chromosomes, expression at ‘housekeeping level’ is presumably maintained by general basic transcription factors. It is well known that expression of fatty acid oxidation enzymes is regulated by peroxisome proliferator-activated receptors (PPARs) (Djouadi and Bastin, 2008; Bastin, 2014). However, to our knowledge, no literature about transcription factors regulating the expression of ETFA and ETFB has been published yet. PPARgene (http://www.ppargene.org/index.php) is a database, which allows prediction of novel PPAR target genes by integrating in silico PPAR-responsive element (PPRE) analysis with high throughput gene expression data (Fang et al., 2016). According to PPARgene, ETFA and ETFB are predicted as PPAR target genes with high confidence. The predictions are based on mouse data, however, since MADD patient cells increase fatty acid oxidation flux upon treatment with the PPAR agonist bezafibrate (see Section 9), a PPAR mediated transcriptional regulation of the ETF genes most likely also takes place in humans.

According to ProteomicsDB, ETFDH shows a similar protein expression pattern as ETFA and ETFB. For the electron donor dehydrogenases, expression of those involved in fatty acid oxidation and amino acid catabolism is highest in most tissues thus likely accounting for the major part of the electron flow to ETF. One exception is LCAD, which like the dehydrogenases involved in choline metabolism and ACAD10 and ACAD11 with more special acyl-CoA substrates, is expressed at significant levels in few human tissues only. This is in contrast to mice, where LCAD plays a more general role and functionally overlaps with murine VLCAD (Cox et al., 2001; Chegary et al., 2009; Maher et al., 2010; Bastin, 2014).

Given the stable expression of ETF, regulation of its activity in electron transfer to the respiratory chain might be accomplished by posttranslational modifications or by interactions with regulatory proteins and compounds. Several posttranslational modifications of both ETF-α and ETF-β subunits have been reported. This comprises carbonylation associated with oxidative stress (Li et al., 2007; Meany et al., 2007), acetylation (Schwer et al., 2009; Zhao et al., 2010), glutarylation (Tan et al., 2014) or succinylation (Rardin et al., 2013) of lysine residues, phosphorylation of threonine and serine residues (Hopper et al., 2006; Bian et al., 2014), and trimethylation of two lysines in beta ETF. While the latter has been shown to affect ETF enzyme activity, the former are observations from large scale proteome studies, mostly of the murine orthologs, that have so-far not been experimentally validated. Trimethylation of lysines at position 200 and 203 of ETF-β is mediated by a specific methylase (Electron transfer flavoprotein β subunit lysine methyltransferase; gene name ETFBKMT) (Rhein et al., 2014; Małecki et al., 2015). The methylated residues are near the loop that mediates recognition and binding to electron donor dehydrogenases. In vitro methylation studies (Małecki et al., 2015) indicated that trimethylation reduced the transfer of electrons from ACDHs. Studies of ETFKMT knock-out mice also supported an inhibitory effect (Shimazu et al., 2018). In conflict with these findings, treatment of cells with a siRNA directed against the methylase resulted in slightly reduced oxygen consumption rates when palmitate was used as a substrate (Rhein et al., 2014). The decreased methylase levels might influence a series of processes and thus the indirect measurement of ETF activity via overall oxygen consumption rates must be interpreted cautiously. Clearly, further studies are required to fully elucidate this matter.

In a proteomics study targeting mitochondrial proteins with unknown function Floyd et al discovered LYRM5 (now called electron transfer flavoprotein regulatory factor 1 (ETFRF1)) as an ETF interactor (Floyd et al., 2016). Further analysis indicated that the interaction may lead to ‘deflavinylation’ of ETF, and thus inactivate it. The exact mechanisms and relevance remain to be determined.

9. Genetic deficiencies and molecular pathogenesis

Genetic deficiencies of ETF cause multiple acyl-CoA dehydrogenase deficiency (MADD, OMIM #231680). The disease was first described by Przyrembel and co-workers in 1976 (Przyrembel et al., 1976). They described an infant with neonatal acidosis, hypoglycemia and strong “sweaty-feet odor”, and demonstrated organic aciduria with markedly elevated levels of glutaric acids and several other organic acids. They named this condition glutaric aciduria type II. Since then, it has become clear that the disorder is due to a primary defect in either ETF or ETF: QO, which result in functional impairment of all ETF-linked flavoprotein dehydrogenases, many of which are acyl-CoA dehydrogenases. Multiple acyl-CoA dehydrogenase deficiency or MADD was therefore suggested as a more appropriate name for the disorders (Goodman et al., 1980; Gregersen et al., 1980). In this section, we will summarize deficiencies in the ETF genes and discuss their molecular pathogenesis.

By now, 54 ETF disease-associated mutations have been deposited in the Human Gene Mutation Database (HGMD) (Supplementary Table S1). Most of the mutations have been identified in MADD patients, however a single genomic gross deletion of approximately 420 kb, which encompasses ETFA, has been associated with microcephaly and intellectual disability (Ibn-Salem et al., 2014), and a de novo ETFB nonsense mutation has been associated with Autism Spectrum Disorder (Iossifov et al., 2014). As shown in Fig. 5, the reported ETF mutations are distributed throughout the ETFA and ETFB genes. The mutations include splice mutations, nonsense mutations, small deletions/insertions, and gross deletions. Missense mutations, where the mutation leads to a single amino acid change, are the most common type of mutations, making up approximately 60% of the reported disease-associated ETF mutations.

Fig. 5.

Disease-associated ETF mutations. ETFA (A) and ETFB (B) gene structures with localization of disease-associated mutations. (C) Distribution of ETFA and ETFB mutations according to functional class with missense mutations in orange (single amino acid substitutions/deletions/insertions) and loss-of-function (LOF) variants in black (splice site, nonsense, out-of-frame and gross deletion mutations). All mutations are from the Human Gene Mutation Database, Professional (HGMD, 27 June 2020). See Supplementary Table S1 for further details. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Although a significant number of missense mutations have been identified in ETF associated with MADD their characterization at the structural, conformational and even functional level is scarce. In 2010 we reported a study where we have compiled a list of missense mutations described for the ETF protein and analyzed each mutation at the structural level aiming to find a correlation between genotype, phenotype and protein structural behavior (Henriques et al., 2010). The use of in silico tools for mutagenesis predictions and structural analysis permitted us to identify structural hotspots within the ETF fold. ETF missense mutations fall essentially in two groups: one in which mutations affect protein folding and assembly; and another group in which mutations impair catalytic activity and disrupt interactions with partner dehydrogenases. In the ETF α-subunit two affected regions were observed: one region that involves essential interactions with ETF-β and another conserved region rich in FAD and MCAD interacting residues. In the case of the ETF β-subunit, it was suggested that there are three hotspot regions: a region rich in cofactor contacts, a region that affects subunit interactions and another involved in the “recognition loop” implicated in interactions with ACDHs. Also, it was suggested that amino acid alterations that lead to drastic changes in the chemical properties of the residues involved in catalytic regions or regions important for intra and inter molecular interactions have a higher probability to result in severe phenotypes (Fig. 6).

Fig. 6.

Structural mapping of disease variants in ETF. (a) Cartoon of the human ETF structure (PDB: 1efv) highlighting the missense mutations listed in Supplementary Table S1. The two figures have been rotated 180° in respect to each other along the y-axis to better show variant position. (b) Structural details around three variants that have been well described (see details in text). The cartoons represent a magnification of a specific region with an identified mutated residue overlaying with the amino acid substitution (adapted from (Henriques et al., 2010)). Figures prepared using Pymol. via its membrane-linked partner ETF-ubiquinone oxidoreductase (ETF:QO).

Furthermore, some ETF variants have been expressed, purified, and characterized to help to elucidate mutation impact on disease phenotype. Salazar et al. reported a study on two disease variants associated with severe phenotypes, the ETFα-Thr266Met, the most frequent mutation found in patients, and ETFα-Gly116Arg. Over-expression of the ETFα-Gly116 mutant in E. coli was shown to be dependent on the coexpression of GroEL and GroES chaperonins, yet the enzyme was catalytically inactive in crude extracts and could not be further purified (Salazar et al., 1997). The ETFα-Thr266Met mutant was shown to have an overall structure similar to the wild type protein, yet the flavin environment is altered by the mutation (Salazar et al., 1997), leading to an enzyme with strongly impaired activity, suggesting that this could be the cause of the severe phenotype. Another severe mutation, ETFβ-Cys42Arg, that directly affects the AMP binding site and intersubunit contacts (Fig. 6b), impairing correct protein folding, could only be moderately expressed in the presence of GroEL/GroES or DnaK/DnaJ/GrpE. However, the protein had no activity probably due to lack of cofactor insertion (Henriques et al., 2010).

The ETFβ-Asp128Asn variant associated with a mild phenotype has also been broadly studied (Olsen et al., 2003; Henriques et al., 2009, 2010). In the first report, describing the heterologous expression, it was shown that the residual activity of the mutant enzyme could be rescued up to 59% of that of wild-type when the transformed E. coli cells were grown at low temperature (30 °C) rather than at 37 °C (Olsen et al., 2003). This was an important indication that environmental factors such as cellular temperature could play a role influencing disease progression. Indeed, in the reported patient disease symptoms were precipitated in connection with a virus infection and fever event. Later this variant was described to have decreased thermal stability and significantly lower specific enzymatic activity. Although the mutation does not impact the overall fold of the protein it likely perturbs FAD interactions (Fig. 6b) (Henriques et al., 2009, 2010). Interestingly, the authors showed that flavinylation could prevent conformational destabilization and activity loss at higher temperatures (Henriques et al., 2009). This effect observed in presence of increased FAD content, that in patients occurs as a consequence of riboflavin supplementation intake (see next section), has been shown for other flavoproteins including for ETF:QO and ACDHs (Olsen et al., 2007; Henriques et al., 2016; Lucas et al., 2020).

Another missense mutation associated with milder forms of MADD is ETFβ-Arg191Cys. For this variant, it was shown that the overall αβα fold topology of native ETF remains unchanged yet there is a substantial decrease in enzymatic activity (Henriques et al., 2010). The authors suggested that the loss of activity could result from impaired interaction with the electron-donor dehydrogenases since the mutation site is at the interface of the complex close to the “recognition loop” (Fig. 6b) (Henriques et al., 2010, 2019).

Four polymorphic variants of ETF have been described (ETFα−171Ile/Thr and ETFβ−154Met/Thr) (Freneaux et al., 1992; Colombo et al., 1994), and it has been proposed that such polymorphisms could constitute susceptibility factors for the disease development in a cohort of VLCAD patients, in which the less stable polymorphic variant was overrepresented (Bross et al., 1999). The ETFα-Thr171has been described as displaying decreased thermal stability in comparison to the others. Later it was shown that this polymorphism is prone to faster FAD release, and that it exhibits increased conformational dynamics during thermal stress (Henriques et al., 2011). Moreover, it was shown that the GroEL chaperonin could rescue ETFα-Thr171 polymorphic variant. These studies are very important to outline how polymorphic variants, which are apparently innocuous, may significantly influence disease progression and severity via indirect effects.

10. Clinical phenotype and therapies in ETF pathology

During fasting, the oxidation of fatty acids and branched-chain amino acids is essential for ATP production, and also provides metabolic intermediates to sustain hepatic ketone body production and gluconeogenesis (Houten et al., 2016). Thus, one of the consequences of ETF deficiency is energy depletion and impairment of glucose homeostasis and ketogenesis. A second consequence are the toxic effects of the substrates and metabolites that accumulate in different organs proximal to the metabolic block, with lipid storage and toxification playing a key role (Frerman and Goodman, 2001).

Based on phenotype, patients with ETF or ETF:QO deficiency can mainly be divided into one of three groups: a neonatal lethal form with congenital anomalies (type I), a neonatal lethal form without congenital anomalies (type II), and a variant form with milder signs and lower mortality rate (type III). The first two groups are classified as severe MADD (MADD-S) and the third as mild MADD (MADD-M). MADD-S disease is usually lethal. It presents during the first few days of life with severe metabolic decompensations, including hypoketotic hypoglycemia, metabolic acidosis, hypotonia and multi-organ failure. MADD-M disease is heterogeneous in terms of symptomatology and severity and most patients survive with appropriate treatment. MADD-M disease may present from childhood to late adulthood and is often triggered by metabolic stress. Episodic vomiting with neuromuscular symptoms with or without hepatomegaly are common features (Frerman and Goodman, 2001).

Many years of research in the molecular causes of MADD suggest that the clinical phenotype to some extent is determined by the severity of the metabolic block at the ETF/ETF:QO site. Already in the eighties, studies of ETF and ETF:QO enzyme activities revealed that the severity of the clinical phenotype does not depend on which enzyme is affected, but rather on the residual enzyme activity, with MADD-M tissue showing a high degree of residual activity, and MADD-S tissue showing no or very low residual activity (Amendt and Rhead, 1986; Loehr et al., 1990). This was confirmed later, when the first gene mutations were identified (Indo et al., 1991; Freneaux et al., 1992; Colombo et al., 1994; Beard et al., 1995), and some relationship between genotype and phenotype was documented (Goodman et al., 2002; Olsen et al., 2003; Schiff et al., 2006). Biallelic loss-of-function (LOF) mutations, which cause no or very low residual enzyme activity, often associate with MADD-S, whereas MADD-M patients carry at least one missense mutation allowing some residual activity to be produced. Within this latter group of patients, environmental or physiological factors, such as diet, riboflavin intake and febrile infections may modulate the expression of the variant genotypes, and as such the clinical phenotype, as discussed above.

In these mild cases therapeutic management comprises a diet low in fat and protein and the avoidance of fasting. Riboflavin supplementation is also a very efficient treatment. The clinical and biochemical efficiency of riboflavin was documented already back in 1982 in two brothers suffering from MADD (Gregersen et al., 1982). Riboflavin supplementation increases the cellular supply of the FAD cofactor, which can mitigate the functional consequence of certain missense mutations that impair flavoprotein folding and/or stability properties, as discussed in detail for ETF variants above. In fact, the majority of MADD-M patients with missense mutations respond well to high doses of riboflavin (Grunert, 2014). Since riboflavin at high doses is well tolerated, early treatment can prevent progression of symptoms and even be lifesaving (Mosegaard et al., 2020). Thus riboflavin supplementation should be initiated as soon as MADD is suspected. Another relevant therapeutic drug is Bezafibrate that is an agonist of the peroxisome proliferator-activated receptors (PPARs), which coordinates transcriptional expression of mitochondrial enzymes, including those involved in fatty acid oxidation. Bezafibrate has been investigated as a promising pharmacological drug to boost the expression of milder missense or stability variants in fatty acid oxidation disorders (Djouadi et al., 2005; Djouadi and Bastin, 2008; Bastin, 2014). Ex vivo studies have demonstrated the ability of bezafibrate to decrease the production of toxic acylcarnitine species in MADD patients, including two patients with ETFB missense mutations and three with biallelic LOF ETFA mutations (Yamaguchi et al., 2012; Yamada et al., 2017). However, clinical evidence of the therapeutic effect of bezafibrate is limited to a single patient with MADD-M (Yamaguchi et al., 2012). A randomized, double-blind, crossover study of bezafibrate in patients with CPT II and VLCAD deficiencies showed lowered low-density lipoprotein, triglyceride, and free fatty acid concentrations, but no changes in fatty acid oxidation, or heart rate during exercise (Ørngreen et al., 2014). The bases for the apparently conflicting findings have been discussed between the respective research groups (Bastin et al., 2015; Ørngreen et al., 2015); however, the benefit of treatment with bezafibrate remains unresolved.

While treatment with riboflavin or bezafibrate are directed primarily towards MADD-M patients - acting by boosting gene expression or folding-stability properties of missense variants - treatment with exogenous ketone bodies (D,L-3-hydroxybutyrate) has been applied to bypass the disturbed ketogenesis in MADD, and may act independently on the severity of the metabolic block at the ETF/ETF-QO site. In fact, ketone body treatment has successfully been used to ameliorate symptoms and biochemical abnormalities in both MADD-M and MADD-S patients, including six patients with genetically confirmed ETF deficiency (Van Hove et al., 2003; van Rijt et al., 2020).

Since patients may be responsive to treatment, early diagnosis is essential. Diagnosis is based on urinary analysis of organic acids and acylglycines using gas chromatography/mass spectrometry (GC/MS) combined with plasma analysis of acylcarnitines using tandem mass spectrometry (MS/MS) (Prasun, 1993). This latter method is also being used for neonatal diagnosis through expanded newborn screening programs that are being offered in several countries. Diagnosis may be challenging in MADD-M cases since the excretion of diagnostic metabolites is considerably less pronounced and often intermittent, being present only during acute episodes (Prasun, 1993). Moreover, since there is more than one cause of MADD, a specific diagnosis can be established only by demonstrating deficiency of ETF or ETF:QO using either specific enzyme activity assays or immunochemical methods (Amendt and Rhead, 1986; Loehr et al., 1990), or by sequence analysis of the ETFA, ETFB and ETFDH genes (Goodman et al., 2002; Olsen et al., 2003). In recent years, genetic defects in other electron transport chain complexes or genetic deficiencies in riboflavin uptake or FAD synthesis/transport have been shown to present with MADD-like biochemistry and phenotype (Bosch et al., 2011; Ho et al., 2011; Vissing et al., 2013; Olsen et al., 2016; Schiff et al., 2016; Kaphan et al., 2018). Thus, because of the ambiguous nature of the MADD biochemistry and since an increasing number of genes are associated with MADD, the use of whole exome or genome sequencing is the strategy of choice for genetic diagnosis.

Establishing advanced cellular and organismal models for any disease is critical to establish proof of concept for potential therapies and phenotypic effects of new mutations, and this is also the challenge for MADD. Initial advances in this direction arose from studies focused on the effects of defects in ETF:QO in zebrafish and in Caenorhabditis elegans that respectively resulted from a non-sense truncation mutation and deleterious point mutations (Chew et al., 2009; Song et al., 2009). Studies of EMS-induced Drosophila missense mutations identified mutations in ETF:QO, were shown to be embryonic lethal. These mutations specifically affect FAD interacting residues, impairing protein assembly and causing the accumulation of acylcarnitines in the fly embryos (Alves et al., 2012). In humans, such mutations would account for the most severe forms of the disease, and the yet unmet challenge is to develop this system in a way that milder disease-causing mutations can be introduced in Drosophila to test their broader phenotypic effects, as well as functional recovery with cofactors and drugs.

Nomenclature

ACDH

Acyl-CoA dehydrogenase

AMP

Adenosine monophosphate

ATP

Adenosine triphosphate

CoQ10

Coenzyme Q10

ETC

Electron transport chain

ETF

Electron transfer flavoprotein

ETF:QO

Electron transfer flavoprotein-ubiquinone oxidoreductase

FAD

Flavin adenine dinucleotide

LOF

Loss of function

MADD

Multiple acyl-CoA dehydrogenase deficiency

NADH

Nicotinamide adenine dinucleotide reduced

ROS

Reactive oxygen species