Mitochondrial Cytochrome c Oxidase Biogenesis: Recent Developments

Abstract

Mitochondrial cytochrome c oxidase (COX) is the primary site of cellular oxygen consumption and is essential for aerobic energy generation in the form of ATP. Human COX is a copper-heme A hetero-multimeric complex formed by 3 catalytic core subunits encoded in the mitochondrial DNA and 11 subunits encoded in the nuclear genome. Investigations over the last 50 years have progressively shed light into the sophistication surrounding COX biogenesis and the regulation of this process, disclosing multiple assembly factors, several redox-regulated processes leading to metal co-factor insertion, regulatory mechanisms to couple synthesis of COX subunits to COX assembly, and the incorporation of COX into respiratory supercomplexes. Here, we will critically summarize recent progress and controversies in several key aspects of COX biogenesis: linear versus modular assembly, the coupling of mitochondrial translation to COX assembly and COX assembly into respiratory supercomplexes.

1. Introduction

One of the defining characteristics of a living system is the ability of many translated proteins to fold and assemble into macromolecular complexes with precision and fidelity. Assembly is particularly complicated for hetero-multimeric complexes, for which the process often requires the assistance of dedicated chaperones and undergoes multiple quality control checkpoints. It is, therefore, not surprising that the failure of complex-components to fold or assemble correctly is the origin of a wide variety of pathological conditions [1]. Uncovering the mechanisms through which such assembly processes take place is one of the grand challenges of modern science from a biological and a biomedical perspective.

Within mitochondria, accurate assembly processes are fundamental for the building of the macromolecular complexes that provide the backbone of aerobic energy production in eukaryotes by the oxidative phosphorylation (OXPHOS) system. The biological roles of these complexes are relevant to cellular adaptation to changing environments as well as to multiple disease scenarios, including encephalo- and cardio-myopathies, cancer, neurodegeneration and the aging process [2]. From a biogenetic point of view, a distinctive level of convolution impinges the assembly of the four OXPHOS complexes whose components are derived from both the mitochondrial and the nuclear genomes. As a case in point, in this manuscript we will discuss the biogenesis of the mitochondrial respiratory chain (MRC) terminal oxidase, the cytochrome c oxidase (COX) enzyme or MRC complex IV (CIV).

1.1. Mammalian COX components, evolution and assembly factors

COX is a copper-heme A terminal oxidase embedded in the mitochondrial inner membrane [3]. It catalyzes electron transfer from reduced cytochrome c to molecular oxygen in a process coupled to the transfer of protons across the inner membrane, thus contributing to the generation of the proton gradient that is used by ATP synthase to drive ATP synthesis. Mammalian COX is a multimeric enzyme formed by 14 subunits of dual genetic origin. The three subunits forming the catalytic core of the enzyme (COX1, COX2 and COX3) are encoded by the mitochondrial DNA, and the remaining by the nuclear DNA. The structure of the core subunits is conserved from α-proteobacteria, the ancestors of mitochondria, to human COX. COX1 and COX3 are highly hydrophobic, integral membrane proteins with no substantial extramembrane domains. COX2, on the other hand, consists of a β barrel extramembrane domain anchored by two transmembrane helices that bind to COX1, opposite from the side of COX1 that binds COX3 (Fig 1A). Subunits COX1 and COX2 contain the redox metal active centers of the enzyme. The extramembrane domain of COX2 extends into the mitochondrial intermembrane space (IMS) to bind soluble cytochrome c. A di-copper center termed Cu_A accepts the electrons from soluble cytochrome c. The Cu_A center is located in a loop region at the bottom of the extramembrane domain of COX2, at the interface of COX1 and COX2 (Fig 1B). From Cu_A, electrons flow to heme a in COX1, which transfers the electrons to the heme a₃-Cu_B binuclear center where O₂ is reduced to water (Fig 1B) [3].

Figure 1. Mitochondrial cytochrome c oxidase: structure, redox metal cofactors and overall assembly pathway.

(A and B) Ribbon diagrams of high-resolution structures of (A) the thirteen-subunit enzyme from bovine (Bos taurus) heart mitochondria [6, 127] (PDB 1OCC). The COX catalytic core subunits are colored in blue (COX1), red (COX2) and pink (COX3). The nucleus-encoded COX subunits are colored in light green. In (B) most of the bovine protein has been removed to show only the redox-active metal centers. A covalent bond between Tyr-371 and the Cu_B ligand His-378 is present but not shown in this structure. Detailed explanations of the Cu_A and Cu_B centers are in the text. (C) The schematic represents a simplified model for the process of monomeric COX assembly based on the modular model recently reported [9]. The catalytic core subunits form subassembly modules with other subunits and assembly factors prior to be incorporated into the main assembly pathway. In general, following their insertion into the inner membrane, COX1 and COX2 are stabilized by substrate-specific chaperones and matured by addition of metal cofactors. Following COX1 maturation, the nuclear DNA-encoded COX4 and COX5A, forming a module argued to be stabilized by the assembly factor HIGD1A are added to COX1 prior incorporation of the COX2-containing, and subsequent a set of three assembly factors (PET100, PET117 and MR-1S) that promote the incorporation of the COX3-containing module and the NDUFA4 subunit to form the COX holoenzyme or Complex IV (CIV). Only a few COX biogenetic factors are represented.

Through evolution, up to eleven smaller subunits (in mammals) have been added to the catalytic core (Fig 1A). This has presumably allowed to cope with evolving energetic requirements of eukaryotic cells over time, to adapt to oxygen-rich environments, and to regulate complex IV activity to coordinate energy production with timely cellular needs [4]. The last subunit of the mammalian complex to be identified was NDUFA4 [5]. NDUFA4 was absent from the crystal structures [6] and had been mistakenly assigned as a subunit of complex I. Adding convolution, mammalian COX has multiple tissue-specific isoforms of nuclear encoded subunits, which portraits the key regulatory role COX has on OXPHOS [4]. Assembly of functional COX is not trivial, since the process is complicated by the dual genetic origin of the COX subunits. Hence, it is not surprising that eukaryotic cells have evolved nucleus-encoded assembly factors and signaling pathways to coordinate mitochondrial-nuclear expression of COX genes, to control the abundance of the subunit and metal cofactor components, and to direct their assembly into a functional enzyme [7, 8]. Already the four-subunit aa₃-type COX of α-proteobacteria requires the assistance of five dedicated conserved assembly proteins for the accumulation of the complex in the membrane. These are COX10 and COX15 that function to synthesize heme A, SURF to possibly deliver heme A, SCO to assemble Cu_A, and COX11 to assemble Cu_B, as it will be detailed in the following sections.

Mitochondria contain more than 30 COX assembly factors (Table I), a list that continues to be extended. For example, immunocapture of COX assembly intermediates has facilitated the identification of the short isoform of the myofibrillogenesis regulator 1 (MR-1S) as a protein that works with the conserved PET100 and PET117 chaperones to assist late stages of COX biogenesis in higher eukaryotes [9]. Over the last 50 years, several model organisms and experimental systems have been used for COX assembly studies. These studies have produced a substantial amount of information regarding the individual assembly factors, some mechanistic detail, and numerous hypotheses and models of the assembly processes. These, along with recent developments and standing controversies, will be discussed in this article, with a primary focus on the mammalian enzyme.

Table I. Mammalian COX subunits and assembly factors.

Alias names are provided for some of the proteins

Table Ia
SUBUNIT ISOFORMS				Subunit contacts within COX monomer	Contact sites within the CI-CIII2-CIV respirasome
LIVER Ubiquitous	HEART	LUNG	TESTIS		With CI subunits	With CIII subunits
Catalytic core subunits
COX1				COX2, COX3*
COX2				COX1*
COX3				COX1*
Supernumerary subunits
COX4-1		COX4-2		COX1t,m, COX5Am
COX5A				COX4-1m, COX6Cm
COX5B				COX1m, COX3m
COX6A1	COX6A2			COX3t
COX6B1			COX6B2	COX2c, COX3c
COX6C				COX2t, COX5Am
COX7A2	COX7A1			COX3t	NDUFB8p ND5o (loose structure)	UQCR11, UQCR1, UQCRB
COX7B
COX7C				COX1t, COX8t	ND5, NDUFB7, NDUFB8p, NDUFB3p NDUFB2o
COX8-2	COX8-1		COX8-3	COX1t	NDUFB7, NDUFB8p, NDUFB3p, NDUFB2o
NDUFA4

Table Ib
ASSEMBLY FACTORS	ROLE
RNA stability and translation
LRPPRC	Human: mitochondrial mRNA stability
TACO1	Human: COX1 mRNA translational activator
Membrane Insertion
OXA1L	Membrane insertion of COX subunits
COX20	COX2 chaperone. Component of the COX2 copper metallation module
COX18	Export of the COX2 C-terminus tail
Copper metabolism and Insertion
COX17	Delivery of Cu(I) to SCO1 and COX11
SCO1 SCO2	Transfer of copper to COX and/or reduction of cysteine residues in COX2
COA6	CX9C-CX10C protein that cooperates with SCO in the metallation of CuA
COX11	Assembly of CuB in the aa3-type heme-Cu oxidases
COX19	Interacts with COX11 as a reductant, critical for COX11 activity
Heme A biosynthesis and insertion
COX10	Farnesylation of heme B
COX15	Oxidation of heme O methyl to formyl
FDX2	Collaborates with COX15 in heme O oxidation
ADR	Collaborates with COX15 in heme O oxidation
SURF1	Catalyzes an assembly step involving COX1; proposed to participate in heme A delivery
Assembly
CMC1	Stabilizes COX1 with COX14 and COA3, perhaps to facilitate the assembly of metal centers
CMC2	Twin CX9C protein of unknown function
COX23	Twin CX9C protein of unknown function
COX14  C12orf62	Binds COX1; required for its stability and assembly
COA3  MITRAC12, CCDC56, COX25	Binds COX1; required for its stability and assembly
COA1  MITRAC15, C7orf44	Required for COX1 stability/maturation/assembly
COX16	Unknown function
HIGD1A	Stabilizes a COX4-COX5A module and perhaps promotes its assembly with COX1
HIGD2A	Supports the formation of a class of COX-containing supercomplexes
MITRAC7	Stabilizes COX1 in the COX1-COX5A-COX4 module
MR-1S	Formation of late COX assembly intermediates
PET100	Formation of late COX assembly intermediates
PET117	Formation of late COX assembly intermediates
PET191	Twin CX9C protein of unknown function

^*Only the contacts with other core subunits are indicated for COXI, COXII, and COXIII

^cCytosolic,

^mmatrix,

^ttransmembrane interactions

^pPorcine respirasome

^oOvine respirasome

1.2. Linear vs. modular COX assembly

Since the pioneering studies on COX assembly during the early 1980s, the process was thought to be linear, with the different subunits and cofactors being added in a sequential and ordered manner. This concept was based on studies that used rat liver mitochondria and followed the incorporation of radiolabeled subunits into the COX holocomplex [10]. The model was later confirmed by analyzing the formation of assembly intermediates in human cultured cells by Blue-Native electrophoresis, which allowed researchers to conclude that de novo assembly initiates around a seed formed by COX1 and proceeds with the formation of several discrete assembly intermediates that accumulate at rate-limiting steps in the process [11, 12].

The concept of linear assembly requires modification, however. Studies using the yeast Saccharomyces cerevisiae have shown that the biogenesis of each of the three subunits of the catalytic core proceeds by a relatively independent process with the participation of subunit-specific chaperones [13]. This promotes the idea of COX assembly as a modular process. Consistent with this concept, the COX catalytic core subunits in human cells also appear to form preassembly modules, which in the case of COX1 and COX2 are fundamental to ensure proper subunit maturation by the incorporation of copper and heme prosthetic groups [14–17]. Further support for modular assembly of human COX came from a recent study that compared COX and COX assembly intermediates immunopurified from mitoplasts of wild-type (WT) and cybrid COX3 mutant cell lines [9]. This study poses that the first module to be formed does not even involve a core subunit, rather two nucleus-encoded subunits, COX4 and COX5A, stabilized by the assembly factor HIGD1A. This initial complex then incorporates the maturation and stabilization module containing COX1 and its chaperones COX14 (alias C12orf62), COA3 (alias CCDC56 or MITRAC12), COA1 (alias C7orf44 or MITRAC15) and SURF1. Taking into consideration the bacterial ancestry of the core subunits, even if the COX4-COX5A module accumulates independently of COX1, as it does in yeast [18], we prefer to maintain the description of COX1 as the seed around which the monomeric COX complex is assembled (Fig 1C). The formation of these two modules, as well as the formation of the COX2-containing module (with subunits COX5B, COX6C, COX7B, COX7C and COX8) and the COX3-containing module (with COX6A, COX6B, COX7A and NDUFA4) seem to be independent of each other [9]. The COX assembly process could still be viewed as a linear process but in which different modules, rather than subunits, are being added in a sequential manner (Fig 1C).

2. Biogenesis of the COX catalytic core subunits

Reviewing studies on COX assembly over the last 50 years, perhaps the greatest area of agreement is in the biogenesis of the core subunits and the overall sequence of metal center assembly into the catalytic core, but even here many questions remain. Studies in model organisms such as α-proteobacteria, the yeast S. cerevisiae and transgenic mice have been essential. Studies on human cell lines derived from patients suffering from mitochondrial disorders owing to mutations in COX subunits or assembly factors are helping to close gaps. The study of individual assembly factors in human cell lines benefits from gene expression silencing approaches as well as the generation of specific gene knockouts (KO) using novel gene editing techniques [14–16]. Here, despite the focus on mammalian COX, studies on model organisms will be summarized when necessary to provide contrasting mechanisms or information not yet available in humans.

2.1. Copper source and delivery to COX-specific metallochaperones

As it will be explained in the following sections, SCO1/2 and COX11 are the assembly chaperones that facilitate copper insertion into the COX2-Cu_A and COX1-Cu_B sites, respectively [19–22]. COX11 and SCO1/2 are anchored to the mitochondrial inner membrane through a transmembrane α-helix and expose the copper-binding site in the IMS, where copper transfer occurs. The SCO and COX11 proteins receive copper from COX17, a conserved IMS-located hydrophobic twin CX₉C protein that was first discovered in S. cerevisiae [23]. COX17 contains a CCxC Cu(I) binding motif [24], where the last cysteine in the CCxC sequence is the first cysteine of a twin Cx₉C motif. While COX17 can bind multiple coppers when its cysteines are fully reduced, in the IMS it is argued to bind and transfer a single Cu(I) ion [25]. Details on copper transfer from COX17 to the COX metallochaperones have been reviewed elsewhere [26].

Despite recent advances, details on mitochondrial copper metabolism and how copper reaches COX17 remain to be fully understood. In yeast and human cells, the mitochondrial matrix contains a pool of copper bound by an anionic fluorescent molecule known as the copper ligand (CuL) [27]. This pool is the source of copper for the metallation of COX17 [28]. CuL has also been found in the cytoplasm and it has been suggested that CuL may recruit copper to mitochondria [28]. Studies in S. cerevisiae have shown that the mitochondrial phosphate carrier protein Pic2 and the mitochondrial inner membrane iron transporter Mrs3 independently transport CuL into the mitochondrial matrix [29, 30]. These copper transporters have functional equivalents in mammals. The human mitochondrial phosphate carrier (PiC) [31], known as SLC25A3, also functions in copper import into human mitochondria [30], and the human mitochondrial iron transporters Mitoferrin-1 and Mitoferrin-2 are each similar to yeast Mrs3. Further work is necessary to clarify whether these carriers transfer CuL into the matrix of mammalian mitochondria. It also remains to be established whether Pic2 and Mrs3 of yeast, and their homologs in mammals, are able to promote copper efflux from the matrix to the IMS, where it would reach COX17.

2.2. COX1 Assembly Line

2.2.1. Is synthesis of human COX1 coupled with cytochrome c oxidase assembly?

COX1 is a highly hydrophobic protein that spans 12 transmembrane domains in the inner membrane. In S. cerevisiae, Cox1 is co-translationally inserted into the membrane by the Oxa1 insertase machinery [32], whereas human OXA1L is apparently not required for CIV biogenesis [33]. Two conserved COX1 chaperones, the proteins Cox14 and Coa3 (Table I and Fig 2A), are single transmembrane proteins with a hydrophilic C-terminus in the IMS [34–36]. They interact with Cox1 and probably direct its insertion into the inner membrane. Certainly, they stabilize and prevent degradation of newly synthesized Cox1 [37, 38], and also chaperone Cox1 in later steps of CIV holoenzyme assembly [17, 37].

Figure 2. COX1 maturation and assembly line.

(A) Model depicting the roles of COX14 (alias C12orf62), COA3 (alias MITRAC12, CCDC56, or COX25) and CMC1 in controlling post-translational events in COX1 biogenesis [14]. According to the model, newly synthesized COX1 would first bind COX14 and COA3, followed by CMC1. CMC1 would promote COX1 stability during or before COX1 maturation and would be released from the growing COX1 complex before the incorporation of COX4-1 and COX5A and additional assembly factors such as SURF1 and MITRAC7 (see explanation in the text). (B) Model for copper delivery to COX1 by COX11 showing a COX11-COX11 homodimer (model 5, cluster 2 of ref. [58]) hovering above fully folded COX1, taken from the structure of the bovine COX (PDB 5b1b). Tethering of the antiparallel dimer by the transmembrane helices of the COX11 monomers (upper ends are shown in gold) dictates that the four S-two Cu(I) cluster of the COX11 dimer (Cu(I) ions in yellow, Cys ligands in red) will face the membrane surface. Linkers of 15 amino acid between the transmembrane helices and the headgroup of the dimer will keep the COX11 Cu cluster near the membrane surface. Cu_B of fully assembled COX1 is shown in magenta while the position of the Cu_B ligands His-290 and His-291 are shown in pink. Cys35 in bacterial COX11 (Cys-121 of human COX11 and Cys-111 of yeast COX11) is not shown, but Cys-35 of each monomer will be present near the exit of each COX11 transmembrane helix from the upper surface of the membrane. It can be seen that His-290 and His-291 need move only a short distance toward the upper surface of COX1 in order for COX11 to transfer Cu(I) to these histidines, by the mechanism proposed in section 2.2.3. (C) Model for COX19 regulation of copper binding to and transfer from COX11 by maintaining Cys-35 in a reduced state. Models of yeast COX11 and COX19 with relevant residues shown as sticks. In COX11, the three relevant cysteines are labeled (the asterisk denotes that numbering of S. meliloti COX11 is based on PDB entry 1SO9). The red, green and blue segments correspond to those identified as interacting with COX19 by Bode et al [61] (in that order along the protein sequence). The red and green segments are the most hydrophobic and conserved, and thus most likely candidates for interaction. In COX19, the putative interaction surface would be a conserved hydrophobic patch built around two TyrLeu dipeptides in the yeast protein. 3D visualization available online at http://lucianoabriata.altervista.org/modelshome.html

In S. cerevisiae, Cox1 synthesis is coordinated with CIV biogenesis via a regulatory mechanism that involves a negative feedback regulatory loop [3, 39]. The Cox1-Cox14-Coa3 complex traps the COX1 mRNA-specific translational activator and COX1 chaperone Mss51, whose release and availability to activate new rounds of COX1 mRNA translation are coupled to Cox1 maturation or assembly with other COX subunits [34]. By this mechanism, Cox1 synthesis is attenuated to approximately 10% of normal in most CIV assembly mutants, with obvious exceptions such as cox14 or coa3 mutants, in which Mss51 cannot be sequestered [34]. A similar mechanism was proposed to exist in human mitochondria, where a complex of COX1 with cardiomyopathy proteins COX14 and COA3 would regulate COX1 synthesis. This possibility was supported by the observation of mildly attenuated COX1 synthesis in [³⁵S]-methionine pulses in fibroblasts from COX14-deficient patients or in HEK293T cells treated with COX14 or COA3 siRNA [17, 38]. However, the putative human Mss51 homolog recently described does not participate in COX1 biogenesis, and although the mammalian-specific COX1 mRNA translational activator TACO1 binds to COX1 mRNA and associates with the mitoribosome [40], it is not a COX1 chaperone [41].

Studies on human CMC1, a twin CX₉C protein that transiently interacts with and stabilizes the early COX1-COX14-COA3 complex, have presented a new perspective [14]. COX1 synthesis proceeds normally in a CMC1-KO human cell line [14]. In wild-type cells, CMC1 is released from the CMC1-COX1-COX14-COA3 complex before the incorporation of COX subunits COX4 and COX5A to the growing subassembly [14]. The CMC1-containing complex accumulates in several mutants affecting later steps of CIV assembly, e.g. COX2 or COX3 homoplasmic mutant cybrids, COX4-, COX5A- or COX10 (heme A biosynthetic enzyme)-silenced cells and a KO of the COX2 chaperone COX20. However, the synthesis of COX1 remains unaffected in these mutants. The COX1-COX14-COA3-CMC1 complex (termed MITRAC) does not accumulate in COX14- or COA3-silenced cells [14]. In these cells, the incorporation of [³⁵S]-methionine into newly synthesized COX1 is nearly normal, but COX1 is rapidly degraded [14]. In humans, therefore, extremely fast degradation of newly synthesized COX1 may partially substitute for a feedback loop that decreases COX1 synthesis.

Other evidence points to a different type of translational control in human cells. If COX1 membrane insertion were co-translational, a failure of this process in the absence of COX14 or COA3 could lead to protein misfolding and, potentially, mitoribosome stalling and premature translation termination. This actually seems to be the case, as COX14 and COA3 were found to act sequentially in the stabilization of specific COX1 transmembrane helices during synthesis and it has been reported that a ribosome nascent COX1 chain, partially inserted into the membrane and bound to COX14-COA3, accumulates when COX4 is depleted [42]. This complex could represent a primed state of the translated COX1 that can be retrieved for assembly, which would enable adaptation of COX1 synthesis to the availability of at least COX4 [42]. In the context of a module-base COX assembly pathway, it seems necessary to explore whether COX1 elongation, as reported, is also regulated by defects in the COX2 or COX3 assembly lines.

2.2.2. Insertion of heme A into COX1

In eukaryotic cells, heme A is synthesized in mitochondria and is present exclusively in COX1 (Fig 1B). The reactive nature of a free heme A molecule makes it likely that a heme-binding protein(s) chaperones the transfer of heme A from its site of synthesis to both of the heme centers in COX1. Whether the heme a and heme a₃ centers are both assembled by the same components remains a key question. Heme A biosynthesis (reviewed in [3]) involves the conversion of heme B to heme O, and then to heme A, by two enzymes, COX10, or heme O synthase, and COX15, or heme A synthase. Both enzymes are integral inner membrane proteins.

The assembly of heme a or heme a₃ centers does not occur during the insertion of COX1 into the membrane. In fact, in R. sphaeroides cells lacking heme A, membrane-inserted apo-COX1 accumulates to significant levels as part of an apo-CIV [43]. Also in yeast mitochondria, newly synthesized Cox1 accumulates in a complex, with Mss51-Coa3-Cox14 and the Hsp70 chaperone Ssc1 [39] that does not contain heme A [44]. This Cox1 pre-assembly complex also forms in yeast cells defective in heme A biosynthesis and in cox1 mutant strains that lack the histidine ligands for heme a [45]. Thus, heme A insertion must occur at a post-translational stage. The heme A insertion stage is thought to commence upon the release of Mss51-Ssc1 from the Cox1 pre-assembly complex and the association first of the assembly factor Coa1 [46], and subsequently of COX subunits Cox5a and Cox6 and the proposed heme A delivery protein Shy1 (the SURF1 homolog of yeast)[45] (Fig 2). In human cells, as discussed earlier, newly synthesized COX1 also accumulates in a complex with COX14, COA3 and CMC1 [17, 37, 38]. Formation of this complex is independent of COX10 and SURF1, suggesting that COX1 in this complex is not hemylated [14]. Release of CMC1 from this complex is coupled to the association of COX1 with SURF1 plus the COX4 and COX5A structural subunits [14], homologs of yeast Cox6 and Cox5a, respectively.

The specific role of Shy1/SURF1 in COX1 hemylation deserves further discussion. The function of the protein is not completely essential since its deletion, from bacteria to humans, results in the retention of at least 25% of functional COX. Studies in R. sphaeroides suggested that the primary role of its single SURF1 protein is to enhance the assembly of heme a₃ [47]. Consistently, purified SURF1 proteins of P. denitrificans (which contains two) were subsequently shown to have the ability to bind stoichiometric amounts of heme A with K_d values of 0.3–0.6 μM [48, 49], values that were increased by ten-fold when a conserved histidine in the C-terminal transmembrane helix of each SURF1 protein was modified [48]. However, mutation of this and other histidines in yeast Shy1 did not significantly affect its function [50], suggesting that yeast Shy1 might not bind heme A.

In all organisms, a second heme A insertase, one which populates the heme a site, must exist. It seems possible that this second heme insertase could be promiscuous and capable of populating the heme a₃ site in the absence of SURF1, thereby explaining the accumulation of some normal COX in the absence of SURF1/Shy1. In yeast, Shy1 associates with Cox1 downstream of the apo-Cox1-Mss51 preassembly complex [45]. The Cox1 assembly complex containing Shy1 is less stable when isolated from cells lacking heme A, suggesting that the insertion of heme A occurs within this subassembly [45]. Rather than acting as a heme A insertase, eukaryotic SURF1 could associate with COX1 to stabilize the heme a₃ site during heme A delivery. This is supported by the finding in S. cerevisiae that overexpression of COX subunits Cox5a and Cox6 (human COX4 and COX5, respectively), which do not bind heme A, significantly suppresses the respiratory defect of Δshy1 cells [46, 51] presumably by stabilizing Cox1.

In S. cerevisiae, both Cox10 and Cox15 form oligomers that are important for their respective functions [52, 53]. Cox15 oligomerization is key not only to heme A biosynthesis but also to its transfer to maturing COX [53]. Cox15 oligomerization may be promoted by interaction with Pet117, a conserved small matrix protein peripherally associated with the inner membrane [54]. In this way, Pet117 supports heme A synthesis and together with Cox15 could participate in heme A delivery to Cox1. During the process, Pet117 could associate with and stabilize a COX1-containing assembly intermediate. Consistently, human PET117 has been identified in assembly intermediates, interacting with late COX assembly factors such as PET100 [9]. The requirement of human PET117 for COX assembly comes from the finding that mutations in this protein cause COX deficiency associated with neurodevelopmental regression [55]. Whether human PET117 interacts with COX15 and promotes heme A synthesis and delivery to COX1 remains to be investigated.

2.2.3. Assembly of Cu_B

Cu_B is composed of a single copper ion bound by three invariant histidine ligands [56] (Fig 1B). His-240 (bovine numbering) also forms a post-translational covalent bond with a nearby tyrosine, forming the amino acid redox center unique to heme-Cu oxidases (Fig 1B).

Cu_B assembly absolutely requires the copper chaperone COX11 [21], composed of a C-terminal soluble domain located in the IMS, plus a single anchoring transmembrane helix, connected by a 15 amino acid flexible linker [57] (Fig 2B). The soluble domain contains a conserved CFCF motif [88]. The purified, soluble domain of yeast COX11 forms a dimer that binds Cu(I) at a stoichiometry of one copper per monomer [19]. Modeling and X-ray absorption spectroscopy predict a dimer of antiparallel monomers with two Cu(I) bound by four S, from adjacent CFCF motifs, at the interface of the two monomers [58] (Fig 2B). Mutagenesis of yeast and R. sphaeroides COX11 confirms that both cysteines of the CFCF sequence are essential for copper binding and, therefore, for Cu_B assembly [19, 59, 60]. A third conserved cysteine, Cys-111 in yeast, is also essential for Cu_B assembly [19]. However, this cysteine, located within the flexible linker of the chaperone, is not required for COX11 to bind normal amounts of Cu(I) in all COX11 proteins [59].

Investigations in S. cerevisiae have shown that Cu_B site formation does not occur co-translationally, but only once Mss51 has been released from the apo-COX1 stabilization complex, near or at the Shy1-containing COX1 assembly intermediate [45]. As must be true for heme A insertion, the delivery of copper to the buried histidine ligands of the Cu_B center requires structural flexibility of COX1. In one conception of Cu_B assembly [59], the Cox11 dimer [58] “floats” above the membrane, with the 2Cu(I)-4S center facing the membrane surface (Fig 2B). The two flexible linkers connecting the dimer to the transmembrane helices allow the 2Cu(I) cluster to approach one of the two Cys-111 residues near the membrane surface. Within Cox1, a loop between transmembrane helices 7 and 8 moves several angstroms to bring the two upper histidine ligands of Cu_B to the outer surface of COX1. With proximity established, the thiol of Cys-111 is envisioned to first interact with the 2Cu(I)-4S cluster of the Cox11 dimer, and then with the histidines of the nascent Cu_B center, to form transition-state coordination complexes that transfer a single Cu(I) from Cox11 to Cox1 (Fig 2B) [59]. Support for this hypothesis comes from a recent study [61] showing that the principal function of Cox19, a twin CX₉C protein essential for COX assembly, is to keep Cys-111 reduced (Fig 2C), as required in the mechanism described above.

Currently, it can only be assumed that the outlined Cu_B assembly mechanism will apply to human mitochondria. Yeast COX19 consists of a helical hairpin structure that forms a hydrophobic surface characterized by two highly conserved Tyr-Leu dipeptides, which are essential for COX19 function [61]. Human COX19 lacks the N-terminal dipeptide but it is replaced by Phe-Met, thus conserving the hydrophobicity and steric hindrance properties (Fig 2C). Importantly, human COX11 and its copper-binding motif are well conserved [57]. Recently, assessment of COX assembly in siRNA-mediated COX11-silenced HEK293T cells indicated particularly decreased steady-state levels of COX2 and COX6b, whereas unmetalated COX1 and COX4 were less affected, since they accumulated in stable early assembly intermediates [14].

2.3. COX2 Assembly Line

2.3.1. Is synthesis of human COX2 regulated by downstream COX assembly events?

Within the context of the modular COX assembly model proposed in human cells [42], coordination of COX2 synthesis to the availability of nucleus-encoded assembly partners and membrane stabilizing factors could be expected. In support of this possibility, COX2 synthesis is decreased in fibroblasts from patients with mutations in the COX2-specific copper chaperone SCO2, but not in its homologue SCO1 [62]. However, COX2 synthesis is not affected in HEK293T cells with knockouts for either COX20 [16] or COX18 [15] or with a knockdown for SCO2 [42]. Cell type-specific effects and experimental design could account for the conflicting reports.

2.3.2. Stabilization and membrane insertion of COX2

COX2 consists of two N-terminal transmembrane domains followed by a soluble globular domain that protrudes into the IMS and contains the di-copper Cu_A center. Specialized proteins required for membrane insertion of COX2 were first identified in S. cerevisiae (reviewed in [3]). Unlike human COX2, the yeast protein is synthesized as a precursor (pCox2) with a cleavable N-terminal extension. pCox2 interacts with the OXA1 machinery, which facilitates export of its N-terminal domain across the inner membrane. pCox2 then interacts with Cox20, a Cox2-specific chaperone that presents it to the inner membrane peptidase complex for removal of the N-terminal extension. Subsequently, Cox18, together with the inner membrane proteins Mss2 and Pnt1, promotes the translocation of the Cu_A-containing globular domain of Cox2 across the membrane. Cox18 resembles some Oxa1 family members but its role is specific for the translocation of Cox2 [63].

Both COX20 and COX18 are conserved from yeast to humans. The characterization of COX20-KO and COX18-KO HEK293T cell lines has demonstrated their requirement for COX2 biogenesis in human cells [15, 16] (Fig 3A). The N-tail of newly synthesized human COX2 is translocated across the membrane and its first transmembrane helix then stabilized by interaction with COX20 [16], a transmembrane protein found to be mutated in several cases of dystonia-ataxia syndrome [64]. There is no evidence that OXA1L acts as a COX2 N-tail membrane insertase as it does in yeast. Knockdown of OXA1L in HEK293 cells affects the biogenesis of the F₁F_o-ATP synthase and complex I without altering the abundance of complexes III or IV [33], and OXA1L does not co-immunoprecipitate with COX20 [16]. Given that COX20 is an integral inner membrane protein with small domains located in the matrix [65] (Fig 3A), it seems plausible that COX20 could guide the co-translational insertion of the COX2’s N-terminal transmembrane helix into the membrane. COX20 would then stabilize COX2 when only its N-terminal transmembrane helix is embedded in the inner mitochondrial membrane and its second transmembrane helix and globular domain are still in the mitochondrial matrix (Fig 3A and B).

Figure 3. COX2 maturation and assembly line.

(A) Model depicting the roles of COX20 and COX18 in COX2 stabilization and membrane insertion [15, 16]. The release of COX18 from the complex coincides with the incorporation of the Cu_A maturation module formed by the copper chaperones SCO1 and SCO2 and the CX₉C protein COA6. (B) Model depicting a possible action of COX20 on COX2 membrane insertion. Left: Acting at an earlier stage, COX20 could stabilize COX2’s TM helices when it is inserted in the membrane only through TM1 while TM2 and the globular domain are on the mitochondrial matrix. Right: At a later stage, COX20’s TM helices could stabilize COX2’s TM helices when they are both inserted in the membrane, before and/or during and/or after flipping of COX2 through the IMS, before its assembly onto the growing oxidase. In the globular domain of COX2 (purple) the purple spheres indicate the location of the CuA site, but note that at these stages the copper ions have actually not been loaded yet. 3D visualizations of these models are available online at http://lucianoabriata.altervista.org/modelshome.html. (C) Models for mammalian COX2 metallation by SCO1-SCO2. The schemes depict five published proposals for the assembly of CuA (explanations are given in section 3 and reviewed in [26]). The possible involvement of COA6 is not depicted. (D) A possible structural model for the electron and/or copper transfer reactions between apo- or copper-SCO and apo-COX2 proteins (cyan and yellow, respectively) built from the x-ray structure of the disulfide-mixed complex between Bradyrhizobium japonicum TlpA and COX2 (also available in 3D online). The depicted apo-COX2 is from B. japonicum (PDB ID 4TXV chain B) and the SCO protein is human SCO2 (PDB ID 2RLI). A structure of T. thermophilus copper-loaded COX2 (PDB ID 1EHK) is shown for the sake of comparing the compactness of the CuA-binding loop. All spheres are relevant cysteines, while other copper ligands are shown as sticks.

Considering COX20 as a helix hairpin whose first helix is as long as required for membrane insertion and whose second helix is roughly twice as long [65], we can speculate that the latter is meant to stabilize COX2’s second transmembrane helix while in the matrix [65] (Fig 3A and B). An alternative, or second, role for human COX20 might be stabilization of the two transmembrane helices of fully membrane-inserted COX2 either before it is flipped to translocate its globular domain to the IMS, and/or after translocation but before being assembled into the growing oxidase [65] (Fig 3B).

Translocation of COX2’s Cu_A-containing globular domain is carried out by COX18 as it promotes insertion of the second transmembrane helix [15]. A structural model for COX18 can be confidently built through homology modeling, revealing five transmembrane helices with the positively charged Arg94 residue buried inside the lumen formed by the TM helices [65], which corresponds to a functional conserved site in membrane protein insertases proposed to “grab” substrates to translocate them through the membrane [66]. COX18 is released from the COX20-COX2-COX18 complex probably coinciding with the incorporation of SCO1 to the complex, because a small fraction of SCO1, but not of SCO2 or COA6 that complete the Cu_A maturation module (discussed below), was co-immunoprecipitated with COX18 [15].

2.3.3. Assembly of the Cu_A center

The Cu_A center is located within a loop region at the bottom of the COX2 β-barrel globular domain, close to the surface of COX1. The heart of the Cu_A center is a symmetric cluster of two copper ions bridged by two S thiol ligands from cysteine residues. One copper ion is further coordinated to a histidine imidazole and a methionine S thioether, and the other to a histidine imidazole and a backbone carbonyl group (Fig 1B). The two copper ions are bridged by the thiol S giving rise to a unique electronic structure that enables their function as a one-electron carrier [67]. The structural features of the Cu_A center dictate some requirements for its assembly, which takes place in the IMS. First, oxidation of apo-Cu_A’s cysteines prevents copper binding; hence Cu_A assembly requires a disulfide reductase. Second, two copper ions are required for assembly, and given that single-copper chaperones are involved, two are required.

In the mature oxidase, most Cu_A ligands lie within a region of COX2 that interacts strongly with a shallow depression on the surface of COX1, such that the accessibility of Cu_A to other proteins is very limited. This implies that chaperone-mediated Cu_A assembly must take place before COX2 associates with COX1. A relevant structural feature in this context is the high degree of flexibility of the Cu_A-binding loops in apo- but not in copper-loaded COX2 [68, 69]. This ensures exposure of the copper ligands to facilitate copper delivery by the chaperones, but rigidity once the Cu_A site is formed as required for efficient electron transfer [70].

The key copper chaperones involved in the delivery of copper to Cu_A in the heme-Cu oxidases are SCO proteins. Originally identified in S. cerevisiae [20], these are membrane-bound copper chaperones present in bacteria and in mitochondria [71]. They consist of single globular domains located in the mitochondrial IMS and anchored to the membrane by a single α-helix. Each SCO protein binds a single copper, either Cu(I) or Cu(II). Copper binding involves two cysteine S from a CXXXC thioredoxin-like motif plus a histidine imidazole N located in a β-hairpin inserted into the thioredoxin fold [72]. The cysteines of the CXXXC motif of human SCO1 form a reversible disulfide with a redox potential at the lower end of those measured for thioredoxins [73]. Thus, SCO proteins could both reduce the disulfide of the nascent Cu_A center and deliver copper. A structural view on how this might proceed is available from the X-ray structure of a disulfide-mixed adduct between TlpA thioredoxin and apo-COX2 soluble domain from Bradyrhizobium japonicum [74]. This structure shows that in order for apo-COX2’s cysteines to engage disulfide adducts with TlpA, the copper-binding loop of apo-COX2 needs to partially unfold, as suggested based on the dynamics of apo-COX2 [68]. If the complex between apo-COX2 and SCO proteins is assumed to adopt a similar orientation such that SCOs’ cysteines reach apo-COX2’s cysteines, then the β-hairpin characteristic of SCO proteins results in a position that suggests a function in unfolding of the copper-binding loop in apo-COX2 (Fig 3D), similar to functions proposed for β-hairpins in other systems.

In vivo and in vitro studies using human, yeast or bacterial cells and purified proteins, have led to at least five possible scenarios for Cu_A assembly, which we have recently reviewed [26]. Focusing on human cells, several key observations led to the following model for COX2-Cu_A assembly: 1. human mitochondria contain two SCO proteins, SCO1 and SCO2, which are closely related; 2. each of the SCO proteins are essential for Cu_A formation by performing non-overlapping functions [62] and mutations in either SCO1 [75] or SCO2 [76] result in severe mitochondrial disorders; 3. human SCO1 and SCO2 cannot complement for each other indicating that they have separable functions, although the two SCO proteins interact and can exchange copper with each other [25]; 4. oxidized apo-SCO1 can be metalated by COX17, in a reaction in which the thiols of COX17 reduce the disulfide bond of apo-SCO1 and simultaneously transfer a copper into the reduced SCO protein [25]; 5. demetalated SCO2 appears to oxidize demetalated SCO1, since the overexpression of SCO2 leads to increased SCO1_oxid/SCO1_red [62]; i.e. human apo-SCO2 appears to have thiol-disulfide oxidoreductase activity. These observations led to the four-step model [62] presented as model 1 in Fig 3C. 1). Metalated SCO2 and metalated SCO1 bind to apo-COX2. SCO2 performs a thiol-disulfide exchange reaction with the cysteines of the Cu_A center and simultaneously transfers one copper into the Cu_A site. This leaves SCO2 oxidized, with a disulfide between the cysteines of its CXXXC thioredoxin-like site. 2) SCO1 transfers the second copper into the Cu_A center, leaving its CXXXC site with two thiols. 3) The disulfide of apo-SCO2 oxidizes the thiols of apo-SCO1, a reaction facilitated by their redox potentials and their proximity since both proteins are bound to COX2. 4) Reduced apo-SCO2 is re-metalated by COX17, while oxidized apo-SCO1 is both reduced and metalated by COX17 via a thiol-disulfide exchange-copper transfer reaction [25].

A recent in vitro biochemical study [77] reported that SCO1 is a metallochaperone that selectively transfers Cu(I) ions based on loop recognition, whereas SCO2 is a copper-dependent thiol reductase of the Cu_A cysteine ligands in COX2. Hence, in this scenario, the reduction of the apo-Cu_A disulfide by SCO2 Cu(I) is carried out by the metal and not by the thiols (Fig 3C-model 2). SCO2 was found to have disulfide reductase activity, but by a chemistry different than a thiol-disulfide exchange reaction. While the previous model [62] proposed that both SCO1 and SCO2 could deliver copper to Cu_A, biochemical evidence suggests [77] that only SCO1 is able to act as a metallochaperone to the Cu_A center. The latter model (Fig 3C-model 2) considers a reaction stoichiometry consistent with the finding that SCO proteins may be active as dimers in vivo. The model is also consistent with the experiments in human cells, except by the fact that apo-SCO2 is not able to oxidize apo-SCO1 in vitro. This reaction may be mediated in the mitochondria by other cofactors.

Another copper-binding protein of mitochondria, not present in prokaryotes, may participate in Cu_A assembly. COA6 is a soluble protein located in the IMS (Fig 3A), where it binds Cu(I) with an affinity similar to that of SCO proteins [78]. Several findings implicate COA6 in Cu_A assembly. Deletions or knockdowns of COA6 in yeast [79, 80] and human cells [78, 79, 81] decrease COX accumulation, but this can be rescued by the addition of exogenous Cu(II) [80, 81] suggesting that another copper chaperone can be driven to substitute for COA6. COA6 interacts with newly synthesized COX2 in both yeast and human mitochondria [78, 79]. In human mitochondria, COA6 is argued to interact selectively with SCO2 [79] but a separate study reports the interaction of human COA6 with SCO1 [78]. Another study has suggested that SCO1 could interact with COX2 prior to the action of SCO2-COA6, perhaps to stabilize the protein upon release of COX18 from the COX2 translocation complex [15], while COX20 could remain chaperoning COX2 during the overall maturation process [15, 16]. In conclusion, however, the role of COA6 in the process remains unknown, and biochemical studies are required to complement in vivo studies.

While the details of Cu_A assembly in humans are being deciphered, the possibility of alternative assembly pathways within the same mitochondrion should not be ruled out.

3. Rate-limiting steps in cytochrome c oxidase biogenesis

The concerted accumulation of COX subunits in mammalian cells is largely regulated by post-translational degradation of unassembled subunits. In some cases, some sort of translational regulation, as for COX1 described earlier, couples protein synthesis with the availability of assembly partners, thus contributing to the stoichiometric accumulation of COX subunits. As a key catalytic subunit, COX1 biogenesis undergoes multiple levels of regulation. In another example, the COX1-COA3-COX14-CMC1 (MITRAC) complex has been proposed to receive early-assembling imported nucleus-encoded subunits from the mitochondrial presequence translocase TIM23 via TIM21 [17]

In addition to passing the final folding quality control checkpoints, proteins in a complex such as COX require proper oligomeric assembly. Assembly factors are frequently referred to as assembly chaperones because they promote the native conformation and stability of their substrates, either subunits or assembly intermediates. In a modular model of COX assembly, formation of each module could be considered as an assembly checkpoint. For example, formation of the COX1-COA3-COX14-CMC1 complex is essential to promote COX1 stability, maturation and a conformation that will allow the protein to progress in the assembly pathway by interacting with the COX4-COX5A sub-module [9, 14, 17]. Independently of translational control, lack of COA3 or COX14 is accompanied by proteolytic removal of newly synthesized COX1.

Some of the molecular processes that promote and regulate the progression of assembly within and amongst the several COX modular intermediates remain unclear; nonetheless assembly factors must play key roles. For example, human cells depleted of either SURF1 or SCO2, accumulate subassemblies containing at least COX1-COX4-COX5A, but this subassembly does not accumulate in COX10-deficient cells [82]. These data indicates that heme A incorporation into COX1 occurs prior to its association with COX2, and that COX2 metallation is essential for its further assembly. A question remaining is how the COX1-COX4-COX5A subassembly remains stable when the COX2 module is not available for incorporation. One answer is MITRAC7, a bona fide checkpoint protein. MITRAC7 is a small inner mitochondrial transmembrane protein with its C terminus facing the IMS [83]. It is a COX1-specific chaperone that interacts with late forms of MITRAC that contain COX4 and COX6A. Increased MITRAC7 levels stabilize and trap COX1 in MITRAC, blocking progression in the assembly process. In contrast, MITRAC7 deficiency leads to turnover of newly synthesized COX1 by a yet undefined mechanism [83]. MITRAC7 role is then to protect the COX1 module from degradation before the COX2 module is incorporated.

Relatively scarce information is available regarding the late steps of assembly in which addition of nuclear encoded subunits completes the formation of the functional holoenzyme. In S. cerevisiae, a high molecular weight assembly intermediate that includes Cox7, Cox7a and Cox8 (human COX7A-COX6C-COX7C) is associated with the chaperone Pet100 in yeast wild-type cells [84]. The same subcomplex accumulates in the absence of Pet100, but without Pet100 it cannot interact with other subassemblies or subunits [84]. PET100 is a small protein (~10 kDa) conserved in humans, where mutations in the corresponding gene result in Leigh’s syndrome associated with defective COX assembly and function [85, 86]. Immunoprecipitation studies showed PET100 to interact with COX7A2 in HEK293 cells [87] and BN-PAGE analysis showed that imported PET100 assembles into a ~300 kDa complex, distinct from the endogenous COX [85], whose composition was not investigated. Recent studies based on immunocapture of COX assembly intermediates identified PET100 t association with the COX assembly factors MR-1S and PET117 to assist late stages of COX biogenesis in human cells [9], probably promoting the association of submodules formed by nucleus-encoded subunits to COX2 or COX3 and/or the incorporation of these modules to the COX1 within the holocomplex assembly line (Fig 1C).

4. Are there multiple, parallel pathways for COX biogenesis?

A growing body of information indicates that COX biogenesis may be a heterogeneous process. Evidence for this is seen in two features of mammalian COX: the presence of tissue-specific isoforms of nuclear encoded subunits, and the assembly of COX into different structures, e.g. as a monomeric complex, or as part of the respirasome (Fig 4A) or other respiratory supercomplexes (SCs).

Figure 4. Contact sites of COX (complex IV or CIV) subunits with subunits from complexes CI and CIII in the context of the mammalian respirasome.

(A) Ribbon diagrams of high-resolution structures of the tight form of the CI-CIII₂-CIV respirasome from ovine (Ovis aries) heart mitochondria [102] (PDB 5J4Z). The three component complexes are colored in light blue (CI), light yellow (CIII) and light green (CIV). In deep blue are colored CI subunits in proximity and proposed to interact with CIV subunits, colored in deep green. (B) Magnified view of deep blue colored CI subunits (ND5 and NDUFB7) proposed to interact with deep green colored CIV subunits COX8 and COX7C [102]. (C) Different view of the structure presented in (A), rotated 156°, with deep yellow colored CIII subunits in proximity and potentially interacting with a CIV subunit colored in deep green. (D) Magnified view of deep yellow colored CIII subunits (UQCRC1, UQCR11 and UQCRB) of the active CIII monomer proposed to interact with the deep green colored CIV subunit COX7A [102].

4.1. Tissue-specific variations

A distinctive feature of mammalian cells and tissues is the existence of tissue-specific or stress-induced subunit isoforms, which have been proposed to perform a regulatory role in energy production and therefore adaptation to tissue-specific metabolic demands [4]. To present, six isoforms have been described for the nuclear subunits of COX in mammalian cells: three liver/heart-type pairs of subunits (COX6A1/COX6A2, COX7A1/COX7A2, and COX8-1/COX8-2), the lung-specific isoform COX4-2, and two testes-specific isoforms, COX6B and COX8-3 (Table Ia). The different isoforms have slightly altered structural features that could affect the organization of the holocomplex and the functioning of the enzyme. In fact, heart (or muscle) isoforms are expressed in tissues that have high aerobic capacity and an abundance of mitochondria, whereas liver (or non-muscle) isoforms are found in tissues like brain, liver and kidney that contain fewer mitochondria. The best-characterized pair of isoform subunits is that of COX4-1 and COX4-2. The COX4-2 isoform is expressed in lung, trachea, and placenta, and at low levels also in brain and heart [88]. A measured increase in COX electron transfer efficiency in lung and trachea has been proposed to be an energetic and protective adaptation to reduce free radicals in highly oxygenated tissues [89]. COX4-2 and its homologous isoform in yeast, termed Cox5b, are also expressed under hypoxic conditions (1% oxygen), as explained below [90].

It is tempting to speculate that the existence of COX subunit isoforms may contribute to the tissue specificity observed in patients suffering from mitochondrial disorders associated with COX deficiency. For example, loss of COX assembly factors, such as SCO2 and SURF1, leads to tissue-specific assembly defects [12, 91]. Deficiencies in these and other COX assembly factors result in tissue-specific variations in COX steady-state levels, activity, composition of COX subcomplexes, and COX assembly into supercomplexes [12, 91].

The presence of tissue-specific isoforms suggests the potential presence of tissue-specific assembly factors or isoforms, yet to be discovered, and deviations from the canonical assembly pathway. Since more than one isoform exist in a variety of tissues, intramitochondrial COX heterogeneity seems to be granted. Modulation of COX heterogeneity may have evolved to meet different metabolic tissue requirements, but how it occurs at the molecular level remains in most cases to be understood.

4.2. COX biogenesis under stress

Cells possess the ability to sense changes in the environment and to adapt accordingly. Post-translational modification of COX subunits (reviewed in [92, 93]) and the existence of COX isoforms contribute to the sensing and response to modulate energy production according to changing environmental oxygen levels or endogenous oxidative stress. A case in point is subunit pair COX4-1/COX4-2, discussed earlier, which are homologues of yeast Cox5a/Cox5b.

Original studies in yeast showed that the expression of COX5a/5b is regulated transcriptionally [94]. Expression of the normoxic COX5a isoform is repressed under hypoxia whereas COX5b isoform expression is repressed under normoxia and de-repressed in response to hypoxia [94] or oxidative stress [95]. In mammalian cells, COX4-1 and COX4-2 are also regulated by oxygen concentrations, although by a different mechanism. Although some controversies remain, the hypoxia inducing factor 1 (HIF-1) complex, a transcriptional activator that functions as a master regulator of oxygen homeostasis in all metazoan species, regulates COX subunit isoform switch [96], although only at very low oxygen tensions (<1%) found under pathological conditions [97]. According to this model, HIF-1 would activate the expression of COX4-2 and, at the same time, of mitochondrial LON protease, responsible of COX4-1 proteolysis [96]. However, other transcriptional factors have been proposed to regulate COX4-2 expression under moderate hypoxia (1–4%). Thus, different mechanisms allow the switch of COX4 isoforms to optimize electron transfer reactions in response to changing energetic conditions. From an assembly perspective, isoform switch is not expected to occur by subunit substitution at the individual complex level, but by controlling the proportion of complexes that contain each isoform.

4.3. COX biogenesis in the context of supercomplexes

COX accumulates in mammalian mitochondria not only as a free enzyme but also as a ternary unit within the respirasome (formed by a complex I –CI- monomer, a complex III dimer –CIII₂- and a CIV monomer: CI-CIII₂-CIV (Fig 4A)) and additional respiratory chain supercomplex assemblies involving CI and/or CIII [98–100].

Studies of the S. cerevisiae CIII₂-CIV_1–2 SCs by cryo-EM and single particle analysis have suggested that CIII and CIV make contact through subunits Cox5a and Qcr6 at the IMS, whereas in the hydrophobic layer, larger spaces found between CIII and CIV were suggested to be filled with phospholipids [101]. In mammalian cells, the respirasome structure (CI-CIII₂-CIV) has been analyzed in bovine, ovine and porcine heart mitochondria and in all cases COX is positioned at the distal end of the membrane arm of CI and adjacent to the CIII dimer [102–105] (Fig 4A). The interaction between CIII and CIV is defined by the contact of COX7A with CIII subunits UQCR11, UQCRC1 and UQCRB in all the structures (Fig 4C and D). The main contact between complexes I and IV occurs between subunits ND5 and COX7C in all species. However, COX7C and COX8 subunits were described as close to NDUFB7, NDUFB8 and NDUFB3 in porcine mitochondria [103] or to NDUFB7 and NDUFB2 in ovine mitochondria [102] (Fig 4A and B). Besides, analysis of a loose structure of the ovine respirasome, showed that COX7A switches contact from CIII to CI, to interact with ND5 [102]. Consistently, an interaction between COX7A and CI subunit NDUFB8 was also observed in the porcine respirasome [103].

Although how supercomplexes assemble remains intriguing, two major models have been proposed. It has been observed that in S. cerevisiae mitochondria, COX subassemblies might interact with respiratory chain CIII components at early stages in the process of supercomplex assembly [106]. In support of this view, nuclear DNA-encoded subunits newly imported into mitochondria isolated from COX-deficient human fibroblasts can integrate not only into the COX holoenzyme by associating with pre-existing subunits but also into SCs by associating with intermediate assembly complexes [107]. Furthermore, the first respirasome assembly pathway proposed, based on studies of di novo assembly of respiratory enzymes in cultured human cells, also supported this model [108]. In this pathway, only when monomeric COX is fully assembled and accumulated up to a certain threshold, newly synthesized COX subunits and subassemblies incorporate into larger structures containing CI and CIII intermediates [108]. However, other studies have reported a temporal gap between the formation of the individual complexes and that of the SCs [109]. Also, more recent di novo assembly studies performed by complexome analysis have allowed the authors to conclude that SC formation follows at least after CI assembly is complete [110], although their kinetics study did not include crucial data on what occurs between 8 h and 24 h following di novo assembly initiation. The proposed dynamic formation and remodeling of MRC complexes and SCs in vivo [109] add complication to the subject. A combination of multiple approaches following assembly di novo and in vivo in an array of control cell types would be required to definitely settle the question or to generate a model that would integrate the different observations into a common picture. It could well be that as recently described for the bacterial ribosomes, the entire SC assembly process could be dynamic and “re-routed” through different pathways as needed, which would fit with the plasticity model for MRC enzyme organization [109].

At least in S. cerevisiae, COX heterogeneity occurs at the organelle level and affects SC assembly and the role of potential SC assembly or stabilization factors [111–113]. The identification and characterization of the Respiratory superComplex Factor Rcf1, involved in the accumulation of CIII₂-CIV and CIII₂-CIV₂ SCs, indicated that this protein is required for the incorporation of a peripheral CIV subunit (Cox13, human COX6A) [111] and the efficient addition of another one (Cox12, human COX6B) [112]. Importantly, COX monomers lacking this subunit do not assemble into the CIII₂-CIV₂ SC. Besides, Rcf1 associates with Cox3, predicted to be at the CIII-CIV interface, and with ADP/ATP carrier (AAC) proteins, and stabilizes CIII-CIV SCs [112]. Nonetheless, Rcf1 does not seem to be a stoichiometric COX component, but to regulate a late maturation step, perhaps participating in the incorporation of cardiolipin [114], the phospholipid that stabilizes interactions between SC components [115, 116]. Rcf1 is a member of the hypoxia inducible gene 1 (HIG1) family and could favor the formation of COX forms and SCs that perform best under hypoxic conditions. Rcf1 has two human homologs, HIGD1A and HIGD2A. Human HIGD1A forms an early COX assembly intermediate with COX4-1 and COX5A [9], perhaps to facilitate their interaction with COX1, and has been proposed to be a positive regulator of COX activity under hypoxia in cardiomyocytes [117]. It will be important to investigate whether hypoxic COX4-2 also forms an independent complex with COX5A and HIGD1A, to better understand how COX4 subunit switch is regulated. Because other studies have identified association of HIGD1A with CIII [118], a possible role for this protein in assisting or stabilizing SC assembly is not discarded. Human HIGD2A is able to partially substitute for Rcf1 in the yeast deletion model [111], and knockdown of HIGD2A in mouse muscle myoblasts impairs the formation of COX-containing SCs [113], which suggests a conserved function. Assembling SCs containing different COX subunits and isoforms could contribute to adjust cellular energy production to different metabolic requirements.

Efforts to identify SC Assembly Factors resulted in the finding of COX7A2L (SCAFI or COX7RP) [119], although its characterization in mouse models has yielded conflicting results [119–123]. There is a consensus that the absence of COX7A2L prevents the formation or stability of the CIII_2-CIV SC, but it is now becoming clear that the CI-CIII_2-CIV respirasome can accumulate normally in some mouse tissues and human cell lines lacking COX7A2L [122, 124]. Interestingly, COX7A2L mutations in mice lead to downstream tissue-specific differences in SC formation, where CI-CIII_2-CIV_2-n respirasomes are more affected in liver than in heart [122], and where even the CI-CIII_2-CIV respirasome seemed to be unstable in some tissues [125], results not reproduced by other groups [121, 122]. This prompted speculations about possible effects of tissue-specific COX isoforms that remain to be structurally demonstrated. COX7A2L was found to interact with CIV, but preferentially with CIII in human cells [124]. However, in an intriguing model, interchanges of different isoforms of CIV subunits have been proposed to drive CIV assembly into different SC structures [125]. COX7A2 subunit present in monomeric COX was proposed to be replaced by COX7A2L in SC containing CIII and CIV, and by COX7A1 on CIV dimers [125]. Furthermore, COX6A1 would favor the free CIV form, while COX6A2, which is located at the interface of the two monomers [6], would stabilize the CIV dimer.

Neither COX7A2L nor HIGD1A/2A have been identified in cryo-EM structures of the respirasome [102–104]. However, consistently with the data discussed above in mice, analysis of the ovine respirasome structure [102] allowed to speculate that COX7A1 (present in the interface of CIII and CIV in the respirasome structure) could actually be substituted by COX7A2L yielding two structures which differ in the presence or absence of interaction between CIII and CIV [102]. Finally, the role of COX7A2L and HIGD2A in SC assembly has been recently tested in situ under live cell conditions, showing how depletion of either COX7A2L or HIGD2A caused a release of CIV from SCs [126], thus further linking these two proteins to COX assembly into SCs.

5. Concluding remarks and perspectives

The integration of layers of information arising from studies in vivo and in vitro, in multiple research models has so far allowed for the generation of multiple hypotheses regarding defined steps of COX biogenesis that are progressively being integrated into a detailed working model. For example, the concept of linear assembly has been progressively substituted by the concept of modular assembly. While the catalog of COX assembly factors must be near to completion, new reports keep updating the list. New assembly checkpoint proteins have been recently identified. Many challenges, however, remain for full understanding of COX biogenesis, especially to disclose the specific role of most assembly factors at the molecular level, the identity of the heme A insertase (perhaps the very heme A synthase, after all), whether and how COX subunit isoforms affect holoenzyme assembly and formation of COX-containing supercomplexes. Although human COX deficiencies have not been extensively discussed here, it is obvious that solving the puzzle of COX biogenesis will facilitate the understanding of enigmatic open questions such as the heterogeneous tissue specificity of COX deficiencies in patients suffering from mitochondrial disease.