JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
Available online at www.sciencedirect.com
ScienceDirect www.elsevier.com/locate/jprot
Review
Exploring mitochondrial system properties of neurodegenerative diseases through interactome mapping☆ James Vlasbloma,1 , Ke Jina,b,c,1 , Sandy Kassira,1 , Mohan Babua,⁎ a
Department of Biochemistry, Research and Innovation Centre, University of Regina, Regina, Saskatchewan S4S 0A2, Canada Banting and Best Department of Medical Research, University of Toronto, Toronto, Ontario M5S 3E1, Canada c Terrence Donnelly Center for Cellular and Biomolecular Research, University of Toronto, Toronto, Ontario M5S 3E1, Canada b
AR TIC LE I N FO
ABS TR ACT
Available online 18 November 2013
Mitochondria are double membraned, dynamic organelles that are required for a large number of cellular processes, and defects in their function have emerged as causative factors for a
Keywords:
growing number of human disorders and are highly associated with cancer, metabolic, and
Epistatic interactions
neurodegenerative (ND) diseases. Biochemical and genetic investigations have uncovered
Mitochondrial diseases
small numbers of candidate mitochondrial proteins (MPs) involved in ND disease, but given
Networks
the diversity of processes affected by MP function and the difficulty of detecting interactions
Neurodegenerative diseases
involving these proteins, many more likely remain unknown. However, high-throughput
Protein–protein interactions
proteomic and genomic approaches developed in genetically tractable model prokaryotes and
RNA interference
lower eukaryotes have proven to be effective tools for querying the physical (protein–protein) and functional (gene–gene) relationships between diverse types of proteins, including cytosolic and membrane proteins. In this review, we highlight how experimental and computational approaches developed recently by our group and others can be effectively used towards elucidating the mitochondrial interactome in an unbiased and systematic manner to uncover network-based connections. We discuss how the knowledge from the resulting interaction networks can effectively contribute towards the identification of new mitochondrial disease gene candidates, and thus further clarify the role of mitochondrial biology and the complex etiologies of ND disease. Biological significance Biochemical and genetic investigations have uncovered small numbers of candidate mitochondrial proteins (MPs) involved in neurodegenerative (ND) diseases, but given the diversity of processes affected by MP function and the difficulty of detecting interactions involving these proteins, many more likely remain unknown. Large-scale proteomic and genomic approaches developed in model prokaryotes and lower eukaryotes have proven to be effective tools for querying the physical (protein–protein) and functional (gene–gene) relationships between diverse types of proteins. Extension of this new framework to the
☆ This article is part of a Special Issue: Can Proteomics Fill the Gap Between Genomics and Phenotypes? ⁎ Corresponding author. Tel.: + 1 306 585 4192; fax: +1 306 337 2409. E-mail address: [email protected] (M. Babu). 1 These authors contributed equally. 1874-3919/$ – see front matter © 2013 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.jprot.2013.11.008
9
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
mitochondrial sub-system in human will likewise provide a universally informative systems-level view of the physical and functional landscape for exploring the evolutionary principles underlying mitochondrial function. In this review, we highlight how experimental and computational approaches developed recently by our group and others can be effectively used towards elucidating the mitochondrial interactome in an unbiased and systematic manner to uncover network-based connections. We anticipate that the knowledge from these resulting interaction networks can effectively contribute towards the identification of new mitochondrial disease gene candidates, and thus foster a deeper molecular understanding of mitochondrial biology as well as the etiology of mitochondrial diseases. This article is part of a Special Issue: Can Proteomics Fill the Gap Between Genomics and Phenotypes? © 2013 Published by Elsevier B.V.
Contents 1. 2. 3. 4.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defects in nuclear-encoded MPs lead to ND diseases . . . . . . . . . . . . . . . . . . . . . . . . . . MPs are tractable pharmacological targets for therapeutic intervention . . . . . . . . . . . . . . . . Current proteomic strategies for charting networks of MP complexes in ND diseases . . . . . . . . . 4.1. Identifying mitochondrial PPIs and complexes through AP–MS . . . . . . . . . . . . . . . . . 4.1.1. .Affinity tags for MP purification . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2. .Improving mitochondrial PPI detection . . . . . . . . . . . . . . . . . . . . . 4.1.3. .Protein identification from the affinity purified mitochondrial samples by MS 4.1.4. .Post-processing scoring steps to predict MP complexes . . . . . . . . . . . . 4.2. Uncovering PPIs for ND-linked MPs by biochemical fractionation coupled with MS . . . . . . 4.3. Two-hybrid systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Quantitative MS based neuroproteomics approaches . . . . . . . . . . . . . . . . . . . . . . 5. Pathway-focused analysis by high-throughput genomics . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Assaying GIs in arrayed format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Screening GIs in pooled format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Defining functional pathways involving MPs from the PPI and GI maps . . . . . . . . . . . . . . . . 7. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.
Introduction
Mitochondria are complex, dynamic, and vital organelles that mediate several fundamental cellular processes — including metabolism, respiration, ion homeostasis, and apoptosis [1,2]. These critical processes are widely conserved, from human to unicellular model organisms such as the budding yeast Saccharomyces cerevisiae [3,4]. The process of mitochondrial biogenesis and inheritance is critical for eukaryotes, and it has been estimated that 1 in 5000 humans suffers from a mitochondrial disease [3]. Due to the complex architecture and integrative role of mitochondria in diverse cellular processes, mitochondrial dysfunction is emerging as a causative factor and hallmark for a wide range of human diseases both inherited and acquired, including cancer, cardiomyopathies, and neurodegenerative (ND) diseases [1,2,5]. These diseases are directly linked to mutations in mitochondrial proteins (MPs) and have been estimated to affect more than 50 million adults in the United States alone (Source: Mitochondria Research Society). In
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
9 11 12 12 12 13 14 14 15 15 16 16 17 18 20 20 20 21 .21
view of the pervasiveness of these diseases, there is an urgent need for more effective treatments and therapies, as current therapies can only provide partial relief of mitochondrial disease symptoms [6]. Although mitochondria possess their own machinery for DNA replication, transcription, and translation, only 13 MPs are encoded by human mitochondrial DNA [1]. Based on bioinformatic, proteomic, and genomic surveys [7–10], nearly 1500 distinct MPs are estimated to be encoded within a cell's nucleus [1], and must be imported into mitochondria via specialized protein translocases that sort these nuclear proteins into one of four mitochondrial sub-compartments: outer membrane (OM), inter-membrane space (IMS), inner membrane (IM), or matrix [2,11] (Fig. 1A). Each of these sub-compartments is further organized into regions containing specific and mostly unique subsets of MPs that determine their functional identity and biochemical capabilities. Within the mitochondria, active mechanisms exist to coordinate the assembly of nuclear- and mitochondrial-encoded proteins
10
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
into macromolecular complexes, and the integrity of these complexes is monitored by key proteolytic quality control pathways inside the organelle [12]. To date, mitochondrial research has mainly proceeded by the study of biosynthetic genes and pathways in isolation, and, while this reductionist-based approach has been quite successful in advancing the state of knowledge of mitochondrial function, a higher level understanding of how the various components of mitochondrial machineries function together in pathways and complexes, particularly within the mitochondria and between mitochondria and extra-mitochondrial (i.e., proteins functioning outside of mitochondria) gene products has so far remained elusive. This is in part due to the technical difficulties in assaying mitochondrial gene products, and the
A
lack of suitable high-throughput methods to reliably detect these interactions in human cell lines. Like any other modular biological subsystem, cellular processes in mitochondria are mediated by extensive and complex networks of physical (protein–protein) and functional (gene–gene, or genetic) interactions [13–15], and so a mechanistic understanding of MP function and biogenesis depends critically on a comprehensive knowledge of the biochemical organization, composition, and dynamics of mitochondrial multi-protein complexes. In recent years, high-throughput protein–protein interaction (PPI) and genetic interaction (GI) approaches have been developed and refined in evolutionarily diverse model organisms, including yeast [16–20], bacteria [21–23], worm [24,25], fly [26,27], and recently, mammalian
Number of disease-linked MPs
B
0
50
100
150
200
PL Metabolic Neurodegenerative
Matrix (0, 12) Ribosomes (0 0, 1))
OM (8, 0)
IMS (2, 0) mtDNA Porins ER BM CJ Cristae
Cancer
Mitochondrial diseases
IM (22, 0)
Cytosol (5, 9)
Hematological Cardiovascular
LS (35)
Psychiatric
Others (54)
Immunological
AD
25
Number of ND-linked MPs
50
0
Metabolism Cellular respiration
0)
D
Number of ND-linked MPs 0
(1
C
Endocrine
)
MR (7)
Nuc Nu N ucleu le s
CMT (7
Ophthalmological ND-linked MPs (soluble) ND-linked MPs (membrane)
(11
)
Developmental
PD
Localization unclear (32, 33)
Miscellaneous
1
2
3
Mitochondrial carrier Aldehyde dehydrogenase
Apoptosis Unclear Organization Miscellaneous Protein folding
Signal transduction
PFam domain family
Biological processes
Transport Pyridine redox Protein kinase AAA Flavin oxidase SCO1/SenC HSP60 Translation Cell cycle Transcription
NADH Dynamin
Fig. 1 – The role of MPs in human diseases. (A) Schematic of mitochondrial subcellular compartments, listing in parentheses the number of membrane (left) and soluble (right) Swiss-Prot annotated MP targets associated with four major features: the matrix, inter-membrane space (IMS), inner membrane (IM), and outer membrane (OM). We considered a protein as a membrane protein only if one of the three algorithms (i.e., TMHMM, Phobius, and SCAMPI) predicted at least one transmembrane helix. BM, boundary membrane; CJ, cristae junction; PL, phospholipids; and mtDNA, mitochondrial DNA. (B) Major diseases linked with MPs (see Supplementary Table S2 for details). The number of disease-linked and ND-linked MPs (pie chart, in parentheses) that are associated with the corresponding disease was determined according to the annotations in disease databases including OMIM, CGP, HGMD, GAD, and others. LS, Leigh syndrome; AD, Alzheimer's disease; PD, Parkinson's disease; CMT, Charcot–Marie–Tooth; and MR, mental retardation. (C) Major biological processes associated with ND-linked MPs, according to Swiss-Prot annotations. (D) Assignment of ND-linked MPs to the most representative Pfam domain family.
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
systems [28–34,155]. These methods offer a promising avenue for exploring the landscape of human mitochondrial interactions pertaining to ND disease-causing proteins. Although protein–protein associations involving subsets of human neurological disease-linked MPs have been detected by various experimental means, affinity purification (AP) or biochemical fractionations coupled with mass spectrometry (MS), and two-hybrid methods have been proven in experiments involving hundreds or even thousands of proteins in mammalian cell lines [29,31,35]. While both approaches are biased towards particular types of interactions, they appear to be largely complementary. AP–MS or biochemical fractionation coupled with MS is better suited for the detection of stable protein complexes, while Y2H is optimized for detecting lower abundance, transient, and condition-specific direct interactions between two proteins [36]. It is likely that these methods, which we discuss at length in this review, will play a vital role in identifying the mitochondrial PPI network, and hence inform further investigations of the links between MPs and ND diseases. Complementary to the physical interaction data provided by PPI networks, GIs reflect functional relationships between genes and are defined when the phenotype that results from mutating one of the genes is modulated by a mutation in the second gene. Development of synthetic genetic array (SGA) technology in yeast has enabled the construction of GI maps either on a genome-wide scale [37], or in focused assays of specific biological systems that can involve hundreds of genes [16], including one specifically targeted to MPs [13]. This latter study identified a cluster of four mitochondrial genes, FCJ1, AIM5, AIM13, and AIM37, that all had similar patterns of GIs with many mitochondrial membrane proteins. Further immuno-precipitation experiments revealed that their protein products were subunits of a complex, termed mitochondrial organizing structure (MitOS), with two additional proteins, Mos1 and Mos2. As the MitOS complex localized to the IM, and knocking out any of the subunits drastically distorted mitochondrial morphology, this complex is thought to function in organizing the IM. Despite the elucidation of these large-scale GI maps in model organisms [21,24,27,37], the systematic identification of a high-resolution GI map for mitochondrial ND disease-causing genes in mammalian systems has so far been limited. However, recently RNAi (RNA interference) based technologies in mammalian cells have enabled large-scale genetic screens to be routinely performed in an arrayed or pooled setting [28,30,32,38], and have been used to interrogate genes related to chromatin remodeling [30,32] and cellular susceptibility to the toxin ricin [39]. Generation of a comprehensive mitochondrial-focused GI map covering the ND disease-related genes in mammalian cells will describe how parallel pathways, and the ensuing functional redundancy, contribute to the overall robustness of ND disease-associated mitochondrial processes. Further integrating this network with a PPI network covering the corresponding gene products will likely provide even more detailed information than the sum of either dataset alone, as was demonstrated previously in yeast [40]. Accordingly, this review is focused on emerging experimental and computational methods that are applicable towards elucidating physical and functional landscapes
11
underlying mitochondrial biology. We begin with a review of the role of mitochondria in ND disease, and discuss the current state of functional annotations in mitochondrial interaction databases. We then present an overview of the experimental and computational approaches for detecting and characterizing PPI and GI networks, with a particular focus on their potential benefits and caveats. Finally, we discuss strategies for integrating the physical and functional interaction data together and with other datasets to produce a more detailed picture of mitochondrial organization, which can aid the development of new therapeutics for ND diseases.
2. Defects in nuclear-encoded MPs lead to ND diseases Genes linked to a given disease have a higher propensity to participate in the same biochemical pathway, as evidenced by the strong global association between protein connectivity and disease activity [41,42]. Although mitochondrial compartments are separate, their processes and functions are highly interconnected and integrated with extra-mitochondrial pathways. Thus, it is not surprising that mutations in nuclear-encoded MPs contribute to the pathogenesis of common ND disorders such as Alzheimer's disease (AD), Parkinson's disease (PD), amyotrophic lateral sclerosis (ALS), hereditary spastic paraplegia, and Charcot–Marie–Tooth (CMT) disease [2,5]. Previous ad hoc investigations on the defects in proteins mediating key mitochondrial processes, including protein import, mitochondrial dynamics (e.g., fission and/or fusion), mitophagy, oxidative phosphorylation (OXPHOS), protein folding, reactive oxidative species (e.g., superoxide and H2O2), and quality control (e.g. endoplasmic reticulum (ER)-associated degradation) machineries have provided insights into the physiological functions and pathogenic mechanisms of several human ND diseases [2,15,43]. For example, a misfolded mutant of superoxide dismutase (SOD1) that binds physically with an outer mitochondrial voltage-dependent anion channel membrane protein, VDAC1, blocks the VDAC1 channel through the lipid bilayer, and thereby increases susceptibility to amyotrophic lateral sclerosis (ALS), a ND disease characterized by the loss of motor neurons [44]. Similarly, mutations in three quality control MPs — a regulatory kinase, PINK1 (also known as PARK6); an E3 ubiquitin ligase, PARKIN (also known as PARK2); and a molecular chaperone, DJ1 (also known as PARK7) that together form a complex (referred as PPD for PARKIN/PINK1/ DJ-1) both in vitro and in vivo have been shown to lead to the autosomal recessive early-onset familial form of PD. In addition to the PPD complex formation, these MPs play an important role in the degradation of misfolded PARKIN substrates [45,46], as well as in mitochondrial homeostasis and mitophagy [47,48]. Recent findings [49] indicate that excess mitochondrial fission mediates neurotoxicity induced by complex I inhibition in toxin models of PD, implicating alteration in the dynamics of mitochondrial fission/fusion as a contributor to PD pathogenesis. Furthermore, mutations in an OM fusion protein, MFN2, cause the ND CMT disease, and defects in the IM fusion protein, OPA1, which forms a complex with OM fusion proteins MFN1 and MFN2, are linked to dominant optic atrophy [50]. These examples illustrate not only the importance of mitochondrial
12
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
dynamics, but also the distribution of mitochondria in neuronal cells [51], owing to their highly polarized nature and particular biochemical requirements [52]. Although biochemical studies have focused on individual disease-causing components [53–55], it is not clear how the processes that are physically interconnected with the rest of the proteins governing mitochondrial function contribute to mitochondrial disease progression. An exhaustive search of literature sources [9,56] and public databases [57–59], including Uniprot, OMIM (Online Mendelian Inheritance in Man), HGMD (Human Gene Mutation Database), CGP (Cancer Genome Project), and GAD (Genetic Association Database) revealed that more than 600 non-redundant nuclear-encoded MP-coding genes are linked to various mitochondrial diseases solely due to mutation consequences [60,61], including 124 genes that are involved in various ND diseases (Supplementary Tables S1 and S2; Fig. 1B). Examination of the sub-cellular localization of these 124 ND-linked MPs from UniProt, as well as the results of various signal peptide and trans-membrane helix prediction algorithms, including TMHMM [62], Scampi [63], and Phobius [64], suggest that 69 are localized to the mitochondrial membrane (58 experimentally verified, 11 computationally-predicted) while the other 53 are soluble and/or membrane associated (Fig. 1A). These 124 ND-linked MPs contain numerous members of the mitochondrial carrier, aldehyde dehydrogenase, and protein kinase protein families (Fig. 1C), as annotated by PFam [65], and span a variety of functional bioprocesses: 50 are implicated in metabolism, 26 in cellular respiration, 8 in apoptosis, 9 in transport, and 4 in organization and dynamics (Fig. 1D). An additional 7 proteins are annotated as having unclear function, providing an opportunity for the discovery of novel functions, for example, by examining the functions of their neighbors within the interaction network following the guilt-by-association principle [66]. Despite the role of these MPs in ND diseases, the coverage of mitochondrial interactions for these proteins by various proteomic approaches is minimal (Fig. 2A; Supplementary Table S3). Hence high-throughput network based (i.e., proteomic and genomic) approaches, which we discuss below, can help to provide a comprehensive and organizing overview of the bioprocesses and pathways mediated by MPs, and facilitate the generation of hypothesis driven questions.
3. MPs are tractable pharmacological targets for therapeutic intervention Given their prominent role in disease progression, mitochondria are emerging as attractive targets for therapeutic intervention [67,68]. For example, etifoxine is a ligand for the mitochondrial OM translocator protein, TSPO, and has been shown to be effective towards the treatment of peripheral nerve injuries and axonal neuropathies [69]. Similarly, a transgenic mouse with familial ALS treated with the TSPO ligand olesoxime showed substantial enhancement in motor performance and delayed disease onset [70]. Moreover, other small molecules including dimebon, piracetam, and simvastatin that target β-amyloid protein (A β), mitochondrial membranes, and anti-apoptotic proteins (Bcl-2), respectively, have been identified to be effective in the treatment of AD [71].
Likewise, there are several other potentially promising therapeutic compounds affecting quality control circuits and mitochondrial biogenesis and pathways, which are outlined in great detail as the most traceable targets in the following reviews [68,72]. Although many small molecules tend to disrupt mitochondrial PPIs [71], the underlying molecular mechanisms of how this targeting occurs are not yet fully uncovered, except for a few compounds targeting certain interacting proteins that are associated with pathophysiological disorders resulting from mitochondrial defects. For instance, disrupting the interaction between the HK (hexokinase) and VDAC (voltage-dependent anion channel) proteins at the mitochondrial OM has been shown to preferentially kill tumor cells, both in vitro and in vivo [73,74]. This promising lead resulted in the development of metabolic inhibitors, such as methyl jasmonate and HK2 peptide, which disrupt the HK-VDAC interaction [67]. A detailed discussion on how various MPs or subsystems function as therapeutic targets exceeds the scope of this review; however, readers are referred to some excellent in-depth reviews on this topic [67,68,71,75]. Despite their importance in the etiology of NDs, identifying new therapeutic MP targets is often difficult, time-consuming, and expensive. Hence, the scope of drug discovery for targeting mitochondrial diseases should be moving from a protein-centric view to a more global networkbased view, wherein new disease gene targets can be identified based on the physical and functional connectivity of the MPs against a set of known drug signatures targeting the mitochondrial activity.
4. Current proteomic strategies for charting networks of MP complexes in ND diseases In this section, we provide a survey of large-scale experimental proteomics methods in human cells that we and others have developed in recent years, which can be optimized for deciphering the mitochondrial interactome of ND diseaselinked MPs. These are summarized below.
4.1. Identifying mitochondrial PPIs and complexes through AP–MS AP–MS for protein identification has proven to be one of the most successful methods for identifying stable protein interactions and complexes, enabling protein function to be characterized on a large-scale in various model organisms [18,19,23,25,26]. Although AP–MS studies in mammalian mitochondrial systems have been limited, the recent development of an efficient strategy based on combined lentiviral [76] and Gateway cloning [77] technologies has extended the scope of AP–MS to a larger scale in multiple cell types for mapping mammalian protein interaction networks [31]. AP–MS depends on the fusion of a protein-coding sequence with an epitope tag compatible with protein purification. Although the tandem affinity purification (TAP)-tag has been used successfully in mapping yeast PPIs [18,19], it has been shown to yield low amounts of purified proteins in mammalian cells [78,79]. However, in recent years, the efficiency of
13
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
Number of PPIs between ND-linked MPs
A
20
40
C-terminal VA-Tag Proteo Y Strep-III 6xHis
Co-fractionation
r
R1
Vp
Mitochondrial target cDNA
pDONR (entry clone)
Pu
ψ
Two-hybrid Co-crystal structure AC-luminescence
MAPLE vector (destination clone)
ro
AP-MS (AC-MS)
ccdB R2 hPGKpr
CM
AC-Western Biochemical activity
5’ L
sin pUC Amp f1 3’ R T L ori ori
TR
FRET PCA Reconstituted complex
C
2xTEV 3xFlag
60
RRE
PPI detection methods
0
B
LR clonase Expression clones
C-terminal VA-Tag
Proteo Y Strep-III 6xHis
2xTEV 3xFlag
Viral-based
c c c
c c
Contaminants TEV protease
HEK293T
Binding c
Anti-FLAG M2 agarose beads
TEV protease c c
Strep-Tactin sepharose beads Elution Protein complex Enzyme digestion
Lentivirus
Protein Identifications Bait Prey A Prey B Prey C Prey D
Transfection reagent (e.g. Lipofectamine)
Cell line (e.g. HEK293) Puromycin selection (3x)
Search against human protein sequences
Puromycin selection (3x)
Stable cell line with VA-Tag (~80% confluency)
Peptide separation and ionization Peptide mixture
Non-viral-based
Package Envelope
Orbitrap (Velos-Pro) mass spectrometer
Fig. 2 – Proteomic approaches for querying physical interactions among MPs linked to ND diseases. (A) Methods used to detect ND-linked PPIs to date. Experimentally determined PPIs were retrieved from BioGRID [153] and iRefWeb [154]. (B) Lentiviral and non-lentiviral based tagging using the MAPLE vector. The Gateway compatible pDONR plasmid containing the mitochondrial target cDNA (entry clone) is integrated into the MAPLE vector (destination clone) using a Gateway LR clonase reaction to generate expression clones. The MAPLE vector containing the target cDNA is then integrated into the HEK293 cell line either via lentiviral or non-viral based tagging. At left, lentiviral tagging requires transfection of HEK293T cells with envelope and packaging viral plasmids to generate lentiviruses containing mitochondrial target cDNA, which are then used to infect HEK293 cells. At right, non-viral tagging does not require any viral elements but only transfection reagents, such as Lipofectamine, are used to transfect HEK293 cells. After three rounds of puromycin selection, stable cell lines are generated that can be later expanded. For details on the MAPLE vector, see Mak et al. [31]. (C) Overview of the VA-tag tandem affinity purification coupled with mass spectrometry procedure. Due to the insolubility of many ND-linked MPs, different detergents can be used to achieve efficient purification.
several new dual-affinity tagging systems in mammalian cells has been compared to the traditional TAP system. For example, a comparison between the TAP and dual protein G-streptavidin binding peptide (GS)-TAP tag in mammalian cells has been shown to increase the protein-complex yield by an order of magnitude over the TAP-tag, but this approach does not yield sufficient affinity purified protein from the tagged baits required for MS [79]. Another strategy is to make use of commercial antibodies that bind the epitope tag of the target protein with high specificity, but the unavailability of antibodies for many human MPs makes this approach less feasible, as generating an antibody against a target epitope through monoclonal hybridoma technology is expensive, laborious, and time-consuming [79].
4.1.1.
Affinity tags for MP purification
The advent of a ~ 12 kDa versatile affinity (VA)-tag constructed in-frame with a Gateway cassette consisting of 3 × Flag, 6 × hexahistidine (His), and 2 × Streptactin (Strep) epitopes, with Flag and His separated by dual TEV protease cleavage sites in the MAPLE (mammalian affinity purification and lentiviral expression) vector, offers a promising strategy for efficient isolation of protein complexes involving ND-related MPs in mammalian cells (Fig. 2B). Most notably, the customized combinatorial tagging design allows for efficient single, dual, or triple affinity purifications based on the amount of starting cells. In our experience, the single-step purification procedure not only minimizes the loss of weakly associated mitochondrial interacting proteins during washing, but also
14
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
reduces the amount of starting material required for purification, and is ultimately more effective than two-step purifications using anti-FLAG and Strep-Tactin sepharose affinity resins (Fig. 2C). To query interactions involving MPs, the VA-epitope tag can be integrated in-frame with the target bait protein through viral or non-viral based transfection approaches at the carboxy terminus, as the mitochondrial localization of many nuclear-encoded MPs frequently requires an aminoterminal mitochondrial targeting signal [80]. The viral based tagging approach relies heavily on the generation of lentiviruses in the human embryonic kidney (HEK293T) cell line, which are subsequently transduced into target cells of any type, including neuronal specific cell lines such as SH-SY5Y, NTera2, PC12, and B12, that are widely used as models for studying human neurogenesis [81,82]. Because the lentivirus can target a broad range of host cells, it makes it feasible to transduce the lentiviruses into any of the desired human cell types. However, caution is required when the protein is expressed from the exogenous promoter (e.g., the constitutive CMV (cytomegalovirus)-driven promoter), as over-production can alter the normal physiological state of the protein, resulting in false positive identifications of MP interactions. This can be mitigated by performing multiple replicate purifications of the same tagged bait protein, so that putative interactions not found in the replicate purifications can be flagged or filtered. Both lentivirus and retrovirus-mediated transductions have been shown to be effective in controlling the level of overexpression of the tagged proteins to a certain extent [83], but the use of inducible expression vectors that can toggle the expression of the fusion protein, such as the doxycyclineinduced tetracycline-responsive element promoter [31,79], will be best-suited as the cytotoxicity caused by protein overexpression can be avoided. While this time consuming and expensive lentiviral based procedure has the advantage of providing a high-level of heterologous gene expression and establishes stable expression of the target protein in a given cell type upon chromosomal integration, it results in low numbers of stable clones as the transfection relies solely on the insertion of foreign DNA into the genome [83]. Conversely, the cost effective and rapid non-viral based tagging approach based on the use of chemical reagent-mediated [e.g., liposome (Lipofectamine or Lipofectamine 2000) and non-liposome (Fugene)] transfection is a widely adapted method (Fig. 2B) that achieves high transient and stable transfection rates (i.e., an average of 70 to 95% efficiency in model cell lines) as early as one or two days after transfection [83]. Since these transfection reagents work in the presence of serum, they also typically improve cell growth and viability and reduce the toxicity effects of the transfection [83]. However, in contrast to lentivirus, the major caveat of the lipid-based transfection is that it can dramatically vary depending on the choice of cell types. For example, while the transfection is effective in dividing cells, it can result in very low transfection efficiency (i.e., < 30%) in post-mitotic cells, such as mature neurons [83,84], limiting the application of this reagent in transfecting primary neural progenitor or post-mitotic neuronal cells. For specific details on these methods, refer to the following in-depth reviews [83,84].
4.1.2.
Improving mitochondrial PPI detection
Successful affinity purification requires efficient isolation of mitochondria using differential centrifugation and detergent solubilization, which is often difficult to achieve for membrane proteins without disrupting the lipid microenvironment to the extent that physical interactions are lost. Mild detergents allow many of these proteins to be purified, but the optimal detergent and concentration are generally different for each of the tagged bait purifications [19,85]. To improve coverage, one approach is to employ multiple detergents, as was done recently in the S. cerevisiae model organism to query over 1500 integral and peripheral membrane proteins, with a demonstrated sensitivity of ~77% compared to a sensitivity of just ~50% when only a single detergent was used [19]. Given the insoluble nature of many ND-related MPs, the selection and use of multiple detergents may therefore be necessary to achieve adequate coverage. Nonetheless, the relatively harsh conditions required for purifications can preclude the identification of lower affinity or transient interactions, and therefore a chemical cross-linking [e.g., dithiobis (succinimidylpropionate) (DSP)] step is sometimes beneficial to stabilize interactions prior to purification. For example, Yoon and colleagues [86], have shown that in the presence of chemical cross-linking by DSP, the immunoprecipitated BHK-21 cells transfected with tail-anchored, Myc-tagged, human FIS1 mitochondrial fission protein co-precipitated with the dynamin-like fission protein 1, DLP1. This additional chemical cross-linking step can capture interactions between flexible regions of solubilized proteins by covalently linking the functional groups of amino acid side chains. MS analysis of cross-linked sites can detect pairs of linked residues within a protein, or between two proteins that are adjacent in space. The length of the cross-linker itself constricts pair-wise intra-molecular interaction sites, and hence provides information about the identity of interacting partners in protein–protein interfaces [85]. When combined with AP–MS, cross-linking provides high-confidence protein interaction data with very little chemical background noise. The tagging and purification process in mammalian cells pose numerous challenges, but the cross-linking offsets, to some degree, material costs and time for capturing less stable interactions while generating a high-quality dataset.
4.1.3. Protein identification from the affinity purified mitochondrial samples by MS An important aspect of AP–MS is the mass spectrometry configuration, including both the choice of instrumentation and the algorithms and databases used for identifying the proteins. The coupling of liquid chromatography to tandem mass spectrometry (LC–MS/MS) via electrospray ionization (ESI) has become the dominant configuration for AP–MS, owing to its high sensitivity and amenability to automation. In this configuration, the trypsin digested affinity purified eluate is separated by liquid chromatography and input to the mass spectrometer, which takes an initial precursor scan and selects those ions corresponding to selected peaks within this spectra. These are isolated and fragmented, and analyzed again to produce fragment spectra. While these peaks may be selected manually, the process may also be automated in a data-driven mode where peptides corresponding to, for example, the most abundant peaks are automatically selected for further
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
fragmentation. LC–MS/MS has been used extensively in large scale proteomics in yeast [18,19], bacteria [23,87], fly [26], and human cell lines [29,31]. Nonetheless, the introduction of new linear ion trap — Orbitrap hybrid mass spectrometers has substantially increased resolution (>100,000 full width at half maximum, at least 100× greater than earlier model LTQ instruments), mass accuracy (104 to
Available online at www.sciencedirect.com
ScienceDirect www.elsevier.com/locate/jprot
Review
Exploring mitochondrial system properties of neurodegenerative diseases through interactome mapping☆ James Vlasbloma,1 , Ke Jina,b,c,1 , Sandy Kassira,1 , Mohan Babua,⁎ a
Department of Biochemistry, Research and Innovation Centre, University of Regina, Regina, Saskatchewan S4S 0A2, Canada Banting and Best Department of Medical Research, University of Toronto, Toronto, Ontario M5S 3E1, Canada c Terrence Donnelly Center for Cellular and Biomolecular Research, University of Toronto, Toronto, Ontario M5S 3E1, Canada b
AR TIC LE I N FO
ABS TR ACT
Available online 18 November 2013
Mitochondria are double membraned, dynamic organelles that are required for a large number of cellular processes, and defects in their function have emerged as causative factors for a
Keywords:
growing number of human disorders and are highly associated with cancer, metabolic, and
Epistatic interactions
neurodegenerative (ND) diseases. Biochemical and genetic investigations have uncovered
Mitochondrial diseases
small numbers of candidate mitochondrial proteins (MPs) involved in ND disease, but given
Networks
the diversity of processes affected by MP function and the difficulty of detecting interactions
Neurodegenerative diseases
involving these proteins, many more likely remain unknown. However, high-throughput
Protein–protein interactions
proteomic and genomic approaches developed in genetically tractable model prokaryotes and
RNA interference
lower eukaryotes have proven to be effective tools for querying the physical (protein–protein) and functional (gene–gene) relationships between diverse types of proteins, including cytosolic and membrane proteins. In this review, we highlight how experimental and computational approaches developed recently by our group and others can be effectively used towards elucidating the mitochondrial interactome in an unbiased and systematic manner to uncover network-based connections. We discuss how the knowledge from the resulting interaction networks can effectively contribute towards the identification of new mitochondrial disease gene candidates, and thus further clarify the role of mitochondrial biology and the complex etiologies of ND disease. Biological significance Biochemical and genetic investigations have uncovered small numbers of candidate mitochondrial proteins (MPs) involved in neurodegenerative (ND) diseases, but given the diversity of processes affected by MP function and the difficulty of detecting interactions involving these proteins, many more likely remain unknown. Large-scale proteomic and genomic approaches developed in model prokaryotes and lower eukaryotes have proven to be effective tools for querying the physical (protein–protein) and functional (gene–gene) relationships between diverse types of proteins. Extension of this new framework to the
☆ This article is part of a Special Issue: Can Proteomics Fill the Gap Between Genomics and Phenotypes? ⁎ Corresponding author. Tel.: + 1 306 585 4192; fax: +1 306 337 2409. E-mail address: [email protected] (M. Babu). 1 These authors contributed equally. 1874-3919/$ – see front matter © 2013 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.jprot.2013.11.008
9
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
mitochondrial sub-system in human will likewise provide a universally informative systems-level view of the physical and functional landscape for exploring the evolutionary principles underlying mitochondrial function. In this review, we highlight how experimental and computational approaches developed recently by our group and others can be effectively used towards elucidating the mitochondrial interactome in an unbiased and systematic manner to uncover network-based connections. We anticipate that the knowledge from these resulting interaction networks can effectively contribute towards the identification of new mitochondrial disease gene candidates, and thus foster a deeper molecular understanding of mitochondrial biology as well as the etiology of mitochondrial diseases. This article is part of a Special Issue: Can Proteomics Fill the Gap Between Genomics and Phenotypes? © 2013 Published by Elsevier B.V.
Contents 1. 2. 3. 4.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Defects in nuclear-encoded MPs lead to ND diseases . . . . . . . . . . . . . . . . . . . . . . . . . . MPs are tractable pharmacological targets for therapeutic intervention . . . . . . . . . . . . . . . . Current proteomic strategies for charting networks of MP complexes in ND diseases . . . . . . . . . 4.1. Identifying mitochondrial PPIs and complexes through AP–MS . . . . . . . . . . . . . . . . . 4.1.1. .Affinity tags for MP purification . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2. .Improving mitochondrial PPI detection . . . . . . . . . . . . . . . . . . . . . 4.1.3. .Protein identification from the affinity purified mitochondrial samples by MS 4.1.4. .Post-processing scoring steps to predict MP complexes . . . . . . . . . . . . 4.2. Uncovering PPIs for ND-linked MPs by biochemical fractionation coupled with MS . . . . . . 4.3. Two-hybrid systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Quantitative MS based neuroproteomics approaches . . . . . . . . . . . . . . . . . . . . . . 5. Pathway-focused analysis by high-throughput genomics . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Assaying GIs in arrayed format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Screening GIs in pooled format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Defining functional pathways involving MPs from the PPI and GI maps . . . . . . . . . . . . . . . . 7. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.
Introduction
Mitochondria are complex, dynamic, and vital organelles that mediate several fundamental cellular processes — including metabolism, respiration, ion homeostasis, and apoptosis [1,2]. These critical processes are widely conserved, from human to unicellular model organisms such as the budding yeast Saccharomyces cerevisiae [3,4]. The process of mitochondrial biogenesis and inheritance is critical for eukaryotes, and it has been estimated that 1 in 5000 humans suffers from a mitochondrial disease [3]. Due to the complex architecture and integrative role of mitochondria in diverse cellular processes, mitochondrial dysfunction is emerging as a causative factor and hallmark for a wide range of human diseases both inherited and acquired, including cancer, cardiomyopathies, and neurodegenerative (ND) diseases [1,2,5]. These diseases are directly linked to mutations in mitochondrial proteins (MPs) and have been estimated to affect more than 50 million adults in the United States alone (Source: Mitochondria Research Society). In
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . .
9 11 12 12 12 13 14 14 15 15 16 16 17 18 20 20 20 21 .21
view of the pervasiveness of these diseases, there is an urgent need for more effective treatments and therapies, as current therapies can only provide partial relief of mitochondrial disease symptoms [6]. Although mitochondria possess their own machinery for DNA replication, transcription, and translation, only 13 MPs are encoded by human mitochondrial DNA [1]. Based on bioinformatic, proteomic, and genomic surveys [7–10], nearly 1500 distinct MPs are estimated to be encoded within a cell's nucleus [1], and must be imported into mitochondria via specialized protein translocases that sort these nuclear proteins into one of four mitochondrial sub-compartments: outer membrane (OM), inter-membrane space (IMS), inner membrane (IM), or matrix [2,11] (Fig. 1A). Each of these sub-compartments is further organized into regions containing specific and mostly unique subsets of MPs that determine their functional identity and biochemical capabilities. Within the mitochondria, active mechanisms exist to coordinate the assembly of nuclear- and mitochondrial-encoded proteins
10
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
into macromolecular complexes, and the integrity of these complexes is monitored by key proteolytic quality control pathways inside the organelle [12]. To date, mitochondrial research has mainly proceeded by the study of biosynthetic genes and pathways in isolation, and, while this reductionist-based approach has been quite successful in advancing the state of knowledge of mitochondrial function, a higher level understanding of how the various components of mitochondrial machineries function together in pathways and complexes, particularly within the mitochondria and between mitochondria and extra-mitochondrial (i.e., proteins functioning outside of mitochondria) gene products has so far remained elusive. This is in part due to the technical difficulties in assaying mitochondrial gene products, and the
A
lack of suitable high-throughput methods to reliably detect these interactions in human cell lines. Like any other modular biological subsystem, cellular processes in mitochondria are mediated by extensive and complex networks of physical (protein–protein) and functional (gene–gene, or genetic) interactions [13–15], and so a mechanistic understanding of MP function and biogenesis depends critically on a comprehensive knowledge of the biochemical organization, composition, and dynamics of mitochondrial multi-protein complexes. In recent years, high-throughput protein–protein interaction (PPI) and genetic interaction (GI) approaches have been developed and refined in evolutionarily diverse model organisms, including yeast [16–20], bacteria [21–23], worm [24,25], fly [26,27], and recently, mammalian
Number of disease-linked MPs
B
0
50
100
150
200
PL Metabolic Neurodegenerative
Matrix (0, 12) Ribosomes (0 0, 1))
OM (8, 0)
IMS (2, 0) mtDNA Porins ER BM CJ Cristae
Cancer
Mitochondrial diseases
IM (22, 0)
Cytosol (5, 9)
Hematological Cardiovascular
LS (35)
Psychiatric
Others (54)
Immunological
AD
25
Number of ND-linked MPs
50
0
Metabolism Cellular respiration
0)
D
Number of ND-linked MPs 0
(1
C
Endocrine
)
MR (7)
Nuc Nu N ucleu le s
CMT (7
Ophthalmological ND-linked MPs (soluble) ND-linked MPs (membrane)
(11
)
Developmental
PD
Localization unclear (32, 33)
Miscellaneous
1
2
3
Mitochondrial carrier Aldehyde dehydrogenase
Apoptosis Unclear Organization Miscellaneous Protein folding
Signal transduction
PFam domain family
Biological processes
Transport Pyridine redox Protein kinase AAA Flavin oxidase SCO1/SenC HSP60 Translation Cell cycle Transcription
NADH Dynamin
Fig. 1 – The role of MPs in human diseases. (A) Schematic of mitochondrial subcellular compartments, listing in parentheses the number of membrane (left) and soluble (right) Swiss-Prot annotated MP targets associated with four major features: the matrix, inter-membrane space (IMS), inner membrane (IM), and outer membrane (OM). We considered a protein as a membrane protein only if one of the three algorithms (i.e., TMHMM, Phobius, and SCAMPI) predicted at least one transmembrane helix. BM, boundary membrane; CJ, cristae junction; PL, phospholipids; and mtDNA, mitochondrial DNA. (B) Major diseases linked with MPs (see Supplementary Table S2 for details). The number of disease-linked and ND-linked MPs (pie chart, in parentheses) that are associated with the corresponding disease was determined according to the annotations in disease databases including OMIM, CGP, HGMD, GAD, and others. LS, Leigh syndrome; AD, Alzheimer's disease; PD, Parkinson's disease; CMT, Charcot–Marie–Tooth; and MR, mental retardation. (C) Major biological processes associated with ND-linked MPs, according to Swiss-Prot annotations. (D) Assignment of ND-linked MPs to the most representative Pfam domain family.
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
systems [28–34,155]. These methods offer a promising avenue for exploring the landscape of human mitochondrial interactions pertaining to ND disease-causing proteins. Although protein–protein associations involving subsets of human neurological disease-linked MPs have been detected by various experimental means, affinity purification (AP) or biochemical fractionations coupled with mass spectrometry (MS), and two-hybrid methods have been proven in experiments involving hundreds or even thousands of proteins in mammalian cell lines [29,31,35]. While both approaches are biased towards particular types of interactions, they appear to be largely complementary. AP–MS or biochemical fractionation coupled with MS is better suited for the detection of stable protein complexes, while Y2H is optimized for detecting lower abundance, transient, and condition-specific direct interactions between two proteins [36]. It is likely that these methods, which we discuss at length in this review, will play a vital role in identifying the mitochondrial PPI network, and hence inform further investigations of the links between MPs and ND diseases. Complementary to the physical interaction data provided by PPI networks, GIs reflect functional relationships between genes and are defined when the phenotype that results from mutating one of the genes is modulated by a mutation in the second gene. Development of synthetic genetic array (SGA) technology in yeast has enabled the construction of GI maps either on a genome-wide scale [37], or in focused assays of specific biological systems that can involve hundreds of genes [16], including one specifically targeted to MPs [13]. This latter study identified a cluster of four mitochondrial genes, FCJ1, AIM5, AIM13, and AIM37, that all had similar patterns of GIs with many mitochondrial membrane proteins. Further immuno-precipitation experiments revealed that their protein products were subunits of a complex, termed mitochondrial organizing structure (MitOS), with two additional proteins, Mos1 and Mos2. As the MitOS complex localized to the IM, and knocking out any of the subunits drastically distorted mitochondrial morphology, this complex is thought to function in organizing the IM. Despite the elucidation of these large-scale GI maps in model organisms [21,24,27,37], the systematic identification of a high-resolution GI map for mitochondrial ND disease-causing genes in mammalian systems has so far been limited. However, recently RNAi (RNA interference) based technologies in mammalian cells have enabled large-scale genetic screens to be routinely performed in an arrayed or pooled setting [28,30,32,38], and have been used to interrogate genes related to chromatin remodeling [30,32] and cellular susceptibility to the toxin ricin [39]. Generation of a comprehensive mitochondrial-focused GI map covering the ND disease-related genes in mammalian cells will describe how parallel pathways, and the ensuing functional redundancy, contribute to the overall robustness of ND disease-associated mitochondrial processes. Further integrating this network with a PPI network covering the corresponding gene products will likely provide even more detailed information than the sum of either dataset alone, as was demonstrated previously in yeast [40]. Accordingly, this review is focused on emerging experimental and computational methods that are applicable towards elucidating physical and functional landscapes
11
underlying mitochondrial biology. We begin with a review of the role of mitochondria in ND disease, and discuss the current state of functional annotations in mitochondrial interaction databases. We then present an overview of the experimental and computational approaches for detecting and characterizing PPI and GI networks, with a particular focus on their potential benefits and caveats. Finally, we discuss strategies for integrating the physical and functional interaction data together and with other datasets to produce a more detailed picture of mitochondrial organization, which can aid the development of new therapeutics for ND diseases.
2. Defects in nuclear-encoded MPs lead to ND diseases Genes linked to a given disease have a higher propensity to participate in the same biochemical pathway, as evidenced by the strong global association between protein connectivity and disease activity [41,42]. Although mitochondrial compartments are separate, their processes and functions are highly interconnected and integrated with extra-mitochondrial pathways. Thus, it is not surprising that mutations in nuclear-encoded MPs contribute to the pathogenesis of common ND disorders such as Alzheimer's disease (AD), Parkinson's disease (PD), amyotrophic lateral sclerosis (ALS), hereditary spastic paraplegia, and Charcot–Marie–Tooth (CMT) disease [2,5]. Previous ad hoc investigations on the defects in proteins mediating key mitochondrial processes, including protein import, mitochondrial dynamics (e.g., fission and/or fusion), mitophagy, oxidative phosphorylation (OXPHOS), protein folding, reactive oxidative species (e.g., superoxide and H2O2), and quality control (e.g. endoplasmic reticulum (ER)-associated degradation) machineries have provided insights into the physiological functions and pathogenic mechanisms of several human ND diseases [2,15,43]. For example, a misfolded mutant of superoxide dismutase (SOD1) that binds physically with an outer mitochondrial voltage-dependent anion channel membrane protein, VDAC1, blocks the VDAC1 channel through the lipid bilayer, and thereby increases susceptibility to amyotrophic lateral sclerosis (ALS), a ND disease characterized by the loss of motor neurons [44]. Similarly, mutations in three quality control MPs — a regulatory kinase, PINK1 (also known as PARK6); an E3 ubiquitin ligase, PARKIN (also known as PARK2); and a molecular chaperone, DJ1 (also known as PARK7) that together form a complex (referred as PPD for PARKIN/PINK1/ DJ-1) both in vitro and in vivo have been shown to lead to the autosomal recessive early-onset familial form of PD. In addition to the PPD complex formation, these MPs play an important role in the degradation of misfolded PARKIN substrates [45,46], as well as in mitochondrial homeostasis and mitophagy [47,48]. Recent findings [49] indicate that excess mitochondrial fission mediates neurotoxicity induced by complex I inhibition in toxin models of PD, implicating alteration in the dynamics of mitochondrial fission/fusion as a contributor to PD pathogenesis. Furthermore, mutations in an OM fusion protein, MFN2, cause the ND CMT disease, and defects in the IM fusion protein, OPA1, which forms a complex with OM fusion proteins MFN1 and MFN2, are linked to dominant optic atrophy [50]. These examples illustrate not only the importance of mitochondrial
12
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
dynamics, but also the distribution of mitochondria in neuronal cells [51], owing to their highly polarized nature and particular biochemical requirements [52]. Although biochemical studies have focused on individual disease-causing components [53–55], it is not clear how the processes that are physically interconnected with the rest of the proteins governing mitochondrial function contribute to mitochondrial disease progression. An exhaustive search of literature sources [9,56] and public databases [57–59], including Uniprot, OMIM (Online Mendelian Inheritance in Man), HGMD (Human Gene Mutation Database), CGP (Cancer Genome Project), and GAD (Genetic Association Database) revealed that more than 600 non-redundant nuclear-encoded MP-coding genes are linked to various mitochondrial diseases solely due to mutation consequences [60,61], including 124 genes that are involved in various ND diseases (Supplementary Tables S1 and S2; Fig. 1B). Examination of the sub-cellular localization of these 124 ND-linked MPs from UniProt, as well as the results of various signal peptide and trans-membrane helix prediction algorithms, including TMHMM [62], Scampi [63], and Phobius [64], suggest that 69 are localized to the mitochondrial membrane (58 experimentally verified, 11 computationally-predicted) while the other 53 are soluble and/or membrane associated (Fig. 1A). These 124 ND-linked MPs contain numerous members of the mitochondrial carrier, aldehyde dehydrogenase, and protein kinase protein families (Fig. 1C), as annotated by PFam [65], and span a variety of functional bioprocesses: 50 are implicated in metabolism, 26 in cellular respiration, 8 in apoptosis, 9 in transport, and 4 in organization and dynamics (Fig. 1D). An additional 7 proteins are annotated as having unclear function, providing an opportunity for the discovery of novel functions, for example, by examining the functions of their neighbors within the interaction network following the guilt-by-association principle [66]. Despite the role of these MPs in ND diseases, the coverage of mitochondrial interactions for these proteins by various proteomic approaches is minimal (Fig. 2A; Supplementary Table S3). Hence high-throughput network based (i.e., proteomic and genomic) approaches, which we discuss below, can help to provide a comprehensive and organizing overview of the bioprocesses and pathways mediated by MPs, and facilitate the generation of hypothesis driven questions.
3. MPs are tractable pharmacological targets for therapeutic intervention Given their prominent role in disease progression, mitochondria are emerging as attractive targets for therapeutic intervention [67,68]. For example, etifoxine is a ligand for the mitochondrial OM translocator protein, TSPO, and has been shown to be effective towards the treatment of peripheral nerve injuries and axonal neuropathies [69]. Similarly, a transgenic mouse with familial ALS treated with the TSPO ligand olesoxime showed substantial enhancement in motor performance and delayed disease onset [70]. Moreover, other small molecules including dimebon, piracetam, and simvastatin that target β-amyloid protein (A β), mitochondrial membranes, and anti-apoptotic proteins (Bcl-2), respectively, have been identified to be effective in the treatment of AD [71].
Likewise, there are several other potentially promising therapeutic compounds affecting quality control circuits and mitochondrial biogenesis and pathways, which are outlined in great detail as the most traceable targets in the following reviews [68,72]. Although many small molecules tend to disrupt mitochondrial PPIs [71], the underlying molecular mechanisms of how this targeting occurs are not yet fully uncovered, except for a few compounds targeting certain interacting proteins that are associated with pathophysiological disorders resulting from mitochondrial defects. For instance, disrupting the interaction between the HK (hexokinase) and VDAC (voltage-dependent anion channel) proteins at the mitochondrial OM has been shown to preferentially kill tumor cells, both in vitro and in vivo [73,74]. This promising lead resulted in the development of metabolic inhibitors, such as methyl jasmonate and HK2 peptide, which disrupt the HK-VDAC interaction [67]. A detailed discussion on how various MPs or subsystems function as therapeutic targets exceeds the scope of this review; however, readers are referred to some excellent in-depth reviews on this topic [67,68,71,75]. Despite their importance in the etiology of NDs, identifying new therapeutic MP targets is often difficult, time-consuming, and expensive. Hence, the scope of drug discovery for targeting mitochondrial diseases should be moving from a protein-centric view to a more global networkbased view, wherein new disease gene targets can be identified based on the physical and functional connectivity of the MPs against a set of known drug signatures targeting the mitochondrial activity.
4. Current proteomic strategies for charting networks of MP complexes in ND diseases In this section, we provide a survey of large-scale experimental proteomics methods in human cells that we and others have developed in recent years, which can be optimized for deciphering the mitochondrial interactome of ND diseaselinked MPs. These are summarized below.
4.1. Identifying mitochondrial PPIs and complexes through AP–MS AP–MS for protein identification has proven to be one of the most successful methods for identifying stable protein interactions and complexes, enabling protein function to be characterized on a large-scale in various model organisms [18,19,23,25,26]. Although AP–MS studies in mammalian mitochondrial systems have been limited, the recent development of an efficient strategy based on combined lentiviral [76] and Gateway cloning [77] technologies has extended the scope of AP–MS to a larger scale in multiple cell types for mapping mammalian protein interaction networks [31]. AP–MS depends on the fusion of a protein-coding sequence with an epitope tag compatible with protein purification. Although the tandem affinity purification (TAP)-tag has been used successfully in mapping yeast PPIs [18,19], it has been shown to yield low amounts of purified proteins in mammalian cells [78,79]. However, in recent years, the efficiency of
13
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
Number of PPIs between ND-linked MPs
A
20
40
C-terminal VA-Tag Proteo Y Strep-III 6xHis
Co-fractionation
r
R1
Vp
Mitochondrial target cDNA
pDONR (entry clone)
Pu
ψ
Two-hybrid Co-crystal structure AC-luminescence
MAPLE vector (destination clone)
ro
AP-MS (AC-MS)
ccdB R2 hPGKpr
CM
AC-Western Biochemical activity
5’ L
sin pUC Amp f1 3’ R T L ori ori
TR
FRET PCA Reconstituted complex
C
2xTEV 3xFlag
60
RRE
PPI detection methods
0
B
LR clonase Expression clones
C-terminal VA-Tag
Proteo Y Strep-III 6xHis
2xTEV 3xFlag
Viral-based
c c c
c c
Contaminants TEV protease
HEK293T
Binding c
Anti-FLAG M2 agarose beads
TEV protease c c
Strep-Tactin sepharose beads Elution Protein complex Enzyme digestion
Lentivirus
Protein Identifications Bait Prey A Prey B Prey C Prey D
Transfection reagent (e.g. Lipofectamine)
Cell line (e.g. HEK293) Puromycin selection (3x)
Search against human protein sequences
Puromycin selection (3x)
Stable cell line with VA-Tag (~80% confluency)
Peptide separation and ionization Peptide mixture
Non-viral-based
Package Envelope
Orbitrap (Velos-Pro) mass spectrometer
Fig. 2 – Proteomic approaches for querying physical interactions among MPs linked to ND diseases. (A) Methods used to detect ND-linked PPIs to date. Experimentally determined PPIs were retrieved from BioGRID [153] and iRefWeb [154]. (B) Lentiviral and non-lentiviral based tagging using the MAPLE vector. The Gateway compatible pDONR plasmid containing the mitochondrial target cDNA (entry clone) is integrated into the MAPLE vector (destination clone) using a Gateway LR clonase reaction to generate expression clones. The MAPLE vector containing the target cDNA is then integrated into the HEK293 cell line either via lentiviral or non-viral based tagging. At left, lentiviral tagging requires transfection of HEK293T cells with envelope and packaging viral plasmids to generate lentiviruses containing mitochondrial target cDNA, which are then used to infect HEK293 cells. At right, non-viral tagging does not require any viral elements but only transfection reagents, such as Lipofectamine, are used to transfect HEK293 cells. After three rounds of puromycin selection, stable cell lines are generated that can be later expanded. For details on the MAPLE vector, see Mak et al. [31]. (C) Overview of the VA-tag tandem affinity purification coupled with mass spectrometry procedure. Due to the insolubility of many ND-linked MPs, different detergents can be used to achieve efficient purification.
several new dual-affinity tagging systems in mammalian cells has been compared to the traditional TAP system. For example, a comparison between the TAP and dual protein G-streptavidin binding peptide (GS)-TAP tag in mammalian cells has been shown to increase the protein-complex yield by an order of magnitude over the TAP-tag, but this approach does not yield sufficient affinity purified protein from the tagged baits required for MS [79]. Another strategy is to make use of commercial antibodies that bind the epitope tag of the target protein with high specificity, but the unavailability of antibodies for many human MPs makes this approach less feasible, as generating an antibody against a target epitope through monoclonal hybridoma technology is expensive, laborious, and time-consuming [79].
4.1.1.
Affinity tags for MP purification
The advent of a ~ 12 kDa versatile affinity (VA)-tag constructed in-frame with a Gateway cassette consisting of 3 × Flag, 6 × hexahistidine (His), and 2 × Streptactin (Strep) epitopes, with Flag and His separated by dual TEV protease cleavage sites in the MAPLE (mammalian affinity purification and lentiviral expression) vector, offers a promising strategy for efficient isolation of protein complexes involving ND-related MPs in mammalian cells (Fig. 2B). Most notably, the customized combinatorial tagging design allows for efficient single, dual, or triple affinity purifications based on the amount of starting cells. In our experience, the single-step purification procedure not only minimizes the loss of weakly associated mitochondrial interacting proteins during washing, but also
14
JO U R N A L OF PR O TE O MI CS 10 0 (2 0 1 4 ) 8 – 24
reduces the amount of starting material required for purification, and is ultimately more effective than two-step purifications using anti-FLAG and Strep-Tactin sepharose affinity resins (Fig. 2C). To query interactions involving MPs, the VA-epitope tag can be integrated in-frame with the target bait protein through viral or non-viral based transfection approaches at the carboxy terminus, as the mitochondrial localization of many nuclear-encoded MPs frequently requires an aminoterminal mitochondrial targeting signal [80]. The viral based tagging approach relies heavily on the generation of lentiviruses in the human embryonic kidney (HEK293T) cell line, which are subsequently transduced into target cells of any type, including neuronal specific cell lines such as SH-SY5Y, NTera2, PC12, and B12, that are widely used as models for studying human neurogenesis [81,82]. Because the lentivirus can target a broad range of host cells, it makes it feasible to transduce the lentiviruses into any of the desired human cell types. However, caution is required when the protein is expressed from the exogenous promoter (e.g., the constitutive CMV (cytomegalovirus)-driven promoter), as over-production can alter the normal physiological state of the protein, resulting in false positive identifications of MP interactions. This can be mitigated by performing multiple replicate purifications of the same tagged bait protein, so that putative interactions not found in the replicate purifications can be flagged or filtered. Both lentivirus and retrovirus-mediated transductions have been shown to be effective in controlling the level of overexpression of the tagged proteins to a certain extent [83], but the use of inducible expression vectors that can toggle the expression of the fusion protein, such as the doxycyclineinduced tetracycline-responsive element promoter [31,79], will be best-suited as the cytotoxicity caused by protein overexpression can be avoided. While this time consuming and expensive lentiviral based procedure has the advantage of providing a high-level of heterologous gene expression and establishes stable expression of the target protein in a given cell type upon chromosomal integration, it results in low numbers of stable clones as the transfection relies solely on the insertion of foreign DNA into the genome [83]. Conversely, the cost effective and rapid non-viral based tagging approach based on the use of chemical reagent-mediated [e.g., liposome (Lipofectamine or Lipofectamine 2000) and non-liposome (Fugene)] transfection is a widely adapted method (Fig. 2B) that achieves high transient and stable transfection rates (i.e., an average of 70 to 95% efficiency in model cell lines) as early as one or two days after transfection [83]. Since these transfection reagents work in the presence of serum, they also typically improve cell growth and viability and reduce the toxicity effects of the transfection [83]. However, in contrast to lentivirus, the major caveat of the lipid-based transfection is that it can dramatically vary depending on the choice of cell types. For example, while the transfection is effective in dividing cells, it can result in very low transfection efficiency (i.e., < 30%) in post-mitotic cells, such as mature neurons [83,84], limiting the application of this reagent in transfecting primary neural progenitor or post-mitotic neuronal cells. For specific details on these methods, refer to the following in-depth reviews [83,84].
4.1.2.
Improving mitochondrial PPI detection
Successful affinity purification requires efficient isolation of mitochondria using differential centrifugation and detergent solubilization, which is often difficult to achieve for membrane proteins without disrupting the lipid microenvironment to the extent that physical interactions are lost. Mild detergents allow many of these proteins to be purified, but the optimal detergent and concentration are generally different for each of the tagged bait purifications [19,85]. To improve coverage, one approach is to employ multiple detergents, as was done recently in the S. cerevisiae model organism to query over 1500 integral and peripheral membrane proteins, with a demonstrated sensitivity of ~77% compared to a sensitivity of just ~50% when only a single detergent was used [19]. Given the insoluble nature of many ND-related MPs, the selection and use of multiple detergents may therefore be necessary to achieve adequate coverage. Nonetheless, the relatively harsh conditions required for purifications can preclude the identification of lower affinity or transient interactions, and therefore a chemical cross-linking [e.g., dithiobis (succinimidylpropionate) (DSP)] step is sometimes beneficial to stabilize interactions prior to purification. For example, Yoon and colleagues [86], have shown that in the presence of chemical cross-linking by DSP, the immunoprecipitated BHK-21 cells transfected with tail-anchored, Myc-tagged, human FIS1 mitochondrial fission protein co-precipitated with the dynamin-like fission protein 1, DLP1. This additional chemical cross-linking step can capture interactions between flexible regions of solubilized proteins by covalently linking the functional groups of amino acid side chains. MS analysis of cross-linked sites can detect pairs of linked residues within a protein, or between two proteins that are adjacent in space. The length of the cross-linker itself constricts pair-wise intra-molecular interaction sites, and hence provides information about the identity of interacting partners in protein–protein interfaces [85]. When combined with AP–MS, cross-linking provides high-confidence protein interaction data with very little chemical background noise. The tagging and purification process in mammalian cells pose numerous challenges, but the cross-linking offsets, to some degree, material costs and time for capturing less stable interactions while generating a high-quality dataset.
4.1.3. Protein identification from the affinity purified mitochondrial samples by MS An important aspect of AP–MS is the mass spectrometry configuration, including both the choice of instrumentation and the algorithms and databases used for identifying the proteins. The coupling of liquid chromatography to tandem mass spectrometry (LC–MS/MS) via electrospray ionization (ESI) has become the dominant configuration for AP–MS, owing to its high sensitivity and amenability to automation. In this configuration, the trypsin digested affinity purified eluate is separated by liquid chromatography and input to the mass spectrometer, which takes an initial precursor scan and selects those ions corresponding to selected peaks within this spectra. These are isolated and fragmented, and analyzed again to produce fragment spectra. While these peaks may be selected manually, the process may also be automated in a data-driven mode where peptides corresponding to, for example, the most abundant peaks are automatically selected for further
JO U R N A L OF P ROTE O MI CS 1 00 (2 0 1 4 ) 8 –2 4
fragmentation. LC–MS/MS has been used extensively in large scale proteomics in yeast [18,19], bacteria [23,87], fly [26], and human cell lines [29,31]. Nonetheless, the introduction of new linear ion trap — Orbitrap hybrid mass spectrometers has substantially increased resolution (>100,000 full width at half maximum, at least 100× greater than earlier model LTQ instruments), mass accuracy (104 to
