Generation of a mitochondrial protein compendium in Dictyostelium discoideum.

The social ameba Dictyostelium discoideum has emerged as a powerful model to study mitochondrial genetics and bioenergetics. However, a comprehensive inventory of mitochondrial proteins that is critical to understanding mitochondrial processes has yet to be curated. Here, we utilized high-throughput multiplexed protein quantitation and homology analyses to generate a high-confidence mitochondrial protein compendium consisting of 936 proteins. Our proteomic approach, which utilizes mass spectrometry in combination with mathematical modeling, was validated through mitochondrial targeting sequence prediction and live-cell imaging. Our final compendium consists of 936 proteins. Nearly, a third of D. discoideum mitochondrial proteins do not have homologs in humans, budding yeasts, or an ancestral alphaproteobacteria. Additionally, we leverage our compendium to highlight the complexity of metabolic reprogramming during starvation-induced development. Our compendium lays a foundation to investigate mitochondrial processes that are unique in ameba and to understand the functions of conserved mitochondrial proteins in D. discoideum.

Introduction

Dictyostelium discoideum, a social ameba, is a well-established model organism to study eukaryotic cellular processes such as cell motility, chemotaxis, and differentiation (Bozzaro, 2013). Under normal nutrient conditions, D. discoideum grows axenically through binary fission (Kessin, 2001). However, upon starvation, amebae secrete cAMP, which attracts neighboring cells to aggregate together and form a multicellular mound. Cells in a mound move collectively as a slug toward light, heat, or humidity to find a suitable environment. The slug eventually matures into a fruiting body consisting of two major types of differentiated cells, spore cells that will start a new life cycle and stalk cells that form a stalk to hold the spore aloft (Kay, 1982). As many of the aforementioned biological processes are intertwined with cellular energetics, investigation of mitochondrial biogenesis and functions is an emerging area in D. discoideum research (Francione et al., 2011; Pearce et al., 2019).

The D. discoideum mitochondrial genome is ∼56 kb, circular, double-stranded DNA that encodes two ribosomal RNAs, 18 transfer RNAs (tRNAs), five open reading frames without annotated function, and 38 proteins including 18 subunits of the electron transport chain complexes and 15 ribosomal proteins (Ogawa et al., 2000). Phylogenetic studies reveal that Amoebozoa diverged before Opisthokonta, but after Plantae (Baldauf and Doolittle, 1997), and are more closely related to animals than plants. Notably, the Dictyostelium mitochondrial genetic system possesses a few differences from metazoans (Pearce et al., 2019). D. discoideum mtDNA has four introns in the cox1/2 genes and utilizes universal codons (Angata et al., 1995; Ogawa et al., 2000), a common feature of most plants’ mitochondria (Jukes and Osawa, 1990; Cho et al., 1998). The universal genetic code and the lack of a full set of tRNA genes on the Dictyostelium mitochondrial genome indicate that some nuclear-encoded tRNAs are likely imported into mitochondria to support the organellar translation. Additionally, the electron transport chain in Dictyostelium contains an additional component compared to its mammalian counterparts: an alternative oxidase (AOX) (Pearce et al., 2019), which is found across eukaryotic clades except vertebrates (McDonald et al., 2009). AOX is highly expressed during vegetative growth, but its expression level is markedly reduced upon starvation, suggesting that metabolic reprogramming potentially occurs during starvation-induced development (Jarmuszkiewicz et al., 2002). Interestingly, either reduction of mtDNA content or disruption of the rps4 locus (encoding mt-ribosomal protein S4) on mtDNA impairs aggregation and slug phototaxis but has no impact on vegetative growth (Chida, 2004; Chida et al., 2008), suggesting that mtDNA, and most likely an intact oxidative phosphorylation system, is essential to initiate the development program. On the other hand, pharmacological inhibitions of either Complex I or Complex V can induce aggregation, even though mitochondrial respiration appears to increase at the beginning of starvation (Kelly et al., 2021). Therefore, the interplay between mitochondrial function and Dictyostelium development remains to be explored.

Despite the growing interest in using D. discoideum as a model organism to study many conserved mitochondrial processes and some unique biology, a comprehensive list of the mitochondrial proteins has yet to be established. A recent proteomic study detected 294 proteins in D. discoideum mitochondria (Mazur et al., 2021), which we believe is far from complete. Nuclear-encoded mitochondrial proteins, which constitute over 90% of the total mitochondrial proteome, are synthesized in the cytoplasm and then imported to mitochondria. It is estimated that the import of ∼60% of these proteins relies on a positively charged, N-terminal mitochondrial targeting sequence (MTS) (Vögtle et al., 2009). Computational approaches that integrate machine learning and known biological data are frequently used to predict mitochondrial targeting based on the presence of an MTS (Almagro Armenteros et al., 2019). However, this method is insufficient to capture all mitochondrial proteins, as most proteins on the outer membrane and in the inner membrane space, and some inner membrane proteins rely on alternate translocation mechanisms (Bolender et al., 2008). An alternative computational approach leverages sequence homology to known mitochondrial protein compendiums that were generated using mass spectrometry (MS)-based proteomic discovery (Pagliarini et al., 2008; Morgenstern et al., 2017). However, to compensate for the rapid evolution of the mitochondrial genome, nuclear-encoded mitochondrial proteins evolve faster than other nuclear-encoded proteins (Cole et al., 1995; Sloan et al., 2014; Havird et al., 2015; Li et al., 2017; Yan et al., 2019). Thus, some Dictyostelium mitochondrial proteins may escape the homology search, and Amoebozoa-specific mitochondrial proteins will certainly be missed.

In this study, we combined quantitative proteomics and mathematical modeling to identify over 850 high-confidence mitochondrial proteins, which were validated through both in silico and fluorescent microscopy analyses. We further complemented the proteomics-based mitochondrial protein discovery with bioinformatic approaches to create a compendium of 936 D. discoideum mitochondrial proteins. We also discuss conserved D. discoideum mitochondrial proteins that may be used as the basis for validating mitochondrial proteins in other organisms, as well as unique features of the mitochondrial proteome in D. discoideum.

Results and discussion

Mitochondrial protein discovery using quantitative proteomics

To identify putative D. discoideum mitochondrial proteins, we searched for Dictyostelium homologs of 1,136 human mitochondrial proteins listed in the Human MitoCarta 3.0 (Rath et al., 2021), and retrieved 616 proteins (Figure 1A, Table S1). This number is much less than known mitochondrial proteins in humans (1,136) and baker’s yeast (901) (Morgenstern et al., 2017; Rath et al., 2021). We posited that the mitochondrial proteomes might be highly divergent between D. discoideum and humans, and many Dictyostelium mitochondrial proteins might be missed from this bioinformatic curation. We, therefore, took a proteomic approach to directly identify mitochondrial proteins in D. discoideum (Figure 1A). From AX2 axenic cultures, we prepared mitochondria isolates—both crude and highly purified—through Percoll gradient ultracentrifugation. We performed tandem mass tag (TMT) liquid chromatography-mass spectrometry (LC-MS) on both mitochondria isolates and included AX2 whole-cell lysate as the control. A total of 6,892 proteins were captured in all samples (Table S2).

Figure 1

Curation of a comprehensive mitochondrial proteome in Dictyostelium discoideum

(A) A homology search against human mitochondrial protein sequences yielded a list of 616 putative D. discoideum mitochondrial proteins. To identify additional mitochondrial proteins, we performed quantitative MS for proteins identified in the whole-cell lysate, as well as in crude and purified mitochondrial samples. Microscopy was used to validate MS and enrichment analyses. After correcting the proteomic dataset based on microscopy results, we incorporated the homology and proteomic analyses to yield a comprehensive mitochondrial protein compendium.

(B) Volcano plots displaying the −log10 (p value) versus log2 (relative enrichment ratios) of proteins in crude mitochondrial or pure mitochondrial versus whole-cell lysate samples. Data are presented as means (n = 3).

A limitation of identifying organellar proteins from their subcellular fractions alone is that high-abundance contaminants are often co-purified and result in false-positive hits. To address this issue, we assessed the probability of a protein localizing to mitochondria by comparing its relative enrichment in mitochondrial preparations to a list of 47 authentic mitochondrial proteins that include components of electron transport chain complexes and enzymes in the citric acid cycle (Table S3). We first calculated the ratio of a protein’s abundance in the mitochondria isolates, both crude and highly purified, versus its abundance in the whole-cell lysate. The resulting value, indicating its enrichment in mitochondrial preparations, was further normalized to the average enrichment ratio of the 47 reference mitochondrial proteins to compute the relative enrichment ratio (RER). Overall, a protein’s RER in crude mitochondria isolate is in accordance with that in purified mitochondria (Figure 2A). However, the distribution of RERs appears continuous in crude mitochondria (Figure 1B), but clusters into two distinct populations, reflecting their different enrichments in purified mitochondria (Figure 1B), which allowed us to determine a proper RER threshold for mitochondrial proteins using mathematical modeling. Thus, we proceeded to only analyze the purified mitochondria RER.

Figure 2

Prediction of mitochondrial localization based on relative enrichment analysis

(A) Crude mitochondrial RER versus pure mitochondrial RER for all samples quantified in the proteomic discovery experiment. The dashed line represents equal crude and pure RERs. Data are presented as means (n = 3).

(B) Distribution of core D. discoideum mitochondrial ETC, OXPHOS, and TCA cycle proteins (n = 47) based on RER.

(C) Summary table of pure mitochondrial relative enrichment ratios for all proteins in the protein discovery experiment.

(D) The distribution of mitochondrial (n = 861) and non-mitochondrial proteins (n = 6031) based on RER follows a Gaussian mixture model, in which a cutoff value (red, RER = 0.392) separates two normal curves.

(E) An RER cutoff of 0.392 (red) effectively separates proteins based on the distribution of proteins with a predicted mitochondrial targeting sequence (MTS, n = 235) or, as a negative control, a signal peptide sequence (SPS, n = 565). The clustering also fits the distribution of proteins with no sequence prediction (Other, n = 6092).

In principle, a true mitochondrial protein would be co-purified with the reference mitochondrial proteins in pure mitochondrial isolates, and its RER should be 1.0. However, the RER distribution of these 47 reference proteins (Figure 2B) appears as a normal curve centered around 1.0, suggesting that many mitochondrial proteins may have an RER below 1.0. Among all proteins profiled using TMT-based LC-MS, only 259 have an RER higher than 1.0 in purified mitochondria (Figure 2C). We posit that mitochondrial proteins might be degraded to different extents, based on their intrinsic stability, during the procedure of mitochondrial purification, which involves overnight ultracentrifugation. Therefore, it is necessary to determine a proper RER value to differentiate mitochondrial proteins from non-mitochondrial proteins. We applied a Gaussian mixture model (GMM) to bin all proteins into two clusters: non-mitochondrial and mitochondrial proteins based on their RER values (Figure 2D). We chose an RER cutoff of 0.392 (Figure 2D, Table S4), such that all proteins with an RER greater than or equal to this value had a 100% probability of belonging to the mitochondrial cluster. Based on this analysis, a total of 861 proteins were assigned as putative mitochondrial proteins (Table S5).

Validation of mitochondrial protein discovery based on quantitative proteomics

To validate the accuracy of RER-based mitochondrial protein discovery, we first assessed the recovery rate of putative mitochondrial proteins in silico. Many mitochondrial proteins possess an N-terminus mitochondrial targeting sequence (MTS) that directs the import of nuclear-encoded mitochondrial proteins into the mitochondrial matrix (Backes and Herrmann, 2017). Overall, 3.4% of all proteins recovered in the proteomics experiment contain an MTS (Table S5). Importantly, we recovered 94% of these MTS-bearing proteins using the RER cutoff of 0.392 (Figure 2E, MTS). As a negative control, 2.2% of proteins that were destined to other organelles such as the ER, Golgi, lysosomes, vacuoles, or secretory pathway (SP) had an RER greater than or equal to 0.392 (Figure 2E, SPS). Furthermore, all remaining proteins with no sequence prediction also fit into these clusters (Figure 2E, other). These analyses demonstrate that an RER cutoff value of 0.392 effectively separates mitochondrial proteins from non-mitochondrial proteins.

We also surveyed the localization of 81 proteins recovered in LC-MS (Table S6), using fluorescent microscopy. These proteins were selected on the basis that their subcellular localization has not been annotated previously as mitochondrial, and their RERs are randomly distributed from 0.1 to 1.5. Each protein was tagged with GFP at its C-terminus and co-expressed with an MTS-mCherry fusion protein, which marks mitochondria in D. discoideum AX2 cells. Among the 81 proteins, 91% of proteins with an RER higher than 0.392 showed complete or partial mitochondrial localization (Figures 3A and 3C), demonstrating a strong positive correlation between RER value and probability of mitochondrial localization (Figure 3A). Moreover, logistic regression analysis on the localization pattern of these 81 proteins predicts that an individual protein has more than a 79% probability of localizing to the mitochondria if its RER is higher than 0.392 (Figure 3B). Although the microscopy analysis demonstrated that few non-mitochondrial proteins were recovered in our compendium, 12% of proteins with an RER less than 0.392 showed mitochondrial localization (Figure 3A).

Figure 3

Validation of proteomic discovery and enrichment analyses

(A) Dot plot with interquartile ranges of proteins that had mitochondrial localization (green, n = 35), partial mitochondrial localization (gray, n = 18), or were excluded from the mitochondria (red, n = 28). ∗, p ≤ 5 × 10−5.

(B) Predicted probabilities and confidence intervals as analyzed by logistic regression for mitochondrial (green), partial mitochondrial (gray), and either complete or partial mitochondrial (blue) localization. The dashed line represents the RER cutoff for predicting mitochondrial localization (0.392).

(C) Representative live-cell confocal images of vegetative stage D. discoideum expressing GFP-tagged genes of interest leveled with cytochrome oxidase c subunit IV tagged with mCherry (Mito-mCherry). Scale bars: 5.00 μm. Arrowheads denote areas of partial localization.

A comprehensive mitochondrial protein compendium in D. discoideum

To further improve the coverage and accuracy of our mitochondrial protein discovery, we first revised the proteomics list based on the in vivo microscopy validation. We removed three non-mitochondrial localizing proteins and added five mitochondrial localizing proteins.

We also examined the proteins with a 95% or greater predicted probability of localizing to the mitochondria based on the GMM (RER Table S4). For these 66 proteins, we retrieved five that had a predicted MTS and/or were homologs of human mitochondrial proteins (Tables S1 and S5). Additionally, there were 41 D. discoideum homologs of human mitochondrial proteins that were not captured in LC-MS. For these 46 proteins, we removed any that demonstrated more sequence homology to a non-mitochondrial human protein than a protein in the Human MitoCarta3.0. We posit that these D. discoideum proteins may share a conserved domain, and therefore emerged in the homology analyses, but do not share the same function. We also removed any proteins that were non-mitochondrial based on previous evidence. We added the remaining 21 proteins to the list, as well as 52 proteins with mitochondrial gene ontologies that had not emerged during the proteomic or homology analysis. The final compendium consists of 936 high-confidence mitochondrial proteins in D. discoideum (Table S7).

Characterization of the D. discoideum mitochondrial proteome

Out of the 936 D. discoideum mitochondrial proteins, 504 and 404 proteins have homologs in the mitochondrial proteome of humans and Saccharomyces cerevisiae, respectively (Table 1). Only 286 D. discoideum mitochondrial proteins have homologs in Rickettsia prowazekii (Table 1), an α-proteobacterium that is closely related to the mitochondrial ancestor. Overall, a total of 300 proteins, representing 32.1% of the D. discoideum mitochondrial proteome, have no homologs in the whole proteome of humans, S. cerevisiae, or R. prowazekii (Table S7), indicating that a large fraction of D. discoideum mitochondrial proteins were evolved de novo after the divergence of Amoebozoa. Moreover, 74 D. discoideum mitochondrial proteins (7.9%) have no homologs in Dictyostelium purpureum (Table 1), a closely related species of social ameba, further substantiating the fast-evolving nature of the ameba mitochondrial proteome.

Table 1

Comparison of mitochondrial proteome in D. discoideum to other organisms

	D. discoideum (936)
H. sapiens (1,136)	504
S. cerevisiae (901)	404
R. prowazekii (832)	286
D. purpureum (?)	862

D. discoideum mitochondrial proteins with homologs in H. sapiens or S. cerevisiae mitochondrial proteomes, or the complete proteomes of the alphaproteobacteria R. prowazekii or D. purpureum. “?” reflects that the number mitochondrial proteins in D. purpureum is currently unknown.

There are also 77 D. discoideum mitochondrial proteins (8.2%) with human homologs that had not been annotated as mitochondrial proteins (Table 2, Table S7). Among these 77 proteins, 62 were also not annotated as mitochondrial proteins in budding yeast, including 27 that had homologs in S. cerevisiae. Given the estimated false-discovery rate of our compendium, the localization of these proteins needs to be experimentally assessed. Nonetheless, there are a few examples, such as the RNB domain-containing protein (DDB_G0288469) and tRNA-binding domain-containing protein (DDB_G0349377), both of which have a predicted MTS and are likely targeted to the mitochondrial matrix. The human homologs of DDB_G0288469, DIS3-like exonuclease 2, and DDB_G0349377, rhomboid-related protein 4, were not included in the human compendium (Rath et al., 2021), despite evidence that the yeast homolog of DIS3-like exonuclease two localizes to the mitochondria (Morgenstern et al., 2017), and that rhomboid-related protein 4 has been partially shown to localize to the mitochondria. The mitochondrial localization of their D. discoideum homologs substantiates these two proteins might indeed localize to the mitochondria and indicates that our compendium can complement previous studies toward a more comprehensive discovery of mitochondrial proteins in other organisms.

Table 2

The functional categorization of all D. discoideum mitochondrial proteins

	overall	unique	diff. localized
protein import & biogenesis	25	8	2
protein maturation & folding	24	13	0
protein degradation	26	1	6
DNA-related processes	39	16	8
mRNA-related processes	10	8	1
tRNA-related processes	44	6	2
ribosome & translation	91	34	2
others of gene expression	29	14	1
respiratory chain assembly	41	11	0
respiratory chain subunits	62	14	2
energy metabolism	44	6	2
lipid metabolism	45	5	5
nucleotide metabolism	5	0	0
amino acid metabolism	37	2	2
coenzyme/cofactor biosynthesis	17	1	1
Fe/S proteins biosynthesis	15	0	0
carriers & channels	44	5	0
regulation & signaling	26	6	10
morphology & dynamics	25	3	6
homeostasis & stress response	29	6	1
developmental processes	14	5	2
Others	95	63	11
Unclassified	150	128	13
Total	936	355	77

The number of proteins in each biological process for all mitochondrial proteins (overall, n = 936), those that lack human homologs (unique, n = 355), and those that have human homologs, but homologs were not annotated as mitochondrial proteins (diff. localized, n = 77) are indicated.

Additionally, we categorized the D. discoideum mitochondrial proteome using PANTHER biological function or protein family classifications (Table 2, Table S7). Proteins involved in mitochondrial gene expression and metabolism comprise the largest fractions of all mitochondrial proteins, over 22% for each category. Other proteins are involved in mitochondrial protein homeostasis, the electron transport chain, redox signaling and metabolism, and regulation of mitochondrial morphology and dynamics. A large fraction of D. discoideum mitochondrial proteins, approximately 16%, have no classified functions based on PANTHER analyses (Table 2).

D. discoideum-specific mitochondrial proteins

Different from human mtDNA, D. discoideum mitochondrial genome contains several introns and has numerous overlapping genes, which require complex post-transcriptional processing. Consistent with this notion, proteins involved in gene expression consisted of a large fraction of D. discoideum-specific mitochondrial proteome (Table 2). On the contrary, few metabolism proteins emerged in the list (Table 2), suggesting that metabolic processes are highly conserved between D. discoideum and metazoans. Here, we expand upon a few of the unique features of the D. discoideum mitochondrial protein compendium.

Mosaic nature of mitochondrial ribosomes

Mitochondrial ribosomes (mitoribosomes), ribosomal assembly factors, and other proteins involved in translation represented 9.7% and 9.6% of the overall and unique mitochondrial protein compendium, respectively. While mitoribosomes are thought to be evolved from bacterial ribosomes, these two differ greatly regarding their structure, function, as well as their composition of proteins and RNAs. We identified 50 proteins that are predicted to be mitoribosomal proteins, including 13 proteins that did not share significant homology with any H. sapiens, S. cerevisiae, or R. prowazekii proteins (Table S8). Interestingly, D. discoideum mitoribosomal proteins belong to families across several taxonomic groups: 35 proteins belong to mammalian mitoribosomal protein families (28 and 39S), one belongs to eukaryotic cytosolic ribosomal protein families (60S), nine belong to yeast mitoribosomal protein families (37 and 54S), two belong to chloroplast or bacterial ribosomal protein families (30 and 50S), one is from archaea, and two are universally conserved among prokaryotes and eukaryotes. In D. discoideum, it has previously been shown that cytosolic ribosomes tether to the mitochondrial outer membrane for cotranslational import (Ahmed et al., 2006). Hence, the recovery of the 60S ribosomal protein L6 could be the result of the association of cytoplasmic ribosome with the mitochondrial outer membrane rather than its localization in the matrix (Gold et al., 2017). Nonetheless, the presence of proteins representing multiple mitoribosome lineages suggests that there may be D. discoideum- or Amoebozoa-specific mechanisms to process mitochondrial transcripts and regulate mitochondrial translation. Further validation of these findings is necessary as the composition and structure of the D. discoideum mitoribosome have yet to be resolved.

Mitochondrial DNA and RNA processing factors

Among the list of unique proteins are 49 candidate mtDNA and mtRNA processing factors including six endonucleases and a pentatricopeptide repeat (PPR)-containing protein A (PtcA, DDB_G0293882). Bioinformatic analysis suggests that PtcA belongs to the mitochondrial group I intron splicing family. PPR proteins, defined by tandem PPR domains, are implicated in several different mitochondrial gene expression processes including translation initiation, and ribosomal stabilization (Manna, 2015). The number of PPR proteins that are encoded in an organism varies greatly: terrestrial plants, such as Arabidopsis thaliana, have upwards of 450 PPR proteins, while humans have 7 (Lurin et al., 2004; Lightowlers and Chrzanowska-Lightowlers, 2013). D. discoideum has 12 PPR domain-containing peptides, including PtcA, reflecting a greater complexity of D. discoideum’s mitochondrial genome compared to that of metazoans (Manna et al., 2013).

Divergent evolution path of lipid biosynthesis

The mevalonate pathway, which produces five-carbon blocks for the synthesis of diverse biomolecules such as cholesterol and coenzyme Q10, is an essential and highly conserved process in eukaryotes, archaea, and some bacteria. In animals and fungi, the mevalonate pathway takes place in ER, and 3-hydroxy-3-methylglutaryl (HMG)-coenzyme A (CoA) reductase (HMGR), a key enzyme in this pathway that converts HMG-CoA to mevalonate, localizes in the ER and peroxisomes (Chin et al., 1984; Keller et al., 1986; Burg and Espenshade, 2011). HMGR2, one of two HMG reductases in D. discoideum, is recovered in our compendium and contains a predicted MTS, suggesting that it likely localizes to the mitochondrial matrix. Likewise, HGSA, one of the two HMG-CoA synthases, also emerged as a mitochondrial protein. Additionally, the isozymes of HMGR2 and HGSA—HMGR1 and HGSB, respectively—were not recovered in our compendium and have predicted signal peptide sequences. Taken together, this suggests that mevalonate metabolism may redundantly take place in the mitochondria in addition to in the ER in D. discoideum, highlighting the evolutionary divergence of some metabolic pathways that originated from the common mitochondrial ancestor.

Implication of mitochondrial function in multicellular development

D. discoideum with reduced mtDNA or a disruption of the gene encoding mt-ribosomal protein S4 displays no defect in vegetative growth but has impaired starvation-induced development, suggesting that mitochondrial respiration is necessary for multicellularity (Chida, 2004; Chida et al., 2008). However, contrasting evidence has demonstrated a significant decrease in mitochondrial respiration after starvation, and accordingly, a decreasing expression of many respiration complexes (Kelly et al., 2021). To understand potential regulations of mitochondrial function in multicellular development, we retrieved RNA sequencing data using the Dictyostelium gene expression database, dictyExpress (Table S9; Parikh et al., 2010; Stajdohar et al., 2017).

Overall, there was a decrease in the expression of mitochondrial genes within our compendium over the 24-h developmental time course (Figure 4A). A similar pattern is observed in proteins that are involved in mitochondrial DNA maintenance and gene expression including many ribosomal proteins. Interestingly, despite the decrease in gene expression machinery (Figures 4B and 4C), 80% of the mitochondria-encoded genes in the dataset (32 of 40) were upregulated (log2FC ≥ 1) in the first 4 h after starvation induction (Figure 4D). Furthermore, in examining all respiratory chain complexes, seven nuclear-encoded ETC subunits had a higher expression level (log2FC ≥ 1) at or after 8 h of starvation, besides the 11 nuclear-encoded subunits that show a burst of expression in the first 4 h after the starvation (Figure 4E, Table S9). The complex pattern of mitochondrial gene expression, particularly the upregulation of electron transport chain complex subunits during the development suggests potential roles of mitochondrial respiration in Dictyostelium development, and that both nuclear and mitochondrial-encoded proteins are likely implicated in these processes. Future characterization of the mitochondrial proteome at different developmental stages would be valuable to complement these findings, as changes in a gene’s expression do not always correspond with its protein level or activity (Kotsifas et al., 2002).

Figure 4

Expression profile of mitochondrial genes during D. discoideum development

Heatmaps showing normalized RNA-seq analysis (Z score) values of (A) all mitochondrial proteins within the compendium and (B–D) specific mitochondrial proteins during the 24-h starvation-induced development cycle. Each column of heat maps represents the time in hours after developmental induction. Rows are clustered by similarity in gene expression profiles.

In conclusion, we generated the most comprehensive list of mitochondrial proteins in D. discoideum to date. Our compendium lays the foundation for future studies to understand the functions of conserved mitochondrial proteins in health and diseases using D. discoideum as the model. It also provides an entry to study many fascinating mitochondrial processes that are unique in ameba. Additionally, through comparative genomics, our compendium will complement mitochondrial protein discovery in other organisms and may shed light on the evolution of the mitochondrial proteome and processes.

Limitations of the study

To reduce the false discovery of non-mitochondrial proteins, we set a stringent RER cutoff at 0.392 based on the mathematical modeling. Our microscopy-based validation demonstrates that several proteins with RERs less than 0.392 actually localize to mitochondria (Figure 3A), while a few others with RER above this threshold do not (Figure 3A). Although the in silico dataset correction improved the coverage of our compendium, some bona fide mitochondrial proteins without homologs in the Human MitoCarta 3.0 that have RER values below the cutoff would be missed in the compendium. In addition, our proteomic discovery method does not effectively discern mitochondria-associated proteins from bona fide mitochondrial proteins that localize to the matrix, intermembrane space, or are inserted in the inner or outer membrane. Future updates to our compendium should examine the submitochondrial localization of the inventoried proteins. Lastly, our mitochondria protein discovery is limited to the vegetative stage of D. discoideum. Future characterization of the mitochondrial proteome in other developmental stages would lead to an improved understanding of mitochondrial regulation of D. discoideum development and differentiation.

STAR★Methods

Key resources table

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Hong Xu (hong.xu@nih.gov).

Material availability

All plasmids generated in this study are available upon request.

Experimental model and subject details

D. discoideum AX2 cells were obtained from the Jin lab and originally derived from the Carole Parent lab, National Cancer Institute, NIH. Cultures were maintained in HL5 medium (Formedium) at 22°C (Fey et al., 2007).

Method details

D. discoideum transformation

Transformants were generated via electroporation as previously described (Gaudet et al., 2007), with modifications. After electroporation (BioRad Genepulser), cells were incubated on ice for 10 min. Subsequently, cells were transferred from the cuvette with 2 mL of HL5 and plated onto 12-well tissue culture plates. After 24 h, transformants were selected with Genectin and/or Blasticidin S (Thermo Fisher Scientific, 10 μg/mL each) in HL5.

Protein mass spectrometry

Mitochondrial isolation

Cells were harvested at a concentration of 1-3 x 106 cells/mL and resuspended at 2 × 107 cells/mL in 800 μL of Reagent A from the Mitochondrial Isolation Kit for Cultured Cells (Thermo Fisher Scientific) on ice. Cell lysis and crude mitochondrial preparation were performed as previously described with modifications (Glancy and Balaban, 2011; Graham, 1999). Cells were lysed with 35 strokes of a Dounce homogenizer followed by the addition of an equal volume of Reagent C. Whole-cell lysate samples were stored at -80°C or were centrifuged three times (700 x g, 10 min, 4°C) to purify the mitochondria. For each centrifugation step, the supernatant was transferred to a fresh 1.5 mL tube. The crude mitochondrial lysate was used immediately for purification or was stored at -80°C.

To generate purified mitochondrial isolates, Percoll gradient centrifugation was performed as follows. Lysis suspension (1-2 mL) was added to the top of a Percoll (Cytiva) and Development Buffer (DB) (5 mM Na2HPO4, 5 mM KH2PO4, 1 mM CaCl2, 2 mM MgCl2, pH = 6.5) solution (8 mL, 30% Percoll) in a 10 mL ultracentrifuge tube. Ultracentrifugation (68,000 x g, 40 min) yielded three distinct layers. The top layer, containing contaminants, was discarded. The middle layer, containing mitochondria, was transferred into a fresh 2 mL tube. Aliquots of the mitochondrial suspension were topped off with 500 μL of DB, then centrifuged (13,000 x g, 10 min, 4°C). Following centrifugation, the supernatant was aspirated, and the mitochondria-containing pellet was maintained on ice. To lyse the mitochondria, the pellet was washed with 2 mL of DB, centrifuged (10,000 x g, 10 min, 4°C), and resuspended in 2 mL of DB with urea (8M). Protein yield was quantified via Bradford Assay (BioRad) according to the manufacturer’s protocol.

Relative protein quantification

Resuspended cell pellets were lysed via pulsed sonication, then sequentially reduced, alkylated, and digested overnight with trypsin. The protein digests were labeled with 10-plex Tandem Mass Tag (TMT) reagents (Thermo Fisher Scientific) (Dayon et al., 2008), then were pooled and desalted. To separate the peptide mixtures into 24 fractions, high pH reversed-phase liquid chromatography was performed (Yang et al., 2012). Each fraction was analyzed on an Orbitrap Lumos (Thermo Fisher Scientific) nanoLCMS system.

Peptide and proteins were identified as described in He et al. (2020). In brief, the resulting LCMS raw data were searched against a database downloaded from dictybase.org using the Sequest HT algorithm on the Proteome Discoverer 2.4 platform (Thermo Fisher Scientific). Three groups of samples were normalized to 47 reference mitochondrial proteins.

Relative enrichment (RE) was defined as the ratio of a protein’s enrichment from a crude or purified mitochondria sample over its enrichment from a whole-cell lysate sample. To calculate the relative enrichment ratio (RER), REs were normalized such that the average RER of 47 known mitochondrial (TCA cycle, ETC, or OXPHOS) proteins is 1 (Table S3). The RER presented is a median value of three biological replicates.

Mathematical modeling

To classify proteins as mitochondrial or non-mitochondrial based on their RER, the RER distribution of isolated proteins was fit to a two-component Gaussian mixture model (GMM) in R (Yu, 2022). All proteins with an RER greater or equal to the cutoff value (RER = 0.392) have a 100% posterior probability of belonging to the mitochondrial cluster (Table S4).

Bioinformatic analyses

Homology analyses

Protein sequence homology was established by BLAST+ (Altschul et al., 1990, 1997; Camacho et al., 2009) expect <0.001 and bit-score > 40 (Pearson, 2013), or by HMMER sequence e-value < 0.01. Subcellular localization was predicted using TargetP-2.0 (Almagro Armenteros et al., 2019). Biological functions for all proteins in the D. discoideum and human mitochondrial proteome were manually categorized from biological function or protein family classifications provided by PANTHER (The Gene Ontology Consortium et al., 2021).

In silico dataset correction

Two strategies were implemented to supplement the mitochondrial protein discovery. The top D. discoideum homolog of human mitochondrial proteins (Table S1, Rath et al., 2021) were curated. Proteins within this list were integrated into the final mitochondrial compendium if they also had a 95% or greater predicted probability of localizing to the mitochondria based on the mixture modeling (Table S4), unless one of the following criteria were met. First, the protein had a human homolog that was non-mitochondrial with a higher degree of sequence identity, indicating that the D. discoideum protein likely had a similar domain as the human MitoCarta protein but had a different function. Second, the literature indicated that the protein was nonmitochondrial or the homology evidence was a poor indicator of a similar function. Additionally, D. discoideum proteins annotated with the gene ontology term “mitochond∗” on AmiGO (Carbon et al., 2009) or dictyBase were selected.

RNA sequencing data normalization and visualization

Normalized RNA-seq data from Parikh et al. (2010) was retrieved using dictyExpress (Stajdohar et al., 2017). For the 936 proteins in the mitochondrial compendium, only 932 corresponding genes were present in the dataset. To compare the gene profiles, data were normalized (z-score) to a mean of 0 and standard deviation of 1 using the scale function in R. For the overall mitochondrial expression profile, genes and timepoints were ordered using hierarchical clustering (heatmaps.2). For the profiles of individual biological processes, scaled data (Table S9) were imported into the matrix visualization software Morpheus (http://software.broadinstitute.org/morpheus) and ordered via hierarchical clustering with one minus Pearson’s correlation as the distance metric and average as the linkage method. Gene upregulation was determined by a log2 fold-change ≥ 1 of non-normalized values.

Library generation for imaging verification

To evaluate the efficacy of our mitochondrial protein identification, 98 proteins were selected to be GFP tagged so that their localization could be assessed via fluorescence microscopy (Table S6). None of the proteins selected had a gene ontology annotation that indicated mitochondrial localization. Four proteins selected for verification had homologs listed in the human mitochondrial proteome.

All genes were synthesized by Gene Universal. Of the 98 genes submitted for synthesis, 85 were generated as inserts in pDM323: a D. discoideum extrachromosomal expression vector with G418 resistance and a C-terminal GFP tag (Veltman et al., 2009); 6 genes were generated as inserts in the shuttle vector puC57 and were cloned into pDM323 between BglII and SpeI sites using the In-Fusion HD Cloning Kit (Takara Bio USA) and confirmed by sequencing. Fragment or plasmid purification was performed using the Zymoclean Gel DNA Recovery Kit (Zymo Research) or High Efficiency 5α Competent E. coli (New England Biolabs) in combination with the QIAprep Spin Miniprep Kit (Qiagen).

The other 7 genes were unable to be synthesized, such that only 92 proteins were screened. Of these 92 genes, only 81 were successfully expressed in D. discoideum. The RER of the proteins that were verified were as follows: 14 proteins with a RER > .75, 11 proteins with a RER = 0.75-0.5, 24 proteins with a RER = .5-0.25, and 32 proteins with RER < 0.25. To observe mitochondrial localization, the mitochondrial targeting sequence of respiratory cytochrome oxidase c subunit IV fused with mCherry (Mito-mCherry) was cloned into pDM326, a D. discoideum extrachromosomal expression vector with Blasticidin S resistance (Veltman et al., 2009). All primers for cloning are listed in Table S10.

Live-cell imaging

To image cells in the axenic phase, cells (200 μL) were transferred to 8-well glass chambers 7 to 10 days post-transformation. Cells were allowed to adhere to the bottom of the chamber for 30 min before the media was aspirated. The media was replaced with 1x PBS after three washes (200 μL for all). Confocal images were collected on a PerkinElmer Ultraview system (Zeiss Plan-apochromat 63x/1.4 oil lens, Volocity acquisition software, Hamamatsu Digital Camera C10600 ORCA-R2, Immersol immersion oil). Images (0.5 μm z-step) were analyzed with ImageJ (National Institutes of Health) and formatted in Adobe Photoshop and Adobe Illustrator.

Quantification and statistical analysis

All data were presented as the mean ± SD unless otherwise indicated. p values were calculated in R using a one-way analysis of variance (ANOVA) followed by Tukey’s post-hoc test to test for the effect of RER on mitochondrial localization. Statistical significance of difference was considered when p < 0.05.

To predict the probability of localization based on RER, outliers were identified and removed from the microscopy validation dataset based on the interquartile method (median +1.5 SD). Data were analyzed using a binomial logistic regression (glm function in R) with excluded from the mitochondria as the reference level, and partial mitochondrial localization, mitochondrial localization, or combined (in which the partial and mitochondrial outcomes are collapsed) as the outcome levels. The fit of each binomial logistic regression was compared via McFadden’s pseudo-R2 (nagelkerke function in R) to determine which predicted probability to report. Predicted probabilities and 95% confidence intervals were calculated (predict function in R) to compare outcomes and were reported in the text for the model with the highest pseudo-R2.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
5α Competent E. coli (High Efficiency)	New England Biolabs	Cat#C2987U
Chemicals, peptides, and recombinant proteins
HL5 medium including glucose	Formedium	Cat#HLG0102
Genectin Selective Antibiotic	Thermo Fisher Scientific	Cat#10131035
Blasticidin S HCl	Thermo Fisher Scientific	Cat#A1113902
Percoll density gradient media	Cytiva	Cat#17089101
Sodium Phosphate Dibasic	Sigma-Aldrich	Cat#S0876
Potassium Phosphate Monobasic	Sigma-Aldrich	Cat#P5655
Calcium Chloride 2M	Quality Biologicals	Cat#351-130-721
Magnesium Chloride 1M	Quality Biologicals	Cat#351-033-721
Urea	Sigma-Aldrich	Cat#U5128
Critical commercial assays
Mitochondrial Isolation Kit for Cultured Cells	Thermo Fisher Scientific	Cat#89874
Protein Assay Kit	Bio-Rad	Cat#5000001
TMT10plex Isobaric Label Reagent Set	Thermo Fisher Scientific	Cat#90110
In-Fusion HD Cloning Kit	Takara Bio USA	Cat#638920
QIAprep Spin Miniprep Kit	Qiagen	Cat#27106X4
Zymoclean Gel DNA Recovery Kit	Zymo Research	Cat#D4002
Deposited data
Raw and analyzed proteomic data	This paper	PRIDE ProteomeXchange: PXD029101
Code for statistics and mathematical modeling	This paper	Zenodo: http://doi.org/10.5281/zenodo.6471311
Original images and supplementary tables	This paper	Mendeley Data: https://doi.org/10.17632/yhjnbnsgz3.1
Experimental models: Organisms/strains
D. discoideum: AX2 Parent strain	Jin lab, National Institute of Allergy and Infectious Diseases, NIH	N/A
Oligonucleotides
Primers for cloning	This paper, Integrated DNA Technology	Table S10
Primers for sequencing	This paper, Integrated DNA Technology	Table S10
Recombinant DNA
GFP plasmid	dictyBase	pDM323
mCherry plasmid	dictyBase	pDM326
Gene libraries for verification	This paper, Gene Universal	Accession numbers in Table S6
Software and algorithms
Adobe Photoshop and Illustrator	Adobe	http://www.adobe.com/creativecloud.html
AmiGO	Carbon et al. (2009)	http://amigo.geneontology.org
BLAST+	Camacho et al., 2009; Altschul et al., 1990, 1997	http://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/
dictyExpress (RNA sequencing data)	Stajdohar et al., 2017; Parikh et al., 2010	http://dictyexpress.research.bcm.edu/landing/
GraphPad Prism 9	GraphPad	http://graphpad.com
HMMER v3.3.2	Howard Hughes Medical Institute	http://http://hmmer.org
ImageJ	National Institutes of Health	http://imagej.nih.gov/ij/
mixR	Yu (2022)	http://github.com/GaryBAYLOR/mixR.git
Morpheus	The Broad Institute	http://software.broadinstitute.org/morpheus
PANTHER	The Gene Ontology Consortium	http://pantherdb.org
Proteome Discoverer v2.4	Thermo Fisher Scientific	http://www.thermofisher.com/us/en/home/industrial/mass-spectrometry/liquid-chromatography-mass-spectrometry-lc-ms/lc-ms-software/multi-omics-data-analysis/proteome-discoverer-software.html
RStudio v1.4.1717	RStudio	http://www.rstudio.com
TargetP-2.0	Almagro Armenteros et al., 2019	http://services.healthtech.dtu.dk/service.php?TargetP-2.0
Volocity Acquisition	Quorum Technologies Inc.	http://www.volocity4d.com/volocity-acquisition