Mitochondrial Microproteins from Discovery to Function

Abstract

Because of the uniqueness of the mitochondria, and the fact that they have their own genome, mitochondrial microproteins (MDPs) are similar to but different than nuclearly encoded microproteins. The discovery of more and more microproteins from this organelle and the importance of mitochondria to cellular and organismal health make it a priority to study this novel class of proteins in search of possible therapeutic targets and cures. This review covers the history of mitochondrial microprotein discovery, describes the function of each MDP, and concludes with future goals and techniques to help discover more MDPs.

Mitochondrial Microproteins

The field of microproteins has been growing at an exponential rate with new microproteins being discovered in both the nuclear and mitochondrial genome. In our review we will focus on microproteins encoded by the mitochondrial genome. Historically, these microproteins were called mitochondrial derived peptides (MDPs), but the field has moved to describe them as mitochondrial derived microproteins (still MDPs to match historical precedence) to meet nomenclature standardization. We will review both the techniques used to discover these novel microproteins as well as their biological implications and function.

Mitochondria and mitochondrial genome

The mitochondria are special from a genetic point of view in that they are the only organelle that has their own, circular genome, and their genome is uniparentally inherited from the mother in most organisms (bivalves being a notable exception[1]). The genome is the last remnants of its past life as an independent, archaea bacteria. Indeed, mitochondria have their own translational machinery and use a different genetic code than the nuclear genome. It is also highly compact with no exons, and it is believed to be transcribed mainly as an operon with many genes transcribed as a single, long transcript [2]. Classically the mitochondrial genome (mtDNA) codes for only 13 proteins, 22 tRNAs, and 2 rRNAs, but this view is changing with the discoveries made in our lab and others (Figure 1). Specifically relevant to MDPs, there is still controversy about which codons are used as stop codons in the mitochondria. Although it was previously thought (circa 1981) that the AGA and AGG codons coded for stop codons due to the lack of a corresponding arginine tRNA, more recent papers have begun to dispute this and suggest that these codons instead cause a frameshift leading to usage of the normal stop codons [3-7]. Furthermore, two papers (one in Biorxiv) from the Breton lab have offered evidence that tRNA-arginine may be imported into the mitochondria [4,8]. While this controversy only affects two of the classical mitochondrial proteins, it has major implications for identification of future, mitochondrially translated MDPs and should be considered when performing future experiments.

Figure 1: Annotation of the mitochondrial genome.

The outer ring is labeled the classical genes (protein coding genes are grey, tRNAs are blue, the rRNAs are yellow, and the non-coding region/control region are peach). The 2^nd ring is the known MDPs (labeled in orange). The Inner rings are a map of all the theoretically possible open reading frames for microproteins.

From a biological point of view, the mitochondria are critical in a number of different cellular functions and physiological diseases. While mitochondria are known as the “powerhouse of the cell” because they make the majority of all the ATP in the cell, they are also involved in other cellular functions such as uridine synthesis, steroid synthesis [9], apoptosis, and calcium storage [10]. Mitochondrial dysfunction has been implicated in any number of diseases including Alzheimer’s disease (AD), cancer, diabetes, and obesity, although it is not completely known whether these are causative or associative relationships [11]. As central players in cellular and physiological processes, understanding the role of mitochondrial microproteins is critical to both basic biology and translational research looking for possible therapeutics.

The discovery of humanin and the origin of mitochondrial microproteins

Microproteins are encoded by small open reading frames (smORFs) and are typically defined as proteins under 100 amino acids in length [12]. Interestingly, by this definition, 2 of the 13 proteins encoded by the mtDNA would be considered microproteins (MT-ATP8 and MT-ND4L). Coincidentally, the open reading frames of both of these microproteins are the only two that also overlap other genes (MT-ATP6 and MT-ND4 respectively) but are translated in a different reading frame. Almost half of protein-coding genes contain smORFs in their 5’ upstream elements (uORFs), and several annotated long non-coding RNAs (lncRNAs) also contain actively translated smORFs. Despite their prevalence, the protein-coding potential of smORFs has often been deprioritized in functional biology experiments, primarily because most smORFs are evolutionarily young. However, over the past two decades, advancements in sequencing technology have revealed that both old and young smORFs can produce functional microproteins with specific biological mechanisms. One of the earliest discovered microproteins is humanin, which is encoded by the mtDNA [13].

Another unique feature of mtDNA is that there are many degenerate copies and fragments of it found scattered in the nuclear genome. In fact there are over 750 Nuclear-Mitochondrial Sequences (NUMTS) found in the human nuclear genome [14,15]. These NUMTS have been shown to express humanin-like proteins, and future studies will need to be conducted to see if this holds true for other MDPs as well [16]. In addition to NUMTS, whether an MDP transcript is translated in the mitochondria or the cytoplasm is another question. For humanin, although we have shown in rats that it is primarily translated in the mitochondria [17], it is unknown in other species. Because of the difference in codon usage between the mitochondria and cytoplasm, humanin could be a 24 or 21 amino acid MDP, and the other MDPs would also have a different sequence.

The discovery of humanin marked a pivotal milestone in the field of microprotein research. Leveraging death-trap screening results, which utilized a cDNA library constructed from the occipital region of an AD-patient brain, the DNA fragment encoding humanin was initially detected as a novel suppressant of cell death [18-20]. Death-trap screenings identify antagonist genes that protect against deadly insults. In this case, by sequencing recovered plasmids in cells, which survived cell death treatment, the 75-base small open-reading frame (sORF) encoding the 24-residue microprotein was found as a crucial antagonist against V6421-APP induced cell death. Humanin thus became the first mitochondrial-derived microprotein ever to be discovered and characterized and joined a growing class of putative smORF-encoded microproteins.

Since its initial discovery as a neuroprotective molecule, humanin has been largely characterized as a cytoprotective microprotein with potent anti-apoptotic effects [21,22]. Interactions between humanin and Bax, a known regulator of apoptosis, underscored a key pathway through which humanin protects against cell stress [21]. The independent discovery of humanin through binding interactions with IGFBP-3, a potential transporter, further elucidated the function of humanin, suggesting its importance as a novel link between metabolism and AD [23]. Continued exploration of humanin as a mechanistic link between metabolic disorder and AD progression posited the molecule as a crucial regulator of peripheral insulin action, primarily through STAT-3 activation [24]. Additional analyses which considered human mitochondrial genetics proved important for determining humanin function.

Humanin exhibits potential as a biomarker for aging and age-related diseases. We have demonstrated that circulating humanin levels decrease in aged non-human primates and are elevated in offspring of centenarians compared to age-matched controls [25]. Furthermore, carriers of the alternative allele for the single nucleotide polymorphism rs2854128 in the humanin coding region exhibited lower circulating humanin levels, which correlated with cognitive aging in African-Americans [26]. On the other hand, decreased humanin levels have been reported in individuals with coronary endothelial dysfunction [27], type 2 diabetes, and AD [28]. Notably, acute aerobic and high-intensity interval exercise, known for their anti-aging effects, have been shown to increase circulating humanin levels [29,30]. However, some studies have reported inconsistent results [31]. These discrepancies could be attributed to variations in assessment methodologies (e.g., different ELISA kits), species, sex, and ethnicity. Further research is necessary to elucidate the potential of humanin as a biomarker for human health and to standardize measurement techniques across studies.

Mechanistically, humanin is the best studied MDP and several critical amino acid residues have been identified. Amino acids 9-11 and 19-20 are critical for the secretion and protective ability of humanin, and while there is no specific signal peptide sequence in humanin, fusion of humanin to a larger protein is sufficient to cause the larger protein to be secreted [32]. After secretion, humanin is known to activate two different receptors and has been shown to interact with many intracellular protein partners (reviewed in [33]).

Collectively, current findings introduce humanin as a dynamic and potent bioactive microprotein, acting through several key pathways to maintain cell health and resilience, and improve cognition. While cDNA screening tools were effective in the early discovery and characterization of humanin, the ongoing development of additional approaches has been crucial in the discovery of additional MDPs.

2^nd generation discovery of MDPs

After the discovery of humanin in the early 2000s, one informative paper by Mercer et al. demonstrated the existence of dozens of previously uncharacterized small mRNAs transcribed from mtDNA [34]. This study inspired and accelerated mitochondrial microprotein research, and our lab successfully identified seven more mitochondrial microproteins by an in silico DNA sequence-based approach using a simple open reading frame search for start and stop codons in the mtDNA. Since humanin is encoded within the MT-RNR2 gene that codes for the 16S rRNA, we further explored this area and identified a total of six smORFs called small humanin-like peptides (SHLPs) 1–6 [35]. Around the same time, we extended the smORF identification to the 12S rRNA, the other rRNA in the mtDNA. Although we discovered multiple smORFs within the MT-RNR1 gene, one smORF that translates into a 16-amino-acid microprotein caught our attention since it is highly conserved among species and there is a strong Kozak sequence upstream of this smORF. We subsequently named this smORF mitochondrial open reading frame of the 12S rRNA-c (MOTS-c) [36]. Together, we discovered a total of seven mitochondrial microproteins, SHLP1-6 and MOTS-c, by an in silico approach (Figure 2, Table 1).

Figure 2: Physiology of the mitochondrial microproteins.

Mitochondrial DNA-derived smORFs are translated into mitochondrial microproteins by ribosomes in mitochondria or cytoplasm. Then, they regulate gene expression levels as well as protein signaling/expression levels in the cell. Also, they are secreted from the cells and are transported to their target tissues, and exhibit unique biological functions. A total of 11 mitochondrial microproteins, such as Humanin, Gau, MOTS-c, SHLP1-6, SHMOOSE, and MTALTND4, have been discovered and reported.

Table 1: A list of all the published MDPs and New Mitochondrial Encoded Proteins.

MDP	Publication Date	Size (amino acids)	Coordinates	Strand	Translation	Properties
Humanin	2001	21∣24	2633-2698∣2707	+	Mito∣Cyto	Neuroprotective, insulin-sensitizer, IGF-1 interaction, apoptosis
gau	2011	100	6288-6590	−	Cyto	Localized to the mitochondria
MOTS-c	2015	16	1343-1393	+	Cyto	Exercise mimetic
SHLP1	2016	24	2559-2485	−	Cyto	No information
SHLP2	2016	26	2168-2088	−	Cyto	Protective against amyloid-β toxicity, adult macular degeneration (AMD), mitochondrial dysfunction, Parkinson's Disease, and ischemia reperfusion (IR) injury
SHLP3	2016	38	1819-1703	−	Cyto	Cytoprotective
SHLP4	2016	26	2522-2442	−	Cyto	Cell proliferation
SHLP5	2016	24	2854-2780	−	Cyto	No information
SHLP6	2016	20	2990-3052	+	Cyto	Pro-apoptotic
SHMOOSE	2023	58	12234-12410	+	Cyto	Alzheimer's Disease protection
mtALTND4	2023	99	11557-11856	+	Cyto∣Mito*	Cellular and mitochondrial effects
New Mitochondrial Proteins
CYTB-187AA	2024	187	14747-15887	+	Cyto	Involved in ATP production, stem cells, and fertility
mtALTCO1	Biorxiv (2024)	259	6089-6866	+	Mito*	Unknown

^*Alternative mitochondrial translation where AGA and AGG code for arginine

SHLPs

SHLP1-6 are encoded by smORFs around the humanin coding region and share some biological features with humanin [35]. Among them, SHLP2 and SHLP3 increase cell viability and reduce apoptosis, while SHLP6 increases apoptosis in NIT-1 β and 22Rv1 cells. More specifically, SHLP2 exhibited protection against amyloid β-induced toxicity and age-related macular degeneration in cells [37]. We recently demonstrated that SHLP2 was localized in mitochondrial complex 1 by physically interacting with NDUFB1, NDUFB7, and NDUFB9, and exhibited protection against mitochondrial dysfunction in in vitro [38]. Furthermore, a naturally occurring genetic variant of SHLP2, K4R SHLP2 resulting from an m.2158T>C polymorphism, exhibited protection against Parkinson’s disease (PD) by increasing its stability in humans and PD model mice [38]. On the other hand, SHLP2 has also been characterized as a regulator of energy homeostasis. SHLP2 treatment protected the mice from (high fat diet) HFD-induced obesity and insulin resistance by suppressing food intake and increasing thermogenesis [39]. This study also demonstrated that SHLP2 bound to and activated chemokine receptor CXCR7 [39]. In humans, it has been demonstrated that circulating SHLP2 levels were significantly lower in white males with prostate cancer than the controls [40]. Although these studies are characterizing the biological roles of SHLP2 and its underlying mechanisms, much remains to be learned about the biological functions of other SHLPs.

MOTS-c

MOTS-c is the second mitochondrial microprotein that was identified and first reported as a metabolic regulator that prevents aging- and HFD-induced obesity and insulin resistance [36]. After its discovery, MOTS-c has been studied deeply and various biological functions were characterized. Among them, we conducted studies related to skeletal muscle function and identified its exercise mimetic actions: enhancing exercise endurance in aged and HFD-fed mice [41], preventing muscle weakness in aged mice [41], and preventing skeletal muscle loss in aged [41], HFD-fed [42], and immobilized mice [43] by regulating metabolic activity as well as suppressing muscle atrophy signaling pathway[43,44]. Although MOTS-c promotes cellular stress resistance against metabolic stress by directly interacting with NRF2 in human embryonic kidney cells (HEK293) and rat adrenal medulla pheochromocytomas (PC12) [45,46], a direct and functional molecular target in skeletal muscle has not been identified yet.

We previously reported that MOTS-c prevents skeletal muscle loss through the CK2/AKT/FOXO signaling pathway [44]. More recently, we identified CK2 as a direct and functional binding protein of MOTS-c [47]. Systemically administered MOTS-c directly binds to the CK2 alpha subunit in both muscle and fat tissues. However, it increases CK2 activity in muscle while inhibiting it in epididymal fat by differentially modulating CK2-interacting proteins [47]. MOTS-c induced muscle glucose uptake was blunted by both a CK2 inhibitor or CK2 alpha knock-down [47]. Additionally, we demonstrated that a naturally occurring genetic variant of MOTS-c, K14Q MOTS-c, results from a m.1382A>C polymorphism [48] and is associated with an increased risk of type 2 diabetes [43,49] and sarcopenia [43], as well as altered skeletal muscle fiber type composition and athletic performance [50] due to its reduced binding affinity to CK2 alpha [47]. Therefore, a direct interaction with CK2 is one of the underlying molecular mechanisms of MOTS-c action in skeletal muscle.

Circulating MOTS-c levels in humans have been associated with some health conditions. Studies have shown that MOTS-c levels are decreased in individuals with impaired coronary endothelial function [51] and in those with coronary artery disease [52]. Regarding exercise and skeletal muscle, high-intensity interval exercise [41] and a combined training of aerobic and resistance exercise [53] have the potential to increase circulating MOTS-c levels in skeletal muscle. Furthermore, studies show a positive correlation between circulating MOTS-c levels and leg muscle mass and function [54], as well as a negative correlation with circulating myostatin levels in human subjects [44]. However, different studies [55-57] have reported inconsistent results, leaving the relationship between MOTS-c and obesity, metabolic syndrome, or type 2 diabetes unclear. Further studies with a large sample size are necessary to establish the role of MOTS-c as a biomarker of human health.

mtALTND4

In 2023 using a proteogenomics approach, Kienzle et al. discovered another mitochondrial microprotein within the MT-ND4 gene that they have called mtALTND4 [4]. This mitochondrial microprotein influences mitochondrial respiration in HeLa and HEK-293 cells: decreasing the routine oxygen consumption rate, maximum coupled and uncoupled respiration, as well as the spare reserve capacity. Since MTALTND4 exists in human plasma, examining circulating levels in various health conditions could be interesting for future studies. The translation of an microprotein from within another gene has been shown to generally interfere and reduce the translation of the larger gene, although in this specific case it is unknown (reviewed in [58]).

CYTB-187AA

More recently, although not strictly a microprotein with 187 amino acid residues, a new protein called CYTB-187AA has been identified as a translation of an alternative reading frame in the MT-CYTB gene of the mtDNA. This protein is translated in the cytoplasm and has effects on stem cells, ATP production, and fertility of female mice [59].

MiWAS Based Approaches

Although the in silico approach to identifying MDPs was fruitful for our lab, it was imprecise and many theoretical smORFs resulted in microproteins with no biological effects, as would be expected. To further complicate matters, our in silico analysis of all possible smORFs in the mtDNA led to a possibility of 600 different MDPs, a number that would be difficult to screen and verify (Figure 1, inner circle). Instead, leveraging a plethora of genomic data, we developed another method to identify possible MDPs based on human population genomics. Mitochondrial-wide Association Studies (MiWAS) is a statistical method used to estimate the effects of mtDNA variants on phenotypes. MiWAS is the mitochondrial-specific counterpart to conventional genome-wide association studies (GWAS). With the ability to capture genetic variation using cost-effective and high-throughput approaches, GWAS has become widespread over the last decade, leading to the identification of genetic loci associated with diseases [60].

MDPs encoded by noncanonical smORFs include humanin, SHMOOSE, MOTS-c, SHLPs1-6, GAU, and mtALTND4 [4,18,21,23,36,61-63]. Recently, mtDNA variation in the smORFs of humanin, SHMOOSE, MOTS-c, and SHLP2 has been associated with diseases and disease-related phenotypes (Figure 3). One significant example is the smORF for SHMOOSE, which contains a single nucleotide polymorphism (SNP) in its 47th codon [64]. This SNP results in an amino acid change from aspartic acid to asparagine, altering the net charge of SHMOOSE. This SHMOOSE SNP, referred to as SHMOOSE.D47N, was identified using unique MiWAS methods (Figure 3).

Figure 3: The use of MiWAS in MDP discovery.

Questions and discoveries made possible by MiWAS techniques for SHMOOSE and SHLP2.

The key difference between MiWAS and GWAS is the input data. MiWAS uses a matrix of mtDNA variation data, while GWAS uses a matrix of nuclear DNA variation data. This distinction is crucial because the statistical methods for estimating the effects of mtDNA and nuclear DNA variants differ. For example, GWAS estimates the effect of diploid genetic variants based on the number of allelic copies, whereas MiWAS estimates the effect of haploid genetic variants. Mitochondrial DNA is maternally inherited, reflecting a different ancestral timeline compared to the biparentally inherited nuclear genome, necessitating unique statistical approaches to capture the distinctive nature of mtDNA [65-67]. In other words, mtDNA does not follow Hardy-Weinberg principle.

One of the most used tools to carry out GWAS is PLINK, a free open-source software that allows users to estimate the effects of gene variants on phenotypes of interest [68]. When researchers use PLINK for their GWAS, the Hardy Weinberg (dis)equilibrium law is considered as a command by the tool. Since mtDNA does not follow Hardy Weinberg principles, when this command is used, mtDNA variants are filtered from the analysis. Furthermore, since PLINK assumes that the genetic variants are diploid, it estimates the effect in a two-allele manner. However, mtDNA is not diploid, meaning that the effect sizes will be shrunk in half if the user does not explicitly tell PLINK that the input data is mitochondrial. PLINK updated their tool in 2018 to ensure mtDNA variants are not treated as diploid [69].

In MiWAS, one technique is to reduce the dimensionality of the mtDNA variant matrix using principal component analysis (PCA) and then use these principal components as covariables in regression models [70]. In 2010, Biffi et al. reported that mitochondrial PCA recapitulated haplogroup information, and that it outperformed haplogroup-stratified analysis, describing it as inferior to mitochondrial PCA in controlling for the population structure [71]. What PCA is intended to do is correct for mitochondrial genetic ancestry—thereby controlling the studied population substructures—to isolate the effects of single mtDNA variants. For example, the SHMOOSE.D47N association with AD was first reported in 2010 in the Alzheimer's Disease Neuroimaging Initiative Phase 1 cohort [72]. This was a small sample size of a few hundred individuals. The researchers in this group examined the effects of mitochondrial haplogroups and individual SNPs individually. Nine haplogroups were split into four haplogroup clusters 1) HV, 2) JT, 3) UK, and 4) IWX based on their relation to ancestor lineage N. A limitation of this approach is that, by doing so, one reduces the statistical power to estimate effects of the haplogroup because of the significant partitioning (i.e., fewer individuals per group). They reported that haplogroup UK was the strongest associated haplogroup with increased AD risk with an odds ratio of 1.92. However, they also looked at the individual three SNPs that define haplogroup UK, one being SHMOOSE.D47N. Although these SNPs define haplogroup UK, their individual odds ratio were 2.22, 2.03, and 1.99 — this suggests that the substructures of larger haplogroups have meaningful effects on disease risk. This is one of the reasons why PCA is used—to correct for these substructures.

However, PCA can introduce collinearity into regression models, potentially leading to overfitting or underestimating the effects of variants. For example, haplogroup architecture is recapitulated by PCA, as Biffi et. al. reported, and there are mitochondrial SNPs that explicitly determine the haplogroup. Therefore, if the regression model includes a SNP plus a principal component, then it is highly likely the two variables are correlated, and that fitting a regression line using these two variables has potential for over-fitting. Therefore, careful consideration to the population substructure should be considered. Miller et al. (2022) performed both mitochondrial PCA correction and adaptive permutation without PCA correction in four unique cohorts. Two of these cohorts showed significant differences in population structure through mitochondrial PCA, but the other two did not show significant differences in structure through mitochondrial PCA. Therefore, the cohorts that did show significant population structure differences were corrected by PCA, whereas the other two were not. This meant that the estimates captured accounted for differences across population structures within and between population cohorts. Therefore, it is strongly encouraged to examine the population structure through PCA of mitochondrial DNA variants prior to carrying out analysis and interpreting results. Ultimately, GWAS-based methods are associative, and experimental confirmation is truly the only way to validate the observation. These association measures should principally serve as a guide for hypothesizing experiments. In the case of SHMOOSE, it was used to design experiments to determine its function and its levels across neurodegeneration and aging (Figure 3). Similarly, in the case of SHLP2, which originally reported PD-risk gene variants performed without mitochondrial PCA, it was used to characterize the functions of SHLP2 in a PD-specific experimental environment (Figure 3).

Indeed, over the last decade, MiWAS has been instrumental in discovering novel peptides encoded by the mitochondrial transcriptome. A MiWAS on AD classification using data from four cohorts revealed that SHMOOSE.D47N was associated with a 30% increase in AD risk [64]. The MiWAS associative effects of SHMOOSE.D47N on neurodegenerative phenotypes guided the downstream discovery of the actual SHMOOSE peptide. SHMOOSE was targeted and detected biochemically using mass spectrometry and immunological approaches. Two unique fragments of SHMOOSE were detected in neuronal mitochondrial lysates. Antibody-based quantification of SHMOOSE further correlated with age and AD-related phenotypes tau and amyloid beta. SHMOOSE was found to localize to the inner mitochondrial membrane, and its alternative form, SHMOOSE.D47N, exerts different biological effects in the mitochondrial inner membrane, most notably altering mitochondrial inner membrane gene expression and response to amyloid beta toxicity [64].

In addition to SHMOOSE, MiWAS has identified SNPs in other MDPs, including SHLP2 and MOTS-c. SHLP2 contains a SNP in its smORF that changes the 4th amino acid from lysine to arginine (SHLP2.K4R), associating with decreased risk for Parkinson's disease (PD) [38]. This variant results in a more stable version of SHLP2 and, like SHMOOSE, exerts its effects within the inner mitochondrial membrane. Another SNP in the smORF of MOTS-c was found to associate with increased risk for diabetes in a Japanese population. This SNP changes the 14th amino acid of MOTS-c from lysine to glutamine (K14Q) [49]. In a study comparing the metabolic effects of MOTS-c versus the alternative K14Q, mice administered MOTS-c exhibited weight attenuation on a HFD over time, whereas K14Q did not [49].

A subtype of MiWAS involving targeted smORF analyses revealed two distinct variants in the MDP humanin. A haplogroup H-determining SNP within the termination codon of the humanin smORF was found in nearly half of European ancestral individuals and in 1-5% of individuals of African ancestry. In these African ancestral individuals, the SNP resulted in greater cognitive decline and lower humanin levels in plasma, highlighting the ancestral-specific effects of certain mtDNA variants [26]. Similarly, another ethnic-specific mtDNA variant in humanin, changing the 3rd amino acid from proline to serine (P3S), was found to interact with APOE4 to promote lifespan. Humanin.P3S is a haplogroup N1b determinant found in individuals of Ashkenazi descent. Humanin.P3S promotes higher affinity binding to APOE4 in vitro and, in mouse amyloidosis models with human APOE4, reduces AD-related pathology more effectively than the more common humanin counterpart [73].

Heteroplasmy of mitochondrial DNA has been hypothesized to be associate with functional phenotypes. However, heteroplasmy is specific to the cell type being studied and the sequencing technology used. Some reports show heteroplasmy under 0.5%, which means that in order to determine the effects of these variants on phenotypes, a large sample size is necessary, and significant sequencing depth is needed to even capture the low incidence of the heteroplasmy [74]. For instance, in nearly 1 million individuals from the UK Biobank Study, the effects of nuclear DNA variation on heteroplasmy were observed [75]. But this was only possible due to the enormous sample size. For most GWAS studies in the past, not only are the sample sizes smaller but they also were carried out using arrays that are not able to capture heteroplasmy.

Altogether, MiWAS has been a powerful approach to identify and functionalize the effects of mitochondrial variants, especially for previously unannotated microproteins. Additionally, because of the ethnic segregation due to the uniparentally inheritance and lack of recombination in the mtDNA, SNPs in MDPs may be a genetic explanation for some of the differences found between ethnic groups (Table 2).

Table 2: Alternative and Ethnic Specific SNPs that affect MDPs or Newly Discovered Mitochondrial Encoded Proteins.

MDP	SNP ID	AA change	RS	Ethnicity	Correlation
HN	G2706A	None	rs2854128	African-Americans	Increase in cognitive decline [26]
HN	C2639T	P3S	None	Ashkenazi Jews	APOE4 resilience [73]
MTALTCO1	T6221C	S45P	none	African-Americans	Increased Prostate Cancer Risk [8]
MOTS-c	A1382C	K14Q	rs111033358	Japanese	Increase in diabetes [49], Possible longevity [48]
SHLP2	T2158C	K4R	None	Europeans	Protection from PD [38]
SHMOOSE	G12372A	D47N	rs2853499	Europeans	Increase in AD risk [61]
In Different Species
SHLP6	C3017T	Premature stop	None	Rodents	Related to heterothermy [84]

Glossary: GWAS: a genome wide association study that relates a genomic variation with a specific phenotype, lncRNAs: long non-coding RNAs are RNA species that were previously thought to be non-coding, MDP: mitochondrial derived microprotein or peptide that is translated from the mitochondrial genome, MiWAS: a mitochondrial wide association study is a GWAS that is specific to the mitochondrial genome, MOTS-c: mitochondrial ORF of the 12S rRNA type-C was the second MDP published and acts as a exercise mimetic, mtDNA: mitochondrial genome and DNA, ORF: an open reading frame is a section of DNA that codes for a protein, PheWAS: a phenome-wide association study is similar to a GWAS but examines a single genetic variation over many phenotypes, SHLP: small humanin like peptide is a group of MDPs that are found in the MT-RNR2 gene, smORF: a small open reading frame is an ORF that codes for a microprotein, SNP: a single nucleotide polymorphism is a DNA variation at a single nucleotide, uORFs: an ORF found in the 5’ upstream element of another ORF

Future Approaches

The boom of next-generation sequencing has been crucial for discovering previously unannotated microproteins encoded by the nuclear and mitochondrial genomes. Over the past decade, researchers have been equipped with cost-effective tools to leverage transcriptomes, resulting in the discovery of microproteins encoded by unique transcripts that contain novel smORFs. These discoveries have involved three distinct technologies and approaches: proteogenomics, translatomics, and evolutionary analysis (Figure 4).

Figure 4: Current and future approaches in mitochondrial microprotein identification.

Current methods include cDNA screens leading to the discovery of humanin, in silico identification of potential open reading frames (ORFs) leading to the discovery of MOTS-c and SHLPs 1-6, and mitochondrial focused genome wide association screens (MiWAS) leading to the discovery of SHMOOSE. Technological advancements and emerging datasets introduce proteogenomics and translatomics as future methods for microprotein discovery. In addition, emerging technologies such as single-molecule protein sequencing could continue to advance the field.

Proteogenomics involves deep transcriptome sequencing to find unique transcripts, followed by in silico translation of these transcripts to annotate the complete putative proteome of microproteins. These putative microproteins are then used as a search space in mass spectrometry-based experiments to detect their peptide fragments [76-78]. Proteogenomic screening methods allow efficient high-throughput identification of biologically relevant microproteins. However, current mass spectrometry methods have limited ability to detect microproteins, often due to low protein abundance, presenting major challenges for validation of predicted mitochondrial microproteins, particularly those with novel transcripts. To address this barrier, researchers frequently institute a validation technique involving chemically indistinguishable isotopically labeled standards for spectra validation and improved precision [76]. In addition, the further validation of smORF translation with translatomic screening allows for continued precision in determining biologically potent microproteins with novel transcripts. Translatomics involves sequencing ribosome-protected RNA following the chemical stalling of ribosomes in live cells. This process, often referred to as ribosome profiling or Ribo-Seq, freezes ribosomes and isolates their translating RNA to infer which open reading frames are undergoing translation. Translatomics has gained popularity and led to the formation of a consortium to annotate all smORFs. To date, the Phase 1 GENCODE smORFs catalog includes over 7,000 smORFs considered to undergo active translation. Determining which of these smORFs encode bona fide microproteins versus those that function as cis-regulators of transcript regulation is an active field of study [79-81]. While ribo-seq profiling provides novel insight into mitochondrial translation, the method is not without fault. Mainly, many microproteins identified through ribosome profiling are unstable and likely never accumulate to detectable levels, leading to a high rate of false positives. Deep sequencing methods using ribosome-protected mRNA fragments, known as “footprints”, is one currently applicable technique to overcome this barrier, as it allows for higher resolution of stable and functional microproteins [82]. Tandem validation with mass spectrometry provides a clearer picture of microprotein behavior, in which ribosome profiling confirms smORF translation while mass spectrometry indicates microprotein stability.

Evolutionary analysis is another common method of protein discovery, as evolutionary conservation is often used as a proxy for biological function. The peptide hormone ELABELA was discovered and functionalized through a combination of transcriptomics and evolutionary conservation. ELABELA is encoded by a previously annotated long noncoding RNA (lncRNA), but its expression signature in tissues and high sequence identity across species prompted researchers to investigate whether this lncRNA was misannotated, which ultimately proved to be the case [83]. Additionally the gau mitochondrial microprotein was discovered in 2011 by this same method [63]. In addition to looking for amino acid conservation, an approach taken by Gruschus et al. looked for synonymous codon bias [84]. They found evidence of codon bias for humanin and SHLP6 and the N-terminal region of MOTS-c and SHLP4 but not for the other SHLPs. This suggests that these other SHLPs are not well conserved or that they are too evolutionarily young to allow for this type of analysis to detect. Recent advancements introduce a fourth method for protein discovery as well as further microprotein characterization, through real-time single-molecule protein sequencing (Figure 4) [85]. This method involves amino acid annotation using high throughput whole-integrated semiconductor chips to probe for targeted protein sequences, amino acid substitutions, and post-translational modifications. The application of this method for the detection of mitochondrial microproteins could prove to be precise and extensive, while also proving informative in determining post-translational modifications which may impact microprotein function. When paired with mitochondrial genome editing, another emerging technique (in infant stages of development), highly precise and critical functional characterization of novel microproteins will be possible. While captivating, current genome editing methods remain imprecise when compared to nuclear genome editing. Currently, one of the most precise mitochondrial gene editing methods, Hifi-DddA-derived base editing, specifically targets mitochondrial DNA without the use of CRISPR guide RNAs while reducing off-target effects [86]. However, this method has an editing efficiency of roughly 50% and is specifically limited to cytosine deamination. Regardless, such methods illuminate promising directions for the future of microprotein discovery and characterization.

The nomenclature of microproteins can be considered somewhat semantic. Some pipelines describe microproteins in length between 9-150 amino acids, whereas others set a minimum threshold at 15 amino acids and maximum at 100 amino acids. Broadly, microproteins represent an area of short protein and peptide discovery. Peptides offer promising therapeutic targets because of their high sensitivity as ligands. However, for peptides to serve as drugs, they need to be engineered to increase stability and half-life, as their native turnover is often extremely fast. The rise of peptide-based therapeutics, most notably GLP1 analogues, underscores the urgency to study the complete peptidome, to fully capture peptide biology and identify therapeutic targets.

While microprotein research has made significant strides since the boom of next-generation sequencing, a deeper understanding of their roles in biology is now sought (see Outstanding Questions). Because of the unique nature of mitochondria, while there are many overlaps with microproteins in the nuclear genome, there are many distinct features and challenges as well. On one hand, the lack of mRNA splicing and introns simplifies some analyses, but the fact that the mtDNA sits behind an additional double-membrane combined with a lack of efficient genomic tools leads to some deficiencies in follow-up analyses. Future studies and techniques will eventually overcome these hurdles. Similar to the nuclear microproteins, this group of mitochondrial microproteins is a rich field of research that has the potential to provide new clues to biology and translational therapies.

Outstanding Questions Box.

• Which conserved smORFs generate functional microproteins?

• For smORFs that are actively translated yet are evolutionary young, which function through their translated byproducts?

• How can proteomics techniques be optimized to reveal lower-abundant microproteins and those difficult to detect using mass spectrometry?

• Are translated smORFs that generate detectable microproteins functional in the canonical sense?

• If they are functional in a noncanonical sense, why do the overwhelming majority lack high sequence identity across species?

• Are they a result of unique translation in perturbed cellular environments, such as cancer or aging?

• And if so, can these microproteins alter these perturbed cellular environments? Does genetic variation in these smORFs represent selection pressure for their encoded peptides, if they are indeed functioning through their peptides?

• How do MDPs exit the cell and how do they have their effects?

• Are MDPs translated primarily in the mitochondria, cytoplasm, or both?

• What tools can be further developed to enhance mito-riboseq?

• What are the actual stop codons for human mitochondria and what are the implications for MDP research?

• How are MDP transcripts regulated and how do they relate to nuclear transcripts with respect to cis regulation or retrograde signaling to the nucleus.

• Is cytoplasmic tRNA-Arg imported into the mitochondria and is this related to metabolic flexibility?

• What post translational modifications are found on MDPs, and does this explain the discrepancy in size seen on many of the Western blots of MDPs by many different labs?

Highlights.

• New mitochondrial microproteins (MDPs) are being discovered with a myriad of different physiological effects

• As the field matures, more information about the mechanism of action is opening new avenues of research

• New techniques are being developed to help more efficiently identify additional MDPs