Eukaryotic Microproteins

Ribosome Profiling

Ribosome profiling, also known as ribosome sequencing (Ribo-seq) (40, 41), has emerged as the leading method for identification of translated smORFs on a global scale. The broad adoption of Ribo-seq can be attributed to its ability to detect thousands of translated ORFs outside annotated CDSs (20, 42–47), as well as quantitate changes in translation under specific conditions and in disease (21, 48, 49).

Ribosome profiling, or Ribo-seq, refers to the nucleotide-precision identification of footprints of elongating ribosomes arrested with translation elongation inhibitors such as cycloheximide. Arrested ribosomes are then enriched in complex with RNA, followed by nuclease digestion of unprotected RNA, isolation of ribosome-protected footprints from the resulting purified monosomes, and deep sequencing. The resulting data must be further processed to infer translated ORFs. Lastly, rigorous data analysis, analysis of experimental replicates, and high-resolution datasets are essential for calling smORF translation using Ribo-seq, because their short lengths and lower abundance lead to a lower signal-to-noise ratio in smORF-mapped reads (42).

Although Ribo-seq is powerful for profiling the footprints of elongating ribosomes and identifying novel coding regions, elongation inhibitors like cycloheximide are not well-suited to deconvolution of some translation initiation sites, especially for ORFs with multiple start sites or overlapping reading frames (50). However, translation initiation sequencing (TI-seq) offers profiling of the footprints of initiating eukaryotic ribosomes using specific inhibitors (51–54). This allows deconvolution of smORF translation initiation from larger main ORFs in multicistronic messenger RNAs (mRNAs) and is especially important for detecting nested and out-of-frame smORFs filtered out by scoring algorithms relying on parameters like three-nucleotide periodicity of P-site occupancies of ribosomes translating a CDS (which is violated by overlapping, out-of-frame ORFs). TI-seq can be combined with Ribo-seq to identify translated smORFs with higher confidence (55).

Low study-to-study concordance is often observed in translated smORF identifications (14, 56); this is in part due to the low signal-to-noise ratio for smORF ribosome footprints as a result of their translation by monosomes (57, 58), as well as variability in smORF identification by different algorithms even when analyzing the same input dataset (56). Some studies recommend analysis of replicate data (56, 59) and intersecting the results of multiple data analysis software (56) as best practices. A broad effort to curate translated smORFs following these guidelines has led to the first publicly available Ribo-seq translatome resource (14).

Mass Spectrometry

Mass spectrometry–based proteomics pipelines tailored for direct detection of smORF translation products have been devised (15, 22, 63, 61). These methods require specialized sample preparation, mass spectrometry, and data analysis methods for sensitive detection of unannotated proteins and microproteins. These approaches have been recently reviewed in detail (18, 62–64), so we provide a brief overview with a focus on smORF-specific innovations and applications.

A necessary advance was development of methods to enrich microproteins from whole proteomes for sensitive detection. Given their short lengths, when digested with trypsin, microproteins typically generate only one, or few, unique tryptic peptides. As a result, microprotein proteomics is a particularly demanding application that requires the detected proteotypic peptide to be well separated from other ions (that is, avoidance of complex spectra). It is therefore critical to enrich microproteins and remove larger proteins via various biochemical methods (15, 65–71). Deeper coverage of microproteins is further afforded via application of several sample preparation protocols in parallel (66, 67, 70, 71), two-dimensional liquid chromatography–mass spectrometry with offline fractionation (18), subcellular fractionation (72), and chemical labeling (73, 74). In sum, sample preparation is a key consideration for successful detection of microproteins in whole-proteome extracts. Supporting this idea, high-quality evidence for microprotein expression has been obtained from peptidomic analysis of peptides displayed on major histocompatibility complex class I (MHC I) and, in human, human leukocyte antigen class I (HLA I), which present peptides derived from proteolysis of intracellular proteins (75, 76). The sequence constraints for HLA I binding to peptide epitopes, as well as the instrument parameters required for sequencing nontryptic HLA I–associated peptides, may afford a more effective normalization of microproteins to canonical proteins, enhancing their detection.

Historically, microprotein sequences were absent from the proteome databases utilized for protein identification by peptide-spectral matching (PSM). Specialized databases must therefore be constructed to identify microproteins. Most databases for eukaryotic smORF proteomics have been constructed from conceptual translation of transcriptomes or predicted smORF-eomes—which contain the annotated and/or smORF translatomes but also contain many entries that are not really expressed (18). Importantly, this database expansion generates false positive identifications, and thus peptide-spectral matches cannot be considered conclusive evidence for microprotein expression, especially when only one proteotypic peptide is identified. The intersection of proteomics and ribosome profiling has advanced smORF discovery, particularly in the creation of such translatome databases as the above, which ideally would contain no extraneous entries that can lead to false positive identifications and thus proving optimal for microprotein identification via PSM (75).

Overall, both Ribo-seq and mass spectrometry–based smORF discovery platforms can identify only candidate smORFs and microproteins, since both methods are indirect and/or rely on statistical scoring. smORF validation requires molecular and genetic approaches, such as development of specific antibodies, smORF silencing or genetic deletion, and/or epitope tag knock-in at the genomic smORF locus. Nonetheless, proteomics has demonstrated utility not only in providing supporting evidence for smORF translation but also for probing the chemical properties, reactivity, localization, interactions, and modifications of microproteins that are challenging to predict with bioinformatic tools (64, 77, 78).

On the Discordance Between Proteomics and Ribo-seq

Proteomic microprotein discovery methods, when operated in shotgun mode and on whole-cell protein extracts, typically detect one to two orders of magnitude fewer smORFs than ribosome profiling—or less (79)—and it is useful to consider the reasons for this discordance. First, shotgun proteomics, even with optimized protocols, is relatively insensitive to detection of membrane proteins, given their hydrophobic nature that leads to production of few tryptic fragments (72), as well as basic proteins, which are overdigested by trypsin. Both of these motifs can be enriched in smORFs, which are encoded in A/T-rich intergenic regions that tend to result in codons for hydrophobic and aromatic amino acids (80, 81), as well as alternative reading frames, which tend to be enriched in disorder-promoting and charged amino acids (82). Second, due to their short lengths, microproteins tend to be detected by only a single proteotypic tryptic peptide, or only a few (15); not only does this decrease the statistical confidence to identify microproteins via PSM, but it also increases the likelihood of coelution with canonical peptides, which decreases spectral quality for microprotein identification. This problem is exacerbated for low-abundance proteins, which are also refractory to detection with shotgun proteomics due to stochastic selection of parent ions for fragmentation and sequencing (72). Finally, specialized or targeted methods are typically required to detect secreted microproteins (21, 83). As a result, proteomics may have a ceiling to its sensitivity for smORF-encoded microprotein detection. Thus, ribosome profiling offers increased sensitivity for identifying some classes—perhaps the most abundant classes—of translated microproteins.

It is also possible that some translated smORFs do not contribute to the stable cellular proteome. For example, as discussed above, peptides derived from smORFs are displayed on HLA I, suggesting that they are involved in self- versus nonself-discrimination by the adaptive immune system (75, 84–86). Along the same lines, translation products arising from A/T-rich genomic regions tend to generate hydrophobic C-terminal epitopes that may be triaged for proteasomal degradation by the GET complex (81). Furthermore, although yeast de novo genes, some of which are smORFs, exhibit low endogenous expression levels, they can exhibit cellular phenotypes and can increase yeast cell fitness when overexpressed (59, 80). Taken together, these studies suggest that poorly conserved, lowly expressed, and/or unstable microproteins may be biologically significant, either contributing to the immunopeptidome or having adaptive potential on evolutionary timescales. In this sense, the smORF translatome may be distinct from the canonical proteome in stability and functionality (59); this property may speculatively contribute to differences in smORF detection between ribosome profiling and shotgun proteomics, where the former detects all translated smORFs and the latter detects only the most stable cellular microproteins capable of producing proteotypic tryptic peptides. Figure 1 depicts the fates of stable versus unstable microproteins.

Genetic Screens

Genetic screens have been applied to query microprotein functionality at scale. A clustered regularly interspaced short palindromic repeat (CRISPR) screen in human induced pluripotent stem cells and K562 cells targeting ~2,000 smORFs, as well as protein N-terminal extensions, found that ~500 noncanonical ORFs exhibit cell growth phenotypes when knocked out (20), demonstrating for the first time that human smORFs are broadly functional. A subsequent CRISPR screen in cancer cell lines demonstrated viability defects for ~10% of targeted smORFs and noncanonical ORFs; it further employed guide RNA tiling to assign phenotypes to translation of the novel ORFs and not to cryptic DNA or RNA regulatory elements (16). Finally, overexpression screens have been employed in yeast to demonstrate that some smORFs, even when they are species-specific, can improve fitness when overexpressed (80). Genome-wide screens have thus provided clear evidence that smORFs are broadly functional in cells and will continue to find utility in species-, cell-, tissue-, and disease-specific contexts.

Computation and Conservation

Although computational methods have made outsize contributions to prokaryotic smORF identification (87), they have also demonstrated utility for eukaryotic smORF prediction despite the greater complexity of eukaryotic genomes. For example, early studies utilized computational prediction to identify hundreds to thousands of smORFs in mouse (9) and yeast (8). Furthermore, computational methods like PhyloCSF have found application in identifying highly conserved, and thus likely functional, smORFs (70, 88).

smORF CLASSES

Unannotated smORFs have been identified in varied genomic loci (Figures 2 and 3) and categorized accordingly, with implications for the evolution, expression, and function of the encoded microproteins. Indeed, several years ago, an insightful review discussed the classes of smORFs in the context of the functional relevance of their genomic locations (89); herein, we update and add complexity to this framework with a focus on the prevalence, expression, and functional significance of each class.

Noncoding RNA smORFs

Perhaps best studied are smORFs assigned to noncoding genomic regions, such as long noncoding RNAs (lncRNAs), as well as other classes of ncRNA including microRNA precursors, promoter-derived lncRNAs, transposons, endogenous retroviruses, and transcribed pseudogenes (Figure 2). Some estimates suggest as many as 40% of lncRNAs may be translated (45) [and others, few to none (90)]; at the same time, dozens of cytoplasmic, polyadenylated lncRNAs can be found in complex with translating ribosomes in some cell types (91), and even nonpolyadenylated lncRNAs such as MALAT1 can be translated by cytoplasmic ribosomes under specific conditions (92). Despite the ongoing refinement of catalogs of coding lncRNAs, many microproteins encoded in lncRNAs have yielded to functional characterization in human and mouse because of their tractability to genetic perturbation. In general, because polyadenylated lncRNAs are, in a sense, mRNA-like, the microproteins they produce tend to function like small, canonical proteins and peptides. For example, the NBDY gene encoding a 7-kDa microprotein regulator of mRNA decapping and P-body formation was originally annotated as lncRNA LOC550643 (38, 77, 93).

Additionally, microRNA precursor-encoded peptides (miPEPs) have been reported to encode plant microproteins (94). Several circular RNAs (circRNAs) have also been proposed to be translated (95), which would require internal translation initiation. A translated microprotein assigned to orf0, a smORF transcribed in the opposite direction from the promoter for the LINE-1 retrotransposon, has also been predicted (96). Another class of smORFs encoded within antisense transcripts, such as PIGB opposite strand (PIGBOS), which localizes to endoplasmic reticulum (ER)-mitochondria contact sites, has been described (97). Importantly, in order to be translated, lncRNAs must localize to the cytoplasm and associate with polysomes; in fact, acquisition of nuclear export has been proposed as a critical step in the evolution of de novo genes from previously noncoding, nuclear transcripts (98).

smORFs Within mRNA

Translated smORFs can also arise within mRNA and are categorized on the basis of on their location with respect to the annotated CDS (Figure 3). smORFs can regulate translation of the CDS in cis and sometimes also produce a functional microprotein. Translation of smORFs within mRNA can thus result in complex biological outcomes. Because they produce more than one protein product, these genes are dual-coding (33, 99).

First, we consider uORFs, smORFs that initiate at start codons within the 5′ UTR and terminate prior to the start site of the downstream CDS. smORFs can also initiate at an upstream, out-of-frame start codon, extending into an alternative reading frame that partially overlaps the CDS; these are sometimes termed upstream overlapping ORFs [u(o)ORFs]. In most cases, the u(o)ORF terminates at a stop codon in its unique reading frame within the overlapping region and is thus shorter than the co-encoded protein; in some cases, a u(o)ORF can continue past the stop codon of the CDS and terminate downstream. Subsequent to Kozak’s (4) identification of nonfunctional uORFs in hundreds of human transcript leaders, it was revealed (5, 100) that uORFs generally downregulate translation of downstream cistrons via multiple mechanisms, including competitive translation, mRNA destabilization via nonsense-mediated decay, and interaction of the translated uORF peptide product with the ribosome. Specific examples, such as the translational regulation of GCN4 by uORFs in yeast (101) and translational repression of mammalian ATF4 by a uoORF (102), demonstrate that uORF versus CDS translation can be rebalanced in response to environmental conditions or stress. Although u(o)ORF translational regulation is independent of the sequence of the translated product, some u(o)ORFs are conserved at the amino acid level, suggesting that the microprotein products of at least a subset of u(o)ORFs may be stable and functional in trans. One example of a conserved, functional uORF is ASDURF (103). The percentage of uORFs and uoORFs that encode functional microproteins in addition to regulating downstream CDS expression is an important, currently unanswered, question.

Of emerging interest are nested or internal ORFs (iORFs) representing frameshifted sequences fully contained within an annotated protein CDS. Importantly, iORFs are encoded in alternative reading frames relative to the CDS [similar to uoORFs and doORFs (downstream overlapping ORFs)], described below], and thus the amino acid sequence of the encoded microprotein is completely different from the annotated protein it overlaps. If the iORF was in the same reading frame as the CDS, it would represent an internal initiation site and would terminate at the same stop codon as the CDS. Although ribosome profiling has enabled discovery of internal, in-frame start codons with biological significance (104), they are not further considered in this article.

Out-of-frame iORFs have been detected with proteomics, where they can represent up to 30% of identified noncanonical ORFs (15, 22) but may be undercounted by both shotgun proteomics and ribosome profiling. iORFs are also challenging to characterize using genetic tools: That iORFs completely overlap a canonical protein CDS in genomic space means genetic tools to perturb, knock down, or knock out iORFs cannot a priori separate the function of a putative iORF from the CDS it overlaps. Only a handful of iORFs, such as MINAS-60 and alt-FUS (74, 105), have been studied at the molecular level to date. This class of smORFs may therefore merit increased attention in the future, but improved tools are needed for their identification and characterization.

Downstream ORFs (dORFs) that are entirely contained within the 3′ UTR and doORFs that initiate within the CDS in an alternative reading frame and extend into the 3′ UTR have also been reported, although they are typically detected in lesser numbers than uORFs/uoORFs (7, 15). Interestingly, quantitative ribosome profiling suggests that dORF translation is positively correlated with CDS translation, suggesting that dORFs could play a regulatory role (108)—or, alternatively, that ribosomes in the 3′ UTR occasionally remain unproductively associated with transcripts and require “rescue” (107). Accordingly, few, if any, dORF- or doORF-encoded microproteins have been characterized to date, and whether the roles of these classes of smORFs go beyond cis translational regulation remains to be addressed.

Mitochondrial DNA smORFs

Finally, a class of microproteins including humanin, the short humanin-like peptides (SHLPs), SHMOOSE, and MTALTND4 has been identified that maps to the mitochondrial genome (Figure 4) (108, 109). Interestingly, several of the smORFs encoding these microproteins overlap the noncoding mitochondrial ribosomal RNAs and mitochondrial transfer RNA genes, and their expression mechanisms are currently undefined. Surprisingly, mitochondrial DNA–encoded microproteins not only participate in energy metabolism but also regulate diverse aspects of physiology, including inflammation and neurodegeneration, and several are secreted, again by an undefined mechanism (108, 109). The molecular rationale for encoding secreted microproteins in the mitochondrial genome is currently unknown.

Overall, the locations of microproteins within ncRNA, mRNA, and mitochondrial-derived RNAs are of interest with respect to the mechanisms by which microproteins are expressed, translated, and regulated, and also have important implications for the functions that these microproteins carry out.

Metabolism

Several translated smORFs, such as humanin, have emerged as important players in metabolic homeostasis (10, 113) (Table 1). As a mitochondrial-derived peptide (MDP)—a microprotein encoded in the mitochondrial DNA—humanin confers cytoprotection against Alzheimer’s disease. The molecular mechanism of humanin is not fully understood, and multiple targets have been proposed (114), including G protein–coupled formylpeptide receptor-like-1 (115), insulin-like growth factor–binding protein 3 (116), and the trimeric complex of ciliary neurotrophic factor receptor alpha (CNTFR), IL-27 receptor subunit WSX-1, and glycoprotein 130 (113). The physiological effects of humanin are thought to extend beyond neuroprotection to other aging-related diseases, including type 2 diabetes mellitus via activation of signaling pathways such as MEK (MAPK/ERK kinase)/ERK (extracellular signal-regulated kinase), PI3K (phosphatidylinositol 3-kinase)/AKT (protein kinase B), and JAK (Janus kinase)/STAT (signal transducer and activator of transcription) (117). Humanin has been implicated in improving insulin sensitivity, particularly in the context of polycystic ovary syndrome, and increasing health span (118–120). A class of MDPs closely related in sequence to humanin is composed of six SHLPs (121). The most well-studied of these is SHLP2, which has been proposed to be cytoprotective in age-related macular degeneration and Parkinson’s disease, as well as to regulate energy homeostasis and insulin sensitivity, much like humanin (121–123). MOTS-c is yet another MDP with functions in improving insulin resistance and glucose metabolism, specifically in skeletal and heart muscle tissue (124). Finally, the MDP known as MOXI or mitoregulin is required for very long chain fatty acid oxidation (29, 125, 126). The MDPs therefore broadly play roles in metabolism, insulin resistance, and aging, not only suggesting new therapeutic avenues for neurodegeneration and diabetes but also intriguingly linking their mitochondrial DNA origin to energy homeostasis.

Microproteins encoded in both the mitochondrial and nuclear genomes can regulate mitochondrial bioenergetics directly and indirectly. For example, MTLAND4 is a secreted MDP that represses mitochondrial respiration (127). In contrast, mutations in the MDP SHMOOSE, like humanin, are associated with Alzheimer’s disease, and the SHMOOSE microprotein localizes to mitochondria and promotes aerobic respiration (128). In an example with strong mechanistic support, the BRAWNIN microprotein acts in concert with the canonical small proteins UQCC3, UQCC4, and SMIM4 to promote assembly and function of respiratory chain complex III (129), as does mitolamban (130). Interestingly, uORF-encoded microproteins have also been associated with mitochondrial energetics: The inner membrane–localized SLC35A4-MP promotes mitochondrial respiration by an unknown mechanism (131); the glioblastoma suppressor MP31 encoded upstream of PTEN inhibits lactate conversion to pyruvate (132); and the MIEF1 upstream microprotein (now annotated as MIDUO) promotes mitochondrial function, likely via interaction with the mitoribosome (133). Interestingly, mitochondrial microproteins can also have tissue-specific functions beyond metabolism, such as the murine Kastor and Polluks microproteins expressed from a bicistronic mRNA. These outer mitochondrial membrane–localized proteins are required for mitochondrial condensation during spermatogenesis (134). Overall, microproteins both derived from and localizing to the mitochondrion have wide-ranging effects on human metabolism and aging-related diseases.

Cellular Transport

Transporters and pumps compose another class of proteins that can interact with microproteins, with several examples in bacteria (135, 136) as well as in eukaryotes (Table 2). We consider two eukaryotic microproteins that interact with transporters and regulate movement of their substrates across membranes. First, the 14-kDa alt-RPL36 protein, which overlaps the CDS of human ribosomal protein L36 in an alternative reading frame, regulates TMEM24 (78). TMEM24 transports phospholipids across ER-plasma membrane contact sites in response to calcium signaling (137, 138). The lipid substrates of TMEM24, including phosphatidylinositol among others, once in the plasma membrane, function in processes such as insulin secretion, neuronal signaling (137, 138), and Akt activation (78). Importantly, although the molecular details remain to be fully elucidated, phosphorylated alt-RPL36 represses TMEM24 to downregulate inositide levels and Akt signaling. Alt-RPL36 repression does not occur in the absence of its phosphorylation, suggesting a mechanism by which TMEM24 activity could be dynamically regulated.

NEMEP is a recently reported transmembrane microprotein previously annotated as a lncRNA in mouse (139). Its human homolog, encoded by TMEM155, is an annotated small protein (SMIM43) of unknown function. During mouse embryonic development, NEMEP interacts with glucose transporters GLUT1 and GLUT3 to promote glucose uptake, a process required for mesendoderm formation.

Muscle and Heart Function

Muscle and heart contractility is a Ca²⁺ signaling–dependent process that relies on the activity of sarcoplasmid reticulum (SR)/ER Ca²⁺ ATPase (SERCA), a pump that sequesters Ca²⁺ in the sarcoplasmic reticulum to terminate contraction (Table 3). SERCA isoforms are expressed in different tissues and are precisely regulated in space and time by a suite of microprotein effectors, the regulins (140). Phospholamban and sarcolipin have long been known as SERCA inhibitors in heart and skeletal muscle, respectively, whose repression is alleviated by their phosphorylation. Surprisingly, lncRNAs specifically expressed in cardiac and skeletal muscle encode additional, critical SERCA regulatory microproteins, which have been termed micropeptides in the literature. The first discovered novel regulin was myoregulin, which is expressed in all skeletal muscle, where it regulates SERCA1 (26). Genetic deletion of myoregulin improves muscle performance by improving Ca²⁺ cycling. Subsequently, endoregulin and another-regulin were identified and found to inhibit and colocalize with nonmuscle SERCA isoforms SERCA2a in cardiac tissue and SERCA3 in other epithelial cells, respectively (27). Finally, in contrast to the tissue-specific inhibitory regulins, DWORF is a muscle-expressed, lncRNA-encoded micropeptide activator of SERCA, which may act by displacing phospholamban to alleviate its inhibition, although DWORF has also been reported to exhibit direct activation activity (30, 141). DWORF overexpression has been reported to alleviate dilated cardiomyopathy (141, 142). As discussed in the Supplemental Text, distantly related regulins have also been identified in invertebrates, suggesting that this class of micropeptides is broadly conserved.

Microproteins have also been reported to regulate muscle development and regeneration. A key molecular cascade in muscle regeneration, mTORC1 signaling, is downregulated by the lncRNA-encoded SPAR microprotein (143). Highly expressed in skeletal muscles, SPAR interacts with v-ATPase in lysosomal membranes and keeps it strongly bound to the Ragulator complex and Rags, thus inactivating mTORC1 even in the presence of amino acids. The increase in myogenic gene expression downstream of increased mTORC1 activation in SPAR-deficient mice likely explains the faster healing observed postinjury. These same myogenic genes are also required for skeletal muscle formation during development, which itself relies on a microprotein named myomerger or myomixer (144). This microprotein is embedded in the plasma membrane of myoblasts, enabling them to fuse into myofibers making up muscle tissue. Finally, Mustn1 is a microprotein involved in muscle extracellular matrix remodeling (145).

Gene Expression and Genome Maintenance

Numerous mammalian microproteins have been reported to play roles in gene expression both inside and outside the nucleus (Table 4). First, we consider microproteins associated with chromatin. An example is SEHBP, which was identified in association with histone H2B and dysregulates transcription globally when deleted (146). A microprotein encoded in a circRNA, LINC-PINT, has also been reported to promote transcriptional termination via interaction with PAF1c (147). Microproteins are also involved in DNA repair. In particular, isoforms of the CYREN (formerly named MRI) microprotein interact with the Ku heterodimer to promote association of nonhomologous end joining (NHEJ) factors to promote DNA double strand break repair during G1 phase of the cell cycle (148, 149). CYREN is also required to prevent aberrant fusion of telomeres by NHEJ during S and G2 (150). More recently, the lncRNA-encoded DDUP microprotein was found to be upregulated by DNA damage, whereupon it retains RAD18 at sites of DNA damage to promote both homology-directed and postreplication repair (151). DDUP expression also correlates with cisplatin resistance (152), suggesting that DNA repair–promoting microproteins could be therapeutically relevant.

Multiple nucleolar microproteins have been reported to regulate ribosome biogenesis as well as translation by mature ribosomes. For example, C11ORF98 interacts with nucleolar proteins nucleophosmin and nucleolin, although its function is otherwise currently unknown (39). The 148-aa, uORF-encoded alt-LAMA3 protein localizes to the nucleolus and interacts with the PeBoW complex to promote ribosomal RNA transcription (73). MINAS-60, which is encoded in an iORF that overlaps the RBM10 gene, also localizes to the nucleolus and downregulates large ribosomal subunit export into the cytoplasm (74). There are thus multiple microproteins localized to the nucleolus in mammalian cells, and it seems reasonable to speculate that additional microproteins may localize to this membraneless organelle to regulate the complex pathway of ribosome biogenesis. Of course, the job of assembled, mature ribosomes is to translate mRNA, and the onco-microprotein APPLE, which is encoded in an antisense lncRNA, has been reported to regulate this process, influencing translation initiation by promoting the interaction between PACB1 and eIF4G (153). Microproteins thus affect protein synthesis by regulating both ribosome biogenesis and translation.

Finally, gene expression depends on not only transcription but also RNA stability and degradation. The NBDY microprotein directly regulates mRNA decay by binding to the cytoplasmic mRNA decapping complex via the coactivator proteins DCP1A/B and EDC4 (38). Deletion of NBDY broadly dysregulates mRNA half-lives in human cells (93). A substantial fraction of mRNA decapping enzyme (DCP2) substrates are stabilized in NBDY knockout cells, including mRNAs encoding factors involved in inflammation and immunity, suggesting that NBDY is required for their decapping and subsequent degradation (93). However, a smaller subset of DCP2 substrates exhibit dramatic destabilization in cells lacking NBDY (93, 154). Phosphorylation of NBDY promotes dissociation of membraneless organelles termed P-bodies that are associated with the mRNA decapping complex (77); this phenomenon may contribute to its regulation of specific DCP2 substrates, but the mechanistic link between P-bodies and mRNA decapping remains unclear. The example of NBDY demonstrates how a microprotein can regulate protein localization and association to control mRNA stability.

Proteostasis

Proteostasis is the maintenance of proteome integrity via the balance of protein synthesis, quality control, and degradation, and microproteins are associated with this process at multiple levels (Table 5). First, some noncanonical translation products, particularly hydrophobic peptides expressed from noncoding genomic regions, can be targeted for proteasomal degradation by the GET complex (81). Multiple ubiquitin-like microproteins have also been reported (155, 156) and have recently been reviewed (63). For example, the pTINCR ubiquitin-like microprotein has been described by several groups as a tumor suppressor in squamous cell carcinoma (157, 158), highlighting the disease relevance of this class of microproteins. Microproteins are thus regulated by, and also may contribute to, protein degradation.

Microproteins are involved in chaperones and the unfolded protein response (UPR). Notably, the ASDURF microprotein has been identified as the previously unknown sixth subunit of the prefoldin-like subunit of the PAQosome (103), which is required for assembly of large, complex molecular machines (159). ASDURF is upregulated in childhood medulloblastoma (49). In another example, the PIGBOS microprotein has been reported to function in ER-mitochondria junctions to regulate the UPR (97). PIGBOS is encoded in a lncRNA antisense to the PIGB gene. PIGBOS interacts with the CLCC1 protein, and this interaction is required for regulation of the UPR. In its absence, cells are sensitized to ER stress and apoptosis. Microproteins thus contribute to proteostasis at multiple levels, with significance for cancer.

Immunity and Inflammation

Multiple microproteins have been implicated in the immune system, especially in regulation of inflammation and innate immunity (Table 6). For example, the rodent lncRNA Aw112010 encodes a secreted (83) microprotein that is translationally upregulated in M1 polarized macrophages and is essential for innate immune responses against bacterial pathogens (21). In contrast, the Mm47 microprotein is downregulated in lipopolysaccharide-treated myeloid cells but is required for Nlrp3 inflammasome activation (160). More recently, the lncRNA-encoded, ER-localized MAVI1 microprotein was found to interact with mitochondrial MAVS to inhibit type I interferon signaling (161).

Interestingly, two miPEPs are involved in innate and adaptive immune regulation. One of these is MOCCI (162), which, interestingly, is a miPEP whose transcript can be processed to generate miR-147b, which regulates RIG-I antiviral defense and downregulates the NDUFA4 mRNA. The latter is consequential because the MOCCI microprotein replaces NDUFA4 in respiratory complex IV during inflammation to lower mitochondrial membrane potential and reduces reactive oxygen species production, providing cytoprotection and reducing interferon responses in endothelial cells. MOCCI thus provides multilevel regulation of mitochondrial activity, inflammatory signaling, and antiviral gene expression. Another miPEP is encoded in the lncRNA MIR155HG, upstream of pri–miR-155 (163). The micropeptide, termed P155, inhibits antigen presentation on dendritic cells by MHC II by disrupting chaperones. Although P155 would thus be detrimental to adaptive immunity, treatment with synthetic P155 was beneficial in animal models of autoinflammation (163). Overall, microproteins play important roles in immunity and in the future may inform therapeutic approaches to immunomodulation.

Cancer and Cell Cycle Regulation

Perhaps the greatest advances in our understanding of microproteins in human disease pertain to cancer (166) (Table 7). This topic has been extensively and recently reviewed (115, 167, 168). We have already discussed several microproteins that regulate tumor growth above; thus, in this section we highlight several key additional examples of cancer-associated microproteins as well as microproteins that regulate the cell cycle.

In addition to MP31, PINT87aa, and pTINCR, numerous tumor suppressor microproteins have been reported that regulate multiple aspects of cancer cell biology. lncRNA-encoded MIAC has been reported to not only in head and neck squamous cell carcinoma, repress cell growth and migration (169) but also to cause enhanced apoptosis in renal cell carcinoma (170); both phenotypes are proposed to occur through the interaction of MIAC with aquaporin 2 to inhibit actin cytoskeleton processes—a link that may require further mechanistic validation. Similarly, miPEP133—yet another disease-associated human pri-miRNA-encoded microprotein—has been shown to harbor tumor suppressor activity because it increases apoptosis downstream of TP53 transcriptional activation in nasopharyngeal carcinoma cells (171). Esophageal squamous cell carcinoma has been reported to be more invasive during nutrient deprivation by downregulating the lncRNA-derived polypeptide YYB1M (172), which blocks the YY1–androgen receptor interaction during nutrient deprivation, thereby inhibiting eEF2K signaling and promoting apoptosis. Multiple additional tumor suppressor microproteins, such as N1DARP (173), CIP2A-BP (174), and SMIM26 (175), have been reported, suggesting that anticancer microproteins may be numerous.

Like ASDURF and APPLE, a substantial number of oncogenic microproteins have also been reported. SMIM30 originates from an lncRNA, and its oncogenic mechanisms are particularly well-studied. SMIM30 is upregulated in many cancers, where it has been reported to enhance the G1/S transition and thus proliferation (176). Interestingly, SMIM30 is anchored in the ER membrane where it decreases cytosolic Ca²⁺ (176). SMIM30 was further found to promote hepatocellular carcinoma progression by upregulating MAPK signaling via the membrane localization of SRC/YES1 (177). The TRPC5OS microprotein, which is encoded in a normally testis-specific antisense lncRNA, has been reported to be aberrantly expressed in breast cancer cells, where it promotes glucose uptake and cell proliferation (178). Several other recently reported lncRNA-encoded microproteins such as ASAP (179) and PACMP (180), among others, have also been reported to promote tumor growth and/or survival. Overall, microproteins can influence cancer cell signaling and metabolic pathways, conferring growth advantage or inhibition on tumor cells. These findings are likely to place smORFs at the forefront of advances in cancer diagnosis and treatment.

As discussed above, microproteins are overrepresented relative to canonical proteins in the MHC I immunopeptidome and have been directly detected in the immunopeptidomes of melanoma (78) and hepatocellular carcinoma (79). Although these HLA-displayed microprotein fragments may not directly affect cancer cell biology, they are of outsize importance to immunotherapy and have been proposed to inform the design of tumor-specific anticancer peptide vaccines (78, 79).