Abnormal mitochondrial structure and function in brown adipose tissue of SLC35A4-MP knockout mice

Abstract

Uncovering the role of upstream open reading frames (uORFs) challenges conventional views of one protein per messenger RNA and reveals the capacity of some uORFs to encode microproteins that contribute to cellular biology and physiology. This study explores the functional role of a recently identified mitochondrial microprotein, SLC35A4-MP, in the brown adipose tissue of mice. Our findings reveal dynamic regulation of SLC35A4-MP expression during primary brown adipocyte differentiation in vitro and during cold exposure or high-fat diet (HFD)–induced obesity in mice. Using a knockout mouse model, we show that loss of SLC35A4-MP disrupts mitochondrial lipid composition, decreasing cardiolipins and phosphatidylethanolamine in brown adipose tissue from HFD-fed mice. SLC35A4-MP deficiency also impairs mitochondrial activity, alters mitochondrial number and morphology, and promotes inflammation. Knockout mice accumulate acylcarnitines during cold exposure, indicating defective fatty acid oxidation. These findings reveal SLC35A4-MP as a previously unrecognized microprotein in regulating mitochondrial function and tissue lipid metabolism, adding to the growing list of functional endogenous microproteins.

A uORF-encoded microprotein, SLC35A4-MP, modulates mitochondrial morphology and function in brown adipose tissue.

INTRODUCTION

Protein annotation efforts had a blind spot for protein-coding small open reading frames (smORFs) because the features of their peptide and small protein products, or microproteins, suggested that they were unimportant, randomly translated sequences since many were not evolutionarily conserved (1, 2). Moreover, many of these smORFs were in the noncoding sequences of known mRNAs that already had a protein-coding ORF. This violated the long-standing dogma that mRNAs are translated to produce a single protein product (3). Instead, upstream smORFs in the 5′ untranslated region (5′UTR), or upstream ORFs (uORFs), are considered to be the functional genetic element (4, 5).

The classic examples of functional uORFs come from work on the unfolded protein response, where the transcription factors ATF4 and CHOP mRNAs contain uORFs that prevent translation of the downstream ATF4 and CHOP main ORFs (mORFs) (6, 7). In this model, the ribosome translates the uORF and is then displaced from the mRNA to prevent the mORFs from being translated. Upon unfolded protein response, the phosphorylation of the ribosome leads to leaky scanning to bypass these uORFs and promote the translation of the mORFs, providing a rapid posttranscriptional mechanism for regulating protein levels (6).

Informatics analysis of transcriptomes identified that approximately half of all human and mouse mRNAs contain uORFs (8). Additional research into uORFs further identified interesting roles for these genetic elements. For example, a translational analysis of the SOX2 transcription factor revealed that uORFs are critical for increasing the translation of oncogenic mRNAs while overall translation levels are reduced (4). Mechanistic studies showed that eIF2A is responsible for engaging uORFs and maintaining oncogene expression in this work.

Another study revealed the role of a uORF in metabolism by controlling the translation of PGC1a (5). PGC1a drives mitochondrial biogenesis. Measurement of mRNA ribosome occupancy using ribosome profiling (Ribo-Seq) protocols provided empirical evidence for a uORF in the 5′UTR of PPARGC1A, the gene that encodes PGC1a. In vitro experiments with synthetic constructs revealed the posttranscriptional inhibition of PGC1a mORF translation by this uORF, as removal or mutation of the uORF led to markedly increased PGC1a levels, validating that this uORF, like others, is a negative regulator of mORF translation. Evolutionary analysis of this uORF demonstrated its active role in vivo, as bluefin tuna do not contain this uORF, which leads to higher PGC1a levels and increases their mitochondrial levels to maintain body temperatures in the cold ocean (5). Thus, uORFs as cis-regulators of translation are well established.

However, in addition to their regulatory roles in translation, some uORFs are known to produce functional microproteins. While uORFs that act as cis-regulators of translation all seem to down-regulate downstream ORF translation (8), the roles for uORF-derived microproteins are emerging more slowly because the function of each microprotein can be very different. A uORF from mitochondrial elongation factor 1 (MIEF1) mRNA is an excellent example of the work necessary to characterize cellular microprotein function. The uORF on the MEIF1 mRNA encodes the MIEF1 microprotein (MIEF1-MP), a bioactive 70–amino acid microprotein targeted to the mitochondria. This microprotein has been shown to regulate mitochondrial size and fragmentation and is thought to do so through an interaction with the downstream MIEF1 mORF encoded protein (9, 10).

Structural biology (11) and proteomics (10) found that MIEF1-MP binds to and regulates the translation of the mitochondrial ribosome. While it was assumed that the MIEF protein and not the microprotein was responsible for the biological function from this mRNA, quantitative proteomics clearly showed that the MIEF1-MP is expressed at two to sixfold higher cellular levels than MIEF1, making this microprotein the primary translation product from the MIEF1 mRNA. The MIEF1-MP demonstrates that some polycistronic mRNAs primarily encode microproteins, which is expected given the mechanism of translation where the ribosome typically falls off an mRNA at a stop codon (12), in this case, the uORF. This exemplar piqued our interest in determining whether other functional uORF-derived microproteins exist.

This led us to a uORF-encoded microprotein on the SLC35A4 mRNA, or SLC35A4-MP (UniProt entry L0R6Q1, S35U4), a 103-amino acid microprotein (13, 14). SLC35A4-MP is a highly conserved microprotein that contains a single-pass transmembrane domain. Biochemical and imaging studies showed that this microprotein is localized to the inner mitochondrial membrane (IMM) (13). Like IMM proteins, the loss of SLC35A4-MP leads to changes in mitochondrial function related to metabolism, as human embryonic kidney (HEK) 293 cells lacking this microprotein have diminished oxygen respiration.

To date, physiological studies of microproteins have primarily focused on the role of these genes in muscle via the discovery of a family of microprotein regulators in muscle (15–17). The microprotein myoregulin, for example, is encoded by an mRNA that was thought to be noncoding, and this microprotein is embedded within the sarcoplasmic reticulum of muscle cells (15). Via protein-protein interactions within the membrane, myoregulin binds to and modulates the activity of sarcoplasmic/endoplasmic reticulum Ca²⁺–adenosine triphosphatase (SERCA), an ion pump. SERCA controls calcium flux from the cytosol into the sarcoplasmic reticulum, ensuring proper muscle function. Loss of myoregulin in vivo leads to increased muscle function and endurance in mice, highlighting this microprotein’s critical role. Furthermore, myoregulin belongs to a prominent family of microproteins that can inhibit or enhance SERCA function (17).

Because of its role in cellular respiration and localization to mitochondria, we hypothesized that SLC35A4-MP could have essential functions in endogenous metabolism and set out to knock out this microprotein in mice. When dealing with polycistronic RNA, it is vital to consider whether a gene knockout (KO) might affect two proteins instead of one. The evidence for SLC35A4-MP’s translation and function is robust, starkly contrasting the lack of supporting data for SLC35A4 itself. As we detail in our work, there is no evidence for translation of the SLC35A4 protein, as it is undetectable by proteomics and not supported by ribosome profiling data. While ribosome profiling shows consistently strong coverage of the uORF, proteomics data further support this model. Peptides corresponding to SLC35A4-MP are readily detected in large-scale proteomics repositories such as the MassIVE Knowledge Base (MassIVE-KB) (18, 19) and PepCentric (20), both of which aggregate community proteomics data, revealing peptides from all SLC35A family members except SLC35A4. Even with overexpression conditions, identification of mORF peptides required enrichment and optimized workflows, suggesting that this protein is not appreciably produced under physiological conditions. These findings reinforce the interpretation that SLC35A4-MP is the primary functional product of the SLC35A4 locus (13, 21) and motivated us to investigate how this microprotein contributes to mitochondrial biology in metabolically active tissues.

Of course, the evidence of absence is not the absence of evidence, so we looked at other SLC35A family members, homologs of the SLC35A4 protein. We find that SLC35A1, A2, A3, and A5 have proteomic evidence, while SLC35A4 does not. These analyses are important because we find that our KO mouse has lower levels of Slc35a4 mRNA, but since there is no good evidence for SLC35A4 protein, we are confident that our results are reporting on the role of the microprotein. Lastly, even if SLC35A4 is translated, it is unlikely to influence the mitochondrial biology of SLC35A4-MP. The overexpression of tagged SLC35A4 localized to the ER-Golgi, while SLC35A4-MP localized to the IMM (13), making it physically impossible for these two proteins to interact.

A priori, we did not know where SLC35A4-MP would exert its strongest effect, but we focused on tissues where it is highly expressed, including brown adipose tissue (BAT). BAT is central to mitochondrial lipid catabolism and thermogenesis, particularly under metabolic stress conditions, including cold exposure or a high-fat diet (HFD) (22). These conditions provide a relevant physiological context to assess the function of mitochondrial microproteins like SLC35A4-MP. Given the overwhelming evidence supporting the importance of SLC35A4-MP and the lack of evidence supporting SLC35A4 translation, we generated a KO model targeting the Slc35a4 uORF to investigate the role of the microprotein in vivo.

We found that SLC35A4-MP is strongly up-regulated during HFD feeding, cold exposure, and brown adipocyte differentiation, consistent with a role during periods of increased metabolic demand. Loss of SLC35A4-MP led to reduced levels of mitochondrial lipids, including cardiolipin (CL) and phosphatidylethanolamine (PE), increased phosphatidic acid (PA), a lower number of mitochondria, reduced mitochondrial activity, and elevated inflammatory markers in BAT. Acute responses to cold exposure were also impaired in KO animals.

Together, the data reveal SLC35A4-MP as a previously unrecognized regulator of mitochondrial function and lipid metabolism in vivo. As one of the few microproteins to be characterized in mice (23, 24), SLC35A4-MP emerges as a potential regulator of mitochondrial and metabolic health, with possible implications for human metabolic diseases such as obesity and dyslipidemia.

RESULTS

Tissue distribution of the SLC35A4-MP and its regulation by an HFD

The characterization of SLC35A4-MP in cell culture has identified it as a transmembrane microprotein essential to the mitochondrial inner membrane, necessary for optimal mitochondrial function (13). Specifically, the loss of SLC35A4-MP in HEK293 cell lines leads to a significant decrease in mitochondrial respiratory capacity, emphasizing its contribution to cellular metabolic processes. Whether SLC35A4-MP has a similar role in vivo is unknown.

We began to explore the physiological regulation and role of SLC35A4-MP by determining its tissue distribution using SLC35A4-MP–specific antisera (13). We selected tissues from C57BL/6J mice, emphasizing metabolic tissues because of the importance of mitochondrial cellular energetics in these tissues. Under basal conditions, SLC35A4-MP is most highly expressed in the kidney and BAT, followed by modest expression in the spleen, lung, and heart, and low but detectable expression in the brain, testis, epididymal white adipose tissue (eWAT), skeletal muscle, and liver (Fig. 1A). These results were consistent with the tissue expression of Slc35a4 mRNA (13).

Fig. 1. SLC35A4-MP expression across tissues and its up-regulation in BAT by HFD.

(A) Representative SLC35A4-MP blot in various tissues, including brain, kidney, lung, spleen, testis, BAT, eWAT, skeletal muscle (gastrocnemius), heart, and liver from male C57BL6 wild-type mice. Ponceau staining served as a protein level control. (B) HFD and chow diet feeding schematic and protocol for BAT collection were created in BioRender. Saghatelian, A. (2025) https://BioRender.com/9luycjt. (C) Representative SLC35A4-MP blot and (D) quantification in BAT of HFD-fed C57BL/6 male mice (n = 4 per group). (E) qPCR analysis of Slc35a4 mRNA expression in BAT of HFD-fed C57BL/6 male mice (n = 4 to 6 per group). Data presented as means ± SEM, with * indicating statistical significance (P < 0.05, unpaired t test) compared to the chow diet.

On the basis of these preliminary results in HEK293T cells that demonstrated a role for SLC35A4-MP in mitochondrial metabolism, we hypothesized that conditions that influence BAT mitochondrial function could regulate SLC35A4-MP levels. BAT metabolism is known to be dysregulated upon increased lipid deposition and a concomitant decrease in mitochondrial activity in obese mice (25). Therefore, we compared SLC35A4-MP expression in BAT from C57BL/6J mice on a chow diet or an HFD for 12 weeks (Fig. 1B). The data showed that an HFD challenge leads to a marked increase in SLC35A4-MP levels in BAT (2.5-fold; Fig. 1, C and D). Likewise, the HFD conditions increase Slc35a4 mRNA levels (1.5-fold; Fig. 1E) compared to chow-fed mice. HFD regulation of SLC35A4-MP and its transcript levels is not observed in the kidney (fig. S1, A and B) or eWAT (fig. S1, C and D).

SLC35A4-MP KO mice are viable with no overt phenotypes

To understand the role of SLC35A4-MP in BAT, we generated KO mice by CRISPR-Cas9 targeting of the SLC35A4 uORF. A guide RNA (gRNA) was designed where the cutting location is in the second codon of the SLC35A4 uORF to disrupt the expression of a functional uORF (fig. S2A). gRNA, trans-activating CRISPR RNA (tracrRNA), and Cas9 protein (all IDT) were combined to form a ribonucleoprotein (RNP) that was injected into fertilized oocytes from C57BL/6 donor mice (26), which were subsequently transplanted into pseudo-pregnant mice to generate mosaic founder generation (F0) offspring. To segregate mosaic mutations in the F0 offspring, these mice were mated with C57BL/6 wild-type (WT) mice to produce first generation (F1) offspring. Polymerase chain reaction (PCR) and sequencing determined mutations in F1 mice, and a strain with a 10–base pair (bp) deletion was established, which eliminates the SLC35A4 uORF. Routine genotyping was done with the primers indicated in table S1. The mice were healthy and viable with no overt phenotypes and normal breeding, which allowed for the generation of a colony for biological studies. Western blot also confirmed KOs in BAT, demonstrating the loss of the SLC35A4-MP (fig. S2B). While these observations suggested that loss of SLC35A4-MP alone does not result in overt phenotypic abnormalities, we next sought to determine whether the canonical SLC35A4 protein is endogenously expressed from this locus and could contribute to the phenotypes observed in our KO mice. This was relevant since quantitative PCR (qPCR) analysis revealed that the Slc35a4 mRNA was knocked down in these mice (fig. S2C).

SLC35A4-MP is the primary translation product of the SLC35A4 locus

We investigated the mORF protein expression by analyzing public proteomics datasets, ribosome profiling data, endogenous tissues, and FLAG-tagged overexpression systems. To address whether SLC35A4 protein could be expressed endogenously and contribute to our KO phenotypes, we analyzed large public proteomics databases. In the MassIVE Protein Explorer, which contains more than 2 billion spectra from thousands of publicly available mass spectrometry (MS) datasets, SLC35A4-MP was robustly detected with 1676 peptide spectrum matches (PSMs) (fig. S3A). In contrast, no peptides corresponding to the canonical SLC35A4 protein were identified. Other members of the SLC35A family, including SLC35A1, A2, A3, and A5, were all detected in this dataset, indicating reliable coverage for this protein family (fig. S3A). We extended our analysis to the PepCentric database containing 2.3 billion spectra across 76 datasets. Here, SLC35A4-MP was identified in 7842 independent runs, while the canonical SLC35A4 protein, again, remained undetected (fig. S3B). Last, in human adipose tissue proteomics data derived from Wang et al. (27), SLC35A4-MP was detected with six PSMs, and SLC35A3 was identified with a single PSM, while the canonical SLC35A4 protein was not detected (fig. S3C). These findings indicated an absence of canonical SLC35A4 protein expression under physiological conditions. However, to exclude the possibility that this was due to detection limits rather than biological absence, we next asked whether the protein could be detected at all, even under forced expression conditions.

To increase the chance of detection and ensure that SLC35A4 generates detectable tryptic peptides, we overexpressed SLC35A4 with a FLAG tag in HEK293T cells. We analyzed both total lysate and FLAG-immunoprecipitated (IP) samples by proteomics. No peptides corresponding to the canonical SLC35A4 protein were detected in the total lysate. However, in the FLAG IP–enriched samples, using optimized trypsin and chymotrypsin digestion protocols, we identified a small number of peptides (one with chymotrypsin and three in each trypsin replicate) mapping to the SLC35A4 protein (fig. S4, A and B). Although peptide counts and sequence coverage were low, this represents the first detection of mORF-derived peptides using MS. Notably, these peptides were absent from our earlier experiments using standard protocols and were also not found in the community proteomics datasets aggregated in MassIVE-KB and PepCentric. We next asked whether these canonical peptides could be detected in endogenous tissues under physiological conditions.

On the basis of the peptides identified in the IP-enriched overexpression samples, we performed targeted proteomic analysis of kidney tissue from SLC35A4-MP KO mice and their WT mice using tandem mass tag (TMT) labeling. We chose the kidney due to its high expression of other solute carrier family proteins, including SLC35A1, SLC35A2, and SLC35A3, which would provide the best chance to detect endogenous SLC35A4 if translated. We specifically included the canonical SLC35A4 peptides from the HEK293T IP samples in our targeted search. Despite this focused approach, no peptides corresponding to the canonical SLC35A4 ORF protein were detected in the endogenous kidney samples. Given these consistent findings across public datasets and tissue proteomics, we turned to ribosome profiling to independently evaluate the translational potential of the canonical SLC35A4 ORF.

Using processed alignment files from Martinez et al. (28), which profiled ribosome occupancy in brown, white, and beige murine adipocytes, we observed robust ribosome coverage over the uORF of Slc35a4 (fig. S5). In contrast, coverage across the mORF was minimal or absent across all conditions and tissues analyzed. While a few isolated peaks appeared within the mORF region (fig. S5), they were low in magnitude, typically ≤3 normalized counts, and none exceeding 5, and are likely background signals rather than evidence of active translation.

Together, these proteomic and ribosome profiling data demonstrate that the canonical SLC35A4 protein is not expressed at detectable levels under physiological conditions, while SLC35A4-MP is robustly translated. Combined with the loss of SLC35A4-MP observed in our KO mice, these findings confirm that our model specifically disrupts the microprotein. Given the high expression of SLC35A4-MP in metabolically active tissues such as BAT, we next explored how the loss of this microprotein affects tissue metabolism in vivo.

Impact of SLC35A4-MP deletion on metabolic phenotypes, lipid droplets, and the lipidome

We observed that SLC35A4-MP expression is up-regulated in BAT under HFD conditions, suggesting a potential role in metabolic adaptation. To test this, we designed an experiment where SLC35A4-MP KO mice and WT littermate controls were fed a 60% HFD for 12 weeks to assess the impact of SLC35A4-MP loss on metabolism. At the end of this period, we noticed a significant increase in body weight in HFD-fed mice (Fig. 2A) but no genotype differences in female or male mice (Fig. 2A and fig. S6A). Furthermore, we characterized these animals further by measuring adipose tissue weight (Fig. 2B), fasting glucose levels (Fig. 2C), glucose tolerance (Fig. 2, D and E and fig. S6, B and C), and insulin resistance (fig. S7, A and B). However, none of these conditions showed any differences between genotypes.

Fig. 2. Impact of HFD-induced obesity on SLC35A4-MP expression in BAT.

Effects of 12-Week HFD in female SLC35A4-MP KO (KO) and WT mice. (A) Initial and 12-week body weight under HFD feeding (two-way ANOVA followed by Šídák’s multiple comparisons test). (B) Fat tissue weight; (C) fasting blood glucose levels; (D) glucose tolerance test (GTT), also expressed as (E) the area under the curve (AUC); (F) BAT histology: hematoxylin and eosin stain in BAT. Scale bar, 100 μm. (G) Quantification of lipid droplet (LD) size in micrometer. (H to J) LC-MS/MS–based untargeted lipidomics of BAT from SLC35A4-MP KO mice and their controls after 12 weeks of HFD feeding (n = 5 per group). (I) The total sum of triacylglycerol species detected in BAT. (J) Heatmap showing the total sum levels of phosphatidic acid (PA, P < 0.001), acylcarnitines (AcCa), phosphatidylinositol (PI), phosphatidylserine (PS), lysophosphatidylcholine (LPC), phosphatidylglycerol (PG), phosphatidylcholine (PC), lysophosphatidylethanolamine (LPE), lysophosphatidylglycerol (LPG, P < 0.02), phosphatidylethanolamine (PE, P < 0.007), and cardiolipin (CL, P < 0.007). Data presented as means ± SEM, with * indicating statistical significance (*P < 0.05, ****P < 0.0001, unpaired t test) compared to the WT mice. ns, not significant.

While there were no changes in these metabolic phenotypes on an HFD, we did observe marked differences between WT and KO BAT upon histology (Fig. 2F). Histological analysis of BAT revealed a notable reduction in the size of lipid droplets in the absence of SLC35A4-MP (Fig. 2F). We quantified this by looking at the distribution of large and small lipid droplets and found that BAT from WT mice had a significantly higher proportion of large lipid droplets (Fig. 2G). In contrast, the BAT from KO mice had smaller lipid droplets (Fig. 2G). While the deletion of SLC35A4-MP did not influence overall body weight or metabolic parameters, it directly affects BAT lipid droplet sizes and, more generally, lipid metabolism in BAT.

To determine which lipids and lipid pathways are being affected by the loss of SLC35A4-MP, we measured the global lipid profile in these tissues using liquid chromatography coupled with tandem MS (LC-MS/MS) in HFD-fed WT and SLC35A4-MP KO mice (Fig. 2H). The BAT lipidomics results showed decreased total triacylglycerol levels in SLC35A4-MP KO mice (Fig. 2I), consistent with the histological observations of smaller lipid droplets. A global analysis of the lipidomics data identified changes in specific lipid classes. In particular, we observed decreased CLs, PEs, and lysophosphatidylglycerols (LPGs) (Fig. 2J). The data also showed an increase in PAs in SLC35A4-MP KO BAT compared to WT (Fig. 2J and S8A). Lastly, we observed decreased polyunsaturated fatty acids (fig. S8, B to E) between KO and WT mice.

To examine whether these lipidomic changes correlate with any changes in gene expression, we carried out a targeted analysis of several genes in the mitochondrial respiratory complex (Nduf1s, Sdhb, Uqcrc1, Cox4l1, and Atp5pf) (fig. S8, F to K) and the mitochondrial carnitine palmitoyl transferase I (Cpt1b) in BAT from HFD-fed mice using qPCR (fig. S8G). We observed a decrease (~50%) in these mitochondrial respiratory complex gene markers in SLC35A4-MP KO mice (fig. S8, F to K) on HFD. Furthermore, we looked at the regulation of these genes in chow or HFD mice and only observed genotypic differences in their expression in the HFD case but not in chow (fig. S8, H to K). These data support changes in metabolic gene expression in SLC35A4-MP KO mice when challenged with an HFD, consistent with phenotypic differences observed in lipidomics under HFD conditions.

To establish whether the role of SLC35A4-MP in lipid metabolism is specific to BAT, we also used lipidomics to compare the hearts, a mitochondrial-rich tissue expressing SLC35A4-MP, from WT and KO mice (fig. S9A). Hearts from KO mice had increased levels of PA compared to WT mice (fig. S9B) and lower amounts of CLs (P = 0.07; fig. S9C) and polyunsaturated fatty acids (fig. S9, D to G), demonstrating that the loss of SLC35A4-MP in BAT and cardiac tissue has a similar impact on lipid metabolism in these tissues. Overall, these results highlight the role of SLC35A4-MP in regulating lipid metabolism, particularly in mitochondrial-rich tissues like BAT and the heart.

Structural and morphological impact of SLC35A4-MP deletion on mitochondria

CL is a specific mitochondrial lipid required for the proper structure and function of mitochondria (29). Therefore, the changes observed in CL levels may signal differences in mitochondrial structure between WT and KO mice. We interrogated this using transmission electron microscopy (TEM), which provided a clear picture of mitochondrial structures within BAT (Fig. 3A). The images show notable changes in the mitochondria of BAT from SLC35A4-MP KO mice when compared to WT mice. SLC35A4-MP KO BAT exhibited an increase in mitochondrial size (as indicated by mitochondrial profile area) and a decrease in mitochondria compared to either diet (Fig. 3, A to C). In addition, WT mitochondria are dynamic, with changing sizes and numbers that shift between diets (Fig. 3, A to C). The KO mitochondria are unchanged (Fig. 3, B and C).

Fig. 3. Impact of SLC35A4-MP deletion on mitochondrial size, shape, and dynamics in BAT.

Twelve-week HFD or chow diet in male SLC35A4-MP KO and WT mice. (A) Mitochondrial tomography slices: 1.6-nm thickness slices from the center of EM tomography volumes of BAT mitochondria from BAT of male WT mice (left) and SLC35A4-MP KO (right) mice with increased mitochondrial area and decreased numbers (scale bar, 1 μm). (B) Quantification of the mitochondrial profile area (size) (n = 14 to 54 per group), (C) number of mitochondria per unit area of cytoplasm (#mitochondria per square micrometer) (n = 10 per group), (D) mitochondrial shape (ratio major and minor axis) (n = 14 to 54 per group), and (E) cristae density (n = 10 mito per group). Data presented as means ± SEM, with * indicating statistical significance [*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, two-way analysis of variance (ANOVA)] compared to the WT mice.

In addition, the shapes of the mitochondria looked different, with WT mitochondria appearing rounder and KO mitochondria having a more oblong shape (Fig. 3D). We quantified this by calculating the ratios of the major and minor axes across all mitochondria. As expected, the ratios for the WT mitochondria were lower (i.e., more circular) than the KO mitochondria (Fig. 3D). While the changes to the overall shape, size, and number of mitochondria are different, there were no significant differences in mitochondrial cristae density (Fig. 3E), although this value did trend lower in KO HFD-fed mice (P = 0.06; Fig. 3E). These images establish SLC35A4-MP as having a role in mitochondrial size and dynamics in BAT and the response of these parameters to diet.

Quantitative proteomic analysis of SLC35A4-MP KO BAT reveals ECM remodeling and inflammatory pathways

We carried out a quantitative proteomics experiment to elucidate the molecular pathways affected by the loss of SLC35A4-MP. BAT tissue was collected from WT and SLC35A4-MP KO BAT samples under chow and HFD conditions. These samples were then processed using our standard proteomics workflow for TMT analysis (30). We obtained excellent proteome coverage in these samples: More than 2000 proteins were detected. We compared the samples using an unbiased analysis to identify the most significant changes between the samples. For example, when comparing WT and KO samples on chow, we found a small number of proteins with roles in lipid metabolism, including increases in acyl– coenzyme A (CoA) thioesterases and fatty acid–binding protein 3 (FABP3) and decreases in stearyl CoA desaturases (SCD1 and SCD1). Still, the fold changes were modest and did not add much to our understanding.

A similar analysis of the HFD-fed group provided more insight. First, several mitochondrial proteins involved in structural and functional maintenance, including PERM1, CHCHD7, and NDUFS8, were significantly reduced in KO BAT (Fig. 4A). Next, we observed an increased abundance of immune-related proteins, including galectin-3 (LGALS3), transporter associated with antigen processing (TAPBP), and intercellular adhesion molecule 1 (ICAM1), in KO BAT (Fig. 4B), suggesting activation of immune pathways. In parallel, an unbiased analysis identified enrichment of extracellular matrix (ECM) remodeling proteins in KO BAT, such as ASPN, OGN, BGN, COL6A3, DCN, LUM, VWA1, and VWA5A (fig. S10A). These proteomic shifts imply broader tissue remodeling that may result from sustained mitochondrial dysfunction.

Fig. 4. Proteomic and functional changes reveal mitochondrial dysfunction and inflammation in BAT of SLC35A4-MP KO mice.

(A to D) TMT-labeled proteomics analysis of BAT from male SLC35A4-MP KO mice fed an HFD for 12 weeks (n = 4 per group). Proteomic samples were processed using a standard TMT workflow, resulting in high proteome coverage with more than 2000 proteins detected. (A) Levels of mitochondrial metabolic protein Perm1, Chchd7, and Ndufs8 (n = 4 per group) and (B) relative abundance of inflammation-related proteins in BAT, including galectin-3 (Lgals3), macrophage-capping protein (Capg), tapasin (Tapbp), and intercellular adhesion molecule 1 (Icam1). (C) Representative F4/80 immunohistochemistry images with eosin counterstaining (40× magnification). (D) Quantification of F4/80⁺ staining area using ImageJ (% area; n = 5 per group). (E) qPCR analysis of Cd86 gene expression and (F) proinflammatory cytokines in BAT of male and (G) female mice (n = 5 per group). (H to J) Respiratory ratio analysis in BMDMs activated with IL-4 from male SLC35A4-MP KO and WT mice fed with chow diet. (I) Basal respiratory capacity (J) ATP-linked respiratory capacity using Seahorse XF96 (n = 12 per group). (K to M) Gene expression of proinflammatory cytokines in BMDMs activated with IFN-γ + LPS or IL-4, (K) Tnf-a mRNA, (L) Il1b mRNA, (M) Il12b mRNA (two-way ANOVA followed by Šídák’s multiple comparisons test, n = 3 per group). Data presented as means ± SEM, with * indicating statistical significance (*P < 0.05, **P < 0.01, ****P < 0.0001 unpaired t test for all other panels) compared to the WT mice.

We also searched these data in a targeted manner to try to explain some of the observations. For instance, changes in mitochondrial size and number suggest dysregulated fission/fusion (31) in the BAT of SLC35A4-MP KO mice. However, an analysis of mitochondrial fission/fusion proteins, including DNM2, DRP1, MFF, MTRF1L, and MTFP1, showed no differences between genotypes (fig. S10, B and C). Similarly, we looked to see whether differences in CL levels were due to a change in CL synthase. Still, we found that the levels of this enzyme were unchanged between genotypes or diets.

In summary, the proteomics data led to three insights: (i) reduced mitochondrial proteins, (ii) increased immune and ECM remodeling markers in BAT of SLC35A4-MP KO mice on HFD, and (iii) evidence suggesting that differences in mitochondrial size, shape, and dynamics are likely driven by posttranslational mechanisms, which are known to modulate mitochondrial dynamics proteins at the level of activity rather than expression. To gain further insight into the molecular context of SLC35A4-MP within mitochondria, we performed IP followed by proteomics to identify proteins that SLC35A4-MP may interact with or localize near. While not designed to detect direct interactions, this approach provided additional support for SLC35A4-MP localization within mitochondrial compartments involved in energy metabolism.

Proteomics of IP SLC35A4-MP validates its presence in the IMM and colocalization with mitochondrial metabolic proteins

Microproteins often exert their functions through protein-protein interactions, as exemplified by myoregulin binding and inhibiting the calcium channel SERCA (15). Although proteomic strategies have successfully mapped interactions for several soluble microproteins (10), identifying partners for membrane-embedded microproteins remains challenging because of the risk of disrupting membrane integrity during isolation. A split-APEX proximity labeling approach was developed to study the outer mitochondrial membrane microprotein PIGBOS1 (32). However, a similar strategy was not feasible for SLC35A4-MP, as fusion to large tags like APEX disrupted its mitochondrial localization, and no IMM-compatible split-APEX system was available.

To address these challenges, we enriched SLC35A4-MP BAT of WT and SLC35A4-MP KO mice under mild detergent conditions (1% Triton X-100), allowing partial solubilization while preserving mitochondrial membrane integrity. These experiments were performed using an SLC35A4-MP antiserum to IP the endogenous protein, avoiding the need for overexpression of tagged constructs. We observed that tagging and overexpression can alter mitochondrial localization and introduce false positives in IP studies. The results should not be interpreted as identifying specific SLC35A4-MP interactors. Instead, the enriched proteins likely reflect a combination of direct interactions, indirect associations (e.g., via shared complexes), or colocalized proteins on the IMM.

The IP-proteomics analyzed 485 proteins, and we filtered these data to select proteins with >threefold change WT/KO, P value <0.05, with an average TMT signal of >50,000 counts and >1 spectral count (fig. S10D). We were left with 62 total proteins, and 48 of which were mitochondrial based on UniProt annotations. These 48 entries were then fed into StringDB to determine whether they belonged to any known protein complexes using experiments and databases in the active interaction sources settings, highest confidence (0.9), and hiding disconnected networks (fig. S10, D and E). Then, we generated a table that included SLC35A4-MP as a binding partner for all proteins found in the pull-down experiment. We also added information about interactions among proteins in the pull-down experiment based on the results from StringDB. We used this table as input to Cytoscape (v3.10.2) (33) and colored with annotations for fatty acid metabolism, adaptive thermogenesis, oxidative phosphorylation (OXPHOS), and the tricarboxylic acid (TCA) cycle (fig. S10E).

Although this dataset does not provide definitive evidence of direct binding, the enrichment of IMM-associated metabolic proteins supports the localization of SLC35A4-MP to mitochondrial regions involved in β oxidation and energy metabolism. This complements our functional data linking SLC35A4-MP loss to changes in lipid composition, mitochondrial structure, and oxidative capacity in BAT.

The absence of SLC35A4-MP is linked to inflammation in brown fat and activated macrophages

Increasing evidence suggests that lipid metabolism and mitochondrial structure disruptions can promote inflammation (34). Consistent with this, we observed up-regulation of LGALS3, CAPG, TAPBP, and ICAM1 (Fig. 4B), markers associated with inflammation and immune cell infiltration (35–38). Given these changes, we hypothesized that the absence of SLC35A4-MP could lead to local inflammation in BAT of HFD-fed mice.

Supporting this hypothesis, we observed increased macrophage infiltration in the BAT of SLC35A4-MP KO mice, as indicated by F4/80 immunostaining (Fig. 4, C and D). qPCR analysis confirmed this finding, with increased Cd86 mRNA and up-regulation of proinflammatory cytokines (Fig. 4, E to G). Specifically, SLC35A4-MP KO mice exhibited a fivefold increase in the mRNA expression of inflammatory cytokines, including Tnf-a, Il12b, Il6, and Il1b in BAT, across both male (Fig. 4F) and female (Fig. 4G) HFD-fed mice.

Since macrophages are essential for BAT function and metabolism, contributing to tissue innervation and homeostatic energy expenditure (39), we isolated and differentiated bone marrow–derived macrophages (BMDMs) from 12-week-old male SLC35A4-MP KO and WT mice fed with a chow diet. These BMDMs were then activated with interleukin-4 (IL-4) for 24 hours to enhance OXPHOS (40). We observed a decrease in basal oxygen consumption rate (OCR) (Fig. 4, H and I) and a significant reduction in adenosine 5′-triphosphate (ATP)–linked OCR (Fig. 4J) in IL-4–activated BMDMs from SLC35A4-MP KO mice.

Following the observation that SLC35A4-MP deficiency increases inflammatory markers in BAT, we examined its role in BMDMs activated with interferon-γ (IFN-γ) and lipopolysaccharide (LPS). SLC35A4-MP deficiency increased Tnfa, Il1b, and Il12b mRNA expression of inflammatory markers in activated BMDMs (Fig. 4, K to M). To evaluate the effect of SLC35A4-MP on systemic inflammation, a single dose of LPS (10 mg/kg) was administered, and tissues were collected after a 4-hour interval (fig. S11). We observed a 200-fold increase in tumor necrosis factor–α (TNF-α) and a 100-fold increase in IL-6 in the plasma of LPS-stimulated mice compared to nonstimulated controls (fig. S11, A and B). In addition, SLC35A4-MP KO mice showed a modest but insignificant increase in TNF-α (P = 0.06) and IL-6 (P = 0.06) compared to WT LPS-stimulated mice (fig. S11, A and B). In addition, no difference in SLC35A4-MP levels was observed in BAT following LPS stimulation (fig. S11, C and D). In summary, the lack of SLC35A4-MP promotes inflammation in BAT.

SLC35A4-MP levels are correlated with brown adipocyte function and differentiation

Since SLC35A4-MP has a role in lipid metabolism and mitochondrial function in brown fat during metabolic stress, such as HFD-induced obesity, we hypothesize that SLC35A4-MP may regulate brown adipocyte maturation and activation. To explore this hypothesis, we analyzed SLC35A4-MP expression during brown adipocyte differentiation. Western blot analysis revealed a progressive increase in SLC35A4-MP protein levels in immortalized brown preadipocytes undergoing differentiation, accompanied by increased Slc35a4 transcript levels (Fig. 5, A to C).

Fig. 5. SLC35A4-MP and brown adipocyte function.

(A) Representative SLC35A4-MP blot and (B) quantification of SLC35A4-MP levels in immortalized murine interscapular brown preadipocytes during differentiation of mature brown adipocytes by Western blot (n = 3 per group). (C) qPCR analysis of Slc35a4 mRNA expression. (D) Representative Western blot and (E) quantification of SLC35A4-MP levels in BAT from 12-week-old male WT mice after 3 days of cold exposure (chronic cold exposure) (n = 3 per group). (F) qPCR evaluation of Slc35a4 mRNA expression in BAT following chronic cold exposure (n = 3 to 4 per group). (G) qPCR analysis of brown adipocyte markers, Ucp1, Ppargc1a, Prdm16, and Elovl3 mRNA in primary brown adipocytes differentiated from the stroma vascular fraction of BAT derived from 1-month-old male SLC35A4-MP KO and WT mice (n = 3 per group). (H) OCR profile of primary brown adipocytes from SLC35A4-MP KO and WT mice. (I) Quantification of basal, (J) maximal, and (K) ATP-linked respiration (n = 10 per group). Data presented as means ± SEM, with * indicating statistical significance (*P < 0.05, **P < 0.01, ****P < 0.0001, unpaired t test) compared to the WT.

Given the seemingly central role of SLC35A4-MP in BAT, we further explored SLC35A4-MP involvement in BAT activation by cold exposure. We observed that 3-day cold exposed mice, known to promote brown fat activity, resulted in an up-regulation of transcript levels over 20-fold compared to mice maintained at 21°C (Fig. 5, D to F), while SLC35A4-MP protein levels increased ~twofold in BAT of mice after 72 hours at 6°C (Fig. 5, D to F) but were not significant (P = 0.09) (Fig. 5E). To further investigate whether SLC35A4-MP influences thermogenic gene expression, we isolated the stromal vascular fraction (SVF) from BAT of 1-month-old SLC35A4-MP KO and WT mice and differentiated them into primary brown adipocytes. Transcript analysis revealed that Ucp1, Ppargc1a, Prdm16, and Elovl3 levels were significantly reduced in SLC35A4-MP KO adipocytes compared to WT controls (Fig. 5G). Consistent with impaired thermogenic programming, KO adipocytes also exhibited reduced mitochondrial activity, including basal, maximal, and ATP-linked OCR (Fig. 5, H to K).

These findings suggest a potential role for SLC35A4-MP in regulating BAT function under cold stress, warranting further investigation to clarify its precise role and regulatory mechanisms. We exposed mice to acute (6 hours) and chronic (72 hours) cold conditions (Fig. 6, A and B). We found that SLC35A4-MP KO mice were more sensitive to acute cold exposure compared to the control WT mice (Fig. 6A). However, no differences were found between SLC35A4-MP KO and WT mice during the 72-hour chronic cold exposure (Fig. 6B). This suggests that SLC35A4-MP may play a role in regulating metabolic processes in brown adipocytes during periods of increased energy demand.

Fig. 6. The impact of SLC35A4-MP KO in BAT during cold stress.

Effects of acute and chronic cold exposure on 12-week-old male SLC35A4-MP KO and WT mice fed a chow diet. (A and B) Rectal body temperature during cold exposure (n = 6 per group). (C) LC-MS/MS–based untargeted lipidomics of BAT from SLC35A4-MP KO mice and their controls after 3 days of cold exposure (n = 3 per group). (D) Acylcarnitine levels in BAT (two-way ANOVA followed by Šídák’s multiple comparisons test, n = 3 per group). (E) Mitochondrial tomography slices: 1.6-nm thickness slices from the center of EM tomography volumes of BAT mitochondria from BAT of WT mice (left) and SLC35A4-MP KO (right) male mice after chronic cold challenge (scale bar, 1 μm). RT, room temperature. (F) Quantification of the mitochondria profile area (size in square micrometer) (n = 21 to 54 per group), (G) number of mitochondria per unit area of cytoplasm (#mitochondria per square micrometer) (n = 10 per group), and (H) mitochondrial shape (ratio major and minor axis) (n = 21 to 54 per group). Data presented as means ± SEM, with * indicating statistical significance (*P < 0.05, **P < 0.01, ****P < 0.0001, unpaired t test for all other panels) compared to the WT mice.

SLC35A4-MP regulates BAT function during cold challenge

To complement our understanding of how SLC35A4-MP influences lipid metabolism and mitochondrial function in response to cold exposure, we analyzed lipidomics and performed EM on BAT samples from SLC35A4-MP KO and control WT mice. Firstly, lipidomics analysis revealed significant alterations in lipid profiles in SLC35A4-MP KO mice following chronic cold exposure. Specifically, we observed a marked increase in total LPC and some species of LPA 18:1 and LPG 16:0 (Fig. 6C). Acylcarnitines are essential for β oxidation in brown adipocytes during cold exposure (41). We observed cold-induced acetylcarnitine accumulation in BAT of SLC35A4-MP KO compared to their WT counterparts (AcCa 16:0 and AcCa 18:0; Fig. 6D).

To further investigate the impact of chronic cold exposure on mitochondrial morphology, we performed TEM analyses of BAT from WT and SLC35A4-MP KO mice (Fig. 6E). Compared to WT, KO mice displayed a significant increase in mitochondrial size (Fig. 6F) and a reduction in mitochondrial number (Fig. 6G). In addition, shape analysis revealed that WT mitochondria tended to be more circular, whereas KO mitochondria appeared slightly more elongated (Fig. 6H). Together, these data demonstrate that SLC35A4-MP is required to maintain normal mitochondrial morphology and lipid remodeling in BAT during chronic cold exposure.

DISCUSSION

To date, approximately a dozen microprotein KO mice have been generated. The largest group of these targets the structure and function of the muscle. It includes several from the small family of SERCA-interacting microproteins (17), such as myoregulin (15) and DWORF (42, 43), as well as microproteins that can regulate the development (44) and fusion (16, 45) of muscle. The microproteins being investigated in vivo have begun to expand in biological scope and now include microproteins involved in inflammation (46–48), cancer (49, 50), fibrosis (51), and cellular metabolism (24, 52). The microprotein derived from a uORF from the phosphatase and tensin (PTEN) homolog gene, called MP31, is another example of a functional uORF microprotein (24). MP31 regulates cellular lactate metabolism, and loss of function in vivo promotes glioma-genesis due to accelerated lactate metabolism (24). While numerous microproteins have been identified within mitochondria, particularly those involved in lipid metabolism, none have been previously associated with BAT biology or with adaptive responses to HFD and cold exposure. Identifying SLC35A4-MP as a regulator of mitochondrial structure and function in BAT expands the physiological relevance of microproteins in thermogenic tissues.

Conditions of metabolic or temperature stress in mice reveal the impact of SLC35A4-MP in vivo more effectively. While its absence does not affect body weight or glucose tolerance during an HFD, it significantly compromises the integrity and oxidative capacity of brown fat mitochondria. Given that obesity is known to impair BAT function through inflammation, insulin resistance, and impaired sympathetic signaling (53, 54), SLC35A4-MP may play a protective role in preserving mitochondrial structure under metabolic stress. Mitochondrial enlargement and the loss of dynamic regulation are among the most notable phenotypes observed in SLC35A4-MP KO BAT mice on both chow diet and HFD. While modest reductions in proteins such as PERM1, CHCHD7, and NDUFS8 were detected, they are unlikely to fully account for the extent of mitochondrial disruption observed. PERM1 has been linked to thermogenic regulation and mitochondrial organization through interactions with the MICOS-MIB complex (55). CHCHD7 is involved in mitochondrial protein import and structural maintenance (56), and NDUFS8 is a core subunit of complex I essential for OXPHOS (57). Although these proteins play important roles in mitochondrial integrity, their modest reduction suggests that additional mechanisms may contribute to the phenotype. These findings suggest that SLC35A4-MP supports mitochondrial homeostasis not only through its metabolic role but potentially via its spatial localization within mitochondria, which may influence mitochondrial dynamics.

To better understand the spatial context of SLC35A4-MP within mitochondria, enrichment via IP from BAT under mild, nonsolubilizing conditions revealed proteins associated with the IMM, including enzymes involved in lipid metabolism, OXPHOS, and the TCA cycle. Although these experiments were not designed to identify specific interactors, the enrichment of IMM-associated proteins supports the conclusion that SLC35A4-MP localizes to mitochondrial regions involved in energy metabolism. This spatial association aligns with the observed mitochondrial and lipid phenotypes in KO mice and may explain how the loss of SLC35A4-MP disrupts lipid metabolism in BAT.

Phospholipids like CL and PE, essential for cristae architecture and respiratory function (22, 29, 58), are reduced in SLC35A4-MP KO BAT, while PA levels are elevated. These changes suggest altered lipid flux through the CL synthesis pathway or increased CL degradation. Accumulation of PA has been shown to influence mitochondrial dynamics (59–61), including fusion and fission processes. Studies have shown that saturated PA can inhibit the oligomerization of DRP1 on mitochondria, thus preventing mitochondrial fission induced by stress conditions (60). This mechanism suggests that elevated PA levels could potentially promote mitochondrial fusion, leading to the observed larger mitochondrial structures in SLC35A4-MP KO BAT.

In addition to structural remodeling, SLC35A4-MP appears to affect fatty acid oxidation (FAO), particularly under cold-induced thermogenic demand. Cold exposure activates BAT to enhance thermogenesis, which heavily relies on FAO to generate heat (62). In this context, SLC35A4-MP–deficient mice exhibit metabolic signatures consistent with impaired β oxidation, including accumulation of acylcarnitines and reduced thermogenic capacity during acute cold challenges (41). The presence of HADHA (hydroxyacyl-CoA dehydrogenase trifunctional multienzyme complex subunit alpha), a critical β-oxidation enzyme (23, 63), among the proteins coenriched with SLC35A4-MP in the IMM further supports its association with FAO pathways. Together, these findings suggest that SLC35A4-MP helps maintain mitochondrial lipid metabolism during periods of high energetic demand.

Beyond mitochondrial structure and lipid metabolism, our proteomic analysis revealed enrichment of ECM remodeling and immune-related proteins in SLC35A4-MP KO BAT. These pathways have been linked to impaired thermogenesis and mitochondrial dysfunction in adipose tissue under metabolic stress (64), suggesting that they may contribute to the broader phenotype observed in our model. While the loss of SLC35A4-MP causes cell-intrinsic mitochondrial defects in brown adipocytes and macrophages, tissue-level remodeling, including ECM disruption and inflammation, may further compromise BAT function and thermogenic capacity.

In conclusion, SLC35A4-MP emerges as a mitochondrial microprotein that supports lipid and energy metabolism in BAT. Its loss leads to impaired mitochondrial remodeling, disrupted oxidative capacity, and sensitivity to metabolic stress. A proposed model summarizing the cellular and metabolic effects of SLC35A4-MP deficiency is shown in Fig. 7. Its discovery underscores the expanding frontier of mitochondrial microproteins in biology. Future studies are crucial to elucidate the in vivo effects of SLC35A4-MP beyond its impact on BAT, including its potential inflammatory roles and significance in cardiac physiology. Given the growing recognition of numerous conserved microproteins from uORFs, SLC35A4-MP represents a pioneer in studying these understudied protein-coding genes, promising to enhance our understanding of microprotein biology and its implications in metabolic health. Furthermore, along with MP31 (24), SLC35A4-MP continues to argue the importance of evolutionarily conserved uORF-derived microproteins as having fundamental roles in cells and physiology. Given nearly 200 conserved microproteins in a recent dataset (65), this category represents a significant untapped reservoir of functional genes.

Fig. 7. Proposed model of SLC35A4-MP function in BAT.

A schematic model illustrating how SLC35A4-MP supports mitochondrial structure and lipid metabolism in BAT. Loss of SLC35A4-MP leads to altered mitochondrial morphology, reduced oxidative capacity, acylcarnitine accumulation, and inflammation, ultimately impairing thermogenic function. Created in BioRender. Saghatelian, A. (2025) https://BioRender.com/38j8mtz.

MATERIALS AND METHODS

Mice

Mice were housed in a 12-hour light/dark cycle in an animal facility at the Salk Institute, La Jolla, CA. All animal procedures were approved by the Institutional Animal Care and Use Committee at the Salk Institute under protocol number 17-00017. The mice were allowed ad libitum access to water and food. The animals were maintained on a standard chow diet (PicoLab, 5053) or subjected to an HFD (Open Source Diets, D12492). SLC35A4-MP KO mice were generated by CRISPR-Cas9 targeting the SLC35A4 uORF. A gRNA was designed where the cutting location is in the second codon of the SLC35A4 uORF to disrupt the expression of a functional uORF (fig. S2A). gRNA, tracrRNA, and Cas9 protein (all IDT) were combined to form an RNP (26) that was injected into fertilized oocytes from C57BL/6 donor mice (26), which were subsequently transplanted into pseudo-pregnant mice to generate mosaic F0 offspring. To segregate mosaic mutations in the F0 offspring, these mice were mated with C57BL/6 WT mice to produce F1 offspring. PCR and sequencing determined mutations in F1 mice, and a strain with a 10-bp deletion was established, which eliminated the SLC35A4 uORF. Routine genotyping was done with the primers indicated in table S1. A commercial laboratory was also used for mouse genotyping (Transnetyx Inc., Memphis, TN).

Metabolic studies

Glucose and insulin tolerance tests were performed 1 week before euthanasia and followed the protocols described elsewhere (66). Briefly, mice were injected intraperitoneally with glucose (2 g/kg body weight injected intraperitoneally) after overnight fasting or insulin (0.75 U/kg body weight injected intraperitoneally; Humalog, Eli Lilly) after 2-hour fasting. Glucose levels were measured in blood collected from the tail at the indicated times after injection using a digital glucometer. For the cold exposure experiment, the mice were housed individually and transferred to a cold chamber with an ambient temperature of 6°C. The temperature was measured periodically using a RET-3 rectal probe (Physitemp) (66).

LPS-induced inflammatory response test

Mice were administered an intraperitoneal injection of LPS (10 mg/kg) derived from E.coli, serotypes O11:B4 and O55:B5, from Alexis Biochemicals (Enzo). Tissues and blood samples were collected 6 hours postinjection. Proinflammatory cytokines were quantified using the mouse TNF-α and IL-6 ELISA MAX Deluxe Kits, following the manufacturer’s instructions (BioLegend).

Histological analyzes

Tissues were freshly fixed in 10% paraformaldehyde–phosphate-buffered saline (PBS), dehydrated by increasing ethanol concentrations, washed in xylene, and then embedded in paraffin. Paraffin blocks were sectioned and stained with hematoxylin and eosin (H&E) or subjected to immunostaining with the F4/80 antibody (14-4801-82, eBioscience). These procedures were performed using services provided by the UCSD Tissue Technology Core. Brown adipocyte cell size was determined with H&E-stained sections of BAT from three SLC35A4-MP KO and three control mice using ImageJ software. The average diameter of cells was determined for each slide.

Cell lines and culturing

BMDMs were isolated from the femur and tibia of 8-week-old male C57BL6 mice obtained from the Salk Institute, La Jolla, CA, animal facility. Red blood cells were lysed, and the remaining cells were plated at a density of 1.0 × 10⁶ cells/ml in RPMI 1640 culture medium [10% fetal bovine serum (FBS) + 1% penicillin-streptomycin + 1% minimum essential medium (MEM) nonessential amino acids (catalog 11140050, Thermo Fisher Scientific) + 1% pyruvate (catalog 11360070, Thermo Fisher Scientific) + 1% MEM vitamins (catalog 11120052, Thermo Fisher Scientific)] supplemented with macrophage colony-stimulating factor (M-CSF, 20 ng/ml). The culture medium was replaced on day 5, and on day 6, the cells were polarized for 24 hours with IL-4 (20 ng/ml) or LPS (100 ng/ml) + IFN-γ (20 ng/ml).

Brown (9B) preadipocytes (66, 67) were cultured to confluence and subsequently differentiated using Dulbecco’s modified Eagle’s medium (DMEM; Sigma-Aldrich) and 10% FBS. The differentiation medium contained 20 nM insulin, 1 nM triiodothyronine (T3), 0.5 mM isobutyl methylxanthine (IBMX), 1 μM dexamethasone, 2.8 μM rosiglitazone, and 0.125 mM indomethacin. After 2 days of induction, the cells were maintained in DMEM supplemented with 20 nM insulin and 2.8 μM rosiglitazone for an additional 6 days (66).

Primary brown preadipocytes were isolated from the SVF of BAT of WT and SLC35A4-MP KO mice using collagenase IV (1 mg/ml). Cells were cultured in 10-cm plates with DMEM (10% FBS and 1% penicillin-streptomycin) until reaching 80% confluence. The cells were then trypsinized, seeded onto 24-well plates, and upon reaching 100% confluence, induced to differentiate for 2 days with DMEM containing 20 nM insulin, 1 nM T3, 0.5 mM IBMX, 1 μM dexamethasone, 2.8 μM rosiglitazone, and 0.125 mM indomethacin. After induction, the cells were maintained in DMEM supplemented with 20 nM insulin and 2.8 μM rosiglitazone for an additional 6 days (67).

Measurements of OCR

The analysis was conducted using a Seahorse XFe96 Analyzer (Agilent), following a previously described method with minor modifications (13). Briefly, BMDMs were cultured in RPMI 1640 medium supplemented with 10% FBS, 1% penicillin-streptomycin, 1% MEM nonessential amino acids, 1% pyruvate, 1% MEM vitamins, and M-CSF (20 ng/ml). The cells were then plated at a density of 1 × 10⁷ cells per well onto a 96-well Seahorse plate. After 5 days of incubation, the cells were polarized for 24 hours with IL-4 (20 ng/ml). Subsequently, the medium was replaced with prewarmed Seahorse XF base medium, without phenol red (Agilent, 103335-100), supplemented with 10 mM d-glucose, 1 mM sodium pyruvate (Gibco, 11360-070), and 2 mM GlutaMax (Gibco, 35050-061). For primary brown adipocytes, the cells were plated at a density of 100 × 10⁵ cells per well onto a 96-well Seahorse plate and differentiated for 8 days. The day before the experiment, the Seahorse XF cartridge was hydrated with XF Calibrant solution (Agilent Technologies) and incubated overnight at 37°C without CO₂. Adipocytes were incubated for 1 hour in Seahorse XF DMEM (Agilent) supplemented with 25 mM glucose, 2 mM pyruvate, and 1 mM glutamine before measurement. OCR was assessed through sequential injections of oligomycin (1 μM), carbonyl cyanide 3-chlorophenylhydrazone (2 μM), and rotenone/antimycin A (1 μM). The assay cycle consisted of 3 min of mixing and 3 min of measurement. Results were normalized to protein level and measured using the BCA Protein Assay Kit (Pierce BCA, Thermo Fisher Scientific) following the manufacturer’s protocol.

Western blotting

Western blotting and reverse transcription qPCR were performed as previously described (13). Cells or tissues were harvested and homogenized on ice in a lysis buffer (50 mM tris, 150 mM NaCl, 1% Triton X-100, and protease inhibitor cocktail). Lysates (60 to 100 μg) were loaded onto 4 to 12% bis-tris gels (Thermo Fisher Scientific, NW04120BOX) and run in 1× MES-SDS buffer (Thermo Fisher Scientific, B0002) at 165 V for 40 min. The gels were transferred onto polyvinylidene difluoride membranes (Thermo Fisher Scientific, IB24001) using the iBlot 2 Dry Gel Transfer System (Thermo Fisher Scientific, IB21001) under the following conditions: 15 V for 15 min. Membranes were blocked with Intercept [tris-buffered saline (TBS)] blocking buffer (LI-COR, 927-60001) at room temperature for 1 hour before the addition of corresponding primary antibodies used at the indicated dilutions (table S1). Membranes were incubated in primary antibodies overnight on a shaker at 4°C. After removing membranes from primary antibodies, membranes were washed with TBS-T (0.1% Tween-20) and incubated in Alexa Fluor–labeled secondary antibodies for 1 hour (see table S1 for dilutions). Following TBS-T washes, images of membranes were captured using LiCor Odyssey CLx at IR700 and IR800. Subsequent analysis and quantitation were performed using ImageJ.

Reverse transcription qPCR

Total RNA was extracted from tissues with TRIzol. RNA (1 μg) was reverse-transcribed with a high-capacity cDNA reverse transcription kit (Applied Biosystems) according to the manufacturer’s instructions. For qPCR, a 10-μl reaction mixture was prepared using Maxima SYBR Green/ROX qPCR MasterMix (2X) (Thermo Fisher Scientific, K0223) and 250 nM of specific forward and reverse primers. The reactions were run in duplicate using the qPCR QuantStudio 3 qPCR (Thermo Fisher Scientific) under the following conditions: 50°C for 2 min, 95°C for 10 min, and 40 cycles of 95°C for 15 s, 60°C for 20s, and 72°C for 30s. 36B4 or 18S expression was analyzed in parallel with normalizing gene expression. Amplification of specific transcripts was confirmed by analyzing melting curve profiles at the end of each PCR. Real-time PCR primer sequences are listed in table S1.

Preparation of antiserum-linked resin for IPs

Antigen affinity purified SLC35A4 MP antiserum was covalently coupled to Affi-Gel10 N-hydroxysuccinimide–activated resin per the manufacturer’s instructions (Bio-Rad, catalog no. 1536099), 1.2 mg of antiserum to 1.5 ml of bed volume resin. For control resin, a similar rabbit immunoglobulin G (IgG) whole molecule ratio, Rockland, catalog no. 011-0102, was covalently linked to Affi-Gel 10. Following the coupling of Igs to resins by rotating overnight at 4°C, the resins were washed with 10 column volumes of 1 N acetic acid to remove any unbound antiserum and then equilibrated and stored at 4°C in PBS with 0.05% sodium azide as 25% gel slurries.

Forward IPs

Lysates were added to washed IgG- or purified SLC35A4-MP antiserum coupled to Affi-Gel 10 beads and rotated at 4°C overnight. The beads were washed and rotated in TBS-T (0.1% Tween-20) five times, with a minimum of 20 min each time. The 2X sample buffer (Thermo Fisher Scientific, 84788) or SDS loading buffer with 2-mercaptoethanol (Sigma-Aldrich, M6250) was added to the beads for Western blot and MS analysis, respectively. Bound proteins were eluted by heating at 95°C for 10 min. Following centrifugation at 5000 rpm for 2 min, the samples were collected without disrupting the beads.

Transmission electron microscopy

BAT from three mice each of KO chow, KO cold, KO HFD, WT chow, WT cold, WT HFD, and were fixed via cardiac perfusion with 2.5% glutaraldehyde +2% paraformaldehyde in 0.15 M sodium cacodylate at 37°C and then placed on ice in the same fixative for 1 hour. The tissues were rinsed three times on ice with 0.15 M sodium cacodylate buffer containing 3 μM calcium chloride and incubated in a mixture of 1% OsO₄, 0.8% potassium ferrocyanide, and 3 μM calcium chloride in 0.15 M sodium cacodylate for 1 hour on ice. Then, the tissues were washed three times with ice-cold double-distilled water and stained with 2% uranyl acetate for 1 hour on ice before incubating in increasing ethanol solutions (20, 50, 70, and 90% on ice, three times 100% at room temperature). Subsequently, the tissues were infiltrated with a mixture of 50% acetone and 50% Durcupan ACM resin (Fluka) for 6 hours with agitation and then incubated three time 6 to 12 hours in 100% Durcupan with agitation and polymerized for 48 hours at 60°C in an oven. A hacksaw was used to cut out a block about 2 mm across and glued on a dummy block. Thin sections about 70 nm thick were cut using a Leica UCT ultramicrotome. The sections were placed on 200-mesh uncoated thin-bar copper grids. A Tecnai Spirit (FEI; Hillsboro, OR) electron microscope operated at 80 kV was used to record images with a Gatan 2Kx2K charge-coupled device (CCD) camera ranging between 11.5 and 2.9 nm per pixel.

For quantitative analysis, the mitochondrial profile area was measured using ImageJ tracing and measurement tools (National Institutes of Health). Every mitochondrion in the TEM image was measured to avoid bias. The number of mitochondria per unit cytoplasmic area was calculated by counting the number of mitochondria in an image and dividing by the profile area of the cytoplasm in that image measured using ImageJ. The mitochondrial shape parameter was measured by dividing the mitochondrial long (major) axis length by its short (minor) axis length. This measurement used ImageJ on every TEM image’s mitochondrion to avoid bias. This type of measurement was possible because BAT mitochondria are globular in shape, making it straightforward to determine the major and minor axes.

EM tomography

Semithick sections of about 350-nm thickness were cut from the cells’ blocks prepared for TEM with a Leica ultramicrotome and placed on 200-mesh uncoated thin-bar copper grids. Twenty-nanometer colloidal gold particles were deposited on each side of the grid to serve as fiducial cues. The specimens were irradiated with electrons for about 20 min to limit anisotropic specimen thinning during image collection at the magnification used to collect the tilt series before initiating a dual-axis tilt series. During data collection, the illumination was held to near parallel beam conditions, and the beam intensity was kept constant. Tilt series were captured using SerialEM software (University of Colorado, Boulder, CO) on a Tecnai HiBase Titan (FEI; Hillsboro, OR) EM operated at 300 kV. Tilt series were taken at 0.81 nm per pixel. Images were recorded with a Gatan 4Kx4K CCD camera. Each dual-axis tilt series first collected 121 images taken at 1° increments over a range of −60° to 60° and then by rotating the grid 90° and collecting another 121 images with the same tilt increment. After collecting the orthogonal tilt series, to improve the signal-to-noise ratio, two times binning was performed on each image by averaging a 2 by 2 x-y pixel box into 1 pixel using the newstack command in IMOD (University of Colorado, Boulder, CO). The new pixel resolution was 1.62 nm. The IMOD package [https://en.wikipedia.org/wiki/IMOD_ (software)] was used for tilt-series alignment and reconstruction. R-weighted back projection was used to generate the tomographic volumes. Crista density, defined as the cristae’s total membrane surface area divided by the mitochondrion volume, was measured using the ImageJ stereology plug-in on slices from tomographic volumes. Every mitochondrion in the slice was measured to avoid bias.

Lipid extraction and LC/MS

Lipids were extracted using a modified version of the Bligh-Dyer method (68). Tissues (heart and BAT) were lysed in water in a bead beater. Volumes equivalent to 30 mg of heart tissue and 15 mg of BAT were used for lipid extraction. The samples were diluted in water to a final volume of 1 ml and manually shaken in a glass vial (VWR) with 1 ml of methanol and 2 ml of chloroform containing internal standards (13C16-palmitic acid, d7-Cholesterol) for 30 s. The resulting mixture was vortexed for 15 s and centrifuged at 2400g for 6 min to induce phase separation. The organic (bottom) layer was retrieved using a Pasteur pipette, dried under a gentle stream of nitrogen, and reconstituted in 2:1 chloroform:methanol for LC/MS analysis.

Lipidomic analysis was performed on a Vanquish HPLC with a Q-Exactive quadrupole-orbitrap mass spectrometer equipped with an electrospray ion source (Thermo Fisher Scientific). Data were acquired in positive and negative ionization modes. Solvent A consisted of 95:5 water:methanol, and solvent B was 70:25:5 isopropanol:methanol:water. For positive mode, solvents A and B contained 5 mM ammonium formate with 0.1% formic acid; for negative mode, solvents contained 0.028% ammonium hydroxide. An XBridge (Waters) C8 column (5 μm, 4.6 mm by 50 mm) was used. The gradient was held at 0% B between 0 and 5 min, raised to 20% B at 5.1 min, increased linearly from 20 to 100% B between 5.1 and 55 min, held at 100% B between 55 and 63 min, returned to 0% B at 63.1 min, and held at 0% B until 70 min. Flow rate was 0.1 ml/min from 0 to 5 min, 0.3 ml/min between 5.1 and 55 min, and 0.4 ml/min between 55 and 70 min. The spray voltage was 3.5 and 2.5 kV for positive and negative ionization modes, respectively; the S-lens rf level was 65. Sheath, auxiliary, and sweep gases were 50, 10 and 1, respectively. The capillary temperature was 325°C, and the auxiliary gas heater was 200°C. Data were collected in full MS/dd-MS2 (top 10). Full MS was acquired from 150 to 1500 mass/charge ratio (m/z) with a resolution of 70,000, automatic gain control (AGC) target of 1 × 10⁶, and a maximum injection time of 100 ms. MS2 was acquired with a resolution of 17,500, a fixed first mass of 50 m/z, an AGC target of 1 × 10⁵, and a maximum injection time of 200 ms. Stepped normalized collision energies were 20, 30, and 40%.

Lipid identification was performed with LipidSearch (Thermo Fisher Scientific). Skyline verified the mass accuracy, chromatography, and peak integration of all LipidSearch-identified lipids (69). The peak areas were used for data reporting, and the data were normalized using internal standards.

TMT by MS-based quantification

Samples were precipitated by methanol/chloroform and redissolved in 8 M urea/100 mM TEAB (pH 8.5). Proteins were reduced with 5 mM tris(2-carboxyethyl)phosphine hydrochloride (TCEP, Sigma-Aldrich) and alkylated with 10 mM chloroacetamide (Sigma-Aldrich). Proteins were digested overnight at 37°C in 2 M urea/100 mM TEAB (pH 8.5), with trypsin (Promega). The digested peptides were labeled with 16-plex TMT (Thermo Fisher Scientific, product A44522), and pooled samples were fractionated by a basic reversed phase (Thermo Fisher Scientific, 84868).

The TMT-labeled samples were analyzed on an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific). Samples were injected directly onto a 25 cm, 100-μm inside diameter column packed with BEH 1.7-μm C18 resin (Waters). The samples were separated at a 300 nl/min flow rate on an EasynLC 1200 (Thermo Fisher Scientific). Buffers A and B were 0.1% formic acid in water and 90% acetonitrile, respectively. A gradient of 1 to 15% B over 30 min, an increase to 45% B over 120 min, an increase to 100% B over 20 min and held at 100% B for 10 min was used for a 180 min total run time.

Peptides were eluted directly from the tip of the column and nanosprayed directly into the mass spectrometer by application of 2.5-kV voltage at the back of the column. The Eclipse was operated in a data-dependent mode. Full MS1 scans were collected in the Orbitrap at 120-k resolution. The cycle time was set to 3 s, and within these 3 s, the most abundant ions per scan were selected for collision-induced dissociation MS/MS in the ion trap. The TMT samples were analyzed using MS3 analysis with multinotch isolation (SPS3), which was used to detect TMT reporter ions at a 60-k resolution (70). Monoisotopic precursor selection was enabled, and dynamic exclusion was used with an exclusion duration of 60 s.

Protein and peptide identification was done using the Integrated Proteomics Pipeline–IP2 (Integrated Proteomics Applications). Tandem mass spectra were extracted from raw files using RawConverter (71) and searched with ProLuCID (72) against the UniProt mouse database. The search space included all fully tryptic and half-tryptic peptide candidates. Carbamidomethylation on cysteine and TMT on lysine and peptide N-term were considered as static modifications. Data were searched with 50–parts per million (ppm) precursor and 600-ppm fragment ion tolerance. Identified proteins were filtered using DTASelect (73) and a target-decoy database search strategy to control the false discovery rate to 1% at the protein level (74). Quantitative analysis of TMT was done with Census (75) filtering reporter ions with 10-ppm mass tolerance and 0.6 isobaric purity filter.

Protein-protein interaction network analysis of SLC35A4-MP–associated proteins

To generate networks showing binding partners, we selected the hits from the pull-down proteomics experiment with a log₂FoldChange > 1.5 that are known mitochondrial proteins and searched for their protein IDs in STRING (https://string-db.org/). Then, we generated a table that included SLC35A4 as a binding partner for all proteins found in the pull-down experiment. We also added information about interactions among proteins in the pull-down experiment based on the results from STRING. We used this table as input to Cytoscape (v3.10.2) (33) and colored the nodes based on the available functional annotation. We colored the edges between SLC35A4 and another protein in the pull-down experiment.

Analysis of public proteomics datasets for SLC35A family protein detection

Public proteomics datasets were analyzed to assess the detection of SLC35A family proteins across diverse biological contexts. Data from the MassIVE Protein Explorer were retrieved using the Protein Explorer web interface (76), which aggregates PSMs from more than 2 billion spectra across thousands of public MS datasets. PSM counts for SLC35A family proteins were extracted from the global MassIVE repository as well as from a focused analysis of proteomics data derived from human adipose tissue (27). Complementary analysis was performed using the PepCentric database, which compiles proteomics data from 76 datasets totaling approximately 2.3 billion spectra.

For PepCentric (20), the database search interface retrieved MS run counts per protein. Data were visualized using a consistent approach in Python (Matplotlib v3.8), with green-turquoise bars and black outlines to represent detection frequency. Figures display protein names on the x axis and either PSM counts or MS run counts on the y axis, depending on the source dataset. Numeric values were overlaid on the bars for clarity. SLC35A4-MP and canonical SLC35A family members were included in all panels. SLC35A4 (mORF) was explicitly included with zero values to highlight its absence across datasets.

Analysis of Ribo-Seq data from mouse brown, beige, and white adipocytes

We obtained raw Ribo-Seq reads from Martinez et al. (28), generated from primary brown adipocytes isolated from interscapular and subscapular BAT, and white and beige adipocytes derived from subcutaneous WAT of C57BL/6J mice. We performed quality control, trimmed the reads using the FastX toolkit (https://github.com/agordon/fastx_toolkit), and then aligned the trimmed reads to a contaminant database containing rRNA and tRNA sequences using STAR (v2.5.3b) (77). Unmapped reads were subsequently aligned to the mouse reference genome assembly mm10 using STAR. We filtered the aligned reads by removing secondary alignments with “samtools view” (78) and then executed samtools view again to remove multimapping reads. Both sets of alignments were visualized in the Integrative Genomics Viewer alongside a GTF file containing Slc35a4 annotations.

Statistical analysis

Results are expressed as means ± SEM. Comparisons between two groups were performed using an unpaired t test, while two-way analysis of variance (ANOVA) was used for multiple group analyses, followed by appropriate post hoc tests when necessary. Statistical significance was set at P < 0.05, and all analyses were conducted using GraphPad Prism 10.

Supplementary Materials

This PDF file includes:

Figs. S1 to S11

Table S1