The biochemical basis of mitochondrial dysfunction in Zellweger Spectrum Disorder

Abstract

Peroxisomal biogenesis disorders (PBDs) are genetic disorders of peroxisome biogenesis and metabolism that are characterized by profound developmental and neurological phenotypes. The most severe class of PBDs—Zellweger spectrum disorder (ZSD)—is caused by mutations in peroxin genes that result in both non‐functional peroxisomes and mitochondrial dysfunction. It is unclear, however, how defective peroxisomes contribute to mitochondrial impairment. In order to understand the molecular basis of this inter‐organellar relationship, we investigated the fate of peroxisomal mRNAs and proteins in ZSD model systems. We found that peroxins were still expressed and a subset of them accumulated on the mitochondrial membrane, which resulted in gross mitochondrial abnormalities and impaired mitochondrial metabolic function. We showed that overexpression of ATAD1, a mitochondrial quality control factor, was sufficient to rescue several aspects of mitochondrial function in human ZSD fibroblasts. Together, these data suggest that aberrant peroxisomal protein localization is necessary and sufficient for the devastating mitochondrial morphological and metabolic phenotypes in ZSDs.

How peroxisomal biogenesis disorders lead to mitochondrial dysfunction is not well understood. This study reveals that peroxisomal proteins mislocalize to the mitochondria, thereby disrupting mitochondrial function. The mitochondrial quality control protein ATAD1 can rescue this defect.

Introduction

Peroxisomal biogenesis disorders (PBDs) are diseases affecting peroxisomes and cell metabolism (Delille et al, ²⁰⁰⁶). PBDs display decreased peroxisome biogenesis, non‐functional peroxisomes, or loss of the organelle altogether, with accompanying clinical manifestations ranging from mild‐to‐severe phenotypes (Berendse et al, ²⁰¹⁶). The most severe class of PBDs, Zellweger spectrum disorder (ZSD), presents with myriad clinical features, including seizures, hepatomegaly, renal cysts, skeletal abnormalities, mitochondrial defects, impaired hearing and eyesight, and ultimately death in early childhood (Crane, 2014; Klouwer et al, ²⁰¹⁵). Current treatments for PBDs are mainly palliative with a focus on symptom management (Klouwer et al, ²⁰¹⁵; Braverman et al, ²⁰¹⁶).

Peroxisomes are single‐membrane organelles that play an integral role in fatty acid metabolism, bile acid synthesis, and in the scavenging of reactive oxygen species (Smith & Aitchison, 2013). For instance, plasmalogens, a class of lipids that are precursors in myelin sheath formation, are solely synthesized in peroxisomes (Luoma et al, ²⁰¹⁵). Peroxins are a group of proteins that regulate peroxisomal biogenesis, inheritance, and metabolic processes. In addition to the approximately 40 known peroxins, more than 50 different enzymes are imported into the matrix of peroxisomes, where the majority of peroxisomal reactions occur. Given the multitude of metabolic pathways that require peroxisomes, it comes as no surprise that mutations in the majority of peroxin‐encoding genes cause profound consequences as in Zellweger spectrum disorder.

Peroxisomes and mitochondria are physically and functionally coupled. Mitochondria not only share and complement peroxisomal roles in lipid metabolism and reactive oxygen species defense, but have also been shown to participate in peroxisome biogenesis in conjunction with the endoplasmic reticulum (ER) (Schrader & Yoon, 2007; Sugiura et al, ²⁰¹⁷). The fate of peroxins including their regulation in the absence of functional peroxisomal import has not previously been explored in detail. There have been sporadic reports indicating that some peroxins display dual localization to the peroxisome and either the ER or mitochondria (Muntau et al, ²⁰⁰⁰; Sacksteder et al, ²⁰⁰⁰; South et al, ²⁰⁰⁰; Halbach et al, ²⁰⁰⁶; Kim et al, ²⁰⁰⁶; Toro et al, ²⁰⁰⁹; Aranovich et al, ²⁰¹⁴). This localization pattern has not been investigated systematically nor has it been investigated as a potential source of mitochondrial impairment.

In the context of PBDs that exhibit a loss of peroxisomal biogenesis and protein import, we hypothesized that either peroxin abundance is decreased in PBDs or else peroxin proteins—lacking their native destination—will mislocalize. We previously found that the peroxisomal tail‐anchored (TA) protein, Pex15, is targeted to mitochondria when the guided entry of tail‐anchored (GET) protein‐sorting system is impaired (Chen et al, ²⁰¹⁴). The extraction and degradation of mislocalized Pex15, as well as other mislocalized TA proteins from mitochondria, is facilitated by yeast Msp1 and its mammalian ortholog ATAD1, which belong to the AAA⁺ ATPase protein family (Wohlever et al, ²⁰¹⁷; Piard et al, ²⁰¹⁸; Wang et al, ²⁰²⁰). Subsequent work by Castanzo et al (2020) concluded that Msp1 is rather non‐specific in recognizing substrates and simply requires the presence of extended unstructured sequences for substrate engagement (Castanzo et al, ²⁰²⁰). Because of its ability to recognize diverse substrates, it was unclear if Msp1/ATAD1 might also impact peroxin localization when peroxisomes are absent, and if it might have an effect on the mitochondrial phenotype observed in ZSDs.

Although we now know that the genomic mutations linked with ZSDs occur in peroxin genes, these diseases were initially associated with gross mitochondrial abnormalities (Goldfischer et al, ¹⁹⁷³). Today, mitochondrial dysfunction, in the form of either decreased respiration and/or changes in mitochondrial morphology, is a widely recognized phenotype of ZSDs (Shinde et al, ²⁰¹⁸; Argyriou et al, ²⁰¹⁹). In fact, some features of ZSD are thought to result from insufficient mitochondrial capacity to fulfill the needs of highly energetic tissues such as the brain and retina (Kann & Kovács, 2007; Gkotsi et al, ²⁰¹⁴). However, it is unclear why these mitochondrial defects develop and whether they independently contribute to ZSD pathophysiology, or are a secondary consequence of non‐functioning peroxisomes.

Here, we report that loss of functioning peroxisomes does not affect peroxin gene expression or translation, but, instead, results in the mislocalization of a specific subset of peroxins to the mitochondria. This aberrant accumulation of peroxins is causative of the metabolic and morphological dysregulation of mitochondria in ZSDs. Importantly, we were able to rescue several aspects of mitochondrial function in human ZSD fibroblasts by modulating ATAD1 expression, which suggests that peroxin mislocalization is the primary contributor to mitochondrial dysfunction in ZSDs.

Results

Peroxin gene expression is maintained in the absence of functional peroxisomes

Using the budding yeast S. cerevisiae as a model system, we generated strains lacking PEX3 or PEX19, which are peroxisomal biogenesis factors involved in targeting peroxisomal membrane proteins (PMPs) to the peroxisome (Jansen & Klei, 2019). Previously, it was reported that Pex19 binds to newly synthetized PMPs before associating with Pex3 to facilitate the import of PMPs into peroxisomes and thus preventing their aggregation (Jones et al, ²⁰⁰⁴). Deletion of either gene has been reported to cause loss of functional peroxisomes (Hettema et al, ²⁰⁰⁰). We assessed the presence of peroxisomes in each of our strains by expressing a red fluorescent protein fused to the peroxisomal targeting motif (the amino acids SKL), which was imported by peroxisomes present in wild‐type cells, causing punctate‐labeled peroxisomes (Fig 1A, upper panel). Conversely, we did not detect punctate peroxisomes in pex3Δ and pex19Δ cells (Fig 1A, upper panel). Instead, faint fluorescence was visible throughout the cytosol, which indicates an absence of intact peroxisomes capable of protein import (Fig 1A, upper panel). To confirm there was no remaining peroxisomal metabolic function despite the loss of punctate peroxisomes, we grew the strains on oleate‐containing media, as yeast require functional peroxisomes to metabolize this otherwise toxic lipid. As expected, the pex3Δ and pex19Δ strains exhibited a growth defect relative to wild‐type yeast on oleate media (Fig 1B). These results indicate impaired peroxisomal metabolism and suggest the loss of functional peroxisomes in these mutant yeast strains, which supports their use as a model for dissecting the molecular consequences of loss of peroxisomal function.

Figure 1. Peroxin gene expression is maintained in the absence of functional peroxisomes.

1. The indicated yeast strains (wild type, pex19Δ, pex3Δ, msp1Δ, pex19Δmsp1Δ, and pex3Δmsp1Δ) expressing RFP‐SKL (peroxisomal marker) and mitochondria‐targeted green fluorescent protein (mGFP) were grown to mid‐log phase and imaged by fluorescence microscopy. Representative images are shown, scale bar, 2 μm. DIC, differential interference contrast. Enhanced: the red signal intensity was set to “best‐fit” in ZENPro microscopy software analysis.
2. The indicated yeast strains (wild type, msp1Δ, pex19Δ, pex19Δmsp1Δ, pex3Δ, and pex3Δmsp1Δ) were grown to mid‐log phase, back‐diluted to OD600 = 1, and serial dilutions were spotted onto agar plates containing synthetic media supplemented with glucose (S‐D), glycerol (S‐Gly), glycerol and oleate (S‐Gly‐Ole), or oleate (S‐Ole).
3. Wild‐type versus pex19Δ RNA levels. Outliers are indicated. Peroxin genes are indicated in red. The blue shading emphasizes that the RNA levels of the indicated genes are lower in pex19Δ compared to wild type.
4. Cumulative distributions of fold changes (log₂) in RNA between wild‐type and pex19Δ yeast (Kolmogorov–Smirnov test). Peroxin genes are indicated in red.
5. Translational efficiency (TE) of wild type versus pex19Δ. Outliers are indicated. Peroxin genes are indicated in red. The red shading emphasizes that the translation efficiency of the indicated genes is higher in pex19Δ compared to wild type.
6. pex19Δ versus pex19Δmsp1Δ RNA levels. Outliers are indicated. Peroxin genes are indicated in red. The red shading emphasizes that the RNA levels of the indicated genes are higher in pex19Δmsp1Δ compared to pex19Δ.
7. Cumulative distributions of fold changes (log₂) in RNA between pex19Δ and pex19Δmsp1Δ yeast (Kolmogorov–Smirnov test). Peroxin genes are indicated in red.
8. Translational efficiency (TE) of pex19Δ versus pex19Δmsp1Δ. Outliers are indicated. Peroxin genes are indicated in red. The blue shading emphasizes that the translation efficiency of the indicated genes is lower in pex19Δmsp1Δ compared to pex19Δ.

Having validated the model system, we first investigated the regulation of peroxisomal gene expression in wild‐type and peroxisome‐deficient yeast strains using RNA‐seq. Despite lacking this highly conserved organelle, gene expression was remarkably similar in wild‐type and pex19Δ yeast (Fig 1C). In particular, there was no significant difference in steady‐state mRNA levels of peroxin‐encoding genes (Fig 1D). Among the limited mRNA changes, we observed that three zinc‐response genes (ADH4, ZAP1, and ZRT1) were downregulated in pex19Δ yeast relative to wild type, which suggests a possible interplay between peroxisomal function and zinc homeostasis (Fig EV1, EV2, EV3, EV4) (Bird et al, ²⁰⁰⁰). Since the translation of some nuclear‐encoded mitochondrial proteins is coupled to their import into the organelle in yeast (Couvillion et al, ²⁰¹⁶; Grevel et al, ²⁰¹⁹), we hypothesized that an analogous regulatory mechanism might exist for peroxisomal proteins. Therefore, in addition to mRNA levels, we also considered the possibility of regulation at the level of translational efficiency (TE). We performed ribosome‐footprint profiling to measure differences in TE between wild‐type and peroxisome‐deficient pex19Δ yeast and again observed no substantial differences between the two strains (Fig 1E). These results demonstrate that the absence of functional peroxisomes does not significantly impact the steady‐state transcription or translation of peroxin‐encoding genes.

Figure EV1. Peroxin gene expression is maintained in the absence of functional peroxisomes.

1. Wild‐type versus pex19Δ RNA levels. All outliers are indicated.
2. Translational efficiency (TE) of wild type versus pex19Δ. All outliers are indicated.
3. pex19Δ versus pex19Δmsp1Δ RNA levels. All outliers are indicated.
4. Translational efficiency (TE) of pex19Δ versus pex19Δmsp1Δ. All outliers are indicated.
5. ZAP1 target genes abundance in wild type versus pex19Δ.

Figure EV2. Pex13 mediates docking and accumulation of peroxin complexes on mitochondria.

1. Heat map representing the protein abundance of control proteins Pex3, Pex19, and Msp1 in digitonin‐solubilized (3 g/g digitonin/protein ratio) mitochondrial membranes from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor).
2. The indicated strains (wild type, msp1Δ, pex19Δ, pex13Δ, pex19Δmsp1Δ, pex13Δpex19Δ, pex13Δmsp1Δ, and pex13Δmsp1Δpex19Δ) were grown to mid‐log phase, back‐diluted to OD600 = 1, and serial dilutions were spotted onto agar plates containing synthetic media with glucose (S‐D), glycerol (S‐Gly), or oleate (S‐Ole).
3. The indicated strains (wild type, msp1Δ, pex13Δ, pex19Δ, pex19Δmsp1Δ, pex13Δpex19Δ, pex13Δmsp1Δ, and pex13Δpex19Δmsp1Δ) expressing RFP‐SKL (peroxisomal marker) and mitochondria‐targeted green fluorescent protein (mGFP) were grown to mid‐log phase and visualized by fluorescence microscopy. Representative images are shown, scale bar, 2 μm. DIC, differential interference contrast.
4. Mitochondria from the indicated yeast strains (pex19Δ, pex19Δmsp1Δ, and pex19Δmsp1Δ pex13Δ) expressing Mdh3‐V5 (under its endogenous promotor) were either left untreated (nt) (lane 1, 4, 7) or were treated with proteinase K in high fidelity buffer (PK‐HF) (cutting control, lane 2, 5, 8) or low fidelity buffer (PK‐LF) (protease protection assay, lane 3, 6, 9) and separated by SDS–PAGE and immunoblotted for Mia40 (intermembrane space)(α‐Mia40), Mdh3‐V5 (α‐V5), porin (outer membrane) (α‐porin), and Tom22 (outer membrane but partially proteinase exposed) (α‐Tom22).

Figure EV3. Peroxins target to mitochondria in cells derived from ZSD patients and overexpression of ATAD1 rescues their mitochondrial morphology.

1. 25 μg of cell lysate from cWT, cWT+ATAD1, PEX3^−/−, and PEX3^−/−+ATAD1 transfected with GFP or Pex13‐GFP were separated by SDS–PAGE and immunoblotted for ATAD1‐HA‐FLAG (α‐HA), Pex3‐V5 (α‐V5), and GFP (α‐GFP). VDAC (α‐VDAC) was used as a loading control.
2. Fluorescence microscopy of human fibroblast cell lines cWT expressing control‐GFP and stained with MitoTracker™ Deep Red FM. Representative images are shown, scale bar 10 μm.

Figure EV4. Overexpression of ATAD1 rescues mitochondrial respiration and metabolism.

1. 35 μg of cell lysate from cWT, cWT‐ATAD1, PEX3^−/−, and PEX3^−/−‐ATAD1 cells were separated by SDS–PAGE and immunoblotted for ATAD1 (α‐ATAD1). Tubulin (α‐tubulin) was used as a loading control.
2. Scatter plot representation of phosphatidyl–ethanolamine (PE) lipidomics. The average (n = 4, biological replicates) normalized peak intensity of each of the 67 detected phosphoethanolamine (PE) species, log₁₀ pareto‐scaled is visualized. Statistical significance was calculated using Welch’s test, P‐values are indicated as: **P ≤ 0.01, ****P ≤ 0.0001. The horizontal lines represent the mean peak intensity of all 67 PE species in each cell line.

Source data are available online for this figure.

Given that the synthesis of peroxin proteins is maintained in the absence of functional peroxisomes, we reasoned that peroxins would re‐route to alternative destinations to avoid cytosolic aggregation (Matsumoto et al, ²⁰¹⁹). For example, Pex15 has previously been shown to mislocalize to mitochondria dependent on the simultaneous disruption of both the GET system, which mediates its normal targeting, and the mitochondrial quality control factor Msp1 (Chen et al, ²⁰¹⁴; Okreglak & Walter, ²⁰¹⁴). To test if loss of Msp1 modified peroxisomal function, we generated an msp1Δ single‐deletion strain as well as pex3Δmsp1Δ and pex19Δmsp1Δ double‐deletion strains. The peroxisomal phenotypes observed in these backgrounds were indistinguishable from their parental strains (Fig 1A, lower panel, and Fig 1B) with a minor difference in the number and size of the peroxisomal puncta in the msp1Δ strain, as reported previously (Chen et al, ²⁰¹⁴; Weir et al, ²⁰¹⁷). Similarly, deletion of MSP1 alone or in the pex19Δ background had no apparent effect on the steady‐state expression or translation of peroxin‐encoding genes (Fig 1F–H), consistent with the known post‐translational role of Msp1 in protein homeostasis. These results demonstrate that the absence of Msp1 does not significantly impact the steady‐state expression or translation of peroxin‐encoding genes.

Peroxisomal proteins accumulate on mitochondria in the absence of peroxisomes

Since their transcription and translation were maintained, we next explored the fate of peroxin proteins upon loss of peroxisomal protein import. The system established above also allowed us to test if peroxin localization is Msp1 sensitive. We performed quantitative comparative mass spectrometry of mitochondria isolated from yeast lacking peroxisomes using Histodenz^TM‐gradient purification (Fig 2A) (Wühr et al, ²⁰¹²). Peroxisomes and mitochondria are known to form extensive contact sites (Shai et al, ²⁰¹⁸), and as a result, mitochondria isolated from wild‐type and msp1Δ strains were heavily contaminated with peroxisomes thereby preventing any interpretable demarcation between the two organelles. To circumvent this issue, we assessed mitochondrial proteomes only from peroxisome‐deficient strains (pex19Δ and pex19Δmsp1Δ), which still allowed us to determine whether loss of Msp1 caused an increase in mitochondrial‐associated peroxins. Our unbiased screen identified mitochondrial abundance of 12 peroxins (Pex2, 3, 4, 11, 13, 14, 17, 22, 25, 29, 30, 31) in pex19Δmsp1Δ compared with pex19Δ yeast (Fig 2B and C), with Pex11 and Pex13 showing the greatest enrichment proteome wide (Fig 2B and C). In addition, five peroxisomal matrix proteins (Lys1, Mdh3, Pot1, Scs7, and YJL185C) were enriched in mitochondria isolated from pex19Δmsp1Δ yeast (Fig 2C), of which only Pot1 has previously been reported to localize to this organelle.

Figure 2. Peroxisomal proteins accumulate on mitochondria in absence of peroxisomes in yeast.

1. Experimental flow of sample generation for quantitative mass spectrometry.
2. Volcano plot showing the most enriched/decreased proteins in the mitochondrial proteome of pex19Δ versus pex19Δmsp1Δ strains as detected by quantitative mass spectrometry (n = 7, biological replicates; n = 2, technical replicates). Student’s t‐tests were performed using scipy.stats.ttest_ind() (Mckinney, 2010). Effect sizes were calculated for Cohen's d (Sawilowsky, 2009; Cohen, ²⁰¹³). A protein's relevance was determined using a P‐value threshold of < 0.05 with an absolute effect size threshold of 1 or greater (large effect). Gray dots represent all measured proteins, and red dots highlight peroxisomal‐associated proteins.
3. Heat map of peroxin proteins from pex19Δ and pex19Δmsp1Δ stains as detected by quantitative mass spectrometry (n = 5). Protein classes are indicated in the row labels. Values were z‐score normalized by protein.
4. The indicated yeast strains expressing PEX13‐RFP and mitochondria‐targeted green fluorescent protein (mGFP) were grown to mid‐log phase and visualized by fluorescence microscopy. Representative images are shown, scale bar, 2 μm. DIC, differential interference contrast.
5. The indicated yeast strains expressing PEX11‐RFP and mitochondria‐targeted green fluorescent protein (mGFP) were grown to mid‐log phase and visualized by fluorescence microscopy. Representative images with enhanced contrast and brightness are shown, scale bar, 2 μm. DIC, differential interference contrast.
6. Total yeast cell lysate, post‐mitochondrial supernatant, and Histodenz^TM‐purified mitochondria from the indicated strain (wild type, msp1Δ, pex19Δ, pex19Δmsp1Δ, pex3Δ, and pex3Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor) were separated by SDS–PAGE and immunoblotted for Pex13 (α‐V5), cytochrome c (α‐cyt c), and HSP70 (α‐hsp70).
7. Total yeast cell lysate, post‐mitochondrial supernatant, and Histodenz^TM‐purified mitochondria from the indicated strain (wild type, msp1Δ, pex19Δ, pex19Δmsp1Δ, pex3Δ, and pex3Δmsp1Δ) expressing Pex11‐V5 (under its endogenous promotor) were separated by SDS–PAGE and immunoblotted for Pex11 (α‐V5) and porin (α‐porin), cytochrome c (α‐cyt c), and HSP70 (α‐hsp70).

To confirm our findings, we generated RFP fusion constructs for the two most mitochondrially enriched peroxins, Pex11 and Pex13, and determined their localization via microscopy. Both Pex11‐RFP and Pex13‐RFP co‐localized with peroxisomal markers in wild‐type yeast but with a mitochondrial marker in pex19Δ yeast, supporting the idea that peroxins can route to mitochondria when their native destination is absent. This co‐localization was more pronounced in pex19Δmsp1Δ yeast (Fig 2D and E), which is consistent with a role for Msp1 in the removal of these mislocalized peroxins from the mitochondria. Although a carbon‐source‐dependent mitochondrial localization of Pex11 has previously been reported in wild‐type yeast (Mattiazzi Ušaj et al, ²⁰¹⁵), to our knowledge, this is the first demonstration of the mitochondrial localization of Pex13, (and Pex4, 17, 22, 25, 29 via mass spectrometry). Interestingly, neither Pex11 nor Pex13 is tail‐anchored proteins, unlike canonical substrates of Msp1.

To further validate the quantitative mass spectrometry results, we expressed V5‐tagged versions of Pex11 and Pex13 in wild‐type, msp1Δ, pex19Δ, and pex19Δmsp1Δ strains and immunoblotted protein from mitochondria isolated via Histodenz™‐gradient purification, resulting in fractions with minor ER contamination. As expected, strains containing both mitochondria and intact peroxisomes (wild‐type and msp1Δ) showed a strong signal for the V5‐tagged proteins (Fig 2F and G) due to peroxisomal contamination caused by co‐purification of these organelles (Shai et al, ²⁰¹⁸). Mitochondria from the pex19Δ strain displayed a faint signal for Pex13‐V5 and Pex11‐V5, but both proteins were dramatically enriched in the pex19Δmsp1Δ mitochondrial proteome compared to the pex19Δ single mutant (Fig 2F and G).

To exclude a pex19Δ‐specific effect, we performed an identical quantitative mass spectrometry experiment comparing pex3Δ and pex3Δmsp1Δ strains and obtained a very similar pattern of peroxins at mitochondria (Fig 3A and B), with Pex11 and Pex13 again showing an increase in abundance in the pex3Δmsp1Δ double knockout strain (Fig 3B). These mass spectrometry results were subsequently validated by immunoblotting Histodenz™‐gradient‐purified mitochondria from pex3Δ and pex3Δmsp1Δ strains transformed with either Pex13‐V5 or Pex11‐V5 (Fig 2F and G). Together, these data demonstrate that peroxisomal proteins accumulate on mitochondria in the absence of functional peroxisomes.

Figure 3. Peroxisomal proteins are incorporated into the mitochondrial membrane in absence of peroxisomes in yeast.

1. Volcano plot (log₁₀) of the most enriched/decreased proteins in the mitochondrial proteome of pex3Δ versus pex3Δmsp1Δ strains as detected by quantitative mass spectrometry (n = 3, biological replicates, n = 1, technical replicate). Student’s t‐tests were performed using scipy.stats.ttest_ind() (Mckinney, 2010). Effect sizes were calculated for Cohen's d (Sawilowsky, 2009; Cohen, ²⁰¹³). A protein's relevance to the model was determined using a P‐value threshold of < 0.05 with an absolute effect size threshold of 1 or greater (large effect). Gray dots represent all measured proteins. Red dots highlight peroxisomal‐associated proteins.
2. Heat map (log₂) of peroxin protein levels from pex3Δ and pex3Δmsp1Δ as detected by quantitative mass spectrometry (n = 3). Protein classes are indicated in the row labels. Values were z‐score normalized by protein.
3. Mitochondria from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ), expressing Pex13‐V5 (under its endogenous promotor) were either left untreated (lane 1, 2, 3) or exposed to Na₂CO₃ and separated into the membrane pellet (4, 5, 6) or supernatant fraction (7, 8, 9). After separation by SDS–PAGE the fractions were immunoblotted for Pex13‐V5 (α‐V5) and porin (α‐porin).

We next used sodium carbonate extraction to determine whether Pex13 is integrated into the mitochondrial membrane or is simply peripherally associated. We found that Pex13 was present in both the membrane pellet and supernatant fractions in wild‐type yeast (Fig 3C). We expected the presence in the membrane fraction but assume that some Pex13 might be bound to Pex5 while in transit to the peroxisome, thereby explaining the soluble pool (Girzalsky et al, ²⁰¹⁰). In contrast, Pex13 did not associate with the membrane fraction in the pex19Δ strain (Fig 3C, lane 5), consistent with Pex13 being removed by the mitochondrial extractase, Msp1. This is supported by the observation that Pex13 was again present in the membrane fraction in the pex19Δmsp1Δ double‐deletion strain (Fig 3C, lane 6), indicating that in this strain Pex13 is integrated into the mitochondrial membrane.

Pex13 mediates docking and accumulation of peroxin complexes on mitochondria

Peroxins assemble into defined complexes on the peroxisomal membrane to facilitate protein import into the peroxisomal matrix via the peroxisomal importomer complex. Pex13, a key component of the docking subcomplex of the importomer, was the most enriched protein in our proteomics datasets (Figs 2C and 3B). Additionally, Pex14 and Pex17, two other members of this subcomplex, as well as a number of peroxisomal matrix proteins accumulated on mitochondria in pex19Δmsp1Δ yeast (Fig 2B and C). Based on these results, we reasoned that it might be possible for a native and functional importomer complex to assemble on mitochondria in this strain (Williams & Distel, 2006; Tabak et al, ²⁰¹³). To test this hypothesis, we isolated mitochondria from yeast strains expressing Pex13‐V5, which accumulates on the mitochondria in peroxisome‐deficient strains (Fig 4A), and analyzed them using Blue Native (BN)–PAGE with varying amounts of detergent (Fig 4B and C). To detect Pex13‐containing complexes, we titered the amount of digitonin detergent in our mitochondrial purification protocol to optimize the protection of native protein complexes while also disrupting non‐specific interactions, which we monitored using internal controls (Wittig et al, ²⁰⁰⁶) (Fig 4B and C). In wild‐type yeast, at four grams of digitonin per gram of mitochondria, Pex13‐V5 migrated in a complex with an apparent total mass of ∼ 180 kDa (Fig 4C, green box). This mass is consistent with that of the native docking complex (172.5 kDa), which is composed of Pex13, Pex14, Pex17, and Pex8. This complex was not detectable by Western blot in peroxisome‐deficient pex19Δ yeast (Fig 4C, orange box), but was restored upon the additional deletion of MSP1 (pex19Δmsp1Δ) (Fig 4C, purple box). To identify the components of the observed complex at ∼ 180 kDa, we performed complexome profiling (Heide et al, ²⁰¹²). Histodenz™‐purified mitochondria from all three strains were separated using BN–PAGE and each gel lane was subsequently cut into ∼ 60 pieces. Proteins in each gel piece were digested with trypsin, and peptides were analyzed by LC‐MS/MS to determine the relative abundance of each protein (Figs 4D and E, and EV2A). We found that the ∼ 180 kDa region of the gel from wild‐type and pex19Δmsp1Δ strains was similar (green and purple boxes, respectively) and contained Pex11, Pex13, Pex14, and Pex25–which are either components of the traditional docking complex of the peroxisomal importomer (Meinecke et al, ²⁰¹⁰) or the core complex, which has additional transient components (Oeljeklaus et al, ²⁰¹²) (Fig 4D and J). All four peroxins were also detected in mitochondria from pex19Δ yeast but at substantially reduced levels (Fig 4D, orange box). Pex17, which was previously found to interact with Pex14 on the surface of peroxisomes (Huhse et al, ¹⁹⁹⁸), was absent at ∼ 180 kDa but detected at the running front in wild‐type and pex19Δmsp1Δ samples presumably because it is detergent labile and its interaction with the rest of the importomer complex was not maintained throughout the BN–PAGE. These data suggest that native peroxin complexes are capable of forming on mitochondria when peroxisomes are absent.

Figure 4. Pex13 mediates docking and accumulation of peroxin complexes on mitochondria.

1. Total cell lysate (T), post‐mitochondrial supernatant (PMS), and Histodenz^TM‐purified mitochondria (M + P, M) from the indicated yeast strains (wild type, pex19Δ, and pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor) were separated by SDS–PAGE and immunoblotted for PEX13‐V5 (α‐V5) and porin (α‐porin).
2. Digitonin‐solubilized (1 or 4 g/g digitonin/protein ratio, as indicated) mitochondrial membranes from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor) were separated by BN–PAGE with a 3–18% gradient and stained with Coomassie brilliant blue. BHM, bovine heart mitochondria; V₂, complex V‐dimer; S_L, supercomplex III₂IV₂; S_s, supercomplex III₂IV₁; V₁, complex V‐monomer.
3. Digitonin‐solubilized (1 or 4 g/g digitonin/protein ratio, as indicated) mitochondrial membranes from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor) were separated by BN–PAGE with a 3–18% gradient and immunoblotted for PEX13‐V5 (α‐V5). Green box for wild type, orange box for pex19Δ, purple box for pex19Δmsp1Δ, indicates gel area of 300–130 kDa correlated with complexome profiling heat map.
4. Heat map representing the protein abundance of Pex11, 13, 14, 17, 25 in digitonin‐solubilized (3 g/g digitonin/protein ratio) mitochondrial membranes from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor). Green box for wild type, orange box for pex19Δ, purple box for pex19Δmsp1Δ, indicates gel area of 300‐130 kDa correlated to complexome profiling heat map.
5. Heat map representing the protein abundance of Mdh3 and Pot1 in digitonin‐solubilized (3 g/g digitonin/protein ratio) mitochondrial membranes from the indicated yeast strains (wild type, pex19Δ, pex19Δmsp1Δ) expressing Pex13‐V5 (under its endogenous promotor).
6. Mitochondria from the indicated yeast strains (pex19Δ, pex19Δmsp1Δ, and pex13Δmsp1Δpex19Δ) expressing Pex22‐V5 (under its endogenous promotor) (Histodenz^TM‐purified) were separated by SDS–PAGE and immunoblotted for Pex22‐V5 (α‐Pex22‐V5) and porin (α‐porin).
7. Mitochondria from the indicated yeast strains (pex19Δ, pex19Δmsp1Δ, and pex13Δmsp1Δpex19Δ) expressing Pot1‐V5 (under its endogenous promotor) were separated by SDS–PAGE and immunoblotted for Pot1‐V5 (α‐Pot1‐V5) and porin (α‐porin).
8. Mitochondria from the indicated yeast strains (pex19Δ, pex19Δmsp1Δ, and pex13Δmsp1Δpex19Δ) expressing Mdh3‐V5 (under its endogenous promotor) (Histodenz™‐purified) were separated by SDS–PAGE and immunoblotted for Mdh3‐V5 (α‐Mdh3‐V5) and porin (α‐porin).
9. Mitochondria from the indicated yeast strains (pex19Δ, pex19Δmsp1Δ, and pex13Δmsp1Δpex19Δ) expressing Pot1‐V5 (under its endogenous promotor) were either left untreated (nt) (lane 1,4,7) or were treated with proteinase K in high fidelity buffer (PK‐HF) (cutting control, lane 2,5,8) or low fidelity buffer (PK‐LF) (protease protection assay, lane 3,6,9) and separated by SDS–PAGE and immunoblotted for Mia40 (intermembrane space) (α‐Mia40), Pot1‐V5 (α‐V5), porin (outer membrane) (α‐porin) and Tom22 (outer membrane but partially proteinase exposed) (α‐Tom22).
10. Cartoon representing the predicted topology of the peroxisomal importomer assembled on mitochondria.

Next, we tested if this complex is functional and is required for the recruitment of additional peroxins to the mitochondria. We found that the peroxisomal matrix proteins Pot1‐V5 and Mdh3‐V5 co‐fractionated with mitochondria in the pex19Δmsp1Δ strain (Fig 2C), similar to the pattern observed previously with Pex13‐V5 and Pex11‐V5 (Fig 2F and G). We were also able to detect Mdh3 and Pot1 in our complexome dataset (Fig 4E). Mdh3 was present and co‐migrated with the complex in all three strains but with lowest abundance in the pex19Δ strain. Pot1 was detected only in the wild‐type and pex19Δmsp1Δ strains and migrated independently of the ∼ 180 kDa complex (Fig 4E). We hypothesized that these matrix proteins were recruited to the mitochondria via the docking complex. To test this, we disrupted the formation of the docking complex by deleting PEX13 in the pex19Δ, msp1Δ, and pex19Δmsp1Δ yeast strains (Fig EV2B and C) and tracked the localization of three V5‐tagged peroxisomal proteins identified in our screen: a peroxisomal membrane protein, Pex22, and two peroxisomal matrix proteins, Pot1 and Mdh3. In the absence of a functional docking complex (pex13Δmsp1Δpex19Δ), the abundance of mitochondrial Pex22‐V5 and Pot1‐V5 was reduced (Fig 4F and G), and the localization of Mdh3 to mitochondria was abolished (Fig 4H). Pot1 and Mdh3 could interact with the docking complex on the trans‐side or even translocate into the mitochondria. Treatment of mitochondria with proteinase K revealed that Pot1‐V5 was protected from proteinase activity only when docking complex formation is possible (pex19Δmsp1Δ) (Fig 4I, lane 6) and not when the formation is impaired by the deletion of Pex13 (pex13Δmsp1Δpex19Δ) (Fig 4I, lane 9). The same pattern was observed for Mdh3 (Fig EV2D). This suggests that Pot1 and Mdh3 enter the mitochondria when the docking complex can be formed. This Pex13‐dependent and Msp1‐sensitive localization of Mdh3 and Pot1 to mitochondria is particularly intriguing as Mdh3 and Pot1 are both soluble, peroxisomal matrix proteins. Although Pot1 was previously identified as an intermembrane space protein with an as yet unknown moonlighting function (Vögtle et al, ²⁰¹²), Mdh3 has not been reported to localize to mitochondria. Taken together, these results suggest that, in the absence of peroxisomes, a functional peroxisomal docking and import complex assembles on mitochondria and this is critical for the recruitment of peroxisomal proteins to their new location, the mitochondria.

Peroxins accumulate in mitochondria in cells derived from Zellweger spectrum disorder patients

Thus far, we have used yeast as a model system in which to study ZSDs. Although peroxisome function and peroxin proteins are highly conserved among eukaryotes, we wanted to test whether our findings in yeast are representative of human disease. To do this, we obtained a PEX3‐deficient (herein referred to as PEX3^−/−) fibroblast cell line derived from a patient with ZSD. To create an isogenic control cell line, we complemented this cell line by expressing wild‐type PEX3 (cWT). In line with published data, this restores peroxisome formation and function (Sugiura et al, ²⁰¹⁷). We then overexpressed ATAD1—the human ortholog of yeast Msp1—to create a total of four cell lines: cWT, cWT+ATAD1, PEX3^−/−, and PEX3^−/−+ATAD1. Finally, in order to track peroxin localization, we expressed PEX13‐GFP (the human homolog of yeast Pex13) in all four cell lines using a lentiviral vector (Fig EV3A).

PEX13‐GFP displayed a distinct punctate pattern in cWT and cWT+ATAD1 cells in agreement with its expected peroxisomal localization (Fig 5A). Strikingly, in PEX3^−/− cells, the PEX13‐GFP signal overlapped entirely with the mitochondrial network. However, in PEX3^−/−+ATAD1 cells, GFP did not co‐localize with mitochondria, but was instead visible throughout the cytosol and in the nucleus (Fig 5A)—a pattern reminiscent of cells expressing GFP alone (Fig EV3B)—likely explained by the observation that the PEX3^−/−+ATAD1 cells contain primarily free GFP that has been cleaved from PEX13‐GFP presumably after extraction from the membrane (Fig EV3A, lane 8). We performed an analogous set of experiments in a PEX16‐deficient cell line derived from a second ZSD patient. In this case, we transiently expressed PEX13‐GFP and PEX10‐GFP, a peroxin that is also a component of the importomer complex (Chen, 2015), and found that both peroxins co‐localize with mitochondria (Appendix Fig S1A and B). These experiments confirm that, as observed in yeast—albeit that robust mislocalization in yeast also requires deletion of MSP1, specific peroxins localize to mitochondria in human cells lacking functional peroxisomes—in this case, cells derived from patients with ZSD. Furthermore, overexpression of ATAD1 results in a reduction of peroxin localization to mitochondria, akin to its function in removing tail‐anchored and other proteins from the outer mitochondrial membrane (Matsumoto et al, ²⁰¹²; Chen et al, ²⁰¹⁴).

Figure 5. Peroxins target to mitochondria in cells derived from ZSD patients and overexpression of ATAD1 rescues their mitochondrial morphology.

1. Fluorescence microscopy of human fibroblast cell lines derived from ZSD patients–cWT, cWT+ATAD1, PEX3^−/−, and PEX3^−/−+ATAD1–expressing PEX13‐GFP and stained with MitoTracker™ Deep Red FM. Representative images are shown, scale bar, 10 μm.
2. Electron microscopy of human fibroblast cell lines cWT, cWT+ATAD1, PEX3^−/−, and PEX3^−/−+ATAD1. Representative images are shown, scale bar 500 nm.
3. Quantification of electron microscopy imaging. Mitochondrial length and width were analyzed using the Fiji length analysis tool. Per cell line > 50 mitochondria were graphed. Statistical significance was calculated using a non‐paired t‐test, P‐values are indicated as: ns (not significant), **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. The horizontal lines represent the mean length and width in each cell line, respectively.
4. Quantification of electron microscopy imaging. Three independent experiments were quantified. Mitochondria were counted (cWT n = 232, PEX3^−/− n = 266, and PEX3^−/−+ATAD1 n = 257) and binned into 3 categories: mitochondria with many cristae and low electron density (Type 1), mitochondria with fewer cristae (Type 2), and mitochondria with high electron density and/or almost no cristae (Type 3).

Overexpression of ATAD1 rescues mitochondrial dysfunction in human ZSD fibroblasts

As mentioned above, ZSD correlates closely with mitochondrial dysfunction. Based on our previous observation, we next asked if overexpression of ATAD1 improved mitochondrial function in PEX3^−/− cells. Our fluorescent microscopy images (Fig 5A) lack the necessary detail and resolution to reveal morphological changes. Therefore, we used electron microscopy to assess gross mitochondrial morphology in all four cell lines (cWT, cWT+ATAD1, PEX3^−/−, and PEX3^−/−+ATAD1) (Fig 5B). Both mitochondrial length and width were reduced in PEX3^−/− cells compared with wild‐type controls but were partially restored to wild‐type levels following ATAD1 overexpression (Fig 5C). Based on osmium staining, mitochondria appeared more protein dense in mitochondria from PEX3^−/− cells compared with wild‐type controls, consistent with aberrant protein accumulation on mitochondria (Fig 5B) (Tapia et al, ²⁰¹²). PEX3^−/− cells also lacked detectable cristae, which are a hallmark of healthy mitochondria (Fig 5B). To more systematically quantify mitochondrial morphology, we outlined three categories: mitochondria with many cristae and low electron density (type 1), mitochondria with intermediate cristae and low electron density (type 2), and mitochondria with few or no cristae and high electron density (type 3). Whereas ∼ 80% of mitochondria in wild‐type cells match the criteria for types 1 and 2, more than 80% of mitochondria in PEX3^−/− cells were instead assessed as type 3. Overexpression of ATAD1 largely rescued the morphological defects in the PEX3^−/− background (PEX3^−/−+ATAD1), substantially decreasing the percentage of type 3 mitochondria (Fig 5D).

Given that overexpression of ATAD1 largely restored mitochondrial morphology in peroxisome‐deficient cells to that observed in wild‐type controls, we next set out to determine whether mitochondrial metabolic function is also restored. Using a Seahorse XFe Flux Analyzer to measure mitochondrial respiration, we found that PEX3^−/− cells had both a lower basal rate of respiration and a lower maximal respiratory capacity than wild‐type cells (Fig 6A–D). ATAD1 overexpression in PEX3^−/− cells restored both of these parameters to wild‐type levels, which suggests that the ATAD1‐dependent removal of peroxins is sufficient to rescue mitochondrial respiratory function (Fig 6A–D). To further characterize the effect of ATAD1, we generated an ATAD1‐knockout cell line in the PEX3^−/− background (PEX3^−/−ATAD1^−/−) (Fig EV4A). These cells exhibited an even greater reduction in mitochondrial respiration compared to their already compromised parental control (Fig 6A–D). These results suggest that endogenous levels of ATAD1 maintain some mitochondrial function in ZSD (as observed in the PEX3^−/− cell line); however, this level of ATAD1 activity is insufficient to combat the deleterious effects resulting from the overwhelming accumulation of peroxins on mitochondria.

Figure 6. Overexpression of ATAD1 rescues mitochondrial respiration and metabolism.

1. Bioenergetic profile of human fibroblast cell lines. OCR (pmol/min/norm.unit) for cWT, cWT+ATAD1, PEX3^−/−, PEX3^−/−+ATAD1, cWT‐ATAD1^−/−, and PEX3^−/− ATAD1^−/− cells plotted against time (XF96e Extracellular Flux Analyzer, Mito‐Stress‐Test). 1 μM Oligomycin A, 0.15 μM FCCP, and 1 μM rotenone + 1 μM antimycin A (final concentrations) were sequentially delivered to the XF96e assay medium through injection ports in the sensor cartridge. The graph is the compilation of two independent assays, error bars show mean ± s.d. (n = 4, biological replicates).
2. Bar graph representation of basal respiration measured in (B) for cWT, cWT+ATAD1, PEX3^−/−, PEX3^−/−+ATAD1, cWT‐ATAD1, and PEX3^−/−‐ATAD1 cells. Error bars show mean ± s.d. (n = 4, biological replicates). Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), **P ≤ 0.01, ***P ≤ 0.001.
3. Bar graph representation of maximal respiration measured in (B) for cWT, cWT+ATAD1, PEX3^−/−, PEX3^−/−+ATAD1, cWT‐ATAD1 and PEX3^−/− ‐ATAD1 cells. Error bars show mean ± s.d. (n = 4, biological replicates). Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), *P ≤ 0.05, **P ≤ 0.01, ****P ≤ 0.0001.
4. Bar graph representation of calculated respiratory spare capacity (uncoupled respiration minus basal respiration) measured in (B) for cWT, cWT+ATAD1, PEX3^−/−, PEX3^−/−+ATAD1, cWT‐ATAD1, and PEX3^−/−‐ATAD1 cells. Error bars show mean ± s.d. (n = 4, biological replicates). Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001.
5. Scatter plot representation of cardiolipin lipidomics. The average (n = 4, biological replicates) normalized peak intensity of each of the 11 detected cardiolipin species, log₁₀ pareto‐scaled is visualized. Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), *P ≤ 0.05, **P ≤ 0.01. The horizontal lines represent the mean peak intensity of all 11 caridiolipin species in each cell line.
6. Scatter plot representation of a subpopulation of the phosphoethanolamine (PE) lipidomics. The average (n = 4, biological replicates) normalized peak intensity of each of the 20 phosphoethanolamine (PE) species (decreased in PEX3^−/−), log₁₀ pareto‐scaled, is visualized. Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), *P ≤ 0.05, ***P ≤ 0.001, ****P ≤ 0.0001. The horizontal lines represent the mean peak intensity of all 20 PE species in each cell line.
7. Scatter plot representation of ether‐phospholipid lipidomics. The average (n = 4, biological replicates) normalized peak intensity of each of the 21 detected ether‐phospholipid (plasmalogen) species, log₁₀ pareto‐scaled, is visualized. Statistical significance was calculated using Welch’s test, P‐values are indicated as: ns (not significant), ***P ≤ 0.001, ****P ≤ 0.0001. The horizontal lines represent the mean peak intensity of all 21 plasmalogen species in each cell line.

Source data are available online for this figure.

To measure other aspects of the restoration of mitochondrial function and metabolism by ATAD1, we performed whole‐cell lipidomics (Source data – Figs 6 and EV4). Compared with their wild‐type counterparts, PEX3^−/− cells showed specific depletion of two lipid classes associated with active mitochondrial metabolism: cardiolipins and phosphatidylethanolamines (PE) (Dudek, 2017; Calzada et al, ²⁰¹⁹). Overexpression of ATAD1 in PEX3^−/− cells restored all cardiolipin species to wild‐type levels (Fig 6E); however, mass spectrometric analysis of all 67 detectable PEs revealed significant differences between PEX3^−/− and both cWT and PEX3^−/−+ATAD1 cells (Fig EV4B). These differences were attributable to changes in the levels of a subpopulation (∼ 30%) of PEs that were lower in the PEX3^−/− cells compared to cWT controls and which increased in some cases beyond wild‐type levels in PEX3^−/−+ATAD1 cells (Fig 6F). The active transport of phosphatidylserines into the mitochondria is one of four pathways that contribute to the synthesis of PEs. Accordingly, the recovering subpopulation presumably represent PEs that rely on the active transport of phosphatidylserines into the mitochondria for their synthesis (Appendix Fig S2). Plasmalogens, on the other hand—which are precursors in myelin synthesis—were decreased in PEX3^−/− cells and were not rescued by ATAD1 overexpression (Fig 6G). This is expected as plasmalogen synthesis requires functional peroxisome metabolism, and although these data clearly show that ATAD1 can mitigate mitochondrial dysfunction in these cells, it cannot compensate for loss of peroxisomal functions. In all, these results show that the aberrant accumulation of peroxins on mitochondria in ZSDs causes defects in mitochondrial morphology, respiration, and lipid metabolism, which can be rescued by overexpression of ATAD1.

Discussion

A diverse array of eukaryotic cells and organisms can survive without functional peroxisomes (Breitling et al, ²⁰⁰²). Yet, human PBDs present with severe and rapidly progressive phenotypes that greatly limit patients’ quality of life and lifespan (Klouwer et al, ²⁰¹⁵). We therefore sought to understand the molecular underpinnings of ZSDs, specifically the etiology of the associated mitochondrial dysfunction.

Previous studies in yeast have found that the transcription and translation of mitochondria‐destined proteins are reduced when mitochondrial‐import pathways are impaired (Dennerlein et al, ²⁰¹⁷), raising the possibility that a similar regulatory paradigm could exist for peroxins. However, RNA‐seq and ribosome‐footprint profiling showed that peroxin gene expression as well as peroxin mRNA translational efficiency are unaffected by stable, genetic loss of peroxisomes. Of note, a previous study reported redistribution of peroxin mRNA from the cytosol to the periphery of peroxisomes when cells are shifted to media that induces rapid peroxisomal biogenesis (Zipor et al, ²⁰⁰⁹). So, although we do not see changes in translational efficiency upon loss of peroxisomes, this does not exclude the possibility that peroxins as a class of mRNA might demonstrate coordinated regulation of translational efficiency upon changes to the cellular environment and/or metabolic needs.

Although we use the term “mislocalization” to describe our observations of peroxins on mitochondria, we cannot differentiate between an active “re‐routing” process or a more passive process of mislocalization. Two main mechanisms of peroxisomal biogenesis are widely accepted: (i) multiplication of peroxisomes initialized by a growth and division pathway and (ii) de novo peroxisome generation involving the ER (Hoepfner et al, ²⁰⁰⁵; Tabak et al, ²⁰¹³; Hettema et al, ²⁰¹⁴; Motley et al, ²⁰¹⁵). A third mechanism has been postulated in which mitochondrial‐derived vesicles contribute to peroxisome biogenesis. This proposes purposeful mitochondrial routing of peroxins for the first time (Sugiura et al, ²⁰¹⁷). It is interesting to note, that the presumed accumulation of peroxins in the ER membrane, which has previously been shown in some ZSD models, is observed without any apparent interference in ER function while we highlight here that the accumulation of peroxins at mitochondria comes with severe functional consequences. Our findings of peroxisomal proteins on mitochondria, albeit in a pathophysiological background, might provide new perspectives on peroxisomal biogenesis, peroxisomal import, and how mitochondrial quality control mechanisms regulate these processes. Further exploration is required to determine whether peroxisomal proteins actively route through the mitochondria or are passively mislocalized, particularly within the context of a healthy cell environment.

Using both microscopy and high‐throughput proteomics, we found that many peroxins—most prominently Pex13, Pex11, Pex14, Pex2, Pex17, and Pex25—accumulate on the mitochondrial outer membrane when peroxisomes are absent. This accumulation led to the formation of the peroxisomal docking complex, a subcomplex of the peroxisomal import machinery (McNew & Goodman, 1994; Walton et al, ¹⁹⁹⁵; Meinecke et al, ²⁰¹⁰; Romano et al, ²⁰¹⁹), on mitochondria. The formation of this complex was shown to play a pivotal role in recruiting both peroxisomal membrane proteins and peroxisomal matrix proteins, such as Mdh3 and Pot1, to the mitochondria. To our knowledge, this is the first evidence of a membrane protein translocase assembling into a functional complex on a non‐native organelle. Insofar as Msp1/ATAD1 protects mitochondria in ZSD cells by dismantling the mislocalized peroxisomal import apparatus, a drug that interferes with this import process might achieve similar therapeutic benefits. In other words, it may be that the only peroxisomal import that occurs in ZSD cells is aberrant import into mitochondria; in that case, inhibiting peroxisomal import could actually benefit ZSD patients. A small‐molecule inhibitor of trypanosomal PEX14 was recently described to kill T. brucei parasites by blocking peroxisomal protein import (Dawidowski et al, ²⁰¹⁷). Perhaps repurposing similar compounds could be an unconventional pharmacological approach in ZSD patients. Further work will be needed to uncover if the peroxisomal import machinery has a distinct array of substrates when localized to the mitochondria versus the peroxisome.

Although peroxins represented the clearest hits from our proteomics data, some other proteins are worthy of further discussion. Particularly, we found three zinc‐related proteins in the RNA‐seq dataset (Zrt1, Zap1, and Adh4) that were differentially regulated in pex19Δ compared with wild‐type yeast. Zrt1 is a zinc transporter, Zap1 is a zinc‐sensitive transcription factor and zinc sensor, and Adh4 is a zinc‐sensitive alcohol dehydrogenase. In addition to these three proteins, we also saw that known Zap1 target genes (including ADH4) were similarly more abundant at the RNA level in wild‐type yeast than in pex19Δ yeast (Figs 1E and EV1E). Although we are not aware of a well‐characterized connection between peroxisome function and Zn²⁺ content, several peroxins contain zinc‐binding domains, and mutations of these domains in Pex2, Pex10, and Pex12 cause particularly severe peroxisomal defects and PBD phenotypes (Gootjes et al, ^2004a; Gootjes et al, ^2004b; Blomqvist et al, ²⁰¹⁷). Therefore, it is possible that there is a regulatory connection between Zn²⁺ homeostasis and peroxin function.

Despite loss of a highly conserved organelle, gene expression and translational efficiency measurements were remarkably similar between wild‐type and pex19Δ yeast. However, as with the proteomics data, there are a small number of mRNAs that display differential translational efficiency in these two strains. In particular, CMC4 and TOM7 appear to be translated more efficiently in pex19Δ yeast than in wild‐type yeast (Fig 1E and H). The protein products of both of these genes are involved in or dependent on the Mia40/ERV1 mitochondrial import pathway (Tokatlidis, 2005), which we found to be affected in our pex mutant strains (Fig 2B). Why these two particular mitochondrial genes would be post‐transcriptionally regulated in this way is unclear. This suggests to us that different mitochondrial import pathways could be affected by peroxin mislocalization and warrants further investigation.

We are particularly intrigued by the implications of our following findings on Msp1/ATAD1. We show that overexpression of ATAD1 is sufficient to restore mitochondrial morphology, respiration, and metabolism in cells derived from ZSD patients. Removal of PEX13, which we use as a representative peroxin, from the outer mitochondrial membrane suggests that ATAD1 very likely has non‐traditional substrates beyond the landscape of TA proteins (Castanzo et al, ²⁰²⁰). Factors such as free fatty acids or bile acid accumulation were eliminated as causative for the mitochondrial impairment (Shinde et al, ²⁰¹⁸). We are the first, however, to connect the peroxisomal protein mislocalization to the mitochondrial phenotype. Beyond the direct benefits to mitochondrial dysfunction at large, improvements to cells' metabolic function might allow more aggressive, combinatorial therapies to help alleviate the plethora of other ZSD symptoms.

In summary, we identify the aberrant accumulation of peroxins as the primary biochemical basis for mitochondrial dysfunction in ZSD and that the augmentation of the cells' own mitochondrial quality control, in this case the extractase ATAD1, is sufficient to remove peroxins and restore several aspects of mitochondrial function in human ZSD fibroblasts. We propose that this newly identified interplay between mitochondria and peroxisomes has significant impact for our understanding of the pathophysiology of ZSD and will inform future therapeutic strategies.

Materials and Methods

Reagents and Tools table

Reagent /resource	Reference or source	Identifier or catalog number
Antibodies
HA epitope	BioLegend	MMS‐101P‐500
V5 epitope mouse	Abcam	ab27671 [SV5‐Pk1]
V5 epitope rabbit	Abcam	ab9116
FLAG epitope	Sigma Aldrich	F7425
HIS epitope	Dr. Adam Hughes	N/A
VDAC	Dr. Dennis Winge	N/A
Por1	Abcam	ab110326
Hsp70	Dr. Kostas Tokatlidis	N/A
Cytochome b	Dr. Kostas Tokatlidis	N/A
GFP	Abcam	ab290
ATAD1	Antibodies Incorporated/NeuroMab, Davis, CA, USA	Cat # 75‐157
Donkey anti‐Rabbit IgG (H + L) Highly Cross‐Adsorbed Secondary Antibody, Alexa Fluor 680	Invitrogen #A10043	AB_2534018
Goat Anti‐Mouse IgG (H&L) Antibody Dylight™ 800 Conjugated	Rockland #610‐145‐002‐0.5	AB_10703265
Chemicals, peptides, and recombinant proteins
Lithium acetate dihydrate	Sigma Aldrich	L6883
Oleic acid	Sigma Aldrich	O1008
TRI Reagent	Ambion	AM9738
Rnase I	Ambion	AM2294
Micro Bio‐Spin P6 gel columns	Bio‐Rad	7326222
Superscript III Reverse Transcriptase	Invitrogen	18080044
T4 polynucleotide kinase	NEB	M0201S
T4 RNA ligase I	NEB	M0437
T4 RNA ligase II, truncated KQ	NEB	M0373H
[gamma‐32P]ATP	Perkin‐Elmer	NEG035C001MC
Rnasin Plus	Promega	N2611
Complete Mini EDTA‐free Protease Inhibitor tablets	Roche	4693159001
Cycloheximide	Sigma‐Aldrich	C1988
Tigecycline	Sigma‐Aldrich	PZ0021
SUPERaseIn	ThermoFisher	AM2694
Proteinase K	ThermoFisher	25530049
Dynabeads MyOne Streptavidin C1 beads	ThermoFisher	65001
Mitotracker CMXRos	Life Techonolgies	M7512
Mitotracker deep red FM	Life Technologies	M22426
β‐mercaptoethanol	Sigma Aldrich	M6250 (SHBC3203V)
N‐ethylmaleimide	Sigma Aldrich	E3876 (SLBV2285)
Digitonin special‐grade (water‐soluble)	Gold Biotechnology	D‐180 (3936.112917A)
Lyticase from Anthrobacter luteus	Sigma Aldrich	L4025 (SLB4644)
Mammalian Protease Inhibitor Cocktail	Sigma Aldrich	P8340
Protease inhibitor cocktail (for use with yeast)	Sigma Aldrich	P8215 (026M4064V)
Histodenz™	Sigma Aldrich	D2158
L‐Glutamine	Sigma Aldrich	G3126
D‐Glucose	Sigma Aldrich	G8270
Sodium Palmitate	Sigma Aldrich	P9767
Bovine Serum Albumin (Ultra Fatty Acid Free)	Roche	CN 03 117405 001
Gibco™ DMEM, high glucose, GlutaMAX™ Supplement, pyruvate	Thermo Fisher	10569010
Trypsin	Thermo Fisher	9057
TMT10plex Isobaric Label Reagent Set plus TMT11‐131C Label Reagent	Thermo Fisher	A34808
Formaldehyde 16%, ultra pure EM grade	Polysciences Inc	#18814‐10
Critical commercial assays
Ribo‐Zero Gold rRNA Removal Kit (Yeast)	Illumina	MRZY1306
Pierce BCA Protein Assay Kit	Thermo Scientific	23225 (RL242698)
Pierce Quantitative Colorimetric Peptide Assay	Thermo Scientific	23275
Mito Stress test Kit	Agilent	103010‐100
Software and algorithms
Prism	GraphPad Software	Version 5,6,7,8
FIJI	ImageJ	Version 1.0
ZenPro	CarlZeiss Software
Seahorse XF FLux Analyzer	Software
Other
Seahorse cartridges / plates	Agilent	101085‐004
Yeast strains
pex19Δ msp1Δ	(Chen et al, 2014)	JRY2722
MAT a/Iα  his3 leu2 lys2 met15 trp1 ura3
pex19Δ msp1Δ	(Chen et al, 2014)	JRY2723
MAT a/Iα  his3 leu2 lys2 met15 trp1 ura3
WT (W303)	(Chen et al, 2014)	JRY245
MAT a  his3 leu2 lys2 met15 trp1 ura3
msp1Δ	(Chen et al, 2014)	JRY1816
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex19Δ	(Chen et al, 2014)	JRY2995
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex19Δ msp1Δ	(Chen et al, 2014)	JRY2726
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex3Δ	(Chen et al, 2014)	JRY2895
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex3Δ msp1Δ	(Chen et al, 2014)	JRY2897
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex13Δpex19Δmsp1Δ	This study	JRY4461
MATa/Iα his3 leu2 lys2 met15 trp1 ura3
pex13Δ	This study	JRY4470
MATIα his3 leu2 lys2 met15 trp1 ura3
wild type	This study	JRY4462
MAT a  his3 leu2 lys2 met15 trp1 ura3
msp1Δ	This study	JRY4463
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex19Δ	This study	JRY4464
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex13Δ	This study	JRY4465
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex19Δmsp1Δ	This study	JRY4466
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex13Δpex19Δ	This study	JRY4467
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex13Δmsp1Δ	This study	JRY4468
MAT a  his3 leu2 lys2 met15 trp1 ura3
pex13Δpex19Δmsp1Δ	This study	JRY4469
MAT a  his3 leu2 lys2 met15 trp1 ura3
Cell lines
Fibroblast cell line PEX3, PBD400‐T1	Gift from Steven J. Steinberg, Johns Hopkins University School of Medicine, Baltimore, MD, USA; available through Coriell Institute	PBD400‐T1
Fibroblast cell line PEX16, R176ter	Gift from Steven J. Steinberg, Johns Hopkins University School of Medicine, Baltimore, MD, USA; available through Coriell Institute	GM06231
cWT (pex3‐: pex3+ atad1‐)	This study	JRL‐cWT (pex3‐: pex3+ atad1‐)
cWT+ATAD1 (pex3‐: pex3+ atad1+)	This study	JRL‐cWT+ATAD1 (pex3‐: pex3+ atad1+)
Pex3−/− (pex: pex3‐ atad1‐)	This study	JRL‐Pex3−/− (pex: pex3‐ atad1‐)
Pex3−/− +ATAD1 (pex: pex3‐ atad1+)	This study	JRL‐Pex3−/− +ATAD1 (pex: pex3‐ atad1+)
cWT atad1‐ (pex: pex3+ atad1‐/atad1‐)	This study	JRL‐cWT atad1‐ (pex: pex3+ atad1‐/atad1‐)
Pex3−/−+ATAD1 (pex: pex3‐ atad1+)	This study	JRL‐Pex3−/− +ATAD1 (pex: pex3‐ atad1+)
HEK293T 293 [HEK‐293]	ATCC® CRL‐1573™
Control fibroblast	(Chen et al, 2014)	YCC‐Control fibroblast

Methods and Protocols

Yeast strains

Saccharomyces cerevisiae W303 (MATa/α, his3 leu2 met15 trp1 ura3) was used to generate all knockout strains. Haploid single and double mutants were generated using a standard PCR‐based homologous recombination method. Briefly, the drug selection cassette (KanMX4, hphMX4, or natMX4) flanked by ∼ 45‐bp fragments upstream and downstream of the gene of interest was PCR amplified and transformed into the wild‐type diploid strain (JRY245) (Wach et al, ¹⁹⁹⁴; Goldstein & McCusker, ¹⁹⁹⁹). JRY4461 was generated by crossing JRR4470 with JRY2726. Haploid strains were generated by sporulation and tetrad analysis. The genotypes of the strains were verified by standard genotyping PCR and are listed in the Key Resources Table. Yeast transformation was performed using the standard TE/LiAc method (Gietz et al, ¹⁹⁹²). Transformed cells were grown in synthetic complete glucose (SD) medium lacking the appropriate amino acid(s) for selection purposes. Medium used in this study is synthetic minimal medium supplemented with different carbon sources: 2% glucose, 2% glycerol, 2% ethanol, or 0.1% oleate. Solid plates contain 2.2% (w/v) agar.

Plasmid construction

To produce plasmids expressing yeast peroxins or Msp1, the ORF and its upstream ∼ 600‐bp promoter region were PCR amplified from yeast genomic DNA and ligated into pRS426, or pRS416 vectors containing either no tag or a C‐terminal His6/HA3‐, V5‐, or GFP/RFP‐tag. The peroxisomal fluorescent marker construct was made by fusing RFP protein (pRS416‐monoRFP) with the peroxisomal targeting sequence (SKL) at the C‐terminus. The generation of human ATAD1 constructs is described in (Chen et al, ²⁰¹⁴). The Retro‐XTMQ Vectors (Clontech) were used to generate mammalian constructs. The human ATAD1 ORF fused to a C‐terminal HA or GFP‐tag, or human PEX3, PEX11, PEX12, or PEX13 ORFs fused to N‐terminal GFP were ligated into pQCXIP or pQCXIZ vectors.

Yeast growth assays

Growth assays were performed using synthetic minimal media supplemented with the appropriate amino acids and indicated carbon source. For plate‐based growth assays, overnight cultures were back‐diluted to equivalent OD600s and spotted as 5‐fold serial dilutions. Cells were allowed to grow for 2–3 days at 30°C before imaging.

Ribosome profiling

RNA‐seq and ribosome‐footprint profiling (RPF) libraries were prepared as described in (Subtelny et al, ²⁰¹⁴) with small modifications from (Wu et al, ²⁰¹⁹). All culturing was performed with synthetic media containing 2% EtOH and 2% glycerol. Cultures were seeded as a 3 ml starter culture from a fresh YPAD plate and grown overnight. The starter culture was used to seed a 50 ml midi culture, which was grown overnight. The midi culture was used to seed a 500 ml culture at OD600 = 0.1, which was grown for 16 h and harvested at OD600 = 2–2.5 by vacuum filtration and rapid flash freezing in liquid nitrogen. Lysate powder was aliquoted and stored at −80°C. Frozen pellets were mechanically lysed using a Sample Prep 6870 Freezer/Mill (Spex SamplePrep; 10 cycles of 2 min on, 2 min off at setting 10).

For RNA‐seq samples, total RNA was extracted from lysate powder using TRI Reagent (Ambion) according to the manufacturer’s protocol. rRNA was depleted from 5 μg of total RNA using the Ribo‐Zero Gold Yeast rRNA Removal Kit (Illumina) according to the manufacturer’s protocol. For RPF samples, lysate powder was resuspended in a buffer containing cycloheximide and tigecycline in order to arrest translating ribosomes. Monosomes were isolated by treating the lysate with RNase I followed by purification over a 10–50% sucrose gradient. Ribosome‐protected fragments were further isolated by Proteinase K/SDS digestion and phenol/chloroform extraction.

RNA‐seq samples were fragmented by alkaline hydrolysis and size‐selected in the range of 19–40 nt. RPF samples were size‐selected in the range of 19–33 nt, and contaminating rRNA sequences were removed via subtractive hybridization. RNA‐seq and RPF samples were both ligated to adaptors, reverse‐transcribed, and amplified to prepare sequencing libraries. These libraries were sequenced on an Illumina HiSeq 2500 using 60‐bp single‐end reads.

Reads were trimmed of adaptor sequence using cutadapt (v1.8) (5 prime adaptor: 5′ GTTCAGAGTTCTACAGTCCGACGATCNNNNNNNN 3′; 3 prime adaptor: 5′ TCGTANNNNNNTCGGAAGAGCACACGTCTddC 3′). Trimmed reads shorter than 10 nucleotides were discarded. Trimmed reads were aligned to the S. cerevisiae genome (R64‐1‐1, downloaded from www.yeastgenome.org) using STAR (v2.5) with the parameters “‐‐alignIntronMax 1000 ‐‐sjdbOverhang 31 ‐‐outSAMtype BAM SortedByCoordinate ‐‐quantMode GeneCounts” and with “‐‐sjdbGTFfile” supplied with transcript annotations. The transcript annotations included all annotated ORFs and excluded the first 50 nt of each ORF.

Fluorescence microscopy

Wild‐type W303 (JRY245) or derived mutant strains were transformed with plasmid expressing mitochondria‐targeted (ATPase subunit, Su9) GFP (mtGFP), RFP‐SKL (GPD promotor), or plasmids expressing Pex13‐RFP or Pex11‐RFP under their respective endogenous promoter. Samples were prepared by growing yeast cells to early log phase (OD600 = 0.8–1) in synthetic dropout medium with appropriate carbon sources at 30°C. To prepare mammalian cell samples, we cultured cells on chamber slides (Nunc^® Lab‐Tek^® Chamber Slide™ system, Millipore) overnight to allow them to fully attach. Cells were either visualized directly, or stained with 25 nM MitoTracker Red CMXRos (Life Technologies) in FBS‐free medium for 10 min, followed by a 10 min recovery first in serum‐containing media and then in phenol red‐free media at 37°C before imaging. Green signals were acquired by tagging the peroxin (Pex10, Pex12, Pex13) with GFP using a pLenti‐CMV‐GFP‐Puro vector (Campeau et al, ²⁰⁰⁹) as described in (Chen, 2015). Images were acquired using either an Axio Observer Z1 imaging system (Carl Zeiss) equipped with 40x and 100x objectives (oil‐immersion) or a Zeiss LSM 710 laser‐scanning confocal microscope with a 40x or 100x (oil‐immersion) objectives. Digital fluorescence and differential interference contrast (DIC) images were acquired using a monochrome digital camera (Axio‐Cam MRm, Carl Zeiss). ZenPro software (Version 4.8, Carl Zeiss) was used to optimize images (tool: best fit) and set the scale bar (in some images superimposed using Adobe Illustrator). The final images were assembled using Adobe Photoshop CS5.1. The yeast fluorescent images shown in this study are representative images from at least two independent experiments (n ≥ 2). The images of the mammalian cells are representative of at least two independent experiments (n ≥ 2).

Airyscan microscopy

A Zeiss LSM880 with Airyscan (Carl Zeiss,) was used to acquire images in Fig 5A. Mammalian cells were plated overnight on glass bottom fluorodishes (WPI Inc) and imaged at 37°C and 5% CO₂. Z‐stacks were acquired with a 63x 1.4 NA oil objective using Immersol 518 F/30°C immersion fluid (Carl Zeiss) under Fast Airyscan mode (Huff, 2016). Raw images were Airyscan processed using ZenPro software (Carl Zeiss) with “auto” settings. FIJI software (Schindelin et al, ²⁰¹²) was then used to generate a maximum projection image comprising 5‐8 μm (z‐depth) of the sample, the scale bar was inserted and superimposed using Adobe Illustrator, and linear adjustments were made to the brightness and contrast of each image.

Isolation of yeast mitochondria and fractionation techniques

Yeast cells were harvested at OD600 = 2–3 unless indicated otherwise. Preparation of crude and purified mitochondria was performed as described previously (Boldogh & Pon, 2007). Briefly, the yeast pellet was washed once with Milli‐Q water, resuspended, and incubated in TD buffer (100 mM Tris‐SO₄, pH 9.4 and 100 mM DTT) for 15 min at 30°C. Spheroplasting was achieved by incubating cells in SP buffer (1.2 M sorbitol and 20 mM potassium phosphate, pH 7.4) supplemented with lyticase (2 mg/g of cell pellet) (Sigma‐Aldrich) for 75 min at 30°C. Spheroplasts were gently washed in ice‐cold SHE buffer (1.2 M sorbitol, 20 mM HEPES‐KOH, pH 7.4, 2 mM MgCl₂, 1 mM EGTA, and 1 mM PMSF) followed by homogenization in ice‐cold SHE buffer (containing 0.6 M sorbitol) with a Dounce homogenizer (15–20 strokes). The crude mitochondrial fraction was obtained by differential centrifugation. Continuous Histodenz™ gradients were used to purify crude mitochondria. To make a gradient, 2.1 ml of 5, 10, 15, 20, and 25% Histodenz^TM in SHE buffer was layered in 14 mm 89 mm Ultra‐Clear centrifuge tubes (Beckman) and the tubes were equilibrated at 4°C for 3–4 h to allow the Histodenz^TM to diffuse. Crude mitochondria were loaded on top of the chilled gradient and separated at 100,000 g for 1 h at 4°C (SW41 rotor; Beckman). Intact purified mitochondria were recovered from a brown band at around 16% Histodenz™ concentration. Protein concentration was determined using a Bradford protein assay kit (Bio‐Rad).

For sodium carbonate extraction assays, 25 μg of mitochondria was incubated in 25 μl of 0.1 M Na₂CO₃ (pH 11) for 10 min on ice. To separate the membrane pellet and supernatant, samples were centrifuged at 14,800 g for 15 min. Fractions were resuspended in equal volumes of SDS sample buffer. For protease treatment assays, 25 μg of mitochondria was incubated in 30 μl of SHE buffer (untreated control), 30 μl SHE buffer with 4‐8 units of Proteinase K (high fidelity buffer), or 30 μl SHE buffer with 4‐8 units of Proteinase K, 4% DMSO and 2 mM Phenylmethylsulfonyl fluoride (low fidelity buffer) for 3 min on ice. The reaction was stopped by adding 4 × SDS sample buffer, and proteinase K was heat inactivated at 96°C for 6 min. The samples were loaded immediately on SDS‐gels for separation.

Quantitative proteomics

This method was developed in the lab of S.P.G., and method description and summaries might contain similar wording as published elsewhere, including (Van Vranken et al, ²⁰¹⁸). The quantitative proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD025595, Project Name: The biochemical basis of mitochondrial dysfunction in Zellweger spectrum disorder, Project accession: PXD025595, https://www.ebi.ac.uk/pride/archive/projects/PXD025595.

Sample preparation

Cell lysis and protein digestion, and peptide labeling: In order to fully use the sample multiplexing ability, an 10‐plex experiment was designed containing the following proteome‐wide comparisons: pex19Δ (n = 5), pex19Δmsp1Δ (n = 5), and a 10‐plex experiment with pex19Δ (n = 2), pex19Δmsp1Δ (n = 2) (Fig 2B and C), pex3Δ (n = 3), pex3Δmsp1Δ (n = 3) (Fig 3A and B). Cell pellets from each condition were resuspended at 4°C in buffer containing 8 M urea, 50 mM EPPS pH 8.5, 50 mM NaCl, and protease inhibitors. Chilled zirconium oxide beads were added to cell slurries. Cells were lysed using a Mini‐Beadbeater (BioSpec products, Bartlesville, OK) at 4°C in 2 ml screw cap tubes for 5 cycles of 30 s each, with 1 min pauses between cycles to avoid overheating. After centrifugation, clarified lysates were transferred to new tubes. Bicinchoninic acid (BCA) protein assay (Thermo Fisher Scientific) was performed to determine protein concentration. Proteins were then subjected to disulfide reduction with 5 mM tris(2‐carboxyethyl)phosphine (TCEP), (room temperature, 30 min) and alkylation with 10 mM iodoacetamide (room temperature, 30 min in the dark). 15 mM dithiothreitol was used to quench excess iodoacetamide (room temperature, 15 min in the dark). Proteins (200 μg) were then chloroform/methanol precipitated and washed with methanol prior to air drying. Samples were resuspended in 8 M urea, 50 mM EPPS, pH 8.5, and then diluted to < 1 M urea with 50 mM EPPS, pH 8.5. Proteins were digested for 16 h with LysC (1:100 enzyme:protein ratio) at room temperature, followed by trypsin (1:100 enzyme:protein ratio) for 6 h at 37°C. Peptides were quantified using Pierce Quantitative Colorimetric Peptide Assay. TMT‐10 reagents (0.8 mg) were dissolved in 40 μL anhydrous acetonitrile, and 7 μL was used to label 70 μg peptides in 30% (v/v) acetonitrile. Labeling proceeded for 1 h at room temperature, until reaction was quenched using 7 μL 5% hydroxylamine. TMT‐labeled peptides were then pooled, vacuum centrifuged to dryness, and cleaned using 50 mg Sep‐Pak (Waters).

The pooled TMT‐labeled peptide sample was fractionated using BPRP HPLC. We used an Agilent 1260 Infinity pump equipped with a degasser and a single wavelength detector (set at 220 nm). Peptides were subjected to a 50 min linear gradient from 8% to 40% acetonitrile in 10 mM ammonium bicarbonate pH 8 at a flow rate of 0.6 ml/min over an Agilent 300Extend C18 column (3.5μm particles, 4.6 mm ID and 250 mm in length). We collected a total of 96 fractions, then consolidated those into 24 and vacuum centrifuged to dryness. Twelve of the 24 fractions were resuspended in a 5% acetonitrile, 1% formic acid solution. Fractions were desalted via StageTip, dried via vacuum centrifugation, and reconstituted in 5% acetonitrile, 5% formic acid for LC‐MS/MS processing.

Mass spectrometry for quantitative proteomics

Mass spectrometry data were collected using an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) equipped with a Proxeon EASY‐nLC 1000 liquid chromatography (LC) system (Thermo Fisher Scientific). Peptides were separated on a 100 μm inner diameter microcapillary column packed with ∼ 35 cm of Accucore C18 resin (2.6 μm, 150 Å, Thermo Fisher Scientific). Approximately 2 μg of peptides were separated using a 2.5 h gradient of acidic acetonitrile using the multinotch MS3‐based TMT method (McAlister et al, ²⁰¹⁴). The scan sequence began with a MS1 spectrum (Orbitrap analysis; resolution 120,000; mass range 400–1,400 Th). MS2 analysis followed collision‐induced dissociation (CID, CE = 35) with a maximum ion injection time of 150 ms and an isolation window of 0.5 Da. The 10 most abundant MS1 ions of charge states 2–6 were selected for MS2/MS3 analysis. To obtain quantitative information, MS3 precursors were fragmented by high‐energy collision‐induced dissociation (HCD, CE = 55) and analyzed in the Orbitrap (resolution was 50,000 at 200 Th) with a maximum ion injection time of 150 ms and a charge state‐dependent variable isolation window of 0.7 to 1.2 Da (Paulo et al, ²⁰¹⁶).

Data analysis

MS2 mass spectra were processed using a SEQUEST‐based software pipeline (Beausoleil et al, ²⁰⁰⁶; Elias & Gygi, ²⁰⁰⁷; McAlister et al, ^{2012, 2014}; Wühr et al, ²⁰¹²). Spectra were converted to mzXML using a modified version of ReAdW.exe. Database searching used the yeast proteome downloaded from UniProt (The UniProt Consortium, 2015) in both forward and reverse directions, along with common contaminating protein sequences. Searches were performed using a peptide mass tolerance of 20 ppm, and a fragment ion tolerance of 0.9 Da. These wide mass‐tolerance windows were chosen to maximize sensitivity in conjunction with SEQUEST searches and linear discriminant analysis (Beausoleil et al, ²⁰⁰⁶; Huttlin et al, ²⁰¹⁰). TMT tags on lysine residues and peptide N termini (+ 229.163 Da) and carbamidomethylation of cysteine residues (+ 57.021 Da) were set as static modifications, and oxidation of methionine residues (+ 15.995 Da) was set as a variable modification.

Peptide‐spectrum matches (PSMs) were adjusted to a 1% false discovery rate (FDR) (Elias & Gygi, 2007). Linear discriminant analysis was used to filter PSMs, as described previously (Huttlin et al, ²⁰¹⁰), while considering the following parameters: XCorr, ΔCn, missed cleavages, adjusted PPM, peptide length, fraction of ions matched, charge state, and precursor mass accuracy. PSMs were identified, quantified, and collapsed to a 1% peptide false discovery rate (FDR) and then collapsed further to a final protein‐level FDR of 1%. PSMs were quantified from MS3 scans; those with poor quality, MS3 spectra with total TMT reporter signal‐to‐noise ratio that is < 200, or no MS3 spectra were excluded from quantitation. Protein quantitation was performed by summing the signal‐to‐noise values for all peptides for a given protein. Each TMT channel was summed across all quantified proteins and normalized to enforce equal protein loading. Each protein’s quantitative measurement was then scaled to 100.

For statistical analysis of proteomics experiments, the signal‐to‐noise ratios for each of the six channels were scaled to represent the percent of total signal in the six channels. Student’s t‐tests were performed using scipy.stats.ttest_ind() (Mckinney, 2010). Effect sizes were calculated for Cohen's d (Sawilowsky, 2009; Cohen, ²⁰¹³). A protein's relevance to the model was determined using a P‐value threshold of < 0.05 with an absolute effect size threshold of 1 or greater (large effect).

Quantitative proteomics data analysis for plotting

Quantitative proteomics samples were geometric mean‐normalized. Common samples between experiments were not combined. Data for heat maps were z‐score normalized across each protein and plots were generated using the xpressplot.heatmap() function from XPRESSplot (Waskom & Kiani, 2018; Berg et al, ²⁰²⁰) and further formatting was performed using Matplotlib. Volcano plots were generated using the geometric‐normalized datasets with no further normalization. Scatter plots were generated using Matplotlib (Hunter, 2007). For code and raw data, please refer to the “Data and Code Availability” section.

Western blot analysis

For immunostaining, proteins were transferred to nitrocellulose membrane after SDS–PAGE and PVDF membrane after BN–PAGE. The membranes were treated with rabbit/mouse antibodies against the indicated epitope tag or loading control according to standard procedures followed by secondary antibodies with fluorophore labels or HRP. Blots were scanned using a Licor Clx or developed following standard procedures. Final figure assembly was done using Adobe photoshop, where contrast and brightness were adjusted using only linear scaling. Final assembly was done in Adobe Illustrator. Loading control signals were acquired by subsequent blotting, or on a separate gel using the exact same preparation.

BN–PAGE

BN–PAGE was performed by I.W. Method description and summaries might therefore contain similar wording as published elsewhere.

Mitochondrial pellets (1 mg mitochondrial protein) were resuspended in 100 μl solubilization buffer (30 mM HEPES, 150 mM potassium acetate, 10% [v/v] glycerol, and 5% [w/v] digitonin); protocol was performed as in (Klodmann et al, ²⁰¹¹). BN–PAGE was performed as published previously by (Wittig, Braun & Schägger, 2006). Briefly, separation was achieved along a linear gradient of 3–18% polyacrylamide equivalent to a molecular mass range of 50–3000 kDa. BN gels were either directly stained with colloidal Coomassie Blue (Neuhoff et al, ¹⁹⁸⁸) or transferred to PVDF membranes for Western blotting.

Complexome profiling

The complexome profiling mass spectrometry proteomics data and method description have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD024625. Project Name: The biochemical basis of mitochondrial dysfunction in Zellweger Spectrum Disorder, Project accession: PXD024625.

Mammalian cell lines and culture

The fibroblast cell line PBD400‐T1 was derived from a patient with peroxisomal biogenesis deficiency (Zellweger Syndrome). The patient carried a single nucleotide insertion (c542insT) leading to a premature stop codon in the peroxin gene, PEX3. An independent fibroblast cell line was derived from a patient with Zellweger syndrome carrying a terminating mutation, R176ter, in PEX16, (GM06231 cells, Coriell Institute). Both cell lines were a gift from S. Steinberg (Johns Hopkins University School of Medicine, Baltimore, MA) and are now available through the Coriell Institute. Cells were immortalized with pBABE‐hygro‐hTert (a gift from Bob Weinberg (Addgene plasmid#1773; http://n2t.net/addgene:1773; RRID:Addgene_1773)), and sustained growth and survival was taken as readout for successful integration. Patient cells were infected, as described below, with retrovirus (pQXCIZ; Clontech) encoding V5‐PEX3 or an empty vector control, and subsequently with retrovirus (pQXCIP; Clontech) encoding FLAG‐HA‐ATAD1 or an empty vector control.

Human fibroblast cells lines were maintained in DMEM/F12 (Gibco) with 15% FBS (Sigma‐Aldrich). HEK293T (ATCC) cells were maintained in DMEM (Corning) with 10% FBS and 1% penicillin/streptomycin (Gibco) (which was omitted in cells plated for virus production). Cells were cultured at 37°C with 5% CO₂. Retrovirus was produced in HEK293T cells co‐transfected with the vector containing the gene of interest, pUMVC‐Gag‐pol (a gift from Bob Weinberg (Addgene plasmid #8449; http://n2t.net/addgene:8449; RRID:Addgene_8449)), and pCMV‐VSVG (a gift from Bob Weinberg (Addgene plasmid#8454; http://n2t.net/addgene:8454; RRID: Addgene_8454)) in a ratio of 3:2:1 using 1 mg/ml polyethylenimine (Polysciences, Inc.). Retrovirus was harvested from the media 48 h post‐transfection, filtered through a 0.45 μm polyethersulfone membrane, collected, and stored at 4°C. Polybrene (EMD Millipore) was added to a final concentration of 10 μg/ml and a 1:1 mixture of viral supernatant and medium (DMEM + 15% FBS) was applied to the target cells. 24 h post‐infection, target cells were selected using 4 μg/ml puromycin (Thermo Fisher) or 150 μg/ml zeocin (Invivogen) for 5–7 days. Stable cell lines were maintained in the appropriate medium with 0.5 μg/ml puromycin (Thermo Fisher) and/or 20 μg/ml zeocin (Invivogen).

Lentivirus (for making PEX‐13‐GFP and PEX10‐GFP expressing lines) was produced in HEK293T cells that were co‐transfected with the vector containing the gene of interest and second generation envelope and packaging plasmids (psPAX–gift from Didier Trono (Addgene plasmid#12260; http://n2t.net/addgene:12260; RRID:Addgene_12260) and pMD2.G–a gift from Didier Trono (Addgene plasmid#12259; http://n2t.net/addgene:12259; RRID:Addgene_12259) in a ratio of 4:2:1 using polyethylenimine (PEI) (Polysciences, Inc.) (PEI:DNA mass ratio 3:1). Virus‐containing medium was collected after 48 h, filtered through 0.45 mm polyethersulfone membrane, aliquoted, and stored at −80°C. Lentivirus‐containing supernatant was added in a 1:1 ratio with medium (DMEM + 15% FBS) to cells together with polybrene transfection reagent at a final concentration 8–10 mg/ml (EMD Millipore). 48 h post‐infection, GFP+ cells were sorted using the Aria Cell Sorter. Cells were allowed to recover for 1 to 3 days before imaging.

CRISPR‐Cas9 deletion of ATAD1

Oligonucleotides encoding sgRNAs targeting human ATAD1 were annealed and cloned into the Px458 vector. pSpCas9(BB)‐2A‐GFP (PX458) was a gift from Feng Zhang (Addgene plasmid # 48138; http://n2t.net/addgene:48138; RRID:Addgene_48138). Two sgRNAs were used, targeting either exon 2 (sense oligo: 5′‐CACCCCGACTCAAAGGACGAGAAA‐3′) or exon 5 (sense oligo: 5′‐CACCCGGTCAGTGTCGAAGGCTGA‐3′; overhang in bold). Patient fibroblasts (see above) were transfected with one of these two plasmids using Lipofectamine 3000 (Thermo Fisher). Two days later, GFP+ cells were sorted and plated as single cells into each well of 96‐well plates. Clonal cell lines were maintained in DMEM + 15% FBS, expanded, and loss of ATAD1 was assessed by Western blot using a monoclonal ATAD1 antibody (Cat# 75‐157 Antibodies Incorporated/NeuroMab, Davis, CA, USA).

Electron microscopy

Cells were grown on aclar discs as monolayers at confluency of 70% and immediately fixed in 2.5% glutaraldehyde, 1% paraformaldehyde in Cacodilic buffer, pH 7.4 for overnight at 4°C. Next day, the cells were rinsed in cacodilic buffer and postfixed in 2% Osmium tetroxide for 1 h at room temperature. The specimens were rinsed in dH₂O and prestained in uranyl acetate for 1 h at room temperature followed by dehydration in increasing concentrations of ethanol, followed by absolute acetone and infiltrated with epoxy resin (Embed‐812 EMS, cat. # 14120, EMS). The polymerization was done at 60°C for 48 h.

The blocks were cut at 70 nm thickness using Leica UC 6 ultratome (Leica Microsystems, Vienna, Austria) and post‐stained with uranyl acetate for 10 min, and lead citrate for 5 min.

Sections were examined at an accelerating voltage of 120 kV in a JEOL‐1400 plus (JEOL, Japan) transmission electron microscope equipped with CCD Gatan camera at 6,000× magnification at the specimen level.

Approximately 50–100 mitochondria were collected from each specimen, and the experiments were repeated with three cultures in each case. Data were analyzed by a human individuum via blinded visual identification of mitochondrial morphologies (cWT n = 232, PEX3^−/− n = 266, PEX3^−/− + ATAD1 n = 257). Their morphology was defined as follows: mitochondria with many cristae and low electron density (type 1), mitochondria with fewer cristae (type 2), and mitochondria with high electron density and/or almost no cristae (type 3). Representative images are displayed in Fig 5B (and Appendix Fig S3). FIJI software (Schindelin et al, ²⁰¹²) was used to generate the scale bar which was superimposed using Adobe Illustrator. The focus on one mitochondria was achieved by cropping the EM image. Image analysis for Fig 5B was performed using GraphPad Prism6.

Mitochondrial oxygen consumption rate (OCR)

Mitochondrial oxygen consumption rate (OCR) was assessed an XF96e Extracellular Flux Analyzer (Seahorse Bioscience), as described previously (Zhang et al, ²⁰¹²). One thousand cells per well of the indicated human fibroblast cell line were plated in XF96 cell culture microplates with XF Dulbecco’s modified Eagle medium (DMEM) containing 10 mM glucose, 2 mM L‐glutamine (Life Technologies), and 2 mM sodium pyruvate (Life Technologies). OCR was measured at 37°C with 1‐min mix and 3‐min measurement protocol. OCR was analyzed after 30 min incubation in a CO₂‐free incubator. Oligomycin (2 μM), carbonilcyanide m‐fluorophenylhydrazone (FCCP) (1.5 μM), and rotenone + antimycin (1 μM each) were sequentially injected into each well to assess basal respiration, coupling of respiratory chain, and mitochondrial respiratory capacity. OCRs were normalized relative to cell count in each well. Basal respiration and maximal respiration were calculated by subtraction of the non‐mitochondrial respiration from the initial respiration and respiration after FCCP uncoupling, respectively. The respiratory spare capacity was calculated by substraction of the basal respiration from the maximal respiration. All plots are generated from raw XF96e Flux analyzer data from two independent experiments performed on separate days with two biological replicates plated in two technical replicates (2 × 2 samples / day, resulting in n = 4 biological as well as technical replicates), imported into WAVE software, and processed through the XF Mito Stress Test Report. Significance was calculated with Welch’s test in GraphPad Prism8 and plots were created in GraphPad Prism8.

Lipidomics

This method description is widely distributed at the University of Utah and external collaborators. Method description and summaries might therefore contain similar wording as published elsewhere.

Lipidomics sample extraction

Samples were prepared as described in (Matyash et al, ²⁰⁰⁸) with following modifications: Solutions are pre‐chilled on ice. Lipids are extracted in a solution of 225 µl MeOH and 750 µl methyl tert‐butyl ether (MTBE) containing internal standards (Avanti SPLASH LipidoMix, Avanti LM6003 Cardiolipin Mix and Avanti Ceramide LipidoMix all at 10 µl per sample; LPC(26:0)‐d4 at 156 pmol per sample; C18(plasm)‐18:1(d9) PE at 135 pmol per sample; C18(plasm)‐18:1(d9) PC at 128 pmol per sample). For other than occasional vortexing, samples incubate on ice for 1 h. Following the addition of 188 µl of PBS and brief vortexing, the samples rest at room temperature for 10 min. Samples are centrifuged at 16,000 g for 5 min at 4°C to collect the upper phases which then evaporated to dryness under speedvac. The bottom aqueous layer is collected separately and dried under speedvac. Lipid extracts are reconstituted in 500 µl IPA and transferred to an LC‐MS vial for analysis. Concurrently, a process blank sample and pooled quality control (QC) samples are prepared by taking equal volumes (∼ 50 µl) from each sample after final resuspension.

Mass spectrometry analysis of samples

Lipid extracts are separated on a Waters Acuity UPLC CSH C18 1.7 µm 2.1 × 100 mm column maintained at 65°C connected to an Agilent HiP 1290 Sampler, Agilent 1290 Infinity pump, equipped with an Agilent 1290 Flex Cube and Agilent 6530 Accurate Mass Q‐TOF dual AJS‐ESI mass spectrometer. For positive mode, the source gas temperature is set to 225°C, with a drying gas flow of 11 l/min, nebulizer pressure of 40 psig, sheath gas temp of 350°C, and sheath gas flow of 11 l/min. VCap voltage is set at 3500 V, nozzle voltage 1000 V, fragmentor at 110 V, skimmer at 85 V, and octopole RF peak at 750 V. For negative mode, the source gas temperature is set to 300°C, with a drying gas flow of 11 l/min, a nebulizer pressure of 30 psig, sheath gas temp of 350°C, and sheath gas flow 11 l/min. VCap voltage is set at 3500 V, nozzle voltage 2000 V, fragmentor at 100 V, skimmer at 65 V, and octopole RF peak at 750 V. Samples are analyzed in a randomized order in both positive and negative ionization modes in separate experiments acquiring with the scan range m/z 100–1,700. Mobile phase A consists of ACN:H2O (60:40 v/v) in 10 mM ammonium formate and 0.1% formic acid, and mobile phase B consists of IPA:ACN:H2O (90:9:1 v/v) in 10 mM ammonium formate and 0.1% formic acid. The chromatography gradient for both positive and negative modes starts at 15% mobile phase B then increases to 30% B over 2.4 min, it then increases to 48% B from 2.4–3.0 min, then increases to 82% B from 3–13.2 min, then increases to 99% B from 13.2–13.8 min where it’s held until 16.7 min, and then returned to the initial conditioned and equilibrated for 5 min. Flow is 0.4 ml/min throughout, injection volume is 3 µl for positive and 10 µl negative mode. Tandem mass spectrometry is conducted using the same LC gradient at collision energy of 25 V.

Analysis of mass spectrometry data

QC samples (n = 8) and blanks (n = 4) are injected throughout the sample queue and ensure the reliability of acquired lipidomics data. Results from LC‐MS experiments are collected using Agilent Mass Hunter (MH) Workstation and analyzed using the software packages MH Qual, MH Quant, and Lipid Annotator (Agilent Technologies, Inc.). Results from the positive and negative ionization modes from Lipid Annotator are merged then split based on the class of lipid identified. The data table exported from MHQuant is evaluated using Excel where initial lipid targets are parsed based on the following criteria. Only lipids with relative standard deviations (RSD) less than 30% in QC samples are used for data analysis. Additionally, only lipids with background AUC counts in process blanks that are less than 30% of QC are used for data analysis. The parsed excel data tables are normalized to tissue mass, and positive and negative mode data are merged.

Data pretreatment

Data were analyzed using in‐house software to normalize and scale (pareto). Supplemental tables also contain the raw input on which no treatment was performed.

Resource availability

All unique/stable reagents and resources generated within this study are available from the corresponding author Jared Rutter (rutter@biochem.utah.edu), upon request and without restriction.

Author contributions

EN and JR designed the study. EN wrote the first version of the manuscript. JTM, SF, and JR contributed significantly to the final manuscript. EN, SF, JTM, and YCC collected the data. EN conceived of the presented idea, developed the theory, and carried out most of the experiments. YCC created yeast strains and generated plasmid constructs, and imaged human cell lines. SF generated and confirmed plasmid constructs, yeast strains, and generated and confirmed the human cell lines used in this manuscript and imaged yeast cells. JTM performed the ribosome profiling and assembled corresponding figures. SL produced lentivirus and infected cells, KJC performed and analyzed quantitative mass spectrometry under the supervision of SPG, JAM, and JEC performed and analyzed lipidomics experiments. IW. performed and analyzed the Blue Native PAGE and complexome profiling. JMW generated and confirmed the human ATAD1 knockout cell lines. CUK performed Airyscan imaging on the Airyscan (Zeiss) under the supervision of MR‐J, JAB performed data analysis of the quantitative proteomics data and prepared figures for these data. JB, NB, CA, and LC advised on peroxisomal biochemistry and the clinical impact of presented data. All authors commented and approved of the final manuscript.

Conflict of interest

The University of Utah has filed a patent related to ATAD1, of which E.N., Y.C.C, J.B., and J.R. are listed as co‐inventors. All other authors declare no competing interests.

Supporting information

Appendix

Expanded View Figures PDF

Source Data for Figure 6

Source Data for Figure EV4

EMBO reports (2021) 22: e51991.

Data availability

The quantitative proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez‐Riverol et al, ²⁰¹⁹) partner repository with the dataset identifier PXD025595, Project Name: The biochemical basis of mitochondrial dysfunction in Zellweger Spectrum Disorder Project accession: PXD025595, (https://www.ebi.ac.uk/pride/archive/projects/PXD025595). The code and processed data tables used to analyze these quantitative proteomics data and reproduce the associated plots are available at GitHub https://github.com/j‐berg/nuebel_2020. RNA‐seq and ribosome‐footprint profiling data have been deposited at the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE159869 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE159869). The complexome profiling mass spectrometry proteomics data and method description have been deposited with the Project name: The biochemical basis of mitochondrial dysfunction in Zellweger Spectrum Disorder, Project accession: PXD024625, onsortium via the PRIDE partner repository with the dataset identifier PXD024625 (http://www.ebi.ac.uk/pride/archive/projects/PXD024625).