Depletion of LONP2 unmasks differential requirements for peroxisomal function between cell types and in cholesterol metabolism

Peroxisomes play a central role in tuning metabolic and signaling programs in a tissue- and cell-type-specific manner. However, the mechanisms by which the status of peroxisomes is communicated and integrated into cellular signaling pathways are not yet understood. Herein, we report the cellular responses to peroxisomal proteotoxic stress upon silencing the peroxisomal protease/chaperone LONP2. Depletion of LONP2 triggered the accumulation of its substrate TYSND1 protease, while the overall expression of peroxisomal proteins, as well as TYSND1-dependent ACOX1 processing appeared normal, reflecting early stages of peroxisomal proteotoxic stress. Consequently, the alteration of peroxisome size and numbers, and luminal protein import failure was coupled with induction of cell-specific cellular stress responses. Specific to COS-7 cells was a strong activation of the integrated stress response (ISR) and upregulation of ribosomal biogenesis gene expression levels. Common changes between COS-7 and U2OS cell lines included repression of the retinoic acid signaling pathway and upregulation of sphingolipids. Cholesterol accumulated in the endomembrane compartments in both cell lines, consistent with evidence that peroxisomes are required for cholesterol flux out of late endosomes. These unexpected consequences of peroxisomal stress provide an important insight into our understanding of the tissue-specific responses seen in peroxisomal disorders. Supplementary Information The online version contains supplementary material available at 10.1186/s13062-023-00416-3.


Introduction
Peroxisomes are ubiquitous eukaryotic organelles that are central to the regulation of lipid metabolism, redox regulation and signaling [52].The precise function of peroxisomes is coupled to cellular demand and the metabolic wiring within a given tissue [13].Peroxisomal metabolic pathways include catalase-mediated reduction of peroxide to water and beta-oxidation of very long chain or branched chain fatty acids.Once fatty acid chains are reduced to shorter lengths, they are transferred to mitochondria for completion of beta-oxidation [52].In liver, peroxisomes function with mitochondria and the endoplasmic reticulum (ER) to generate bile acids, a process that can utilize up to 90% of newly synthesized cholesterol in rodents [49].In the brain, peroxisomal factors are enriched in astrocytes compared to neurons, and cultured astrocytes but not neurons demonstrate betaoxidation activity [8,18,26,40].In oligodendrocytes, a primary function of peroxisomes is the generation of plasmalogen ether lipids, as these lipids synthesized in peroxisomes and ER are essential for myelin formation by these cells [21].These examples highlight the fact that the mass metabolic flux through peroxisomes varies greatly depending on the tissue and need.Pathologies arise when peroxisomal function is altered by pathogenic variants.Peroxisomal biogenesis diseases are the most severe, manifesting with multisystemic developmental pathologies, growth failure, and early lethality [52].Peroxisomal dysfunction is also associated with ageing, cancer, and neurodegenerative diseases, likely contributing to the pathogenesis of common diseases [50,55].
A key open question is how peroxisomes may communicate their status through stress responses and retrograde signaling pathways.New approaches to monitor transcriptional, lipidomic and proteomic changes in conditions of induced peroxisomal stress, primarily through the targeted loss of proteins essential for peroxisomal import, shed light on generalized organellar dysfunction.While global cellular stress responses are elicited in these conditions, there is a surprising variability in the details of these responses depending on the specific import component under investigation, the cell or tissue type and the model organism used [25,37,53].This hints that peroxisomal stress may not always lead to comparable stress responses, likely reflecting distinct functions of peroxisomal import components and their clients.
The loss of peroxisome import provides an excellent experimental approach to reduce or block peroxisomal function.In this study, we took a different approach to specifically monitor the stress response driven by an accumulation of unfolded proteins within the peroxisomal lumen.Currently, lon peptidase 2 (LONP2) is the only known peroxisomal protease that is conserved in plants, fungi and mammals [12].LONP2 is the product of an early gene duplication of mitochondrial LONP1, which was retained from its bacterial origin [11].Lon peptidases act as both a chaperone and an ATP dependent protease responsible for the degradation and turnover of oxidized proteins in bacteria, mitochondria, peroxisomes and chloroplasts [47,54].Studies have directly investigated the biochemistry and substrates of peroxisomal LONP2, where the protease TYSND1, which is responsible for the processing of peroxisomal enzymes including ACOX1 (Kurochkin et al., 2007), is one substrate identified in mammalian cells [31], and other clients including catalase have been identified in the fungi P. chrysogenum and H. polymorpha [1,2].In addition to LONP2 protease activity, its chaperone function was illustrated by the development of significant peroxisomal protein aggregates upon loss of LONP2 in fungi, thereby demonstrating that loss of LONP2 induced proteotoxic stress [1,2].While the function of LONP2 as a key regulator of peroxisomal protein turnover and homeostasis is widely accepted, the functional consequences of these processes have not been explored in mammalian cells [36].To map the peroxisomal proteotoxic stress response that could be independent from a total loss of protein import, we took an siRNA approach to acutely silence expression of LONP2 in two distinct cell lines, COS-7 (African green monkey kidney cell line transformed with SV40) and the human osteosarcoma cell line U2OS.Imaging, RNAseq and lipidomic data reveal new findings that couple peroxisomal homeostasis to cholesterol trafficking, retinoic acid signaling, ribosome biogenesis and the integrated stress response.

Depletion of peroxisomal LONP2 caused peroxisomal dysfunction in COS-7 cells and U2OS cells
To induce acute peroxisome-specific damage, we silenced LONP2 over 6 days in COS-7 and U2OS cells.Immunoblotting of LONP2-silenced cells after 6 days revealed an accumulation of the auto-cleaved products of the protease TYSND1, products established as LONP2 substrates [31], indicating that the efficiency of proteolytic processing of LONP2 inside peroxisomes was impaired (Fig. 1a and c).While there was an accumulation of these autocleaved forms of TYSND1, the remaining enzyme appeared functional since a known TYSND1 substrate, ACOX1, was processed normally (Additional file 1: Fig. S1).The precursor form of ACOX1 was reduced, perhaps reflecting a reduction in import.Analysis of peroxisomal membrane protein PMP70, the luminal protein thiolase (ACAA1) or catalase did not reveal a reduction in either of the protein levels, indicating that the peroxisomal mass was generally unchanged (Fig. 1a and c, Additional file 1: Fig. S3).Confocal microscopy demonstrated that peroxisomes were less abundant, but individual peroxisomes were elongated and enlarged in both cell lines (Fig. 1b and d).
The mechanisms of peroxisomal luminal and membrane protein import are distinct [9].To monitor the import efficiency of luminal proteins, we transiently transfected cells with a peroxisome lumen-targeting sequence serine-lysine-leucine (SKL) fused with the Cyan Fluorescent Protein (CFP) and labelled peroxisomes with peroxisomal membrane protein PMP70.CFP-SKL was efficiently targeted to peroxisomes in control cells, but no longer imported into peroxisomes in LONP2-silenced COS-7 cells (Fig. 1e and f ) or U2OS cells (Additional file 1: Fig. S2), demonstrating that the knockdown of LONP2 impaired peroxisomal luminal protein import.To evaluate the efficiency of peroxisomal membrane protein import, we used adenoviral expression of peroxisomal biogenesis factor 3 (PEX3) tagged with Yellow Fluorescent Protein (PEX3-YFP), whose import was unaltered in LONP2-silenced COS-7 cells (Fig. 1g).These data establish that depletion of LONP2 in both cell lines leads to alterations in proteolysis of substrates like TYSND1 and inhibits protein import of our reporter CFP-SKL into the peroxisomal lumen without affecting membrane protein insertion.Given that the expression levels of endogenous peroxisomal proteins seen by western blot were mostly unchanged, including the processing of TYSND1 substrates like ACOX1, it suggests that the overall effect of LONP2 silencing on peroxisomal protein content to is relatively mild after 6 days of silencing.
Loss of mitochondrial LONP1 leads to the accumulation of aggregated matrix content and the eventual initiation of autophagy-mediated organellar turnover, or mitophagy [58].Thus, we evaluated whether LONP2 leads to the induction of pexophagy by monitoring levels and localization of LC3, a marker of autophagic membranes, and autophagic cargo adapter p62.As a positive control, we induced non-selective autophagy, using HBSS media, starving the cells of essential amino acids [42].This resulted in an increased number of LC3-p62 puncta on peroxisomes (Fig. 1i, left).However, the number of peroxisomes colocalized with transiently expressed GFP-LC3 and the endogenous p62 was unchanged between non-targeted (NT-) and LONP2-silenced cells after 6 days (Fig. 1i, right).Moreover, LONP2 knockdown did not induce global autophagy, as LC3 isoforms and p62 remained unchanged in COS-7 cells (Fig. 1h).These data are consistent with the analysis of peroxisome morphology, which shows enlarged organelles, and immunoblots that showed no change in peroxisomal proteins like PMP70 and thiolase (Fig. 1a-d).
Peroxisomal import failure can lead to the targeting of peroxisomal proteins to mitochondria with detrimental consequences on mitochondrial function, metabolism, and cristae architecture [30].Specifically, the abrogation of Pex3 or Pex19, core components of peroxisomal membrane insertion, resulted in the mitochondrial import of a series of peroxisomal membrane, and some luminal, proteins [30].In addition, alterations in peroxisomal handling of generated as a by-product of mitochondrial respiration reactive oxygen species (ROS) by catalase or peroxiredoxins, or aberrant beta-oxidation of very long chain fatty acids can induce mitochondrial dysfunction [15].Although Pex3-YFP was correctly targeted to peroxisomes in cells lacking LONP2 (Fig. 1g), we further examined whether LONP2 knockdown altered mitochondrial morphology by using Mitochondrial Network Analysis (MiNa), a semi-automated ImageJ plugin [48].Confocal imaging of mitochondria stained with MitoTracker Deep Red FM demonstrated that mitochondrial network morphology remained unchanged upon silencing of LONP2 (Fig. 1j).Next, we employed the fluorescent dye Mito-SOX that specifically detects ROS, and electrochemical potential-dependent TMRE stain.investigate Confocal microscopy showed unaltered TMRE stain, but enhanced accumulation of MitoSOX-stained superoxide within mitochondria in LONP2-silenced COS-7 cells (Additional file 1: Fig. S4a).Surprisingly, MitoSOX-stained superoxide did not accumulate within mitochondria in LONP2-silenced U2OS cells (Additional file 1: Fig. S4b).Since peroxisomal phenotypes upon silencing LONP2 were consistent between the two cell lines, mitochondrial superoxide accumulation in COS-7 cells likely arises from impaired mitochondrial redox status, rather than is caused directly by peroxisomal proteotoxic stress.
Taken together, the loss of LONP2 alters peroxisomal morphology and numbers without initiation of pexophagy or altering mitochondrial network morphology, and in COS-7 cells results in accumulation of mitochondrial ROS. of the colocalization between peroxisomes (PMP70) and transiently expressed peroxisome luminal protein (CFP-SKL) in NT-and LONP2-silenced COS-7 cells after 144 h.Scale bar, 2 µm.g Representative confocal images showing the colocalization between peroxisomes (PMP70) and transiently expressed peroxisome membrane protein (PEX3-YFP) in NT-and LONP2-silenced COS-7 cells after 144 h.Scale bar, 5 µm.h Immunoblotting for autophagy-related proteins, vinculin was used as a loading control.i Representative confocal images of COS-7 cells grown in HBSS for 24 h (left) and quantification (right) showing the number of LC3-P62 puncta on peroxisomes of NT-and LONP2-silenced COS-7 cells and U2OS cells after 144 h.Cells were grown in either DMEM or HBSS for 24 h and LC3-P62 puncta on peroxisome was visualized with transiently expressed GFP-LC3 and the immunofluorescence of p62 and PMP70.Scale bar, 2 µm.j Representative confocal images and quantification showing the size and mean branch length of mitochondria in NT-and LONP2-silenced COS-7 cells after 144 h.Scale bar, 5 µm.On all plots, dots represent individual cells from n = 3 technically independent experiments depicted in different colours, and triangles represent the mean within an independent experiment.The mean values were used to calculate the average (horizontal bar), s.d.(error bars), and P-values (using a two-tailed unpaired Student's t-test)

Peroxisomal proteotoxic stress leads to ISR and ribosome biogenesis activation in a cell specific manner
The specificity of mitochondrial ROS accumulation to COS-7 cells hinted that the loss of LONP2 may lead to global stress responses, which however may be uncoupled from the primary peroxisomal phenotypes common in both cell lines.To investigate the overall cell responses to LONP2 silencing, we analyzed transcriptomes of NT-and LONP2-silenced cells using RNA sequencing (RNAseq) in COS-7 and U2OS cells.Surprisingly, the gene expression response resulting from the loss of LONP2 between the two lines was drastically different (Fig. 2a and b, Additional file 2: Table S1), despite similar peroxisomal morphology and import phenotypes, and the relatively mild effects on overall peroxisomal protein content (Fig. 1c--g).Unsupervised principal component analysis showed that NT-and LONP2-silenced samples separated clearly in COS-7 but not U2OS line (Fig. 2a).In COS-7, expression levels of 3835 genes were downregulated and 3215 genes were upregulated (|log2(FC)|> 0.5, p.adj < 0.05), while in U2OS expression levels of only 182 and 77 genes were down-and up-regulated, respectively (|log2(FC)|> 0.5, p-value < 0.01) (Fig. 2b, Additional file 2: Table S1, and see Methods for details on thresholding of statistical significance).
As we anticipated a stress response signature upon loss of LONP2, we focused our analysis on the integrated stress response (ISR).The ISR is an evolutionary conserved eukaryotic signaling pathway that acts to attenuate cap-dependent translation via phosphorylation of eukaryotic translation initiation factor (eIF2), while simultaneously inducing one or more ISR effectors [activating transcription factors (ATF) ATF3, 4, 5, C/EBP homologous protein (CHOP) encoded by DDIT3] and their downstream targets [6,32].ISR effectors ATF3-5, CHOP, and several of their targets were induced in LONP2-depleted COS-7 but not U2OS cells (Fig. 2c and  d).Consistent with this signature, we observed a dramatic reduction of protein synthesis (a key consequence of eIF2a phosphorylation during the ISR) in COS-7 but not U2OS cells, detected by puromycin labelling for 10 min [43] (Fig. 2e).Induction of global cellular stress in LONP-2 depleted COS-7 cells was further evidenced by the activation of apoptosis-associated cleaved caspases 3 and 7, increased DNA damage (as monitored by γ2HAX phosphorylation) and induction of a regulator of innate immunity transcription factor NF-kB (Fig. 2d).These effects were not observed upon LONP2 depletion in U2OS cells.Together, these data suggest that proteotoxic peroxisomal stress activates ISR in a cell-type-specific manner.This is also consistent with the increased Mito-SOX staining in COS-7 but not U2OS cells (Additional file 1: Fig. S4b).
We also observed a global upregulation of mRNAs encoding factors implicated in ribosome biogenesis in LONP2-depleted COS-7, but not U2OS cells (Fig. 2f ).Expression of a core mediator of ribosomal assembly RRS1 was also increased at the protein level (Fig. 2g) [16].This was unexpected, since ribosome biogenesis is generally suppressed under stress [23,29,46], and LONP2-depletion in COS-7 cells resulted in strong reduction in protein synthesis (Fig. 2e).Collectively, these findings suggest that in addition to induction of ISR, depletion of peroxisomal LONP2 results in altered ribosome biogenesis in COS-7 cells.
A core goal of stress responses to organellar dysfunction is to restore homeostasis, which often includes upregulation of chaperone proteins, and in this case, potentially peroxisomal proteins.A previous report in C.elegans characterizing the response to PEX5 deletion noted the upregulation of peroxisomal genes driven by the activation of transcription factors PPARα/MED15 orthologues [37].However, in both COS-7 and U2OS, depletion of LONP2 did not result in an upregulation of peroxins or peroxisomal genes (Additional file 1: Fig. S5).

Peroxisomal proteotoxic stress leads to defective retinoic acid signalling
While the ISR was not induced in U2OS cells, we sought to identify common responses to peroxisomal proteotoxic stress between COS-7 and U2OS cells, given that they both exhibited altered peroxisomal morphology and reduced import of the reporter CFP-SKL.Gene ontology analysis of transcripts changed in both cell lines revealed a significant downregulation of the retinoic acid (RA) / vitamin A signalling pathway [33] as a top hit (Fig. 3a).We validated this at the protein level by immunoblotting key proteins involved in RA signaling pathway: cellular all-trans retinoic acid (ATRA) binding proteins (CRABP2) and aldehyde dehydrogenase 1 family member A2 (ALDH1A2) [35] (Fig. 3b).Consistent with the RNAseq data, CRABP2 levels were strongly reduced in both COS-7 and U2OS.However, a second enzyme on this pathway, ALDH1A2, was not changed in U2OS, suggesting that the effects at the protein level in this cell line may be partial.
To test the functional consequence of the downregulation of RA signaling enzymes, we treated cells with all-trans retinoic acid (ATRA) to activate RA signaling and drive transcription of specific gene targets.RT-qPCR analysis confirmed the efficiency of LONP2 silencing in both cell lines (Fig. 3c).Upon ATRA treatment, the transcriptional upregulation of a retinoic acid-responsive gene, Transglutaminase 2 (TGM2), was strongly suppressed in LONP2 knockdown in COS-7 cells, but only mildly altered in U2OS (Fig. 3c).Indeed, levels of TGM2 mRNA were reduced in COS-7 cells treated with siLONP2 even prior to ATRA stimulation.Collectively, these data uncover the impact of peroxisomal dysfunction on retinoic acid signaling pathways.

Lipidomic analysis revealed accumulation of sphingomyelins and cholesterol esters in LONP2 silenced cells
One of the core functions of peroxisomes is in the generation and catabolism of specific lipids.Therefore, to investigate the effect of LONP2 knockdown on the regulation of lipid homeostasis, we performed comprehensive untargeted lipidomics in both COS-7 and U2OS cells.The dataset retained 2,094 mass spectrometry features, defined by their m/z, retention time and signal intensity.Similar to gene expression data, the lipidomic changes resulting from the loss of LONP2 between the two lines were drastically different as can be seen from the score plot of unsupervised principal component analysis (Fig. 4a).Among the 2,094 features of the final dataset, 467 (22%) of them passed our selected threshold (|(FC)|> 1.5, P corr -value < 0.05) as significantly different between the two cell lines under basal condition (Additional file 1: Fig. S6a, Additional file 3: Table S2), whereas upon depletion of LONP2 (vs.siNT), 242 were changed in COS-7 and 234 in U2OS cells (Additional file 1: Fig. S6b).It is notable that the lipid profile differed more markedly between the two cell lines at baseline (~ 22% were significantly different) than upon silencing of LONP2 within each cell line (~ 11% significantly different).These differences in lipidome profiles may suggest that the specific flux or wiring of lipid synthesis and turnover pathways may underlie the sensitivity of COS-7 cells to mild alterations in peroxisomal proteostasis upon silencing LONP2.
Given the untargeted nature of lipidomic analysis, only lipid features of interest have been annotated, which for this study are those that passed the selected threshold of significance for the impact of LONP2 silencing in COS-7 and U2OS cell lines (See Methods for details).A total of 206 lipid features were annotated using MS/MS and data alignment with our in-house database [10] and were found to belong to different lipid subclasses (Fig. 4b).The most significantly changed unique lipid species upon LONP2 silencing in COS-7 cells and U2OS cells, which were annotated using MS/MS (22 and 25 unique lipids, respectively), are shown as box plots (Fig. 4c).
From these data, we conclude that for all three group comparisons conducted in this study, glycerophospholipids, specifically phosphatidylcholines (PC), showed the highest number of significantly up-or downregulated lipid features (red dots; Fig. 4b), reflecting substantial remodeling of their fatty acyl side chains.Under the basal state, sphingolipids, both sphingomyelins and glucosylceramides, as well as cholesterol esters were more abundant in COS-7 cells compared to U2OS cells (SP and ST; Fig. 4b).Upon LONP2 silencing, COS-7 and U2OS cells shared some lipid changes: namely downregulation of PCs (Fig. 4b and orange boxes in Fig. 4c), upregulation of sphingolipids, namely sphingomyelins (SM) such as SM(d34:0) and glucosylceramides (SP in Fig. 4b; Purple boxes in Fig. 4c; Fig. S6c) and of lipids containing a saturated or monounsaturated very long chain fatty acyl side chain (VLCFA ≥ 22 carbon chains in bold, Fig. 4c).
There were, however, important differences between these two cell lines in the response to LONP2 silencing for other lipid subclasses.For example, and of relevance to cholesterol metabolism, cholesterol ester (CE) CE20:1 or CE22:1 showed a striking ~ 4 log 2 (FC) accumulation in LONP2-silenced COS-7 cells (P corr = 0.000091).The accumulation of these CEs in LONP2-silenced U2OS cells was less prominent, and variable (upregulated in 2 samples, unchanged in 3), therefore did not pass our significance threshold (log 2 (FC) = 1.36,P corr = 0.26) (Fig. 4b and c, Additional file 1: Fig. S6c, Additional file 3: Table S2).A prominent set of ceramide species were also upregulated in COS-7 cells, many of which carried acyl chains ≥ 22 carbons.In contrast, ether PC was upregulated in LONP2-silenced U2OS but not COS-7 cells (Brown boxes in Fig. 4c).
Given the striking accumulation of CEs in LONP2silenced COS-7 cells vs. LONP2-silenced U2OS cells, we further investigated the sensing of the excess cholesterol in our system.For this, we analyzed the expression of the ER membrane protein insulin-induced gene 1 (INSIG1), a core sensor within the cholesterol biosynthetic pathway that is rapidly degraded in response to the cholesterol depletion [4,38].Surprisingly, INSIG1 showed downregulation with immunoblotting (Fig. 4d) and RNAseq (log 2 (FC) = 3.34, p.adj = 2.56E-14) in COS-7 cells.In U2OS, INSIG1 mRNA and protein remained unchanged (Fig. 4d,RNAseq log 2 (FC) = 0.17, p.value = 0.44).Thus, concurring with lipidomic data, these results show that in COS-7 cells, cholesterol is accumulated upon peroxisomal proteotoxic stress, but the loss of INSIG1 suggests that ER cholesterol sensing is impaired.

Cholesterol flux analysis shows accumulation in endolysosomes in response to LONP2 depletion
Impaired sensing of excess cholesterol in LONP2 depleted cells prompted us to examine the subcellular localization of the elevated cholesterol.For this, we stained cholesterol with Filipin.Confocal imaging showed cholesterol staining to be abundant in both Nile Red-stained lipid droplets, and endolysosomes marked with lysosomal-associated membrane protein 1 (LAMP1) (Fig. 5a-b).LONP2 knockdown also led to an increased number of lipid droplets (Fig. 5a), consistent with increased triglycerides seen in lipidomics analysis (TG in Fig. 4b).
To monitor the flux of cholesterol in the absence of LONP2, we incubated cells with excess free cholesterol for one hour and stained them with Filipin.As with the endogenous cholesterol, confocal microscopy showed enhanced accumulation of Filipin within endolysosomes in both LONP2-silenced COS-7 and U2OS cells as compared to corresponding controls (Fig. 5c and d, left).Filipin staining was observed in less than 8% of endolysosomes in control COS-7 or U2OS cells, which rose to approximately 30% of endolysosomes in LONP2-silenced COS-7 and U2OS cells (Fig. 5 c and d, right).The accumulation of cholesterol in lysosomes is reminiscent of the phenotypes in human lysosomal storage diseases, such as Niemann-Pick disease type C (NPC) caused by pathogenic variants in NPC1 or NPC2, which block cholesterol export from the lysosome [34].Indeed, electron microscopic analysis of LONP2-silenced COS-7 cells revealed many lamellated structures and electron-dense particles that strongly resembled the abnormal lysosomes observed in NPC cells (Fig. 5e).This included the presence of lipid whorls and lamellar inclusions, consistent with the observation that LONP2 depletion impaired sphingomyelin and cholesterol homeostasis (Fig. 5e).
Our results indicate that LONP2 knockdown impairs the ability of cholesterol to efflux from endolysosomes.Trapping of cholesterol in endolysosomes may lead to cholesterol depletion within the ER, driving INSIG1 degradation (Fig. 4d).This was specific to COS-7 cells, which started with significantly higher levels of sphingomyelins, glucocylceramides and cholesterol esters relative to U2OS cells.This may indicate that the steady-state flux of cholesterol and sphingomyelins are much higher in COS-7 at steady state, rendering them more sensitive to LONP2 depletion.Indeed, only upon addition of excess cholesterol for flux analysis in U2OS cells did we observe similar phenotypes to COS-7 cells with the accumulation of cholesterol in endolysosomes.This is consistent with previous work suggesting that peroxisomes are required for the flux of cholesterol from the late endosome to the ER [5].To investigate whether the lower steady-state cholesterol flux renders U2OS cells resistant to overall cellular responses to peroxisomal proteotoxic stress, we incubated U2OS cells with excess free cholesterol and assessed the impact of cholesterol overloading on the ISR activation at different time courses (2 h, 6 h and 24 h).Although confocal microscopy showed the lysosomal accumulation of Filipin-stained cholesterol at every time course upon silencing of LONP2 (Additional file 1: Fig. S7), LONP2-depletion did not induce the activation of CHOP in cholesterol loaded U2OS cells (Additional file 1: Fig. S8), suggesting that additional factors are required to activate the ISR.Taken together, our data establish a critical requirement for functional peroxisomes in the trafficking of cholesterol and lipid droplet biogenesis.
To exclude the possibility of off-target effects in our LONP2 silencing experiments, COS-7 cells were transfected with different siRNA targeting the 3′ UTR of LONP2 and then rescued by transient expression of siresistant LONP2 cDNA.Immunoblotting revealed that CHOP activation, as well as downregulation of INSIG1 and CRABP2 were fully rescued by the transient expression of LONP2 cDNA (Fig. 5f, left).Cholesterol homeostasis was also rescued functionally, as free cholesterol no longer accumulated at endolysosomes after transfection of LONP2 cDNA for 48 h (Fig. 5f, right).

Specificity of response to siLONP2 relative to loss of Pex5
Previous approaches to investigate peroxisomal stress pathways focused on loss of core peroxins essential for protein import in C. elegans, Drosophila and mammalian cells [25,30,37,53].LONP2 is a chaperone and protease that is important for the turnover and proteostasis within peroxisomes, and, as we demonstrated here, is required for the import of at least some luminal proteins.Therefore, we tested whether the pathways that are induced by LONP2 silencing are also triggered by the loss of core import proteins like PEX5 in our system.As with the loss of LONP2, PEX5-depleted COS-7, but not U2OS cells displayed clear upregulation of ISR effector CHOP, as well as downregulation of the ER cholesterol sensor INSIG1 (Additional file 1: Fig. S9).In contrast to the depletion of LONP2 however, PEX5 silencing failed to bolster expression of the ribosomal assembly protein RRS1 or retinoic acid CRABP2 protein levels (Additional file 1: Fig. S9).These intriguing differences suggest that luminal peroxisomal proteotoxic stress induces distinct responses relative to global peroxisomal protein import failure.

Discussion
Data presented here extend and identify new consequences of peroxisomal dysfunction, including induction of the ISR, disruption of cholesterol regulation and trafficking, and altered retinoic acid signalling.Our data highlight the significant variability in the dependence of different cell types on peroxisomal function.Importantly, our strategy to manipulate the peroxisomal protease/ chaperone LONP2 allowed us to maintain peroxisomal membrane protein import and thus avoid the direct induction of mitochondrial stress through the mitochondrial import of peroxisomal proteins [30].
Our data showed that loss of LONP2 led to changes in peroxisome morphology, where cells had fewer and larger organelles, suggesting that LONP2 function was directly or indirectly required for peroxisomal division.While we expected to observe the induction of pexophagy as part of a peroxisomal quality control response, we saw no induction of autophagy or changes in peroxisomal mass.This suggests that the consequence of LONP2 silencing on peroxisomal protein composition was relatively mild.For example, while we observed an accumulation of LONP2 substrate TYSND1, the functional impact of this was not yet seen on TYSND1 substrate ACOX1.In addition, while the acute transfection of CFP-SKL demonstrated a failure to import in LONP2 silenced cells, the import defects were not sufficient to reduce expression of the endogenous peroxisomal proteins we examined with the exception of the precursor of ACOX1.Our approach allowed us to capture the early stages of peroxisomal dysfunction sufficient to explore both common consequences of LONP2 loss and the dramatic cell-specific induction of the ISR.We did not observe a transcriptional induction of peroxisomal biogenesis (including PPAR genes), a repair mechanism seen in previous investigations into the peroxisome stress response pathways (Additional file 2: Table S1) [37].
Lipidomics, RNAseq and cell biological analysis presented here further establish peroxisomes as a critical contributor to cholesterol homeostasis and intracellular flux.This is consistent with recently published evidence that peroxisomes form direct contact sites with lysosomes to flux cholesterol to the ER [5].To this end, syn-aptotagmin7 (syt7) was identified as the essential tether between peroxisomes and lysosomes to drive cholesterol flux.We observed no apparent changes in syt7 mRNA levels (Additional file 2: Table S1).We thus suspect that the loss/aggregation of an unidentified peroxisomal luminal protein interfered with this pathway of cholesterol flux.The loss of INSIG1 was also seen upon loss of PEX5 in COS-7 cells, indicating that impaired cholesterol trafficking may occur upon diverse peroxisomal dysfunctions.Untargeted lipidomics also revealed an increase in mono-or un-saturated very long chain fatty acyl chains among the significantly changed lipids, suggesting peroxisomal dysfunction in beta-oxidation pathways upon depletion of LONP2.Interestingly, these changes were enriched within ceramide and sphingomyelin species in COS-7, but within ether lipids and glucocylceramides in U2OS.This is further indication that peroxisomes play important, yet differential roles in the wiring of lipid metabolism between the two cell lines.The impact of these elevated lipids on cell survival may be reflected in the increased caspase activation in COS-7, where ceramides are tightly linked to an enhanced apoptosis [28].The increased ether lipids may alter the susceptibility of U2OS cells to ferroptotic death pathways, as highlighted in recent studies linking peroxisomal function to the ferroptosis [57].Future work will continue to explore the contribution of peroxisomal lipid metabolism to cell survival pathways.
Another common pathway altered upon loss of LONP2 (but not PEX5) in both cell lines was a prominent  acid and cholesterol metabolism [41].Retinoic acid binds to the nuclear receptor RXR, which controls gene expression of a host of genes and can form dimers with up to 10 other nuclear hormone receptors including regulators of bile acid metabolism and lipid homeostasis [33].Of relevance to peroxisomes, knockout of RXRα in mouse hepatocytes leads to a sevenfold upregulation of a key regulator of bile acid metabolism CYP7A1, again linking retinoic acid signaling and peroxisomal function [51].Our data suggest a retrograde signaling pathway from peroxisomes into the RXR signaling pathway as well.Future work will explore the physiological, tissue specific regulation to unravel how and why the loss of LONP2 results in the downregulation RA signaling.
Additional responses to LONP2 depletion included the COS-7-specific induction of the ISR, and the increased mRNA transcripts linked to ribosome biogenesis.The ISR is a highly conserved stress response pathway, classical triggers of which include amino acid or heme deprivation,viral infection,and ER stress [32].ISR is seen as a pro-survival mechanism, however recent studies demonstrated the context-specific and time-sensitive role of the response in a variety of pathologies including models of neurodegenerative and mitochondrial diseases [6,20,44].Our study establishes peroxisomal proteotoxic stress as a new ISR trigger, providing a potential target for therapeutic interventions.COS-7 cells lacking LONP2 also demonstrated increased transcription of ribosome biogenesis messages that included rRNA, snoRNA and mRNAs.Interestingly, PEX5 loss did not lead to the increased expression of RRS1, a core regulator of ribosomal biogenesis, yet showed a clear activation of the ISR with CHOP expression.Another recent study showed significant downregulation of mRNA linked to ribosome biogenesis upon loss of peroxins [17], a trend also seen in our Pex5 depletion experiment.Therefore, the ribosome biogenesis response can be uncoupled from the peroxisome-driven activation of the ISR and appears to be specific to the loss of LONP2.
The acute silencing approach allowed us to observe the immediate response to the loss of a core peroxisomal chaperone/protease, avoiding the confounding effects of cellular adaptation in complete loss of function models.Our data suggest that the cell-specific responses reflect a differential dependence on peroxisomal functions in lipid handling between cell types and tissues.Peroxisomes exhibit profound functional plasticity depending on the tissue context, driving bile acid synthesis in the liver to the generation of ether lipids in oligodendrocytes, and scavenging ROS [52].It will be important to further explore the consequences of peroxisomal stress in tissues and disease states, where the study of dynamic lipid handling and flux may be studied in more physiological settings.Understanding the contributions of peroxisomes to cholesterol flux in liver during bile acid synthesis, or within development/cancer model systems where peroxisomes may impact retinoic acid metabolism, could open doors to new therapeutic targets in multiple disease settings from Zellwegers syndrome to metabolic disorders and cancer.

Adenovirus infection
C-terminal YFP-fused PEX3 was subcloned into D2-MCS viral vector (BioVector) under the control of a CMV promoter as described previously [45].Adenoviral PEX3-YFP was used to infect cells at 100 pfu per cell in the presence of 4 μg/ml polybrene (Sigma, H9268-10G) for 16 h.

Live cell imaging
Cells plated in a glass-bottom cell culture dish (MatTek) were incubated with 250 nM TMRE (Sigma) and 1 μM MitoSOX ™ Green (Invitrogen) for 30 min at 37 °C.Cells were washed in DMEM and observed in DMEM containing no phenol red (GIBCO) supplied with 10% FBS and 2 mM L-glutamine, NEAA and 10 mM HEPES pH 7.4 using a spinning disk confocal microscope (described above) with a 100 × objective and EMCCD camera.For quantification analysis, 30 cells in each condition were randomly chosen and examined if the MitoSOX signal was colocalized with mitochondria to calculate percentage of cells with MitoSOX-positive mitochondria.

RNAseq analysis
Total RNA was extracted from each cell using RNeasy Plus (QIAGEN) after 168 h siRNA transfection, and the quantity and quality was determined using Bioanalyzer (Agilent Technologies).Single read sequencing was performed with NextSeq 500 (Illumina) using 75 cycles High Output kit v2.Sequences were trimmed for sequencing adapters and low quality 3' bases using Trimmomatic version 0.35 [3] and aligned to the reference Chlorocebus sabaeus genome version ChlSab1.1 (gene annotation from Ensembl 104) or Homo sapiens genome version GRCh38 (gene annotation from Gencode version 37) using STAR version 2.7.1a [7].
Raw gene expression counts were obtained directly from STAR, and normalization and differential expression analysis were performed using DESeq2 version 1.30.1 [24].Metascape software was used for gene ontology analysis [56].We used unadjusted p-values for U2OS analysis as false discovery rate calculation has low power in datasets with small effect sizes (i.e.low fraction of changed genes in RNAseq) [22], and in our dataset Benjamini-Hochberg procedure resulted in over-represented p.adjusted values (Additional file 2: Table S1).For comparison analyses between COS-7 and U2OS, gene identifiers were converted from C. sabaeus to H. sapiens using the R/Bioconductor package biomaRt (version 2.48.3).

Puromycylation assay
Translation intensity was monitored by the incorporation of puromycin as previously described [43].After 7 days siRNA transfection, cells were pulsed with 10 μg/ml puromycin (Sigma, P8833) at 37 °C for 10 min.Cells were then harvested as described above and blots probed with anti-puromycin antibodies, as described above.

Retinoic acid stimulation
COS-7 cells were incubated in the presence or absence of 5 µM retinoic acid (Sigma, R2625) at 37 °C for 24 h before isolating total RNA for analysis by qRT-PCR.

Untargeted lipidomics
Lipid extraction, sample and data analysis were performed using a previously validated semi-quantitative untargeted lipidomic workflow [10].In brief, lipids were extracted from each cell after 168 h siRNA transfection and spiked with six internal standards: LPC 13:0, PC19:0/19:0, PC14:0/14:0, PS12:0/12:0, PG15:0/15:0, and PE17:0/17:0 (Avanti Polar Lipids).Protein concentration was determined using a colorimetric-based assay based on the Bradford dye-binding method and samples were injected into a 1290 Infinity high-resolution HPLC coupled with a 6530 Accurate Mass quadrupole time-of-flight (LC-QTOF) (Agilent Technologies) via a dual electrospray ionization (ESI) source.Elution of lipids was assessed on a Zorbax Eclipse plus column (Agilent Technologies) maintained at 40 °C using an 83 min chromatographic gradient of solvent A (0.2% formic acid and 10 mM ammonium formate in water) and B (0.2% formic acid and 5 mM ammonium formate in methanol/acetonitrile/methyl tert-butyl ether [MTBE],55:35:10 [v/v/v]).Data acquisition was performed in positive ionisation mode.All samples were processed and analyzed as a single batch.MS quality controls (QCs) were performed by (i) injecting 3 "in-house" QC samples and blanks at the beginning, middle and end of the run and (ii) monitoring six internal standards spiked in samples for signal intensity, mass mass-to-charge ratios (m/z) and retention time (RT) accuracies.Mass spectrometry (MS) raw data processing was achieved as previously described using Mass Hunter B.06.00 (Agilent Technologies) for peak picking and an in-house bioinformatic script [2018, Forest et al.] in both Perl and R languages that we developed for: (i) MS feature peak alignment and retention time (RT) correction; (ii) filter of presence: features retained must be present in 80% of samples from at least one group, thereby setting a maximum of 20 missing values within a group; (iii) normalization of signal intensities using cyclic loess algorithm; (iv) imputation of missing values to 90% of lower values.The resulting final dataset included 2,094 high-quality MS signals, thereafter referred to as features, defined by their m/z, RT and signal intensity (listed in Additional file 3: Table S2).Statistical analyses are performed using the final dataset on log2-transformed signal intensity data.An unsupervised principal component analysis was first applied followed by independent testing of each feature using or two-tailed unpaired Student's t-test with Benjamini-Hochberg correction for the indicated group comparison.Given the untargeted nature of our lipidomic analysis, only lipid features of interest have been annotated to unique lipids, a step that is crucial given that about 50% of all MS features are duplicate ions of the same lipid.For this study, lipid annotation was focused on the features that passed the selected threshold of significance, namely P corr -value (P value after Benjamini-Hochberg multiple testing correction) of < 0.05 and an absolute fold change (FC) │(FC)│ > 1.5 (corresponding to │log2 (FC)│ > 0.58) for the comparisons siLONP2 vs. siNT in COS-7 and U2OS.Lipid annotation was achieved as previously described in details (Forest et al.) by (i) MS/MS analysis for the 100 most significant lipid features in each comparison and (ii) through data alignment with our in-house database, which contains > 500 unique lipids with previously determined MS/ MS spectra for the remaining significant features.Advances in the relationship between regulator of ribosome synthesis 1 (RRS1) and diseases Front Cell Dev Biol 9 620925 17.Huang K, Kim J, Vo P, Miao T, Bai H. Peroxisome import stress impairs ribosome biogenesis and induces integrative stress response through eIF2α phosphorylation.bioRxiv.2021:2020-11.

(
See figure on next page.)Fig. 1 LONP2 knockdown triggered the accumulation of its substrate TYSND1, alterations in peroxisome size and numbers, and luminal protein import failure.a-d Immunoblotting (a, c) and representative confocal images and quantification of peroxisome size and numbers (b, d) in non-targeting (NT)-and LONP2-silenced COS-7 cells (a, b) and U2OS cells (c, d) after 144 h.TYSND1-F and TYSND1-C indicate full-length TYSND1 and the self-cleaved C-terminal region of TYSND1, respectively.Scale bar, 5 µm.e, f Representative confocal images (e) and quantification (f)

Fig. 2
Fig. 2 Cell-specific ISR activation upon silencing of LONP2.RNA sequencing (n = 3 independent experiments per condition), LONP2 silencing for 7 days.a Principal component (PC) score plot.b Volcano plots showing all identified transcripts.c Transcripts encoding integrated stress response (ISR) factors.d Immunoblotting for ISR-related proteins, vinculin was used as a loading control.e Translation assay: cells were incubated with puromycin for 10 min., and incorporation of puromycin into newly synthesized proteins was revealed with anti-puromycin antibodies.GAPDH was used as a loading control.f Transcripts encoding ribosomal biogenesis-related rRNA, snoRNA and mRNA genes.g Immunoblot of ribosomal biogenesis regulator RRS1 in NT-and LONP2-silenced COS-7 cells and U2OS cells (See figure on next page.)

Fig. 3
Fig. 3 LONP2 silencing leads to impaired retinoic acid signaling.a Gene ontology of RNA sequencing data (dataset is described in Fig. 2).Numbers on Venn diagram indicate the number of genes in each category.b Immunoblotting of retinoic acid regulating proteins, vinculin used as a loading control.c COS-7 or U2OS cells were incubated in the presence or absence of 5 µM retinoic acid for 24 h and mRNA transcript levels for LONP2 (left two panels) and TGM2 (right two panels) quantified with qRT-PCR.n = 3 independent experiments, means are depicted in different colours.The mean values were used to calculate the average (horizontal bar), s.d.(error bars) and P-values (using a two-tailed unpaired Student's t-test)

(
See figure on next page.)Fig. 5 LONP2 knockdown induced lipid storage, including aberrant cholesterol accumulation in lysosomes.a Representative confocal images (left) and quantification (right) of lipid droplets (Nile Red) in NT-and LONP2-silenced COS-7 cells after 6 days.Scale bar, 5 µm.b Representative confocal images showing the Filipin-stained intracellular cholesterol (green), Nile Red-stained lipid droplets (magenta) and endolysosomes (blue) visualized with the immunofluorescence of LAMP1.Scale bar, 2 µm.c, d Free cholesterol overload experiment in NT-and LONP2-silenced COS-7 cells (c) and U2OS cells (d) after 7 days.Representative confocal images (left) and quantifications (right) of the colocalization between cholesterol (Filipin) and endolysosomes (LAMP1).Scale bar, 2 µm.e Transmission electron microscopy micrographs of NT-and LONP2-silenced COS-7 cells.Scale bar, 500 nm.f Immunoblotting (left) and image quantification (right) of cholesterol overload experiment in NT-and LONP2-silenced COS-7 cells rescued with transient expression of siRNA-resistant LONP2 cDNA for 48 h.Data from n = 3 technically independent experiments, and their respective means are depicted in different colours.The mean values were used to calculate the average (horizontal bar), s.d.(error bars) and P-values (using a two-tailed unpaired Student's t-test) downregulation of the retinoic acid/Vitamin A pathway.Retinoic acid is a branched chain lipid that plays a key role in a plethora of both essential in development, and disease-related pathways [35].Historically, retinoic acid has been described as a peroxisome proliferator [14] and a possible link between peroxisomes and lipid droplets [39].Intriguingly, RA signaling pathway has been linked to peroxisomal function in the liver, specifically to bile

Fig. 5 (
Fig. 5 (See legend on previous page.) 18. MS Ioannou J Jackson SH Sheu CL Chang AV Weigel H Liu HA Pasolli CS Xu S Pang D Matthies HF Hess J Lippincott-Schwartz Z Liu 2019 Neuronastrocyte metabolic coupling protects against activity-induced fatty acid toxicity Cell 177 1522-1535 e1514 19.Y Kabeya N Mizushima T Ueno A Yamamoto T Kirisako T Noda E Kominami Y Ohsumi T Yoshimori 2000 LC3, a mammalian homologue of yeast Apg8p, is localized in autophagosome membranes after processing EMBO J 19 5720 5728 20.Kaspar S, Oertlin C, Szczepanowska K, Kukat A, Senft K, Lucas C, Brodesser S, Hatzoglou M, Larsson O, Topisirovic I, Trifunovic A.Adaptation to mitochondrial stress requires CHOP-directed tuning of ISR.Sci Adv.2021;7.21.CM Kassmann C Lappe-Siefke M Baes B Brugger A Mildner HB Werner O Natt T Michaelis M Prinz J Frahm KA Nave 2007 Axonal loss and neuroinflammation caused by peroxisome-deficient oligodendrocytes Nat Genet 39 969 976 22. M Krzywinski N Altman 2014 Points of significance: comparing samplespart I Nat Methods 11 215 216 23.MS Lindstrom J Bartek A Maya-Mendoza 2022 p53 at the crossroad of DNA replication and ribosome biogenesis stress pathways Cell Death Differ 29 972 982 24.MI Love W Huber S Anders 2014 Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 Genome Biol 15 550 25.FD Mast J Li MK Virk SC Hughes AJ Simmonds RA Rachubinski 2011 A Drosophila model for the Zellweger spectrum of peroxisome biogenesis disorders Dis Model Mech 4 659 672 26.ZA McGee CM Clemens RL Jensen JJ Klein LR Barley GL Gorby 1992 Local induction of tumor necrosis factor as a molecular mechanism of mucosal damage by gonococci Microb Pathog 12 333 341 27.M Neuspiel AC Schauss E Braschi R Zunino P Rippstein RA Rachubinski MA Andrade-Navarro HM McBride 2008 Cargo-selected transport from the mitochondria to peroxisomes is mediated by vesicular carriers Curr Biol 18 102 108 28.R Nganga N Oleinik B Ogretmen 2018 Mechanisms of ceramide-dependent cancer cell death Adv Cancer Res 140 1 25 29.C Ni M Buszczak 2023 The homeostatic regulation of ribosome biogenesis Semin Cell Dev Biol 136 13 26 30.E Nuebel JT Morgan S Fogarty JM Winter S Lettlova JA Berg YC Chen CU Kidwell JA Maschek KJ Clowers C Argyriou L Chen I Wittig JE Cox M Roh-Johnson N Braverman J Bonkowsky SP Gygi J Rutter 2021 The biochemical basis of mitochondrial dysfunction in Zellweger Spectrum Disorder EMBO Rep 22 e51991 31.K Okumoto Y Kametani Y Fujiki 2011 Two proteases, trypsin domain-containing 1 (Tysnd1) and peroxisomal lon protease (PsLon), cooperatively regulate fatty acid beta-oxidation in peroxisomal matrix J Biol Chem 286 44367 44379 32.K Pakos-Zebrucka I Koryga K Mnich M Ljujic A Samali AM Gorman 2016 The integrated stress response EMBO Rep 17 1374 1395 33.M Petkovich P Chambon 2022 Retinoic acid receptors at 35 years J Mol Endocrinol 69 T13 T24 34.SR Pfeffer 2019 NPC intracellular cholesterol transporter 1 (NPC1)-mediated cholesterol export from lysosomes J Biol Chem 294 1706 1709 35.E Pohl CWE Tomlinson 2020 Classical pathways of gene regulation by retinoids Methods Enzymol 637 151 173 36.LC Pomatto R Raynes KJ Davies 2017 The peroxisomal Lon protease LonP2 in aging and disease: functions and comparisons with mitochondrial Lon protease LonP1 Biol Rev Camb Philos Soc 92 739 753 37. E Rackles M Witting I Forne X Zhang J Zacherl S Schrott C Fischer JJ Ewbank C Osman A Imhof SG Rolland 2021 Reduced peroxisomal import triggers peroxisomal retrograde signaling Cell Rep 34 108653 38.A Radhakrishnan JL Goldstein JG McDonald MS Brown 2008 Switch-like control of SREBP-2 transport triggered by small changes in ER cholesterol: a delicate balance Cell Metab 8 512 521 39.WG Robison Jr T Kuwabara 1977 Vitamin A storage and peroxisomes in retinal pigment epithelium and liver Invest Ophthalmol Vis Sci 16 1110 1117 40.GL Russo G Sonsalla P Natarajan CT Breunig G Bulli J Merl-Pham S Schmitt J Giehrl-Schwab F Giesert M Jastroch H Zischka W Wurst SH Stricker SM Hauck G Masserdotti M Gotz 2021 CRISPR-mediated induction of neuronenriched mitochondrial proteins boosts direct glia-to-neuron conversion Cell Stem Cell 28 524-534 e527 41.A Saeed M Hoekstra MO Hoeke J Heegsma KN Faber 2017 The interrelationship between bile acid and vitamin A homeostasis Biochim Biophys Acta Mol Cell Biol Lipids 1862 496 512 42.G Sargent T Zutphen van T Shatseva L Zhang V Giovanni Di R Bandsma PK Kim 2016 PEX2 is the E3 ubiquitin ligase required for pexophagy during starvation J Cell Biol 214 677 690 43.EK Schmidt G Clavarino M Ceppi P Pierre 2009 SUnSET, a nonradioactive method to monitor protein synthesis Nat Methods 6 275 277 44.EL Spaulding TJ Hines P Bais ALD Tadenev R Schneider D Jewett B Pattavina SL Pratt KH Morelli MG Stum DP Hill C Gobet M Pipis MM Reilly MJ Jennings R Horvath Y Bai ME Shy B Alvarez-Castelao EM Schuman LP Bogdanik E Storkebaum RW Burgess 2021 The integrated stress response contributes to tRNA synthetase-associated peripheral neuropathy Science 373 1156 1161 45.A Sugiura S Mattie J Prudent HM McBride 2017 Newly born peroxisomes are a hybrid of mitochondrial and ER-derived pre-peroxisomes Nature 542 251 254 46.W Szaflarski M Lesniczak-Staszak M Sowinski S Ojha A Aulas D Dave S Malla P Anderson P Ivanov SM Lyons 2022 Early rRNA processing is a stress-dependent regulatory event whose inhibition maintains nucleolar integrity Nucleic Acids Res 50 1033 1051