A deeply conserved protease, acylamino acid-releasing enzyme (AARE), acts in ageing in Physcomitrella and Arabidopsis

Reactive oxygen species (ROS) are constant by-products of aerobic life. In excess, ROS lead to cytotoxic protein aggregates, which are a hallmark of ageing in animals and linked to age-related pathologies in humans. Acylamino acid-releasing enzymes (AARE) are bifunctional serine proteases, acting on oxidized proteins. AARE are found in all domains of life, albeit under different names, such as acylpeptide hydrolase (APEH/ACPH), acylaminoacyl peptidase (AAP), or oxidized protein hydrolase (OPH). In humans, AARE malfunction is associated with age-related pathologies, while their function in plants is less clear. Here, we provide a detailed analysis of AARE genes in the plant lineage and an in-depth analysis of AARE localization and function in the moss Physcomitrella and the angiosperm Arabidopsis. AARE loss-of-function mutants have not been described for any organism so far. We generated and analysed such mutants and describe a connection between AARE function, aggregation of oxidized proteins and plant ageing, including accelerated developmental progression and reduced life span. Our findings complement similar findings in animals and humans, and suggest a unified concept of ageing may exist in different life forms.

R eactive oxygen species (ROS) are by-products of O 2 metabolism and represent a challenge to all aerobic life. ROS play a dual role as potentially lethal oxidants and as signaling molecules 1 . Therefore, aerobic organisms possess sophisticated redox systems to scavenge and detoxify excess ROS. The major sources of ROS in plant cells are the electron transport chains in mitochondria and plastids, but also in peroxisomes and at the plasma membrane 2 . Environmental stresses such as heat, drought, or intense light are factors for increasing ROS production to detrimental levels. Plants possess a repertoire of detoxifying enzymes such as catalases, superoxide dismutases, ascorbate peroxidases, glutathione peroxidase-like proteins, and peroxiredoxins. Electrons for reduction are largely provided via non-enzymatic components such as ascorbic acid, glutathione, and NADPH [3][4][5][6][7] . In addition, a range of heat-shock proteins assist in disaggregation or refolding of damaged proteins [8][9][10] .
Despite conversion into non-toxic derivatives, the continuous exposure to ROS results in oxidation of DNA, lipids, and proteins 5 . On the protein level, ROS lead to irreversible cysteine oxidation, advanced glycation end products, adducts with ROSgenerated reactive aldehydes, and carbonylation of amino acid side-chains 11,12 . If not cleared via proteolysis, an excess of oxidized proteins accumulates to cytotoxic protein aggregates. Plant antioxidant systems and the role of ROS as signaling molecules in abiotic stress responses are well studied 13 . Yet, factors involved in a plant cell's last line of defense, such as the proteolytic systems for the clearance of irreversibly oxidized proteins, are still underexplored.
In humans, AARE malfunction is linked to different types of cancer 26,28,29 and sarcoma cell viability 30 . Moreover, AARE and proteasomal activity correlate and cooperatively prevent cytotoxic aggregate formation [31][32][33] . Several selective inhibitors have been identified 34,35 and blocking of AARE function is considered as anti-cancer treatment 28 . Despite an increasing number of studies on AARE functionality in humans, AAREs in plants are far less characterized.
AARE from Arabidopsis thaliana (AtAARE) and from cucumber have endo-and exopeptidase functions 17 , and silencing of AtAARE increased the levels of oxidized proteins 21 . AtAARE activity was also detected in plastid stroma fractions, although a fusion with a fluorescent reporter did not co-localize with chloroplasts 21 . Suppression of AtAARE via RNAi resulted in an enhanced accumulation of oxidized proteins in roots and enhanced electrolyte leakage in leaves, but a further impact on plant physiology was not described 21 .
Moreover, the complete loss of function of this protease has not yet been reported for any organism, neither bacterium, archaeon, animal, or plant. Thus, although the deep evolutionary conservation of AARE suggests its pivotal role in all major life forms, experimental evidence is far from optimal.
In a proteomics study on protein arginylation we identified an AARE homolog from the moss Physcomitrella 36 . Our further analysis revealed altogether three Physcomitrella AARE homologs (PpAARE1-3). Here, we analyzed the subcellular localization of these PpAARE isoforms and of their homolog from the angiosperm Arabidopsis (AtAARE). We show that an alternative splicing event is targeting PpAARE1 to chloroplasts, mitochondria, and the cytosol. We provide evidence that an alternative translation initiation is sufficient to localize AtAARE to the same three subcellular compartments. Bioinformatic analyses of several genomes suggest that the localization of AARE in chloroplasts and mitochondria is conserved across the plant lineage. Employing combinatorial gene knockouts and protein coimmunoprecipitation we found distinct interactions between these three isoforms and their concerted action on progressive ageing in Physcomitrella. Likewise, an Arabidopsis AARE loss-offunction mutant exhibits enhanced levels of oxidized proteins and accelerated bolting, as a hallmark of plant ageing.
To investigate whether other plants also possess multiple AARE homologs and to infer their phylogenetic relation, we conducted BLASTP 44 searches against protein models from selected species using the protein sequence of AtAARE as a query. We selected the alga Chlamydomonas reinhardtii 45 50 , and identified a single homolog (GAQ80280.1) in this species. We also included proteins of Funaria hygrometrica 51 , a close relative to Physcomitrella from the Funariaceae family 52 , in our search. Finally, the AtAARE sequence was again searched against the Physcomitrella 39 and Arabidopsis 53 proteomes. Homology of the resulting BLAST hits was confirmed if the reciprocal best BLAST hit against A. thaliana was again AtAARE.
In P. trichocarpa, we identified a single homolog for which three distinct splice variants are annotated (Potri.008G160400.1, Potri.008G160400.2, Potri.008G160400.3). These encode different protein isoforms, but two variants seem to encode non-catalytic proteins. AARE enzymes are prolyl-oligopeptidases with a conserved catalytic triad (Ser/Asp/His) in the C-terminal peptidase domain 54,55 . In splice variant 2 (Fig. S2, Potri.008G160400.2) alternative splicing results in the deletion of the catalytic Asp whereas the whole catalytic triad is lacking in splice variant 3 (Potri.008G160400.3). Hence, we consider these splice variants as non-active and disregard them from further discussion.
In  56 . We also did not find any proteomics evidence across several Physcomitrella analyses 36,57-59 for this protein. Therefore, we excluded this gene from further analysis.
We then used phylogenetic reconstruction to investigate the origin of gene duplications within the gene family. As an outgroup, we included the well-characterized rat 16 and human 26 AARE and two isoforms of the Antarctic icefish Chionodraco hamatus 27 . Physcomitrella and F. hygrometrica share three distinct pairs of orthologs hinting at an expansion in the common ancestor of the two species. Our phylogenetic analysis did not resolve AARE subfamilies across kingdoms (Fig. 1a) and we conclude that the gene family expansions observed in rice and in the Funariaceae are lineage-specific events.
In addition, this analysis reveals a closer relationship between PpAARE1 and PpAARE3, which presumably originate from a more recent gene duplication event, compared to PpAARE2. This is supported by the fact that the open reading frames (ORFs) of PpAARE1 and PpAARE3 are represented by a single exon whereas the ORF of PpAARE2 is split across 17 exons, similar to AtAARE (Fig. 1b). This is in congruence with a more recent emergence of intron-poor genes in intron-rich families linked to stress response and developmental processes 60 and in line with intron-less orphan Physcomitrella genes as earliest responders to abiotic stress 61 .
For the three PpAARE genes, several splice variants are annotated, but only two splice variants of PpAARE1 give rise to distinct protein isoforms ( Fig. 1b; PpAARE1_1, PpAARE1_2). Both splice variants are present in Physcomitrella protonema (Fig. 1b, c), according to RT-PCR with splice variant-specific primers (Supplementary Data S1). Likewise, for AtAARE two different ORF definitions exist. With Araport11 62 , a new version of the gene model was introduced exhibiting a longer ORF at the 5′ end (Fig. 1b). We detected the full-length transcript via reverse transcription polymerase chain reaction (RT-PCR, Fig. 1c).
For PpAARE1, alternative splicing in the 5′ end results in an N-terminal truncated variant whereas the longer non-spliced variant encodes an N-terminal plastid transit peptide (cTP) according to TargetP2.0 63 . A cleavage of the transit peptide at the predicted cleavage site (Ala 72 -M 73 , Supplementary Data S2) of PpAARE1 would release exactly the protein encoded by the short splice variant. In contrast, PpAARE2 and PpAARE3 do not harbor any predicted N-terminal targeting signals. Moreover, PpAARE3 is also lacking the WD40 domain that is present in PpAARE1 and PpAARE2 (Fig. 1d).
The extension of the originally annotated ORF of AtAARE also encodes a plastid transit peptide (Fig. 1d). To our knowledge, the longer variant of AtAARE has not yet been investigated, whereas the short variant of AtAARE localizes to the nucleus and the cytosol 21 . In agreement with the latter findings, we could predict a nuclear localization sequence (NLS, KKKK) with LOCALIZER 64 . Thus, targeting of AtAARE to the cytosol and the nucleus, but also to plastids could be enabled by alternative translation initiation. Likewise, PtAARE harbors a plastid transit peptide and a potential alternative translation initiation site downstream of the predicted cTP cleavage site.
Accordingly, we checked for NLS in PpAARE isoforms and found one (KRRP, Supplementary Data S2) in PpAARE1 and PpAARE3, whereas PpAARE2 has none, further supporting our hypothesis that PpAARE1 and PpAARE3 originate from a relatively recent gene duplication event. Accordingly, alternative splicing also generates two distinct transcripts for AARE1 in rice (OsAARE1, Supplementary Data S2), where one variant encodes a potential plastid transit peptide.
For all other plant species, no plastid targeting sequence was predicted, while the C. reinhardtii AARE harbors a mitochondrial targeting sequence (Supplementary Data S2).
PpAARE1 and AtAARE in mitochondria, chloroplasts, and cytoplasm. Organellar targeting of AARE has not yet been reported, although AARE activity was observed in plastidenriched fractions of cucumber 17 . However, chloroplasts, peroxisomes and mitochondria are major hubs of ROS generation 65 , and thus are likely organelles with elevated levels of oxidized proteins. Thus, we investigated whether PpAARE1 and AtAARE would localize to chloroplasts in vivo.
We generated fusion constructs of the PpAARE isoforms and of AtAARE with eGFP for transient expression in Physcomitrella protoplasts. Due to the presence of a predicted plastid targeting peptide for PpAARE1, eGFP was fused in frame to the 3' end of all coding sequences (CDS). Since also peroxisomes are ROSproducing organelles, we used PlantPredPTS1 66,67 to check for the presence of C-terminal positioned peroxisomal targeting signals. None of the selected AARE isoforms were predicted to localize to peroxisomes (Supplementary Data S2). Although AtAARE has a C-terminal CKL tripeptide, which is experimentally verified to mediate peroxisomal targeting 66,68 , the properties of its other C-terminal amino acids most likely prevent peroxisomal targeting. A more recent prediction approach for PTS1-mediated targeting for Arabidopsis proteins 69 further supports this conclusion. Peroxisomal targeting can also be mediated via N-terminal nona-peptides 70,71 , but these motifs are also not present within the first 100 aa in any of the selected AARE sequences. In agreement with these predictions eGFP was fused to the 3′ end of the CDSs.
For PpAARE1 three different fusion constructs were assembled. Among these, a fusion of the CDS of the short splice variant (Pp3c2_30720V3.1) and eGFP was cloned, as well as a fusion of the CDS of the longer splice variant (Pp3c2_30720V3.2) and eGFP. Additionally, we cloned a fusion of eGFP and the sequence part in which both variants differ (M 1 -A 72 , Pp3c2_30720V3.2). This part harbors the plastid transit peptide predicted by TargetP2.0. All fusion constructs were expressed under the control of the Physcomitrella Actin5 promoter 57,72 in a pMAV4 plasmid backbone 73 .
The PpAARE1 isoform derived from the short splicing variant (PpAARE1_1, Fig. 1b) clearly localized to the cytoplasm (Fig. 2). The same localization was observed for PpAARE2 and PpAARE3 (Fig. 2). Despite a predicted NLS, we did not observe a nuclear localization, either for PpAARE1 or for PpAARE3.
The isoform encoded by the longer splice variant of PpAARE1 (PpAARE1_2, Fig. 1b) localized to chloroplasts and surprisingly also to mitochondria (Fig. 2). In contrast to the diffuse cytosolic distribution of PpAARE1, specific foci were observed in chloroplasts. To investigate whether the N-terminal sequence differing between the two PpAARE1 variants (M 1 -A 72 , Pp3c2_30720V3.2) is sufficient to confer dual targeting, we fused this N-terminal sequence 5' to eGFP and observed again dual localization (Fig. 2, PpAARE1_Nt). Full-length PpAARE1 was necessary to localize eGFP to foci within chloroplasts, as the PpAARE1_Nt:eGFP fusion led to a uniform distribution. However, full-length PpAARE1 was also homogeneously distributed throughout the cytoplasm. This indicates the presence of interactors that recruit PpAARE1 to specific sites or complexes within the chloroplasts. Further, we conclude that the N-terminal extension of PpAARE1_2 encodes an ambiguous targeting signal for import into chloroplasts and mitochondria as it is capable of directing the fusion protein simultaneously to both organelles. Based on this data, PpAARE1 is targeted to three organelles: chloroplasts, mitochondria and the cytosol. To independently scrutinize these findings, we measured AARE enzyme activity in the organelles. Using organelle fractionation as previously described 74 , we detected AARE activity in chloroplasts, the cytosol, and mitochondria, although to a lesser extent in the latter (Fig. S3). Thus, in vivo localization of fusion proteins and enzyme measurements after cell fractionation independently confirm the predicted triple localization of AARE in Physcomitrella.
Simultaneous localization of proteins to chloroplasts and mitochondria can be mediated via ambiguous targeting signals which are recognized by both translocation machineries. We evaluated whether ATP2 75 , a tool for the prediction of ambiguous targeting, would recognize PpAARE1 but this was not predicted to be dually targeted. In contrast, AtAARE was predicted to be dually targeted via an ambiguous signal. Thus, we cloned the analogous three fusion constructs for AtAARE and investigated the subcellular localization of their encoded proteins accordingly.
The AtAARE isoform translated from the shorter ORF (AtAARE, Fig. 1b) localized to the cytoplasm (AtAARE_SV, Fig. 2), as observed for PpAARE1_1. This result is partially in agreement with Nakai et al. 21 since we could not observe nuclear localization. Using the fusion construct of the longer AtAARE variant, we observed clear dual targeting of the protein to chloroplasts and mitochondria (AtAARE_LV, Fig. 2), as observed for PpAARE1_2. Here, the eGFP signal was distributed homogeneously in the chloroplasts, in contrast to the foci of PpAARE1:eGFP. Next, we cloned only the N-terminal sequence differing between both variants (M 1 -A 55 , longer ORF definition, Fig. 1b) and fused it to eGFP. In order to investigate whether the exact N-terminal difference between the two AtAARE variants would be sufficient for targeting, the M 56 (same as M 1 in the shorter variant), which is the P1 aa at the predicted cleavage site, was deleted. Using this construct, the eGFP signal localized to chloroplasts and mitochondria (AtAARE_Nt, Fig. 2). The signal within chloroplasts was homogeneously distributed, similar to the longer AtAARE variant. Thus, we conclude that the N-terminal extension of both long PpAARE and AtAARE variants is sufficient for dual targeting of proteins in vivo. Intriguingly, the longer variant of AtAARE localized exclusively to chloroplasts and mitochondria although alternative translation initiation should be possible. This is interesting as alternative translation initiation is also possible in the longer splice variant of PpAARE1 (PpAARE1_2). In the latter also, the fusion protein localizes exclusively to chloroplasts and mitochondria, which excludes the possibility of an alternative translation initiation, at least in protoplasts. There are numerous transcripts in mammals where translation of an upstream positioned ORF suppresses the translation of the downstream main ORF 76 . A similar scenario is conceivable in Physcomitrella. However, it remains unclear how and if translation from the internal start codons is controlled. It is also possible that factors controlling alternative translation initiation of AtAARE are absent in Physcomitrella, at least in a spatiotemporal manner, or they might only be triggered in specific physiological situations. According to our data, the translation of the two variants of PpAARE1 is mainly controlled by alternative splicing and not by alternative translation initiation.
In summary, PpAARE1 and AtAARE localize to three subcellular compartments via an ambiguous targeting signal. In contrast, PpAARE2 and PpAARE3 localize solely to the cytoplasm. Double knockout of PpAARE1/2 reduces lifespan. Null mutants of AARE have not been described in any organism, and the biological role of this protease apart from its catalytic function remained unclear. Hence, we created AARE knockouts in Physcomitrella by deleting the full CDS of each gene via gene targeting ( Fig. 3a) according to 77 . To enable subsequent knockout (KO) of other AARE genes, different selection markers were used. Since three distinct AARE genes exist in Physcomitrella which result in cytosolic proteases, we generated all possible combinations of double KOs and triple KOs to avoid potential compensation of the loss of function.
Plants surviving the selection procedure were screened via PCR for the absence of the respective CDS and correct integration of the KO construct in the target locus (Fig. S4a). Additionally, the number of genomic integrations of the KO construct was measured via quantitative PCR (qPCR) as described 78 (Fig. S4b). At least three independent lines were identified for all single, double and triple KOs ( Fig. S4c-i) and a line with only a single integration in the genome was detected for each KO (j-p, line numbers with stars). Further, haploidy of all lines was confirmed via flow cytometry (Fig. S5a-c) as described 79 . These precautions were made as the transformation procedure may generate plants with multiple integrations 80 , possibly leading to off-target effects. Further, the transformation procedure may lead to diploid plants with altered gene expression 81 . We used haploid lines with a single integration of the KO construct (Fig. S4) for subsequent experiments.
Typically, AARE exopeptidase activity is assayed via N αacetylated amino acids like AcAla or AcMet coupled to a reporter such as para-nitro-anilide (pNA) or 7-amido-4-methylcoumarin Fig. 3 Phenotype analyses, AARE activity and level of oxidized proteins in Physcomitrella WT and KO mutants. a Each isoform was knocked out using a different selection marker. The selection markers confer resistance against G418 (npt), hygromycin (hpt) and zeocin (zeo), respectively. Homologous regions (HR) were chosen in order to remove the full CDS of the corresponding gene upon integration of the selection marker. b Mean AARE activity on AcAla-pNA in Physcomitrella protonema with standard deviation (n = three biological replicates). Significance levels are based on a one-way Anova and subsequent post-hoc test (***p < 0.001) and indicate differences compared to WT. c Phenotypes of gametophores of WT and the different KOs cultivated on KnopME for 4 months. All colonies were grown on the same plate. Upper panel: Bar = 0.5 mm; lower panel: Bar = 2 mm. d Protonema colonies grown on KnopME with 0.02% glucose taken from suspension culture. Bar = 1 mm. Images were taken 9 days after spotting. e Box plot showing the distribution of colony areas (n = 12 colonies per box plot; colony images taken 7 days after spotting). No significant difference of colony areas between WT and the KOs was observed (tested via one-way Anova). f Box plot showing the number of gametophores per colony and day of WT and ΔPpAARE1/2. Counting was performed at indicated days after spotting protonema suspension culture on solid medium (KnopME with 0.02% glucose). Indicated significance levels are based on a two-way Anova with subsequent post-hoc test (***p < 0.001). The box plots (e, f) depict the mean (horizontal bold line) of the data, the interquartile range (box) and the 1.5x interquartile range (whiskers). g Mean AARE exopeptidase activity on AcAla-pNA in gametophores. Depicted is the mean of three independent colonies per line with standard deviations. All colonies were cultivated on the same plate (Fig. S7a). Significance levels are based on one-way Anova and indicate differences compared to WT (***p < 0.001). h Levels of oxidized proteins in gametophores of WT and the different KOs (Coomassie-stained loading control is shown in Fig. S7b). The analysis was repeated three times (additional blots available in Fig. S7c, d). Detection of oxidized proteins was performed with the OxyBlot TM Protein Oxidation Detection Kit (Merck).
(AMC). From these, AcAla-pNA was tested for several eukaryotic AAREs 22,26 , including AtAARE 17 . Here, we analyzed the impact of AARE loss of function on the activity towards AcAla-pNA and AcLeu-pNA. The latter is a substrate of bacterial and archaeal AARE isoforms but also eukaryotic isoforms exhibit cleavage activity on this substrate 16,[22][23][24][25] . On the single KO level, the exopeptidase activity on both substrates was significantly reduced in the ΔPpAARE1 mutant whereas the single KO of the other isoforms did not affect the activity (Fig. 3b). This strong impact of PpAARE1 on the exopeptidase activity was consistent across all transgenic mutant lines (Fig. S6).
One important step in Physcomitrella development is the transition from protonema to gametophores, a developmental progression regulated among others by plant age and the nutrient status 82 . The single KOs were phenotypically inconspicuous on the gametophore level (Fig. 3c). In contrast, gametophores of ΔPpAARE1/2 and ΔPpAARE1/3/2 were severely stunted and colonies were denser compared to wild type (WT) or the other mutants (Fig. 3c). This growth effect is restricted to gametophores since protonema growth on solid medium did not differ between WT and the KOs (Fig. 3d, e). Intriguingly, ΔPpAARE1/2 and ΔPpAARE1/3/2 mutants showed accelerated developmental transition, as they developed gametophores from protonema earlier (Fig. 3d). Since other KO lines did not show this effect, we attributed this to the double KO of PpAARE1/2 and performed a quantitative comparison with WT. Here, gametophores were already observed after 6 days in the double KO of PpAARE1/2, while in WT a similar number of gametophores per colony was observed only after 13 days (Fig. 3f). Consequently, the double KO of PpAARE1/2 causes accelerated developmental progression but gametophores remained ultimately smaller than in WT (Fig. 3c). These effects are not linked to AARE exopeptidase activity since the exopeptidase activity in gametophores was significantly reduced in all lines with a KO of PpAARE1 (Fig. 3g), which mimics the activity profile in protonema (Fig. S6).
To analyze AARE endopeptidase activity, we assessed the total levels of oxidized proteins in gametophores. In this assay, protein carbonyl groups derived from oxidation are derivatized with 2,4dinitrophenylhydrazine (DNPH) to 2,4-dinitrophenylhydrazone (DNP). This irreversible modification is then recognized on Western blots by a primary anti-DNP antibody. Since DNPH can also react with oxidation states of cysteine side chains 83 , this assay detects not only protein carbonylation but general protein oxidation.
With this assay we found that PpAARE2 had the strongest impact on the level of oxidized proteins in gametophores (Fig. 3h) and thus is not linked to exopeptidase activity. This was consistently observed in three independent analyses (Fig. 3h,  Fig. S7). Apparently, PpAARE3 does not have any impact on exopeptidase and endopeptidase activity in gametophores under standard conditions (Fig. 3g, h).
Taken together, PpAARE1 predominantly acts as exopeptidase, while PpAARE2 predominantly acts as endopeptidase, and only the simultaneous loss of both activities in the double knockout mutants has the severest phenotypical consequences.
We found another remarkable difference between WT and mutants with a double KO of PpAARE1/2 in older plants. After 5 months of cultivation, ΔPpAARE1/2 and ΔPpAARE1/3/2 were only viable at the tip of the gametophores (Fig. 4a), whilst most of the colony was already dead. In contrast, gametophores of WT and the other KOs were fully viable. After 8 months, ΔPpAARE1/ 2 and ΔPpAARE1/3/2 were already dead, in contrast to WT and the respective parental lines, which only showed some dead gametophores (Fig. 4b, c).
In summary, mutants with a double KO of PpAARE1/2 exhibit accelerated developmental transition from protonema to gametophore (Fig. 3f), while size and life span of gametophores is strikingly reduced (Figs. 3c and 4a-c). In contrast, these effects are not visible in ΔPpAARE1/3. Therefore, these ageing phenotypes are linked to the concurrent loss of major AARE endopeptidase and exopeptidase activity.
Distinct in vivo interactions of PpAARE isoforms. In different organisms, AARE forms different homomeric complexes such as dimers 14 , tetramers 17 , or hexamers 24 . Thus, we analyzed whether the PpAARE isoforms can interact with each other. Previously, all three isoforms were identified, although the protein modification used for pulldown (N-terminal arginylation) was only identified on PpAARE1 36 . This gave rise to two hypotheses: First, PpAARE2 and PpAARE3 are also targets for N-terminal arginylation, but modified peptides were not identified for these isoforms. Second, the isoforms interact in complexes which were pulled down due to the N-terminal arginylation of PpAARE1. We generated Citrine fusion lines for each isoform via in-frame tagging at the native locus (knock-in, Fig. 5a). The original stop codons of the respective PpAARE CDS were deleted. The Citrinetag was intended for two different analyses: First, it should enable in vivo monitoring of PpAARE isoforms expressed from the native promoter, and second, it is a valid tag for coimmunoprecipitation (Co-IP) via commercially available trapbeads.
In plants with a detectable fusion transcript (Fig. S8a-c) the presence of the target protein was checked via IP and subsequent MS analysis. For PpAARE3:Citrine lines, we detected transcripts in one line and obtained only insufficient coverage and intensity at the MS across several Co-IPs. Thus, these Physcomitrella lines were excluded from further analysis. The co-precipitation of other PpAARE isoforms with the respective bait isoforms was observed in test IPs (Fig. S8d, e) confirming previous MS-Data 36 . All plants harboring the Citrine fusion were phenotypically inconspicuous (Fig. S8f) and haploid (Fig. S8g). Although the fusion proteins were detected in two independent lines for each of the two isoforms (PpAARE1, PpAARE2), we could not observe any Citrine signal within Physcomitrella protonemata or gametophores, probably due to the low abundance of the PpAARE isoforms. Nevertheless, MS intensities and sequence coverage enabled quantitative Co-IPs. The MS data have been deposited in PRIDE 84,85 with the accession codes PXD033854 and PXD038742.
When targeting PpAARE1:Citrine, both other isoforms appeared as significant interacting partners (Fig. 5b, p < 0.01, FDR = 0.01). In a reciprocal Co-IP targeting PpAARE2:Citrine only, PpAARE1 appeared as significant interacting partner. PpAARE3 was not detected in this pulldown. Although lacking a reciprocal IP targeting PpAARE3:Citrine, the data show that PpAARE1 interacts with PpAARE2 and PpAARE3, whereas PpAARE2 only interacts with PpAARE1. Consequently, there are distinct interactions of PpAARE1 with PpAARE2 and PpAARE3 in vivo, possibly resulting in cytosolic heteromeric AARE complexes in Physcomitrella.
AARE affects bolting time in Arabidopsis. In Physcomitrella three AARE genes exist, and the concerted action and interaction of the enzymes affect plant ageing. To evaluate if this is an evolutionary conserved function, we analyzed the situation in Arabidopsis. Here, it was known that silencing of the single AARE gene leads to an accumulation of oxidized proteins, whereas overexpression did not affect their levels 21 . To gain deeper insights, we screened for available Arabidopsis T-DNA mutants.
We identified two T-DNA insertion lines (SALK_071125 and GK-456A10) in the T-DNA Express database at the SIGnAL website (http://signal.salk.edu). SALK_071125 (s68) has a T-DNA insertion in an intron at the 5′ UTR, whereas the T-DNA insertion in GK-456A10 (GK) maps to an intron in the region encoding the catalytic domain (Fig. 6a). We obtained these mutants, identified homozygous populations of GK-456A410 containing the T-DNA insertion by their resistance to sulfadiazine, and confirmed their genotype by PCR. In the case of the Salk line (s68), homozygous plants had lost their resistance to kanamycin, but we confirmed their genotype by PCR.   (Fig. 6a), although very reduced in s68, while the protein was not detectable via Western blot (Fig. 6b). Surprisingly, in WT AARE was detected at around 100 kDa, although the estimated molecular weight is~84 kDa (90 kDa for the longer ORF variant without cleavage of the plastid targeting peptide) which indicates the presence of posttranslational modifications. Phenotypically, neither seedlings nor adult mutant plants showed obvious deviations from WT (Col, Fig. S9).
Next, we assayed AARE exopeptidase function in Arabidopsis WT and the two mutants. The exopeptidase activity on AcAla-pNA was significantly reduced in both T-DNA mutants. In contrast, the activity on AcLeu-pNA did not change significantly, although a slight reduction was observed in the GK line (Fig. 6c). Thus, in agreement with Western blot and activity data, the remaining transcripts detected via RT-PCR (Fig. 6a) are not translated to functional enzymes, suggesting complete loss of AARE function in the mutants. Based on this characterization, we concentrated on T-DNA mutant s68.
Subsequently, we assayed the AARE endopeptidase activity in Arabidopsis. The most striking feature reported were increased levels of oxidized proteins in AARE-silenced Arabidopsis plants 21 which is in line with our findings in Physcomitrella AARE mutants. To corroborate this, we investigated levels of oxidized proteins in the Arabidopsis T-DNA mutant (s68) in comparison to WT (Col) cultivated under short day conditions. In both genotypes, the distribution of oxidized proteins was different to Physcomitrella gametophores (Fig. 6d). Using the same extraction protocol, oxidized proteins in Arabidopsis were mainly of lower molecular weight, whereas in Physcomitrella they were mainly at higher molecular weight (Fig. 3h). Despite these size differences, the level of oxidized proteins was higher in the Arabidopsis mutant than in WT (Fig. 6d), mimicking the situation in Physcomitrella. Together, we found AtAARE exopeptidase and endopeptidase activity to be reduced in our experiments.
Oxidized proteins accumulate during vegetative growth of Arabidopsis and are depleted at the transition to bolting 86 . It was not clear from this study whether the level of oxidized proteins is a signal for bolting, or if the reset of protein oxidation at the bolting transition is a side-effect of enhanced protein turnover during this process. To address this, we assessed the bolting behavior of Arabidopsis WT and mutant plants and found that bolting in mutant plants differed significantly from WT. In WT, bolting started at 24 ± 1 rosette leaves whereas it started at 21 ± 2 rosette leaves in the mutants (Fig. 6e). Consequently, this accelerated developmental transition in the mutants correlates with enhanced levels of oxidized proteins.

Discussion
A universal definition of ageing is difficult due to strongly differing characteristics between and within the domains of life. In humans, the term ageing is inherently linked to degenerative diseases whereas plant ageing is considered as the progression of developmental transitions from seeds to senescence. Nevertheless, a common feature of ageing is the execution of a genetic program that controls growth, development and maturation. In turn, the progression of this genetic program depends on the metabolic state and environmental factors. Among the molecules responsible for ageing are ROS. Their involvement in ageing and diseases was first postulated by Harman 87 and was extended by considering mitochondria as central sources 88 . In humans, mitochondrial malfunction in line with increased ROS levels is a central determinant of ageing and associated pathologies 89 . In plants, mitochondrial ROS increase during ageing of seeds 90 and are major determinants of germination capacity 91 . In photosynthetic tissues chloroplasts are the major source of ROS and their levels increase during ageing [92][93][94] . Plants cannot escape from stress situations that increase ROS production to detrimental levels, and despite several layers of ROS defense, oxidized proteins constitute the major share of modified molecules under stress. Consequently, protein oxidation and subsequent aggregate deposition are hallmarks of ageing 95,96 . The degradation of artificially oxidized proteins by AARE has been demonstrated repeatedly, and silencing of AtAARE in turn increased the levels of oxidized proteins [19][20][21][22] .
However, the contribution of AARE to the progress of ageing remained elusive, although several studies associate AARE function to age-related diseases [97][98][99] . Here, we identified three AARE genes in Physcomitrella. This gene family expansion is an outstanding feature of the moss family Funariaceae, as most organisms we analyzed contain only a single AARE gene, and only a few have two. We analyzed these three isoforms in Physcomitrella and compared selected features to the single isoform in Arabidopsis. Our data reveal specific functions in age-related developmental transitions and life span determination.
Surprisingly, we found triple localization of AtAARE and PpAARE1 to chloroplasts, mitochondria and the cytosol in vivo, suggesting a functional role for AARE in these cellular compartments. In Physcomitrella, the triple localization is mediated via alternative splice variants and in Arabidopsis likely via alternative translation initiation. Although there have been indications of AtAARE being associated to chloroplasts 17,21 , there was no clear evidence for organellar targeting of this protease until now. It is remarkable that the triple localization of AARE is evolutionary conserved between Arabidopsis and Physcomitrella, although likely executed via different molecular mechanisms. This suggests an essential and evolutionary conserved function of AARE activity in the cytoplasm, chloroplasts and mitochondria. As mosses and seed plants diverged more than 500 million years ago 100 , this is a deep evolutionary conservation.
Previously, AARE exopeptidase activity was observed in cucumber chloroplasts 17 , and AARE peptides were found in proteomes of Arabidopsis chloroplasts 101 and mitochondria 102 . In contrast, AARE was not identified in plastid or mitochondrial proteomes of Physcomitrella 57,59 . We found AARE exopeptidase activity in chloroplasts and mitochondria of Physcomitrella. Chloroplasts are a major source of ROS in photosynthetic tissues exposed to light, whereas mitochondria are the major source of ROS in non-photosynthetic tissues or in the dark 93,94,103,104 . Until now, it remained unresolved how plants deplete oxidized proteins from these organelles.
In yeast and mammals, the ATP-dependent LON and AAA proteases are involved in clearance of misfolded and oxidized mitochondrial proteins. Intriguingly, mutants of plant homologs of LON proteases did not show clear effects on the levels of oxidized proteins, but AAA-type FTSH proteases may play a role 105 . Nevertheless, stressors such as heat, drought or intense light compromise photosynthesis 106,107 and mitochondrial respiration 108 , leading to a depletion of ATP and to mitochondrial dysfunction 109 . In turn, energy supply for ATP-dependent defense systems such as heat-shock proteins and AAA-type proteases is severely compromised, leaving the question unanswered how oxidized proteins in chloroplasts and mitochondria can be cleared. Because AARE is an ATP-independent protease, our data suggest that organellar-targeted AARE may act as an ATP-independent defense to prevent or attenuate protein aggregation in the major ROS-producing organelles.
Based on localization prediction, 70% of our selected plant species possess one AARE isoform that localizes to either plastids or to mitochondria. Whether dual targeting via ambiguous targeting also occurs in these species remains to be experimentally validated. Further, we do not exclude that the remaining species also have organellar AARE isoforms, because our predictions may be compromised by incomplete gene models. Such incomplete AARE gene models without a transit peptide were present in earlier genome annotations of Arabidopsis and Physcomitrella 41,53 , whilst complete gene models with transit peptides were only introduced with later versions 39,62 .
It is not yet clear how translation of both variants in Arabidopsis is regulated, but a recent study highlights the importance of alternative translation initiation in shaping different subcellular localizations or functions of proteoforms 110 . This mechanism is also present in Physcomitrella, where dual targeting of FtsZ isoforms to the cytoplasm and chloroplasts is enabled via alternative translation initiation 111 . Thus, localization of PpAARE1 to the cytoplasm is also possible from the longer splice variant. Alternative translation initiation of AtAARE is further evidenced by proteomics data (www.peptideatlas.org/builds/arabidopsis/) 112 . Here, we found evidence for N-terminal acetylation of M 56 which is the initiator methionine of the short variant (PXD012708 113 ).
A reporter fusion of the shorter AtAARE ORF was observed in the cytoplasm and in the nucleus 21 . We did not detect a nuclear localization of any AARE isoform. Using LOCALIZER 64 we identified an NLS in AtAARE, in PpAARE1 and PpAARE3, but not in PpAARE2. In contrast, the human AARE homolog HsACPH does not have a predictable NLS, but nuclear import is mediated via interaction with XRCC1 under stress, where it acts in DNA-damage repair 114 . Similarly, a nuclear localization of AARE might also occur under stress in planta.
In Physcomitrella, AARE1 is the dominant exopeptidase, whereas AARE2 acts predominantly as endopeptidase, and the operation mode of AARE3 remains unresolved. Crystal structures from bacterial and archaeal AAREs revealed two possible entrances for substrates to the catalytic centers 14,115 but the mode of substrate entry is not fully understood. Although the quaternary arrangements of subunits differ between species 115,116 , the secondary structure arrangement is conserved across kingdoms and specific subunit interactions (multimerization) are likely a mechanism to mediate substrate specificity and modulate activity. This could be further used as a switch between endopeptidase and exopeptidase activity and additionally enable access of the catalytic center via structural re-arrangements. Accordingly, the interaction between the distinct PpAARE isoforms may modify substrate specificity and activity.
Our data indicate a partial compensation between PpAARE1 and PpAARE2. Both single KO mutants are phenotypically inconspicuous under normal conditions, although a significant reduction in exopeptidase activity was observed in ΔPpAARE1. PpAARE2 had the strongest effect on the accumulation of oxidized proteins in gametophores on the single isoform KO level, whereas the double KO of PpAARE1 and PpAARE2 showed the most striking effect in all lines as accelerated transition from protonema to gametophores. This phenotype is similar to mutants with disturbed auxin transport 117 . However, colony growth in these mutants was reduced which is different from our double KO mutant (PpAARE1/2). Intriguingly, this double KO results in stunted gametophores and a reduced life span. This phenotype partially resembles a loss-of-function mutant of a central component of autophagy (PpATG3) in Physcomitrella 118 . In ΔPpATG3, gametophores exhibit a reduced life span and colonies are smaller than in WT. In contrast, gametophore size was not affected. Further, photosynthetic capacity in ΔPpATG3 was also reduced, an effect which is apparently not caused by AARE depletion 21 and PpAARE genes were not differentially expressed in ΔPpATG3 118 . We conclude that the reduced life span observed in ΔPpAARE1/2 and ΔPpAARE1/3/2 is not due to an impaired autophagy system causing nitrogen starvation. This is in line with data which opposes autophagy at the onset of senescence in Arabidopsis 119 .
In mammals, elevated levels of oxidized proteins are associated with age-related pathologies, such as Alzheimer's disease, diabetes and different types of carcinomas 120 . If proteolytic clearance fails, further accumulation of oxidized proteins causes protein aggregation, which is a hallmark of ageing in animals 96,121,122 . A connection between protein oxidation and ageing was less well studied in plants. Plastid ROS levels increase during ageing 92 , which is in line with strong oxidation of plastid proteins in ageing leaves 86 . Likewise, protein oxidation marks the developmental transition between vegetative growth and flowering in Arabidopsis 86 . Physcomitrella mutants ΔPpAARE1/2 and ΔPpAARE1/3/2 showed accelerated developmental transition from protonema to gametophores, reduced life span and increased levels of oxidized proteins as signs of accelerated ageing. This is supported by the fact that gametophore tips, which is younger tissue, are viable longer than the older stems in both mutants. In the Arabidopsis AARE T-DNA mutants we found increased levels of oxidized proteins under normal cultivation conditions and an accelerated developmental transition, in this case premature bolting. These findings suggest an evolutionary conserved connection between protein oxidation and ageing.
We provide here a detailed analysis of AARE genes in the plant lineage and an in-depth analysis of AARE localization and function in the moss Physcomitrella and the annual angiosperm Arabidopsis. AARE loss-of-function mutants have not been described for any organism so far. We generated and analyzed such mutants and describe a connection between AARE function, aggregation of oxidized proteins and plant ageing, including accelerated developmental progression and reduced life span. Our findings complement similar findings in humans and animals where AARE malfunction is associated with protein aggregation and age-related pathologies.
To solidify the role of AARE in ageing in different life forms, particularly in plants with contrasting maximum life spans and in animals of different complexity, loss-of-function mutants should be established and analyzed in selected model species. Likewise, a deeper understanding of AARE function in human diseases is desirable. Together, such analyses may contribute to a unified concept of ageing in different life forms. Hydroponic Physcomitrella gametophore cultures were assembled as described 36,127 . Here, a thin layer of protonema from suspension was distributed on gauze mesh (PP, 250 m mesh, 215 m thread, Zitt Thoma GmbH, Freiburg, Germany) capped on a glass ring. The glass rings with protonema-covered mesh gauze were placed in Magenta®Vessels (Sigma-Aldrich, St. Louis, USA) and KnopME medium was added until the protonema-covered gauze mesh was moist. The medium was exchanged every 4 weeks. Gametophores were harvested after 12 weeks.
Gametophore colonies on Agar plates (KnopME) were generated by transplanting single gametophores to new plates. Plates were sealed with Parafilm®.
Generation of AARE knockout lines. Knockout constructs for each PpAARE gene were generated by amplifying genomic regions as homologous flanks. The PCR products were fused to a neomycin (PpAARE1), hygromycin (PpAARE2), and zeocin (PpAARE3) resistance cassettes, respectively, employing triple template PCR as described 128 with primer sequences listed in Supplementary Data S1 using the Q5 polymerase (New England Biolabs, Ipswich, USA). The knockout constructs were released from their vector backbones with XhoI (PpAARE1), BglII (PpAARE2) and DraI (PpAARE3), respectively. Enzymes were purchased from Thermo Fisher Scientific. Digested plasmids were precipitated and sterilized prior to transfection using standard ethanol precipitation method 129 . Transfection of Physcomitrella WT or KO (for consecutive knockouts) protoplasts was conducted via PEGmediated procedure 77,130 . The WT strain as well as the identified PpAARE KO lines are accessible via the International Moss Stock Center (IMSC, www.mossstock-center.org). IMSC accession numbers for the mutants and WT are available in Supplementary Data S4.
Screening of plants surviving antibiotics selection was done by PCR. KO mutant plants surviving the antibiotics selection were checked for the presence of a genomic region which should be removed upon homologous recombination-based integration of the knockout construct (Fig. S4a). In case of the absence of this WT signal, plants were further checked for correct 5′ and 3′ integration of the respective knockout construct using primers listed in Supplementary Data S1.
Protonema growth and gametophore induction. Suspension cultures were started at the same day and disrupted weekly as described. Dry weight was measured in triplicates and suspension density was adjusted to 440 mg dry weight (DW) per litre (mg DW/L) as described 131 . Droplets of 15 µL were distributed on solid medium (Knop ME, 0.02% glucose). Sealed plates were cultivated as described above. Three droplets each per line were distributed on one plate and all lines were grown on the same plate. 4 plates (12 colonies) were used per assay. Colony areas were measured with ImageJ. White pixels counted from binarized colony images were used as quantitative values. Gametophores were counted upon visibility of the first leafy structure on buds.
Generation of PpAARE-Citrine knock-in lines. Knock-in constructs to fuse the coding sequence of Citrine to the C-terminus of PpAARE isoforms via an 18 bp linker 128 at the endogenous genomic locus were cloned via Gibson assembly 132 . All necessary parts were amplified using primers listed in Supplementary Data S1. Additionally, XhoI (PpAARE1), SalI (PpAARE2), and BamHI (PpAARE3) restriction sites were added to the 5′ and 3′ ends of the respective knock-in construct. All parts were assembled into pJet1.2 vector backbone (Thermo Scientific) using the Gibson Assembly®Cloning Kit from New England Biolabs (Ipswich, Massachusetts, USA). Transfection of Physcomitrella WT protoplasts were conducted as described 77,130 by co-transfecting a plasmid containing a neomycin phosphotransferase resistance (nptII) cassette as transient selection marker (pBSNNNEV, Mueller et al.) 57 . The linearized plasmid and the co-transfection vector were purified and sterilized via ethanol precipitation 129 prior to transfection.
The presence of Citrine was checked with primers listed in Supplementary Data S1 and resulting positive lines were further checked for correct 5′ and 3′ integration by PCR using the Plant Phire Kit with primers listed in Supplementary Data S1. All identified fusion lines are available via the International Moss Stock Center (IMSC, www.moss-stock-center.org). IMSC accessions are listed in Supplementary Data S4. qPCR analysis. The copy number of the integrated KO constructs was determined using a qPCR-based method 78 . Genomic DNA was extracted from 100 mg frozen protonema using the GeneJET Plant Genomic DNA Purification Kit (Thermo Scientific, Waltham, USA). DNA concentrations were adjusted to 3 ng/µL for each sample and qPCR was performed with primers for the 5′ and 3′ flanks as well as with primers for the corresponding selection cassette. Additionally, primers for the single copy gene CLF (Pp3c22_2940V3) were used as internal control. Reference lines were WT as well as in-house lines with known single integrations of the used selection cassettes. Primers are listed in Supplementary Data S5. PCR reactions were done using 2x SensiFAST Mix (Bioline, London, UK) and analyzed in a Lightcycler 480 II (Roche, Basel, Schweiz). cDNA preparation. RNA was extracted using the innuPREP Plant RNA Kit (Analytik Jena, Jena, Germany). The extracted RNA was treated with DNAse I (Thermo Scientific) and subsequently reverse transcribed into cDNA using Superscript III Reverse Transcriptase (Life Technologies, Carlsbad, USA).
Fusion constructs for subcellular localization. All constructs were generated using Gibson assembly 132 and integrated into a PpAct5:Linker:eGFP-MAV4 vector backbone 73 . Integration of the different coding sequences was done in frame in front of the Linker:eGFP. All parts for the Gibson assembly (inserts and corresponding vector backbones) were amplified either with Phusion TM polymerase (Thermo Fisher Scientific) or with HiFi polymerase (PCR Biosystems Ltd) according to the manufacturer´s instructions. The primers were designed to have a 25 bp overlap to the corresponding fragment to be fused with. All primers and combinations are listed in Supplementary Data S6. In the case of the N-terminal difference of AtAARE (M 1 -A 55 of AT4G14570.1, gene model provided by TAIR (https://www.arabidopsis.org/)) the Actin5 promoter was replaced by the CaMV35S promoter 133 previously validated in Physcomitrella 134 .
Cloned plasmids were purified using the PureYield™ Plasmid Midiprep kit (Promega, Wisconsin, USA) according to the manufacturer´s instructions. The plasmids were purified and sterilized via ethanol precipitation 129 .
Confocal microscopy. Confocal imaging was performed on transiently transfected live protoplasts using Leica TCS SP8 (Leica Microsystems, Wetzlar, Germany). Immediately before microscopy, MitoTracker™ Orange CMTMRos (Thermo Fisher Scientific) was added to protoplast suspensions to a final concentration of 100 nM. For all imaging experiments, an HC PL APO CS2 63x/1.40 OIL objective was used with a zoom factor of 4. The pinhole was set to 35.4 µm. For excitation, a WLL laser (70%) was used. In a sequential acquisition setup, eGFP and chlorophyll were excited with the same laser beam (488 nm, 2%) and their signals were detected simultaneously, whereas MitoTracker TM was excited with a different laser beam (550 nm, 2%) and its signal was detected separately. The detection ranges were specified as 502-546 nm for eGFP, 662-732 nm for chlorophyll, and 597-636 nm for MitoTracker TM . The images were acquired as z-stacks with the number of optical sections varying with the protoplast size. The voxel sizes of the resulting z-stacks were 0.0903, 0.0903, 0.239 µm in the x-y-z order. For visual representation and analysis, single slices with the best signal-to-noise ratio were selected and extracted from each z-stack using FIJI software.
Cultivation of Arabidopsis. Seeds were surface-sterilized for 4 min in 80% ethanol and subsequently for 1 min in 100% ethanol. Seeds were placed on plates containing ½ MS supplemented with 1% (D + ) sucrose and 0.8% Agar. Alternatively, seeds were directly placed on sterilized soil. Seeds were incubated at 8°C for 2-4 days in the dark for stratification before placing them in growth chambers. Plants were further cultivated under short day conditions at 22°C and 70 µmol photons/m 2 s in an 8 h/16 h light/dark cycle. For bolting assays, plants were placed in a phytochamber at 22°C and 70 µmol photons/m 2 s in a 16 h/8 h light/dark cycle (long day condition). Rosette leaves and days since sowing were counted upon appearance of the shoot in the middle of the rosette.
Screening of Arabidopsis AARE mutants. Arabidopsis thaliana lines with T-DNA insertions in the At4G14570 locus were identified from the public T-DNA Express database at the SIGnAL website (Salk Institute Genomic Analysis Laboratory). Lines GK-456A10, SALK_080653C, SALK_071125C, and SALK_205137C, were obtained from the Nottingham Arabidopsis Stock Centre. Homozygous mutant alleles were verified by PCR using the following primers: forward LB GK 5′-ATATTGACCATCATACTCATTGC-3′ and reverse GK-456 5′-CTTCAAAGAAACACCAATCAG-3′ for the GK-456A10 line, and forward LB-pROK 5′-GCGTGGACCGCTTGCTGCAACT-3′ and reverse Salk_53 5′-TCTTTAGCCGAATCAGTTCCAGA-3′ for the SALK_080653C, SALK_071125C, and SALK_205137C lines. Identified homozygous mutant plants were back-crossed with Arabidopsis Col-0. The F2 generation was screened for homozygous mutant plants using the above listed primer sets to identify the mutant allele or substituting the forward primers with forward WT-GK-456 5′-AAGATGCTTTGCAGTCTC TA-3′ and forward WT-Salk 5′-ACTGCCTTATGATCCATTGTCTC-3′, to identify the GK and SALK lines WT alleles, respectively. RT-PCR was additionally performed to check for the presence of AtAARE transcripts using Taq polymerase on cDNA prepared as described above with primers At1-At4. All primer combinations are listed in Supplementary Data S4.
AARE exopeptidase activity. The enzyme activity assay according to 17 was modified. Here, tissue (80-100 mg) was homogenized in liquid nitrogen and dissolved in 1 mL extraction buffer (50 mM PBS, 1 mM EDTA, 2 mM DTT). After centrifugation at 20,000 × g for 20 min at 4°C, 300 µL supernatant was mixed with 700 µL reaction buffer (50 mM HEPES-KOH, pH 7.5 containing 1 mM AcAla-pNA (Bachem, Bubendorf, Switzerland) or 50 mM HEPES-KOH, pH 7.5, 10% DMSO containing 1 mM AcLeu-pNA (Bachem) and incubated at 37°C for 30-120 min. The reaction was stopped by the addition of 500 µL 30% acetic acid. Absorbance was measured at 410 nm in a photospectrometer. This approach was used to generate the data of Figs. 3b and S6a-d. Later the method was modified (Figs. 3g and S6e-g). Here, tissue was homogenized in liquid nitrogen and dissolved in 100 µL extraction buffer (50 mM PBS, 1 mM EDTA, 2 mM DTT) per 10 mg FW. After centrifugation at 20,000×g for 20 min at 4°C, 5 µL supernatant was mixed with 195 µL reaction buffer (50 mM HEPES-KOH, pH 7.5 containing 1 mM AcAla-pNA (Bachem, Bubendorf, Switzerland)) in a 96 well micro titer plate and incubated at 37°C for 30-120 min. In the case of AcLeu-pNA as substrate, 50 mM HEPES-KOH, pH 7.5 with 10% DMSO containing 1 mM AcLeu-pNA (Bachem) was used as reaction buffer. Every biological sample was measured in three technical replicates. Absorbance was measured at 410 nm. Activity was calculated using a molar absorbance coefficient 21 of 8.8 mM*cm −1 and represents the release of pNA [µmol] per minute normalized to the total protein concentration of the sample. The protein concentration was determined using the A 280 method of a NanoDrop TM (Thermo Fisher Scientific) or with a Bradford assay 135 .
Western blots. Western blots were performed as described 74 using the ECL Advance detection kit (GE Healthcare). The primary antibody against A. thaliana AARE was kindly provided by Dr. Yasuo Yamauchi 17 . The primary antibody was diluted 1:10,000 in TBST Buffer with 2% Blocking (GE Healthcare) and incubated on the membrane for 2 h. As secondary antibody anti-Guinea pig antibody, coupled to a horseradish peroxidase (Agrisera, AS 10 1496) diluted 1:10,000 in 2% TBST with 2% Blocking (GE Healthcare), was applied for 1 h.
Detection of oxidized proteins. Plant tissues were homogenized in liquid nitrogen and proteins were extracted in 50 mM PBS, 50 mM DTT, 1 mM EDTA. In all, 3-6 µg total protein was derivatized with DNPH and subsequently detected with an anti-DNP antibody according to the manufacturer's instruction of the OxyBlot Protein Oxidation Detection Kit (S7150, Sigma-Aldrich). Equal amounts of the derivatized protein samples were employed as loading control on a separate SDSgel and stained with Coomassie or silver staining.
Computational predictions. Predictions for the presence of cleavable targeting peptides were performed with TargetP2.0 63 . Additional predictions of subcellular localizations were performed with LOCALIZER 64 . The presence of peroxisomal targeting signals was predicted with PredPlantPTS1 66,67 . Prediction of protein domains was performed using InterProScan 136 and protein domain annotations according to PFAM 137 were used.
Co-Immunoprecipitation. Co-immunoprecipitation was performed using GFP-Trap Magnetic Particles M-270 (Chromotek, Planegg-Martinsried, Germany) as recommended by the manufacturer with modifications. In all, 300 mg protonema was homogenized in a 2 mL reaction tube using a tungsten and a glass bead. For each line three biological replicates were realized. The extraction buffer was chosen according to the manufacturer's recommendations for plant samples and contained 25 mM HEPES-KOH, pH 7.5, 2 mM EDTA, 100 mM NaCl, 200 nM DTT, 0.5% Triton X-100, 1% plant protease inhibitor cocktail (PPI, P9599, Sigma Aldrich). Ground plant material was dissolved in a final volume of 2 mL ice-cold extraction buffer and incubated for 10 min in a water quench ultrasonic device. Samples were centrifuged at 4°C at 20,000 × g for 30 min. For each sample 25 µL magnetic particle slurry was washed with 500 µL extraction buffer. The sample supernatant was transferred to the cleaned beads and incubated, rotating for 1 h at 6°C. Subsequently, beads were washed with 1 mL extraction buffer without Triton and PPI and again with 500 µL. Beads were then dissolved in 500 µL wash buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA), transferred to a new reaction tube and washed again in 500 µL wash buffer. A RapiGest solution (0.2% in 50 mM Tris-HCl, pH 7.5; RapiGest SF Surfactant (Waters, Milford, MA, USA) was mixed 3:1 with 5 mM DTT in 50 mM Tris-HCl, pH 7.5. 30 µL of the resulting mixture was applied to each sample. Samples were incubated at 95°C for 5 min under continuous shaking. Samples were cooled down to RT and 5 µL of a trypsin (V5117, Promega) solution (0.025 µg/µL in 50 mM Tris-HCl, pH 7.5) were added to each sample. Digestion on the beads was performed for 30 min at 32°C under continuous shaking. Supernatants were transferred to new reaction tubes and the remaining beads were washed twice with 50 µL 5 mM Iodoacetamide solution (in 50 mM Tris-HCl, pH 7.5). The wash supernatants were combined with the trypsincontaining supernatant and incubated over night at 30°C under continuous shaking. Acid-catalyzed cleavage of the RapiGest surfactant was performed as recommended by the manufacturer. Samples were purified using C18-STAGE-Tips as described 138 and eluted from the tip in 30% ACN in 0.1% FA.
Mass spectrometry measurement and data analysis. MS analysis was performed on an Orbitrap Q-Exactive Plus instrument (Thermo Fisher Scientific) coupled to an UltiMate 3000 RSLCnano system (Dionex LC Packings/Thermo Fisher Scientific) as described 139 . Database search and label-free quantitation was performed using MaxQuant software V1.6.0.16 140 . For each Co-IP a specific database was employed containing all V3.3 proteins of Physcomitrella 39 as well as the sequence of the respective fusion protein. Additionally, the contaminant list provided by the software was included. Decoys were generated on the fly by reverting the given protein sequences. Variable modifications were formation of pyro Glu (N-term Q, −17.026549 Da), oxidation (M, + 15.994915 Da), acetylation (N-term, + 42.010565 Da) and deamidation (N, + 0.984016 Da). Carbamidomethylation (C, + 57.021464 Da) was specified as fixed modification. Enzymatic specificity was set to tryptic with semi-specific free N-terminus. An FDR of 0.01 was set for protein identification. LFQ values 141 were used as quantitative values. Interaction analysis was performed in Perseus V1.6.12.0 142 . Missing values were imputed from a normal distribution with a down-shift of 1.8 and distribution width of 0.3. Interaction partners were accepted at an FDR of 0.01 and a p-value < 0.01.
Raw files of the test-IP measurements (Fig. S8e) were processed using Mascot Distiller V2.7.10 and searched against all Physcomitrella protein models V3.3 39 using Mascot Server V2.7.0 (Matrix Science). Processed mgf files from immunoprecipitation experiments targeting N-terminal arginylation (PXD003232 36,143 ) and the test-IPs were searched again against all Physcomitrella protein models V3.3 39 using Mascot Server V2.7.0 (Matrix Science). The precursor mass tolerance was 5 ppm and the fragment mass tolerance was 0.02 Da. Variable modifications were formation of pyro Glu (N-term Q, −17.026549 Da), oxidation (M, + 15.994915 Da), acetylation (Nterm, + 42.010565 Da) and deamidation (N, + 0.984016 Da). Carbamidomethylation (C, + 57.021464 Da) was specified as fixed modification. Enzymatic specificity was set to tryptic with semi-specific free N-terminus. Search results were loaded in Scaffold TM 5 (V5.0.1, Proteome Software) and proteins were accepted at an FDR = 1 and peptides at an FDR = 0.5. A table of identified proteins is accessible in Supplementary Data S7.
Multiple sequence alignment and phylogenetic reconstruction. Homologous protein sequences were aligned with UPP 144 (version 4.4.0) using default parameters and subsequently translated into a codon-aware CDS alignment with PAL2NAL 145 (version 1.4). Based on this multiple coding sequence alignment we reconstructed a maximum likelihood tree with RAxML 146 (version 8.2.12) using the GTRCAT model with 1000 rapid bootstrap samples. The tree was rooted at the split between animal and plant sequences and plotted in R 147 using the packages ape 148 and ggtree 149 .
Statistics and reproducibility. Statistical differences in datasets were analyzed with one-way Anova with subsequent post-hoc test if different lines at same conditions were compared. Two-way Anova with subsequent post-hoc test was performed to analyze differences between lines at different conditions. Sample sizes of biological replicates are specified in the figure legends. Anova and post-hoc analysis was performed in R 147 . Statistical significance was accepted at *p < 0.05, **p < 0.01, and ***p < 0.001.
Reporting summary. Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.