Domain collapse and active site ablation generate a widespread animal mitochondrial seryl-tRNA synthetase

Abstract Through their aminoacylation reactions, aminoacyl tRNA-synthetases (aaRS) establish the rules of the genetic code throughout all of nature. During their long evolution in eukaryotes, additional domains and splice variants were added to what is commonly a homodimeric or monomeric structure. These changes confer orthogonal functions in cellular activities that have recently been uncovered. An unusual exception to the familiar architecture of aaRSs is the heterodimeric metazoan mitochondrial SerRS. In contrast to domain additions or alternative splicing, here we show that heterodimeric metazoan mitochondrial SerRS arose from its homodimeric ancestor not by domain additions, but rather by collapse of an entire domain (in one subunit) and an active site ablation (in the other). The collapse/ablation retains aminoacylation activity while creating a new surface, which is necessary for its orthogonal function. The results highlight a new paradigm for repurposing a member of the ancient tRNA synthetase family.


INTRODUCTION
During gene translation each transfer RN A (tRN A) is aminoacylated with its corresponding amino acid by the cognate aaRS, prior to its transport to the ribosome, where it participates in translation elongation via the recognition of complementary codon triplets in the messenger RNA ( 1 ).
The aaRSs are ancient enzymes that e volv ed contemporarily to the extant genetic code ( 2 ), possibly from a primiti v e two-protein-one-tRNA comple x ( 3 ).Fr om this pr ocess two uni v ersally conserv ed families of a pproximatel y ten enzymes each emerged ( 4 ).All members of each family share a catalytic domain with a common fold, around which different multi-domain structures e volv ed.The e xtended and continuing di v ergence of aaRSs included the incorporation of additional biological functions that are species-specific and rely on the accretion of new and idiosyncratic domains ( 5 ).These non-canonical functions are associated with transcription and transla tion regula tion, nutrient-sensing mechanisms, or eukaryotic regulatory pathways linked to inflammation or cell cycle regulation, among others (6)(7)(8)(9).
In eukaryotes, the complex evolutionary history of aaRSs is further confounded by the convergence of two or three orthologous aaRS sets resulting from the endosymbiotic e v ents that gav e rise to mitochondria and chloroplasts ( 10 , 11 ).In animals, most cytosolic and mitochondrial aaRSs are encoded by distinct nuclear genes ( 12 ).An example of this situation is seryl-tRNA synthetase (SerRS), for which two different nuclear genes exist: serS 1, coding for the cytosolic enzyme (SerRS1), and serS2 , which codes for the mitochondria-localized variant (SerRS2) ( 13 ).
SerRSs belong to the class II aaRS family ( 14 ), and are characterized by the presence of an N-terminal coiled-coil domain that interacts with the pseudo-knot region of the tRNA ( 15 ).Homodimeric SerRSs contain two identical tRNA binding sites that span across the dimer inteface, in which the N-terminal coiled-coil domain from one subunit recognizes the tRNA pseudoknot, while the tRNA acceptor stem binds to the catalytic site of the second subunit.Thus, each monomer contributes its coiled-coil domain to one tRNA binding site, and its catalytic site to the opposite tRNA binding site (15)(16)(17).
The maintenance of mitochondria-specific SerRS subsets is r equir ed to aminoacylate mitochondrial tRNA Ser (mt-tRN A Ser ) w hich are exposed to distinct selecti v e pr essur es and elevated evolutionary rates, leading to a marked divergence characterized by the loss of structural features that are usually highly conserved in canonical cytosolic tRNAs (18)(19)(20).The mt-tRNA Ser  GCU r epr esents one of the most prominent examples for this process, undergoing loss of the long variable (V)-arm, deletion of the entire D-arm, and a substantial remodelling of its T-arm.Biochemical and structural work shows that the degeneration of tRNA Ser GCU was accompanied by a fundamental rewiring of the intermolecular specificity rules underl ying SerRS / tRN A Ser interactions in mammals ( 17 , 21 , 22 ), possib ly e xplaining why dedicated SerRS2 are retained.
In animals, a second paralogous mitochondrial serS gene that codes for the protein SLIMP (Seryl-tRNA synthetase-Like Insect Mitochondrial Protein) e volv ed following the duplication of a serS2 gene early in the evolution of metazoans, and was retained by most animal groups but lost in vertebrates ( 23 ).For clarity, here we will use SerRS2 to refer to all mitochondrial SerRS.In SLIMP-containing species we will refer to the subunits SerRS2 ␣ and SLIMP, to illustrate the fact that these two proteins ma y f orm a heterodimer (SLIMP / SerRS2 ␣) (Figure 1 A).
In Drosophila melanogaster , SLIMP ( Dm SLIMP) is an essential protein that associates with SerRS2 ␣ ( Dm SerRS2 ␣) forming an obligate heterodimer ( Dm SLIMP / DmSerRS2 α) absolutely r equir ed to aminoacylate mt-tRNA Ser .Dm SLIMP depletion induces a loss in respiratory capacity and a distorted mitochondrial morphology in all tissues and de v elopmental stages (23)(24)(25).Remar kab ly, the Dm SLIMP / Dm SerRS2 ␣ comple x is involved in complex regulatory pathways that control mitochondrial DNA replication via an interaction with the LON protease ( 25 ), and cell cycle progression from G1 to S phase ( 26 , 27 ).
In order to understand the evolution of metazoan SerRS2, the structural differences between homodimeric SerRS2 and the SLIMP / SerRS2 ␣ heterodimer, and the ability of Dm SLIMP / Dm SerRS2 ␣ to retain full recognition of tRN A Ser w hile gaining new cellular functions, we characterized the phylogenetics and structural evolution of these proteins, and experimentally studied their tRNA recognition mechanisms.These analyses unveiled a remarkab le assymetric co-e volution of SLIMP and SerRS2 ␣ in all SLIMP-containing species, indicating that the heterodimer described in Drosophila is likely the enzyme variant acti v e in most invertebrates.Most striking in this co-evolution of the two subunits is the concomitant loss of the N-terminal coiled-coil domain in the SerRS2 ␣ subunit and the collapse of the catalytic site of SLIMP.
We show here that the coiled-coil domain in Dm SLIMP remains indispensable for aminoacyla tion, demonstra ting that the Dm SLIMP / Dm SerRS2 ␣ complex has a single aminoacylation site.Notably, the mechanism of tRN A Ser GCU reco gnition by Dm SerRS2 is very similar to that of homodimeric mammalian SerRS2s, e v en allowing cross-species aminoacylation of human mt-tRNA Ser GCU by the Drosophila enzyme.Finally, we show that the aminoacylation activity of the Dm SLIMP / Dm SerRS2 ␣ heterodimer is compatible with the binding of additional molecular partners.

Phylogenetic analyses
For phylogenetic inferences we initially retrie v ed the sequences of SerRS2 homologs from the National Center for Biotechnology Information (NCBI) ( 28 ), or identified them by BLAST searches.Subsequentl y, homolo gs were aligned with Clustal X (Supplementary Figure S2) ( 29 ).Phylogenetic analyses were performed using distance methods and maxim um likelihood (PRO TDIST and PRO TMLg) with PHYLIP 3.6 ( 30 ).The program SEQBOOT was used to create 100 bootstrap replicates of the initial alignments.Subsequently, NEIGHBOR and CONSENSE were used to calculate the confidence limits of each node using the 100 bootstrap r eplicates.Tr ees wer e drawn with DRAWGRAM and DRAWTREE, and visualized with Adobe Illustrator.

Structure modeling and analysis
3D protein structures for both SLIMP and SerRS2 sequences were created by means of ExPASy The regions that correspond to the major domains of each protein type are represented by colored blocks under the alignment, and the sequences are color-shadowed accordingl y. ( B ) Phylo genetic anal ysis of SerRS2, SerRS2 ␣ and SLIMP sequences performed as described.Numbers at branches r epr esent the bootstrap support for each node.The deduced point of SerRS2 duplica tion tha t gave rise to SerRS2 ␣ and SLIMP is marked by a red square.The branches where SLIMP was lost are crossed by red lines to signify the gene loss (see also Supplementary Figure S1).SWISS-MODEL homology modeling, with the crystallized Bos taurus SerRS2 ( Bt SerRS2) structure as a template for all sequences (pdb: 1WLE) ( 23 , 30 ).As of recently howe v er, these structures can also be found in the AlphaFold EBI database v2.To create the heterodimer, Alphafold2 ( 37 ) was operated through ColabFold ( 38 ).3D structur es wer e accordingl y anal ysed by comparing the structural features, the globular domain and coiled-coil, in the structural alignments with BtSerRS2.A coiled-coil determination was performed for all the sequences with MARCOIL ( 39 , 40 ).

Active site analysis
An analysis of the residues potentially involved in the interaction with seryl-adenylate in the acti v e site of SLIMP and SerRS2 was based on multiple sequence alignments as described by Guitart et al. ( 26 ).Sequences were aligned with Clustal X.

Determination of conserved residues involved in tRNA Ser recognition
Models complexed with tRNA were created with PY-MOL for both a Bt SerRS2 homodimer and for a Dm SLIMP / DmSerRS2 α heterodimer.Structural differences between the canonical SerRS2 homodimer and SLIMP / SerRS2 ␣ heterodimer were identified by aligning the structures.Amino acid residues potentially involved in tRN A reco gnition and binding were identified according to the tRNA configuration relati v e to the models.To analyse amino acid conservation, a multiple sequence alignment was performed for SerRS2 and SLIMP sequences with Clustal X. tRNA Ser structures were compared and structural alignments were created to identify differences in tRNA Ser between species containing the canonical SerRS2 homodimer and species containing SLIMP.

Protein cloning and purification
Dm SLIMP, Dm SerRS2 ␣ and Dm SLIMP were cloned into the pQE-70 vector (QIAGEN) for bacterial expres-sion of a C-terminal 6His-tagged protein.Dm SLIMP was found to be mostly insoluble, which significantly reduced the purification yield of this protein (Supplementary Figure S3).The sequence of all plasmids was confirmed by sequencing.Dm SLIMP and Dm SerRS2 ␣ were cloned into pOPINFS e xpression v ectors (Oxfor d Pr otein Pr oduction Facility) as a bicistronic product, resulting in a C-terminal 6His-tagged Dm SerRS2 ␣ and C-terminal Strep II tagged Dm SLIMP separated by a ribosome binding site.BL21 (DE3) cells (Nzythech) were transformed with the resulting plasmids and a starter culture was diluted 1 / 100 in autoinduction media at 37 • C for 2 hours, and 17 hours at 25 • C until the OD600 was stable.After harvest, the cell pellet was resuspended in lysis buffer (20 mM sodium phosphate buffer, 200 mM NaCl, 50 mM imidazole, protease inhibitor cocktail, and DnaseI) and lysed using a cell disruptor (20 000 Psi).The lysate was centrifuged at 24 000 g for 1 h, and the supernatant was filtered with a 0.45 mm filter.The enzymes were purified on HisTrap columns according to the manufacturer's protocol (GE Healthcare Life Sciences).The collected fractions were analyzed by SDS-PAGE and dialyzed (20 mM sodium phosphate buffer, 200 mM NaCl, 1 mM DTT).Proteins wer e stor ed in dialysis buffer with 10% glycerol, and protein concentration was measured with Pierce BCA Protein Assay Kit (Thermo Fisher).

Expression and purification of the mitochondrial lon protease substrate binding domain (lon-SBD)
The substrate binding domain of the mitochondrial Lon protease isoform from Drosophila melanogaster (Lon-SBD; UniProt ID Q7KUT2-2; His93 to Ile432) was cloned into the pCri4a protein expression plasmid (Addgene plasmid #61314) coding for an N-terminal 6xHis-tagged thioredoxin domain, which can be removed upon incubation with Tobacco Etch Virus (TEV) protease.
For immobilized-metal affinity chromato gra phy (IMAC), clarified lysates were supplemented with 10 mM imidazole, and loaded onto a pre-equilibrated gravity flow column containing 1.5 ml of Ni Sepharose 6 Fast Flow resin (Cytiva), washed and eluted with Buffer A containing 20 and 250 mM imidazole, respecti v ely.Elution samples were then rebuffered to Buffer A using a Vivaspin 20 Ultr afiltr ation Unit (MWCO 10 0000, PES.Sartorius) to remove imidazole.Samples were then incubated with in-house produced 6His-tagged TEV protease at a peptidase to substrate ratio of 1:50 (w / w) in presence of 5 mM ␤-mercaptoethanol at 4 • C overnight.The sample was then reloaded onto the same column equilibrated in Buffer A to remove uncleaved target protein and TEV protease.The flow-through of the re v erse chromato gra phy was re-concentrated and further purified using size exclusion chromato gra phy and a Super de x 200 GL column (Cytivia).Protein purity was verified by SDS-PAGE analysis.

In vitro transcription of tRNAs
tRNAs were obtained by in vitro transcription using T7 RN A pol ymerase according to standard protocols (Saint-L éger et al., 2016).Genes encoding tRNA Ser variants were cloned into the pUC-19 vector under the control of a T7 RN A pol ymerase promoter.DN A-templa tes for in vitr o transcription were amplified by PCR using forward and reverse primers complimentary to the T7 promoter and the 3 end of the tRNA gene, respecti v ely.Transcription reactions were performed in 40 mM Tris-HCl pH 8.0, 25 mM NaCl, 25 mM MgCl 2 , 2 g / ml yeast pyrophosphatase (Roche), 1 mM Spermidine, 5 mM DTT, 18 mM GMP, 4 mM each of ATP, CTP, GTP and UTP with 75 g / ml T7 polymerase and DNA template at 37 • C for 6 hours.Reactions were stopped by phenol / chloroform extraction followed by purification of the tRNA by 12% denaturing PAGE.tRNA was eluted from the gel in buffer containing 200 mM NaOAc, 20 mM Tris-HCl, 5 mM EDTA (pH 5.3), followed by ethanol-precipitation.The final tRNA was taken up in Rnase-free water and stored at -80 • C.

Aminoacylation assays
In vitro tRNA aminoacylation assays were performed at 25 • C in 100 mM HEPES pH 7.2, 40 mM KCl, 50 mM MgCl 2 , 1 mM ATP, 0.1 mg / ml -1 BSA, 1 mM DTT, 10 mM serine, 500 Ci / mol -L -[ 3 H(G)]-Serine and 0.5 mM protein.Reactions were initiated by addition of pure enzyme and samples of 22 l were spotted onto W ha tman 3 Mm discs at varying time intervals.Radioactivity (corresponding to amino acid ligated to tRNA substrate) was measured by liquid scintillation (TriCard 2900TR, P ackard).Aminoacylation r eactions with mutant SerRS and tRNAs were carried out in an assay solution containing 50 mM HEPES pH 7.5, 60 mM KCl, 20 mM MgCl 2 , 4 mM ATP, 2 mM DTT, 4 g / ml pyrophosphatase, 10 M cold l-serine and 5 M [ 3 H]-serine (1 mCi / ml).Reactions were initiated by addition of SerRS (0.5 M).At varying time intervals, 5-l aliquots wer e r emoved and applied to a MultiScreen 96-well filter plate (0.45 m pore size hydr ophobic, low-pr otein-binding membrane; Merck Millipor e), pr e-wetted with quench solution (0.5 mg / ml salmon sperm DNA, 0.1 M EDTA, 0.3 M NaOAc (pH 3.0)).After all time points were collected, 100 l of 20% (w / v) trichloroacetic acid (TCA) was added to precipitate nucleic acids.The plate was washed four times with 200 l of 5% TCA containing 100 mM cold serine, followed once by 200 l of 95% ethanol.The plate was then dried, followed by deacylation of bound tRNAs by addition of 70 l of 100 mM NaOH.After 10 min incuba tion a t RT, the NaOHsolution was centrifuged into a 96-well fle xib le PET microplate (PerkinElmer) with 150 l of Supermix scintillation mixture (PerkinElmer).After mixing, the radioactivity in each well of the plate was measured in a 1450 Mi-croBeta Micoplate Scintillation and Luminescence Counter (PerkinElmer).

Phylogenetics of SerRS2, SerRS2 ␣ and SLIMP
To study the evolution of metazoan SerRS2 we performed full length alignments of animal SerRS sequences and performed phylogenetic analyses ( 31 ) which confirmed the existence of two major clades: one populated by cytosolic SerRS, and the second containing mitochondrial SerRS and SLIMP sequences (Figures 1 a-b, and Supplementary Figure S1).The SLIMP sequences are monophyletic and emerge early in metazoan evolution, after a duplication that took place at the root of bilaterian species.Interestingly, SLIMP e volv es faster than SerRS2, but is retained in most invertebrate taxa, including arthropods , molluscs , annelids , tardigrades , echinoderms , and hemichordates (Figures 1 A,  B, and Supplementary Figure S1).This phylogenetic analysis re v eals the distribution of potential SLIMP / SerRS2 ␣ heterodimers in metazoans.

The SerRS catalytic site pocket is lost in SLIMP sequences
Whole-sequence structure-based alignments of Dm SLIMP and Dm SerRS2 ␣ with Bos taurus SerRS2 (80.5% and 89.6% sequence identity, respecti v ely) show that the overall structure of the catalytic core domain of class II aaRSs is well conserved in both Dm SLIMP and Dm SerRS2 ␣ (Supplementary Figure S2).To study the detailed structural features of the Dm SLIMP / Dm SerRS2 ␣ heterodimer we used the structure-based alignments to model its 3D structure using Alphafold Multimer ( 41 ) (Figure 2 A), and manually docked a tRNA molecule using the complex structure of H. sapiens SerRS2 with mt-tRNA Ser GCU (PDB: 7U2B) as a r efer ence ( 17 ).This analysis shows that, while all SerRS2 ␣ conserve the canonical catalytic site ar chitectur e, SLIMP sequences present a complete collapse of the catalytic cavity due to the loss of all conserved residues in the catalytic pocket and adjacent positions (Figure 2 B).This degeneration includes muta tions a t positions involved in the binding of the seryl-adenylate, such as N297 and T316 (R240 and F258 in DmSerRS2 ␣, respecti v ely), or the acceptor stem of tRNA Ser , such as N191 and T312 (K141 and R254 in DmSerRS2 ␣, respecti v ely) (Figure 2 B).These observations are in agreement with the fact that Dm SLIMP on its own is unable to bind ATP or aminoacylate tRNA Ser ( 23 ).This catalytic site degeneration occurs in all SLIMP sequences, suggesting an extended distribution of the SLIMP / SerRS2 ␣ heterodimer among invertebrates.Interestingly, in the starfish Acanthaster planci conservation of catalytic site residues in SLIMP is slightly higher, a first indica tion tha t evolution of SLIMP / SerRS2 ␣ in echinoderms may differ from other species (see below) (Figure 2 B).
These observations, together with the fact that Dm SLIMP is r equir ed for aminoacylation in Drosophila , suggest that SLIMP / SerRS2 ␣ heterodimers present a single tRNA binding surface, where aminoacylation takes place at the catalytic site of the SerRS2 ␣ subunit.To try to identify whether tRNA Ser present idiosyncratic features adapted to the SLIMP / SerRS2 ␣ complex, we built structure-based alignments of animal tRNA Ser sequences.Howe v er, with the exception of a small insertion in the D-arm of tRNA Ser GCU , we found mt-tRNA Ser in SLIMP-containing organisms to be generally similar to other mt-tRNA Ser (Figure 2 C and Supplementary Figure S3).

The coiled-coil domain is lost in SerRS2 ␣ but retained in SLIMP
In vertebrate SerRS2, tRN A reco gnition is achie v ed primarily through interactions with three distinct structural domains, namely: the 'coiled-coil', the 'N-helix', and the 'Ctail' domains ( 22) ( 17 ).The coiled-coil domain is the major tRN A Ser reco gnition element in SerRSs and contacts the identity-defining extended V-arm of cytosolic tRNA Ser as well as the D-and T-loops of mammalian mt-tRNA Ser variants ( 22 , 42 ) to properly position the tRNA molecule relati v e to the catalytic site.The N-helix and C-tail r epr esent mitochondria-specific sequence extensions, to date only described in vertebrate SerRS2s, where they form an interface with the T-stem to promote tRNA binding and aminoacylation ( 22 , 17 ).Our model of the Dm SLIMP / Dm SerRS2 ␣ heterodimer shows that Dm SLIMP conserves the N-terminal coiled-coil domain characteristic of canonical SerRS structures.Strikingly, howe v er, this coiled-coil domain is completely lost in Dm SerRS2 ␣ (Figures 2 A, B, and Supplementary Figure S2).Thus, the degeneration of the catalytic site pocket in Dm SLIMP is accompanied by the loss of the coiled-coil domain in Dm SerRS2 ␣.This asymmetric domain loss extends to the N-terminal extension (lost in Dm SerRS2 ␣ and retained by Dm SLIMP), and to the C-tail domain (lost in Dm SLIMP and retained by Dm SerRS2 ␣) (Supplementary Figure S2).
We used the coiled-coil prediction algorithm MARCOIL ( 39 , 40 ) to determine the presence or absence of the coiledcoil structure in SLIMP and SerRS2 ␣ across species (Figure 3 B).This re v ealed that the concomitant loss of the coiledcoil domain in SerRS2 ␣ and the collapse of the catalytic site pocket in SLIMP extends to all SLIMP-containing species.Interestingly, the echinoderm Str ongylocentr otus purpur atus SerRS2 ␣ retains a vestigial coiled-coil structure.This, together with the conservation of some SLIMP catalytic site residues in the starfish A. planci ( vide supra ), suggests that DmSLIMP / Dm SerRS2 ␣ evolution in echinoderms differs from other invertebrates.

SLIMP's coiled-coil domain is r equir ed f or tRNA r ecognition and aminoacylation
The coiled-coil of SerRS is essential for tRN A Ser reco gnition, and the critical residues involved in this function hav e pre viously been identified for B. taurus and H. sapiens SerRS2 ( 22 , 17 ).Our data show that these positions are conserved throughout most SLIMP sequences, while they are lost along with the coiled-coil domain in most SerRS2 ␣ (Figur e 3 A).Mor eov er, we hav e pre viously shown that purified Dm SLIMP or Dm SerRS2 ␣ alone do not possess aminoacylation activity, while their co-expression generates a heterodimeric Dm SLIMP / Dm SerRS2 ␣ capable of efficiently charging mt-tRNA Ser with serine.Similarly, reconstituting an enzymatically acti v e heterodimeric structure from individually purified Dm SLIMP and Dm SerRS2 ␣ is possible ( 25 ).
To experimentally evaluate the functional importance of the coiled-coil domain of SLIMP, we generated a variant of Dm SLIMP that lacks the N-terminal coiled-coil domain ( Dm SLIMP, see materials and methods), and tested its ability to support tRNA Ser aminoacylation in vitro .The Dm SLIMP / Dm SerRS2 ␣ heterodimer was not efficiently expressed and could not be purified.We then attempted to reconstitute the Dm SLIMP / Dm SerRS2 ␣ heter odimer fr om individually purified Dm SerRS2 ␣ and Dm SLIMP (Supplementary Figure S3).In contrast to reconstituted Dm SLIMP / Dm SerRS2 ␣, the reconstituted Dm SLIMP / Dm SerRS2 ␣ heterodimer showed no aminoacylation activity of Drosophila mt-tRNA Ser (Figure 4 A).
Biochemical and structural analyses have shown that mammalian SerRS2s use a conserved arginine (R146 in human SerRS2) in the coiled-coil domain to specificall y reco gnize the idiosyncratic T-arm ar chitectur e of mt-tRN A Ser GCU .Notabl y, sequence comparisons show that this arginine is conserved in SLIMP (R123).We generated R123 mutants of Dm SLIMP and analysed their effect on cognate tRNA Ser GCU aminoacylation.Both Dm SLIMP R123A and Dm SLIMP R123E abolished aminoacylation by the Dm SLIMP / Dm SerRS2 ␣ heterodimer, the same effect described for the corresponding R146 mutations in bovine and human SerRS2 ( 22 ) ( 17 ) (Figure 4 B).Thus, the Dm SLIMP coiled-coil domain is essential for tRNA aminoacylation, possibly through a mechanism that is conserved between homodimeric and heterodimeric SerRS2.

The tRNA recognition by Dm SLIMP / Dm SerRS2 ␣ is largely conserved with human SerRS2
As shown above (Figure 2 C), a comparison of metazoan tRNA Ser  GCU sequences shows no apparent major difference between SLIMP-containing species and those devoid of SLIMP.For example, the T-arm ar chitectur e of human mt-tRNA Ser GCU (a key structural feature that includes a (6bp) T-stem and a remodelled non-canonical T-loop) ( 17 ) is conserved in Drosophila tRNA Ser GCU (Figure 2 C).This led us to hypothesize that the tRN A reco gnition mechanism may be conserved between hetero-and homodimeric SerRS2.
We tested the ability of heterodimeric DmSLIMP / DmSerRS2 ␣ to aminoacylate mt-tRNA Ser  variants in which either the T-stem was reduced from 6 base-pairs to the canonical 5 bp ( Dm tRNA Ser(GCU) (DTstem)) or the non-canonical T-loop was replaced by a canonical one ( Dm tRNA Ser(GCU) (DT-loop)).Both mutations abolished charging of the tRNA Ser GCU variants, demonstra ting tha t the non-canonical structural features of mt-tRN A Ser GCU reco gnized by mammalian SerRS2 are also important for recognition by heterodimeric Dm SLIMP / Dm SerRS2 ␣ (Figure 4 C) ( 17 ).
Then we asked whether heterodimeric Dm SLIMP / Dm SerRS2 ␣ could cross-aminoacylate human mt-tRNA Ser .Howe v er, virtually no charging activity was observed using a wild-type transcript of human mt-tRNA Ser GCU (data not shown).Mammalian SerRS2 recognizes the discriminator base (N73) of mt-tRNA Ser with a pr efer ence for purine bases ( 21 , 43 ), whereas most invertebrate mt-tRNA Ser sequences contain a U73.We generated a human mt-tRN A Ser GCU (A73U) m utant.Strikingly, this transcript was efficiently charged by Dm SLIMP / Dm SerRS2 ␣ (Figure 4 D).
We also tested whether heterodimeric Dm SLIMP / Dm SerRS2 ␣ could aminoacylate human mt-tRN A Ala (A73U) or mt-tRN A Phe (A73U) transcripts, but no activity could be detected with these tRNAs, demonstra ting tha t the heterodimeric Dm SLIMP / Dm SerRS2 ␣ can efficiently discriminate against non-cognate human mt-tRNAs (Figure 4 D).
Taken together, these results support the hypothesis that the Dm SerRS2 ␣ subunit of Dm SerRS2 recognizes the discriminator base of cognate tRNA Ser with a pr efer ence for pyrimidine bases, and that the coiled-coil domain of Dm SLIMP specifically binds the idiosyncratic T-arm of mt-tRNA Ser GCU using the same recognition mechanism as homodimeric mammalian SerRS2 (Figures 4 B-D).

Dm SerRS2 ␣/ Dm SLIMP aminoacylation activity is compatible with LON binding
The evolutionary loss of the coiled-coil domain in Dm SerRS2 ␣ and the degeneration of the catalytic site pocket of Dm SLIMP necessarily generates a heterodimeric enzyme in which only one tRNA binding surface is used, thus liberating surface area for interactions with other proteins.
Pr evious r eports demonstra ted tha t SLIMP interacts with the substrate binding domain of the mitochondrial protease LON to stimulate the degradation of TFAM ( 25 ).To test whether the interaction of LON with heterodimeric Dm SLIMP / Dm SerRS2 ␣ is compatible with the aminoacylation activity of the enzyme, we performed aminoacylation assays with co-expressed and reconstituted DmSLIMP / Dm SerRS2 ␣ ( vide supra ) in the absence or presence of LON's substrate binding domain.In all cases the aminoacylation activity of Dm SLIMP / Dm SerRS2 ␣ was not affected by the presence of the LON domain in the r eaction mixtur e, demonstra ting tha t their interaction is compatible with the tRNA aminoacylation activity, and supporting the possibility that the loss of one tRNA binding surface allowed SLIMP / mSerRS ␣ to establish new functional interactions (Figure 5 ).

DISCUSSION
In animal mitochondria, two different tRNA Ser forms coexist: the near-canonical tRNA Ser UGA , and the highly diverged and structurally distinct tRNA Ser GCU .The realization that mitochondrial SerRS2 in B. taurus was capable of ef ficiently aminoacyla ting these two very dif ferent substrates ( 22 ) prompted an investigation on the molecular features of SerRS2 that permit such substrate di v ersity.Studies on Hs SerRS2 re v ealed an adapti v e mechanism that allows the enzyme's tRNA binding surface to recognize its two highly di v ergent cognate substrates ( 17 ).
In Drosophila , the aminoacylation of mt-tRNA Ser is performed by Dm SLIMP / Dm SerRS2 ␣, a heterodimeric SerRS2 composed of the monomers Dm SLIMP and Dm SerRS2 ␣ ( 23 , 25 ).How Dm SLIMP / Dm SerRS2 ␣ e volv ed, and how it recognizes tRNA was not understood.Here, we set out to investigate the evolutionary history of animal mitochondrial SerRS2, and the structural and functional di v ersity of this group of enzymes.To achie v e this aim we determined the phylogenetic relationships of animal SerRSs and characterized the structural and functional differences of their mitochondrial forms.
The phylo genetic anal yses pr esented her e show that the SerRS2 duplication that gave rise to SLIMP took place at the root of bilaterian evolution (Figure 1 A, B).The dupli-cated gene that would e v entually become SLIMP was fixed in all insect, echinoderm, hemichordate, mollusc, annelid, and tardigrade species for which complete genome information is available, where it is always accompanied by an additional SerRS2 sequence (SerRS2 ␣).Distance-based phylogenetic trees clearly show that the catalytic domain sequences of SLIMP e volv e at a faster rate than SerRS2 ␣ sequences (Supplementary Figure S1).
The structural analyses based on the heterodimeric Dm SLIMP / Dm SerRS2 ␣ model re v eal the same process of sequence di v ergence and domain loss in all SLIMPcontaining species (Figure 2 A) involving the mutational collapse of the catalytic pocket of SLIMP, and the disappearance of the tRNA-binding N-terminal coiled-coil domain in SerRS2 ␣.Possibly, the formation of non-productive complexes after the collapse of the acti v e site in SLIMP was a factor in the selection of a SeRS2a variant devoid of the coiled-coil domain.These changes led to the complete loss of one of the two acti v e sites found in homodimeric SerRSs, highlighting how gene duplication e v ents can lead to the separation of molecular functions within a protein complex (Figure 6 ).
Sequence comparisons show that the process of asymmetrical domain loss in SerRS2 extends to most invertebrate species (Figures 2 B, C).Howe v er, echinoderm species such as S. purpuratus and A. planci show an a pparentl y higher conservation of canonical SerRS structures.The appar ent r etention of a coiled-coil structur e by S. purpuratus SerRS2 ␣ suggests that the mutational collapse of the catalytic pocket in SLIMP may have preceded the loss of the coiled-coil domain in SerRS2 ␣ (Figure 6 ).Considering that Echinodermata r epr esent the closest sister clade to chordates (where SLIMP was not retained), it is possible that the evolution of the SLIMP / SerRS2 ␣ heterodimer was slower in deuterostomes (echinoderma ta / chorda ta), and was la ter fully re v ersed to a canonical SerRS2 structure in chor dates alone.The comparison of all metazoan SerRS2 sequences also re v eals that, in the nematode species analysed, SerRS2s also lost the coiled coil domain (Supplementary Figure S1).Interestingly, howe v er, we could not detect SLIMP in these nema tode species, indica ting tha t nema tode SerRS2 followed a different, and as yet uncharacterized, evolutionary path.
In Drosophila the evolution of heterodimeric Dm SLIMP / Dm SerRS2 ␣ led to the acquisition by this enzyme of additional interactions and regulatory functions ( 25 ), but these did not affect the aminoacylation function of the heterodimer, as evidenced by our experiments using LON domains (Figure 5 ).Our da ta indica te that a heterodimeric SerRS2 composed of SerRS2 ␣ and SLIMP is probably the established functional enzyme in all protostomes, but it remains to be determined if the non-canonical functions of Dm SLIMP / Dm SerRS2 ␣ extend to other SerRS2 enzymes in SLIMP-containing species.
Our muta tional da ta indica te tha t the tRN A reco gnition mechanism used by the single acti v e site surface of heterodimeric SerRS2 is equivalent to that used by mammalian homodimeric SerRS2 (Figures 4 C, D).This tight conservation of the tRN A reco gnition mechanism is compatible with the large structural variations in mt-tRNA Ser GCU and tRNA Ser UGA isoacceptors, as previously noted in the analysis of the tRNA recognition modes of human SerRS2 ( 17 ).Thus, although the degeneration of mitochondrial tRNA structures may be an consequence of Muller's ratchet ( 44 , 45 ), the sequence and structural drift seen in tRNA Ser preserves the interaction with SerRS2, r einfor cing the idea that tRNA molecules are fixed by their complex set of interactions and are only free to e volv e in simplified scenarios such as the mitochondria, and only in directions permitted by their remaining functional constraints ( 46 , 47 ).
The natural history of animal mitochondrial SerRS illustr ates how centr al to the evolution of this molecular system is the conservation of tRN A reco gnition and aminoacylation mechanisms, which impedes the free drift of all structures inv olved.Ho wever, peripheral regions of the system, be it the tRNA structure or the aaRS ar chitectur e, ar e under lighter selecti v e pr essur e and can e volv e as long as tRNA charging activity is not compromised.As result of this une v en selecti v e pr essur e, large variations in specific structural regions of mt-tRNA Ser and SerRS2 appeared and e volv ed to acquire new interactors and functions in invertebrates.Inter estingly, vertebrate SerRS2 r everted this evolution and returned to the ancestral homodimeric state, through a pro-cess whose understanding will likely r equir e characterizing SerRS2 in echinoderms.

DA T A A V AILABILITY
All supporting data used in this article have been deposited in a public repository, and can be found at the following DOI: https://doi.org/10.6084/m9.figshare.23560404.v1 .

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.

Figure 1 .
Figure 1. ( A ) Multiple sequence alignment of sequences of SerRS2, SerRS2 ␣, and SLIMP.Residue conservation is marked by dots and asterisks.Functionally relevant positions are shown in bold.The regions that correspond to the major domains of each protein type are represented by colored blocks under the alignment, and the sequences are color-shadowed accordingl y. ( B ) Phylo genetic anal ysis of SerRS2, SerRS2 ␣ and SLIMP sequences performed as described.Numbers at branches r epr esent the bootstrap support for each node.The deduced point of SerRS2 duplica tion tha t gave rise to SerRS2 ␣ and SLIMP is marked by a red square.The branches where SLIMP was lost are crossed by red lines to signify the gene loss (see also Supplementary Figure S1).

Figure 2 .
Figure 2. ( A ) Three-dimensional homology-based model of the Dm SLIMP / Dm SerRS2 ␣ heterodimer bound to a tRNA molecule.The Dm SLIMP subunit is shown in green, Dm SerRS2 ␣ in blue and the tRNA in red.Residues in the catalytic pocket of SerRS2 ␣, and the conserved arginine in the coiled coil domain of Dm SLIMP are displayed in CPK format.Notice the absence of a coiled-coil domain in Dm SerRS2 ␣, and the presence of a single tRNA molecule bound by the two enzyme subunits.( B ) Alignment of the sequences of the catalytic pockets of SerRS2, SerRS2 ␣ and SLIMP sequences.( C ) Secondary structur e r epr esentation of animal tRNA Ser GCU structur es illustr ating the over all conservation of these molecules.

Figure 3 .
Figure 3. ( A ) Sequence alignment of most conserved positions in the coiled-coil domains of SerRS2, SerRS2 ␣ and SLIMP sequences, showing the disappearance of these positions in SerRS2 ␣.The corresponding sequence numbering for Dm SerRS2 ␣ and Dm SLIMP are shown in the top two rows (see also Supplementary Figure S2).( B ) Predictions of coiled-coil structures in representati v e sequences of SerRS2, SerRS2 ␣ and SLIMP.

Figure 5 .
Figure 5. Aminoacylation assays of mt-tRNA Ser GCU by Dm SLIMP / Dm SerRS2 ␣ in the presence or absence of LON's substrate binding domain at two differ ent r elati v e concentrations.

Figure 6 .
Figure 6.Proposed structural and functional evolutionary history of metazoan SerRS2.