Type I acetyltransferases encompass E. coli and S. enterica Pat. These enzymes are large enzymes (>80 kDa) composed of an N-terminal domain homologous to nucleotide-diphosphate (NDP)-forming acyl-CoA ligase/synthetase (~700 aa) but with lack in activity. At the C-terminus the enzymes contain a GNAT domain (~200 aa). For the S. enterica Pat (SePat) it was shown that the N-terminal domain binds to acetyl-CoA with a potential regulatory role on GNAT activity as mutations in the N-terminal domain impaired SePat activity (Thao and Escalante-Semerena, 2012). These mutations also cause alterations in the SePat structure as studied by circular dichroism suggesting that the N-terminal domain is important for the structural integrity of SePat (Thao and Escalante-Semerena, 2012).

The type II acetyltransferases are similar in size compared to type I GNAT enzymes but their domain organization is different, i.e., the NDP-forming acyl-CoA synthetase domain is located at their C-termini (~900 aa) while the GNAT catalytic domain (~200 aa) is located in their N-termini (Figure 2B). As described for class I enzymes also in class II enzymes the NDP-forming acyl-CoA synthetase domain is inactive due to the absence of a catalytic histidine residue. For many NDP-forming acyl-CoA synthetase domain containing proteins the formation of oligomers was reported (Marchler-Bauer et al., 2015; Yang et al., 2020). The type II enzyme, PatA from Streptomyces lividans (SlPatA) contains a regulatory C-terminus with a collagen G-P-S motif that is important for its catalytic acetyltransferase activity. It is furthermore under debate if this C-terminal region is important for the observed oligomerization of SlPatA.

Although the exact role of the regulatory NDP-forming acyl-CoA synthetase domain in bacterial class I and class II enzymes is not completely understood, all data suggest that it might be important for binding and sensing of acetyl-CoA to adjust the KAT activity to the cellular metabolic state. Moreover, it might be important for structural integrity of the enzyme and/or for allosteric regulation of the GNAT catalytic activity (Thao and Escalante-Semerena, 2012). For S. enterica SePat it was shown that it binds two molecules of acetyl-CoA, one molecule with the N-terminal and another with the C-terminal domain, respectively. Mutational analyses revealed that the N-terminal NDP-forming acyl-CoA synthetase domain is essential for the catalytic activity in E. coli PatZ and SePat (Thao and Escalante-Semerena, 2011; de Diego Puente et al., 2015).

Type III protein acetyltransferases encompass enzymes that are similar in the domain organization as class I enzymes (Figure 2B). However, they do not contain a NDP-forming acyl-CoA synthetase domain but instead a smaller regulatory domain (~300–400 aa) located at their N-termini. In Mycobacterium smegmatis MsPat (MSMEG_5458) and M. tuberculosis MtPat (Rv0998), a cAMP-binding domain precedes the GNAT domain (Supplementary Figures 1A, 2). The type III cAMP-GNAT from Mycobacterium tuberculosis H37Rv was structurally characterized by X-ray crystallography. The GNAT domain shows presence of the important sequence motifs, such as a glutamate that could act as catalytic base (MSMEG_5458: E234) and the motifs important for acetyl-CoA-binding (Nambi et al., 2010; Lee et al., 2012). The cAMP-binding domain strongly resembles cAMP-binding domains present in eukaryotic proteins such as protein kinase A (PKA) and its isoforms, in phosphodiesterases (PDEs), in cyclic nucleotide-gated ion channels (CNGs) and in the small G protein Rap guanine-nucleotide exchange factor (GEF) EPAC (exchange protein activated by cAMP) (Xu et al., 2011; Jager et al., 2012; Lee et al., 2012; Gancedo, 2013; Steegborn, 2014). Moreover, it is structurally highly similar and homologous to the cAMP-binding site of the CAP (catabolite gene activator protein)/CRP (cAMP responsive protein) from E. coli suggesting that the cAMP-binding sites are evolutionary related (Supplementary Figure 1A; Weber et al., 1982). In CAP/CRP, PKA, and EPAC, binding of cAMP to the regulatory subunits or the allosteric cAMP-binding site, results in conformational changes ultimately leading to activation of protein function (Beebe, 1994; de Rooij et al., 2000; Rehmann et al., 2003a, b; Rehmann, 2006, 2017; Harper et al., 2008). In analogy, it was shown that cAMP results in allosteric activation of MsPat (MSMEG_5458) activity as detected by assessing acetylation of the identified physiological substrate protein USP (universal stress protein) in presence/absence of cAMP (Nambi et al., 2010, 2013; Xu et al., 2011; Lee et al., 2012). Notably, mutation of the catalytic glutamate E234 to alanine in M. smegmatis MsPat (MSMEG_5458) resulted in a reduction in its catalytic activity, while cAMP-binding was not affected by the mutation (Nambi et al., 2010). However, the activity was not completely abolished in the E234A mutant and could be restored upon addition of cAMP. This shows that other mechanisms than the presence of the supposed catalytic base glutamate contribute to activation of the substrate lysine for acetyl-group transfer.

Another domain identified in type III acetyltransferases is the ACT (ACT: aspartate kinase, chorismate mutase, TyrA) domain preceding the C-terminal GNAT domain (Figure 2B). ACT-GNAT acetyltransferases are so far only identified in actinomycetes (Hentchel and Escalante-Semerena, 2015). ACT domains are found in many enzymes involved in regulation of the metabolism, such as in amino acid and purine biosynthesis. ACT domains are structurally adopting a ferredoxin-like βαββαβ topology of which two pairs form an eight-stranded antiparallel β-sheet which is flanked by the four α-helices at one side (Chipman and Shaanan, 2001). These domains bind to small regulatory ligands mostly amino acids. For the PatB enzyme from Streptomyces lividans (SlPatB) and Micromonospora aurantiaca (MaPatB; Micau_1670) and other acetyltransferases in actinobacteria it was shown that binding of L-Cys, L-Arg, and/or L-Asn improved its capacity to acetylate the acetyl-CoA-synthetase (Acs) (Xu et al., 2014; Lu et al., 2017). Bioinformatics analyses suggested the presence of more than 150 potential ACT-GNAT acetyltransferases to be encoded by actinobacteria (Lu et al., 2017). Why this type of GNATs is so broadly distributed in actinobacteria needs further investigation.

Finally, another type III enzyme was identified in Myxococcus xanthus, the protein acetyltransferase MxKat. MxKat was shown to sense NADP⁺ with its N-terminal domain preceding the GNAT domain (Figure 2B and Supplementary Figures 1B, 2; Liu et al., 2015). As also observed for cAMP and amino acid sensing KATs, also in the NADP⁺ sensing KAT MxKat, binding of the ligand NADP⁺ to the N-terminal domain allosterically regulates its acetyltransferase activity (Liu et al., 2015). The primary sequence and structural modeling suggests that the NADP⁺-binding domain adopts a typical Rossmann-fold consisting of two repeats of the topology β-α-β-α-β forming a six-stranded parallel β-sheet with two α-helices on each site of the sheet (Liu et al., 2015; Supplementary Figure 1B). Interestingly, binding of NADP⁺ but not of NAD⁺, NADH or NADPH to the N-terminal Rossmann-fold domain of MxKat resulted in inhibition of the acetyltransferase activity rather than activation as observed upon ligand binding for cAMP- and ACT-GNATs (Supplementary Figures 1, 2; Liu et al., 2015).

Type IV GNATs encompass most bacterial acetyltransferases (Figure 2B). These enzymes do not contain any regulatory domain and consist almost entirely only of the GNAT domain. Examples for the class IV acetyltransferases are the recently identified novel E. coli lysine acetyltransferases RimI, YiaC, YjaB, and PhnO, Rheudopseudomonas palustris RpPat and Mycobacterium smegmatis MsPat (Christensen et al., 2018).

Notably, sometimes the classification of GNATs into type III or IV is not straightforward. As an example, the acetyltransferase AcuA in B. subtilis (BsAcuA) is encoded in the acuABC-operon, which was originally identified to be important for acetoin and butanediol metabolism (Grundy et al., 1993, 1994). AcuC is a classical Zn²⁺-dependent deacetylase with structural homology to mammalian HDACs containing all important sequence motifs important for catalytic activity. The acetyl-CoA synthetase AcsA in Bacillus subtilis (BsAcsA) is encoded in reverse orientation upstream of the acuABC-operon and BsAcuA was shown to acetylate and inactivate BsAcsA and BsAcuC to deacetylate and activate BsAcsA (Gardner et al., 2006). The exact role of BsAcuB is not known. Interestingly, structural modeling with Phyre2 reveals homologies to an ACT domain in the N-terminus and a CBS (cystathionine beta synthase)-domain in the C-terminus (Supplementary Figure 1C; Kelley et al., 2015). The CBS domain is found in enzymes binding to adenosyl-group containing molecules such as S-adenosylmethionine or ATP. To this end, BsAcuB might directly affect BsAcuA and/or BsAcuC function. This needs additional investigation. The fact that BsAcuA activity might also depend on a regulatory ACT domain, albeit not present in one polypeptide chain but supplied in trans by BsAcuB, makes a direct classification of BsAcuB in either type III or type IV difficult. Based on the domain organization the B. subtitlis GNAT BsAcuA might belong to class IV. However, if AcuB is important for AcuA activity, it could also be classified into class III, although the regulatory subunit is supplied in trans rather than in cis. Future studies are needed to clarify the role of BsAcuB for BsAcuA and/or BsAcuC function (Gardner et al., 2006).

Finally, type V GNATs encompasses acetyltransferases that contain a dual arrangement of GNAT domains as exemplified on the KAT Eis (enhanced intracellular survival) from Mycobacterium tuberculosis (Figure 2B). The two GNAT domains cover the N-terminal and central part of the enzyme and at the C-terminal region folds into a five-stranded β-sheet that is surrounded by four α-helices on one side. The crystal structure of Eis shows that it forms a hexamer in solution (Chen W. et al., 2011). The C-terminus is directly involved in catalysis. As the central GNAT domain lacks an Arg involved in binding to CoA phosphates it is likely that it does not show catalytic activity (Chen W. et al., 2011). However, the central GNAT domain might be important for the overall fold of the protein and for the oligomeric state and might therefore also indirectly be needed for full activity of the N-terminal GNAT domain (Chen W. et al., 2011). Evolutionary this domain arrangement might be developed by gene duplication. It resembles the domain organization and function of mammalian KDAC6, which also contains two catalytic domains of which only the C-terminal domain is highly active while the presence of the N-terminal domain stimulates the activity of the C-terminal domain via affecting the protein structural integrity (Zou et al., 2006; Boyault et al., 2007). Eis was shown to act as lysine acetyltransferase for mitogen-activated protein kinase-phosphatase 7 (MKP-7) and additionally as small molecule acetyltransferase acetylating and thereby inactivating aminoglycoside antibiotics. Both activities are important for suppression of the host cell response upon M. tuberculosis infection (Lella and Sharma, 2007; Zaunbrecher et al., 2009; Chen W. et al., 2011; Houghton et al., 2013).

Apart from presence of different regulatory domains all bacterial protein acetyltransferases use similar catalytic strategies to achieve acyl-group transfer to the substrate amino-group as explained in the next paragraph.

The mutation of the postulated conserved catalytic glutamate (E809) in E. coli PatZ (also known as YfiQ or Pka) and S. enterica SePat did reduce catalytic activity but it did not switch off catalytic activity completely. Furthermore, SePat E809Q was not defective in vivo suggesting a catalytic mechanism without this glutamate acting as general base (de Diego Puente et al., 2015). It was reported for SePat but not for PatZ that this E809 is important for Pat structure rather than for catalysis (Thao and Escalante-Semerena, 2012). Also for other GNAT KATs it was shown that a catalytic glutamate is either missing or not important for catalysis and it is suggested that the deprotonation of the substrate lysine is conducted by other residues such as active site histidine residues or that the positive electrostatics in direct vicinity of substrate lysine side chain lowers its pK_a value favoring deprotonation and increasing its nucleophilicity (Angus-Hill et al., 1999; Hickman et al., 1999; de Diego Puente et al., 2015). In fact, a more basic pH could restore the catalytic activity of E. coli PatZ E809A (de Diego Puente et al., 2015). Several bacterial KATs were shown to possess auto-ac(et)ylation activity. For E. coli PatZ it was reported it forms a tetramer in solution which is independent of acetyl-CoA-binding (de Diego Puente et al., 2015). However, auto-ac(et)ylation of lysine residues in the N-terminal domain and the catalytic GNAT domain induces oligomerization into an octamer increasing its catalytic activity (de Diego Puente et al., 2015). For E. coli PatZ the N-terminal domain is essential for enzymatic activity as the isolated GNAT domain was catalytically inactive (de Diego Puente et al., 2015). For the S. enterica enzyme Pat (SePat) acetyl-CoA-dependent oligomerization was reported from a monomer form to a tetramer. In analogy to E. coli PatZ also for SePat this effect was due to acetyl-CoA-binding to the N-terminal NDP-forming acyl-CoA synthetase domain (Thao and Escalante-Semerena, 2011). Also this oligomerization resulted in an increase in its activity. For both enzymes a positive cooperativity was observed for acetyl-CoA-binding resulting in an increase in acetyltransferase activity, suggesting that this might be a general mechanism observed in these bacterial class I GNAT enzymes (de Diego Puente et al., 2015). Notably, also for the class II GNAT Pat from Streptomyces lividans formation of a higher oligomer, an octamer, was observed. It is not clear if this is driven by ac(et)ylation, acyl-CoA-binding or by a different mechanism postulated, i.e., by intermolecular interactions driven by the affinity of the collagen signature G-P-S that was found in the C-terminus and as being essential for its catalytic activity (Tucker and Escalante-Semerena, 2013, 2014).

Future studies will show if and which residues are important to act as general base during catalysis of PatZ, SePat and other class I and class II enzymes. As stated above, oligomerization is an important mechanism to achieve full activity in class I and class II enzymes. To this end, oligomerization might induce conformational changes in the enzymes structure that are essential for the correct arrangement of the catalytic machinery into a competent state for catalysis or for substrate binding. To fully understand, how these enzymes achieve substrate acetylation, structural data of the full length PatZ, SePat, or SlPat will be needed.

The type III enzymes contain N-terminal domains, e.g., cAMP-binding domains or ACT-domains, that mediate oligomerization and regulate the enzymatic activity. Binding of cAMP to the N-terminal domain of Mycobacterium smegmatis MsPat was shown to increase the catalytic activity (Xu et al., 2011; Nambi et al., 2013; Podobnik et al., 2014). This was due to inducing a huge conformational change in the protein upon cAMP-binding that relives autoinhibition and replaces a pseudo-substrate sequence from the protein-substrate binding site. This mechanism reminds of regulation of PKA function and EPAC activation. For the cAMP-regulated protein acetyltransferase from Mycobacterium smegmatis MsPat MSMEG_5458 it was shown that is has a catalytic core which resembles other GNAT proteins (Podobnik et al., 2014). The suggested catalytic base E235, contributes to catalysis. However, mutation of the catalytic glutamate E235 to alanine in M. smegmatis MSMEG_5458 resulted only in a reduction in its catalytic activity. The binding to cAMP was not affected by the mutation (Nambi et al., 2010). However, the activity of MSMEG_5458 E235A was not completely abolished and was restored upon addition of cAMP. This shows again that other mechanisms than the glutamate contribute to catalysis. Mutation of further residues lining the substrate lysine binding site, such as R223 and V225, in M. tuberculosis MtPat Rv0998, also impaired the catalytic activity (Lee et al., 2012). Studies using different acyl-CoA as donor molecules for ac(et)yl transfer, such as propionyl-CoA and butyryl-CoA, showed that MtPat Rv0998 has a high degree of acyl-chain promiscuity as it can efficiently turn over diverse acyl-group donor molecules (Nambi et al., 2013). This might reflect the role of the cAMP-binding site to precisely control promiscuous MtPat Rv0998 activity dependent on the prevalence of cellular cAMP, which is a second messenger produced by adenylyl-cyclase (Lee et al., 2012). Functionally, MtPat was shown to regulate fatty acid and propionate metabolism (Nambi et al., 2013).

As stated above, ACT-domain containing GNATs were only found in actinomycetes so far. Binding of amino acids such as L-Cys, L-Arg, and/or L-Asn was shown to allosterically activate GNAT activity (Xu et al., 2014; VanDrisse and Escalante-Semerena, 2018). ACT-GNATs are involved in regulation of amino acid and purine metabolism. Structural alignment shows that the ACT-GNAT MaPatB (Micau_1670) from Micromonospora aurantiaca and the Streptomyces lividans SlPatB (EFD70633) have a glutamate at the analogous position to HspCAF/HsGcn5 suggesting that this act as catalytic base during catalysis for deprotonation and activation of the substrate amino group (Xu et al., 2014; VanDrisse and Escalante-Semerena, 2018). For both enzymes the acetyl-CoA synthetase (Acs) was shown to be a substrate (Xu et al., 2014).

In contrast, the type III enzyme Myxococcus xanthus MxPAT (Mxan_3215) that contains a Rossmann-fold domain for specific NADP⁺-binding N-terminally to the GNAT domain is negatively regulated by NADP⁺, i.e., the acetyltransferase activity is inhibited upon NADP⁺-binding (Supplementary Figure 1B). For the enzyme M. xanthus MxPAT (Mxan_3215) (Phyre2: V116/S156 are at position E570/D610 in HspCAF) no catalytic glutamate is present as shown by primary sequence alignment or by structural modeling with known catalytic base glutamate containing protein GNATs (Liu et al., 2015). This suggests that also for the NADP⁺-GNAT enzymes other strategies are employed to catalyze acetyl-group transfer. To really show how these NADP⁺ regulated acetyltransferases catalyze the acetyl transfer structural data including complexes with/without NADP⁺, substrate and acetyl-CoA/CoA is needed.

All of these type III acetyltransferases share a similar mode of action. Binding of a ligand, such as NADP⁺, amino acids or cAMP to the N-terminal domain results in allosteric regulation of the GNAT catalytic activity. The outcome can be activation of acetyltransferase activity as observed for ACT-domain containing and cAMP-binding GNATs or also inactivation of the acetyltransferase activity as observed by the NADP⁺-binding GNATs. For all of these enzymes, binding to the ligands binding site is resulting in conformational changes that affect acetyltransferase activity maybe by affecting substrate binding, by altering the electrostatics of the active site or the arrangement of the catalytic machinery in a more/less competent state for catalysis. For all of these type III enzymes, this mode of regulation allows a tight control of the GNAT activity to the availability of the ligands, which are direct indicators of the cellular metabolic state.

The class IV enzymes, RimI, YiaC, YjaB, and PhnO were recently shown to act as N-(ε)-lysine acetyltransferases using an elegant experimental system (Christensen et al., 2018). An E. coli gutted strain was engineered that carried genomic deletions in genes pta, encoding phosphotransacetylase needed for biosynthesis of acetyl-phosphate, in cobB, encoding the sirtuin deacetylase CobB, in acs, encoding acetyl-CoA synthetase and in patZ (yfiQ), encoding acetyltransferase PatZ (Christensen et al., 2018). This strain has a reduced systemic background in non-enzymatic acetylation and it should furthermore accumulate acetylation catalyzed by different novel KATs as the deacetylase CobB is absent. This enabled the identification of RimI, YjaB, YjaC, and PhnO as novel protein lysine acetyltransferases. To validate the activities catalytic mutants were created. These mutants were selected based on structural models (Christensen et al., 2018). For switching off catalytic activity, either the supposed general base glutamate in PhnO, or a tyrosine residue present in all four KATs, was selected that is supposed to act as general acid to protonate the CoA sulfhydryl group to resolve the tetrahedral reaction intermediate (Figure 2D). For PhnO mutation of either the suspected catalytic glutamate (E78A) or the tyrosine (Y128A) completely abolished systemic acetylation of whole cell lysate. For RimI, YiaC, and YjaB the observed systemic acetylation activity was abolished with mutation of the postulated general acid tyrosine (Christensen et al., 2018). Structural analyses using the experimentally determined structures of RimI and YjaB and models obtained with Phyre2 show that these GNATs are structurally very similar (Figure 3A). The hydroxyl groups of a tyrosine (YjaB: Y117, RimI/YjaC: Y115, PhnO: Y128) points toward the acetyl-group of acetyl-CoA substrate and might therefore have a role for substrate binding or during catalysis as general acid as suggested (Christensen et al., 2018). To finally judge this, a structure of the ternary complex consisting of GNAT, substrate and acetyl-CoA would be desirable. This Tyr is replaced by Phe in HsGcn5 and HspCAF. An acidic residue acting as general base cannot be identified in the novel KATs RimI and YiaC (RimI: T65; YiaC: A72) suggesting that another catalytic mechanism is used by these protein GNATs maybe involve the acidic active site electrostatics, other residues or a catalytic water molecule acting as nucleophile for catalysis (Figures 3B,C). The Thr65 in RimI could act as remote base by activating a water molecule for catalysis. A similar mechanism has been observed for GNATs involving a serine residue (Figure 2C; Ud-Din et al., 2015). For the KATs PhnO and YjaB an acidic residue that would have the capacity to act as general base is present as found by alignment of their structures and their primary sequences with HspCAF and HsGcn5 (YjaB: D74, PhnO: E78). Notably, although RimI and YjaB are dimeric, their active sites are at the opposite site of the dimer interface suggesting that oligomerization does not directly influence their activity. However, for human GNAT enzymes presence in multi-protein complexes was shown which also affects substrate specificity and catalytic activity. If these novel KATs are present in complexes with other proteins and if this affects the catalytic activity and/or substrate specificity needs further investigation. Moreover, also substrate binding might result in an arrangement of the catalytic residues into a competent state for catalysis in an induced fit mechanism as was also proposed for human HAT1 (Wu et al., 2012).

Recently, a novel catalytic mechanism was reported for the P. aeruginosa PA3944 GNAT, which shows activity toward polymyxin antibiotics (Baumgartner et al., 2021). This catalytic mechanism involves a catalytic serine residue directly acting as nucleophile resulting in the formation of a covalent acyl-enzyme intermediate (Figure 2C). It was shown that the glutamate originally regarded as catalytic base plays a role in substrate recognition or stabilization (Baumgartner et al., 2021). If this mechanism also applies to certain GNAT protein acetyltransferases needs further investigation.

For the enzyme Eis of Mycobacterium tuberculosis (MtEis), belonging to type V GNATs, the catalytic mechanism to acetylate aminoglycosides was analyzed biochemically and structurally (Chen W. et al., 2011; Houghton et al., 2013). Eis forms a hexamer in solution and only the N-terminal GNAT domain is active as shown by mutational studies (Chen W. et al., 2011; Houghton et al., 2013). As shown for the NADP⁺-GNATs, also Eis does not use a catalytic glutamate for catalysis. Instead MtEis activates a catalytic water, the hydroxide acting as general base, by the C-terminal α-carboxyl group of Phe402 acting as remote base (Chen W. et al., 2011). This water molecule is coordinated by His119 via its main chain amide, and it furthermore orients the aminoglycosides amino group via its main chain carbonyl group. Catalysis proceeds via formation of a tetrahedral intermediate which is resolved by a tyrosine, Tyr126 in MtEis, acting as general acid protonating the CoA sulfhydryl group for collapse of the intermediate (Figure 2D; Chen W. et al., 2011). Notably, the substrate binding cavity is highly negatively charged to allow an efficient electrostatic attraction of the positively charged (poly)amine group containing aminoglycosides at physiological pH.

CobB was originally identified in S. typhimurium LT2 to act in cobalamin biosynthesis and propionate catabolism (Tsang and Escalante-Semerena, 1996). It was shown that CobB possesses weak ADP-ribosyltransferase activity and later CobB was found to have robust NAD⁺-dependent deacetylase activity for acetyl-CoA synthetase in S. enterica (Frye, 1999; Imai et al., 2000; Landry et al., 2000; Smith et al., 2000; Starai et al., 2002). Sirtuins were first described in budding yeast Saccharomyces cerevisiae and it was shown that the yeast SIR2 had a strong impact on replicative lifespan (Kaeberlein et al., 1999). Later, it was shown that SIR2 is an NAD⁺-dependent lysine deacetylase (Imai et al., 2000). In mammals, sirtuins are classified into class III protein deacetylases, while classes I, II and IV are the classical Zn²⁺-dependent lysine deacetylases (KDACs). As KDACs were initially identified as histone deacetylases they were originally named as histone deacetylases (HDACs). Humans encode seven sirtuins that are categorized based on phylogenetic analyses. Eukaryotic sirtuins within class III of lysine deacetylases were classified in sirtuin subclasses I-IV (Olesen et al., 2018). Subclass Ia contains SIRT1, that has a robust deacetylase activity and it primarily located in the nucleus, subclass Ib encompasses the robust lysine deacetylases SIRT2 (cytosolic) and SIRT3 (mitochondrial). The other subclasses show catalytic activities with different acyl-chain preferences (Blasl et al., 2021). The subclass II enzyme SIRT4 is mitochondrial and it was shown to possess weak deacetylase activity but removes longer acyl-chains from lysine side chains and it has mono-ADP-ribosyltransferase (MAR) activity (Frye, 1999; Ahuja et al., 2007). The subclass III enzyme SIRT5, which is also localized to the mitochondrial matrix, has efficient lysine deacylase activity toward negatively charged acylations such as malonyl-, glutaryl-, and succinyl-groups. Subclass IV is divided in the subclass IVa enzyme SIRT6, localized in the nucleus with preferences for longer fatty acylated substrates and with reported MAR activity, and the subclass IVb enzyme SIRT7, which enriched in the nucleoli with activities as deacetylase, desuccinylase, and activity toward longer acyl chains (Liszt et al., 2005).

Phylogenetic analyses revealed that all archaeal sirtuins belong to subclasses III and U (U: undifferentiated) and the bacterial sirtuins are classified into subclasses II, III, M (M: macrodomain-linked), U and the subclass of Sir2-like sirtuins (Olesen et al., 2018). Most prokaryotic sirtuins belong to the mammalian sirtuin subclasses II and III, with subclass III containing the far most prokaryotic enzymes (Frye, 2000). Further bacterial sirtuins belong to subclass M, subclass U and to subclass of Sir2-like enzymes (Frye, 2000; Olesen et al., 2018). Subclass III genes are found in nearly all Gram-negative bacteria and in archaea suggesting that it either developed prior to the divergence of the domains during evolution or that the gene was transferred by lateral transfer from bacteria to archaea or vice versa (Frye, 2000). So far, no prokaryotic subclass I enzyme could be identified (Frye, 2000). The fifth subclass, subclass U, is found in several firmicutes, i.e., Gram-positive bacteria, such as B. subtilis, Staphylococcus aureus, and in the Gram-negative species Thermotoga maritima (Frye, 2000). These enzymes were classified into the separate subclass U as these enzymes show sequence motifs that are in between subclasses II/III and I/IV (Frye, 2000). The B. subtilis sirtuin SrtN was phylogenetically analyzed to also belong to subclass U (Gardner and Escalante-Semerena, 2009; Greiss and Gartner, 2009; Olesen et al., 2018). The catalytic activity of SrtN is similar to mammalian SIRT4 acting as lipoamidase (Ahuja et al., 2007; Anderson et al., 2017a; Rowland et al., 2017). Both, E. coli CobB and B. subtilis SrtN are active as lipoamidase affecting the lipoylation levels and activities of pyruvate dehydrogenase and α-ketoglutarate dehydrogenase as also reported for mammalian SIRT4 (Mathias et al., 2014; Rowland et al., 2017). Recently, the enzyme Sir2La from Lactobacillus acidophilus NCFM was identified as first subclass U sirtuin with dual activity as efficient deacylase, debutyrylase and depropionylase (Olesen et al., 2018).

The macrodomain-linked sirtuins SirTMs (subclass M) were found to be particularly present in pathogenic organisms such as pathogenic bacterial families including Chlostridiaceae, Enterococcaceae, Lachnospiraceae, Spirochaetaceae, and Veillonellaceae and in diverse pathogenic fungal families (Rack et al., 2015). They are encoded in an operon together with a macrodomain and in bacterial genera Lactobacillus, Staphylococcus and Streptococcus additionally with GcvH-L (glycine cleavage system H-like) and with LplA2 (lipoate protein ligase A).

The E. coli enzyme CobB shows homologies and similar substrate preferences as mammalian SIRT5 (Du et al., 2011; Fischer et al., 2012). In analogy, it was reported that CobB possesses a robust deacetylase, de-2-hydroxyisobutyrylase, desuccinylase and demalonylase activity (Figure 1C; Peng et al., 2011; Colak et al., 2013; Dong et al., 2019). E. coli CobB was shown to act as deacetylase for acetyl-CoA synthetase (Acs) resulting in activation of Asc activity (Zhao et al., 2004). However, in contrast to SIRT5, which only possesses weak deacetylase activity, CobB shows comparable deacetylase and desuccinylase efficiencies (Zhao et al., 2004; Colak et al., 2013). Recently, CobB was shown be an efficient lipoamidase in analogy to mammalian SIRT4 acting on important lipoylated metabolic complexes, such as the α-ketoglutarate dehydrogenase (KDH) complex, the pyruvate dehydrogenase (PDH) complex and the glycine cleavage (GCV) complex and as de-2-hydroxyisobutyrylase modulating enolase activity (Rowland et al., 2017; Dong et al., 2019). Furthermore, CobB was shown to act as de-homocysteinylase and it is supposed to revert protein lysine propionylation (Mei et al., 2016; Sun et al., 2016).

SirTMs do not possess any deacylase activity, which is most likely due to the absence of the catalytic histidine residue (Rack et al., 2015; Olesen et al., 2018). The SirTMs from the Gram-positive bacterial pathogens Staphylococcus aureus and Streptococcus pyogenes possess an mono-ADP-ribosyltransferase (MAR) activity toward the also in the operon encoded protein GcvH-L. This activity is dependent on prior lipoylation catalyzed by LplA2 (Rack et al., 2015). The operon encoded macrodomains, such as YmdB from E. coli, are able to reverse the MARylation acting as MAR hydrolase (Rack et al., 2015). For some of these operon encoded macrodomains, such as YmdB from E. coli and MacroD from S. aureus, an activity as O-acetyl-ADP-ribose deacetylase was reported, resulting in formation of ADP-ribose and acetate (Supplementary Figure 2; Chen D. et al., 2011). Thereby, this constitutes a system to regulate the cellular levels of O-acetyl-ADP-ribose that is formed during sirtuin-catalyzed deacetylation. In mammalian cells ADP-ribose was shown to act as a second messenger amongst others acting on ADP-ribose-gated calcium channels (Russo et al., 1998; Perraud et al., 2001; Borra et al., 2002). What exactly the physiological role of the O-acetyl-ADP-ribose and/or ADP-ribose is in bacterial cells, if these act as signaling molecules or second messengers, needs further investigation. One possibility is that ADP-ribose is the physiologically active second messenger formed by deacetylation of O-acetyl-ADP-ribose originating from sirtuin catalysis as postulated for eukaryotes (Borra et al., 2002; Chen D. et al., 2011). Physiologically, SirTMs were shown to be involved in reactive-oxygen species (ROS) stress response in bacteria (Rack et al., 2015). If these operon encoded SirTMs have additional substrates needs further investigation.

Acetyl-CoA is the central metabolite and a second messenger, integrating all main metabolic pathways involved in carbohydrate, fatty acid and protein metabolism (Figure 5A). Acetyl-CoA is produced under aerobic conditions in carbohydrate metabolism by oxidation of glucose through glycolysis and subsequent oxidative carboxylation of pyruvate by the pyruvate dehydrogenase (PDH) complex. Under anaerobic conditions bacteria can also convert pyruvate to acetyl-CoA by pyruvate-formate lyase (PFL) or by pyruvate-ferredoxin-oxidoreductase (PFO). Moreover, acetyl-CoA is formed in β-oxidation of fatty acids and by decomposition of ketogenic amino acids either directly (Leu, Ile, Trp) or indirectly via formation of acetoacetyl-CoA and subsequent conversion to acetyl-CoA by thiolase (Leu, Lys, Phe, Trp, Tyr). Most amino acids are glucogenic and can be converted to glucose via gluconeogenesis, which can be metabolized to indirectly yield acetyl-CoA (Figures 5A,B; Krivoruchko et al., 2015). The amount of product formed can only be modified by altering the concentrations of the reaction partners or the products, while the equilibrium of the reaction is not affected. Acetyl-CoA can also be produced by acetyl-CoA synthetase (Acs) in an irreversible ATP-consuming reaction that is composed of two half reactions: (1) formation of acetyl-adenylate (acetyl-AMP) from acetate and ATP, (2) reaction of acetyl-AMP with coenzyme A (CoASH) to form acetyl-CoA and AMP (Figure 5B; Starai et al., 2002; Wolfe, 2005; Krivoruchko et al., 2015). Next to this Acs a second acetyl-CoA synthetase is reported in some archaea such as Pyrococcus furiosus and eukaryotic protists such as Entamoeba histolytica that coverts acetate and CoASH in a single step to acetyl-CoA (Reeves et al., 1977). A reversible reaction to produce acetyl-CoA is via conversion of acetate and phosphate derived from ATP to yield the high-energy metabolite acetyl-phosphate (AP) and ADP by acetate kinase (AK) and conversion of AP to acetyl-CoA by phosphotransacetylase (Pta) (Figure 5B). Notably, this pathway is important for acetyl-CoA production in cells under conditions of carbon overflow in which high cellular acetate concentrations (>5 mM) are accumulating (Krivoruchko et al., 2015; Schilling et al., 2015). In contrast, under conditions of low acetate concentrations (~0.2 mM) acetyl-CoA synthetase (Acs E. coli: K_M 0.2 mM; Acs S. enterica: 6 mM; Acs M. tuberculosis: 1.2 mM) produces acetyl-CoA for anabolic reactions (Figure 5B; Brown et al., 1977; Kumari et al., 1995, 2000; Reger et al., 2007; Li et al., 2011). Notably, it is surprising that the K_M-values toward acetate reported for Acs of different bacterial species vary strongly considering the high degree of sequence identity E. coli: 0.2 mM; S. enterica: 6 mM; M. tuberculosis: 1.2 mM) (Brown et al., 1977; Reger et al., 2007; Li et al., 2011). Finally, acetyl-CoA is produced by ATP-citrate lyase in reductive tricarboxylic acid cycle (reverse TCA cycle) that is used by some autotrophic bacteria such as Chlorobium species for carbon dioxide assimilation (Wahlund and Tabita, 1997; Kanao et al., 2001, 2002; Hugler and Sievert, 2011). In eukaryotes ATP-citrate lyase catalyzes the conversion of citrate, that is transported from the mitochondrial matrix into the cytosol, and CoASH to form oxaloacetate and acetyl-CoA in the cytosol (Wei et al., 2020). Besides from these main routes of acetyl-CoA production, other specific routes were described in specific bacterial strains. As an example, an alternative acetate-driven TCA cycle was shown to exist in diverse bacterial species that live as symbionts in animals and insects. In this cycle the acetate-succinate-CoA transferase (ASCT) replaces succinyl-CoA synthetase resulting in formation of acetyl-CoA and succinate from succinyl-CoA and acetate (Kwong et al., 2017).

In mammals, KATs were shown to use different acyl-CoA molecules as donors for protein acetylation, such as propionyl-CoA, butyryl-CoA, malonyl-CoA, hydroxyisobutyryl-CoA, and crotonyl-CoA (Figure 1C). Next to acetyl-CoA, GNATs in eukaryotes and prokaryotes were reported to also catalyze acylation using different acyl-CoA molecules as donor molecules for acyl group transfer such as butyryl-CoA and propionyl-CoA (Figure 1C). As an example, SePat uses acetyl-CoA to acetylate acetyl-CoA synthetase (Acs) and propionyl-CoA to propionylate propionyl-CoA synthetase (PrpE). In both cases acylation in the C-terminus of acyl-CoA synthetase results in its inactivation (Garrity et al., 2007; Takenoya et al., 2010). Recently, crotonylation was identified as widespread acylation in Streptomyces roseosporus (Sun et al., 2020). The GNAT acyl-transferase Kct1 from S. roseosporus was identified as efficient crotonyl-transferase for the glucose kinase converting glucose to glucose 6-phosphate resulting in inhibition of its catalytic activity (Sun et al., 2020).

So far bacterial protein acetyltransferases were not systematically analyzed concerning their acyl-chain preferences and future studies are needed to elucidate the activities and specificities for diverse bacterial acetyltransferases. These data will unravel the physiological roles of the respective GNATs and will furthermore show how substrate specificity is created for the diverse GNAT types. In bacteria diverse other acyl-CoA molecules apart from acetyl-CoA are produced in metabolism (Figure 1C). Malonyl-CoA can be produced from acetyl-CoA by carboxylation and under consumption of ATP by the biotin-dependent enzyme acetyl-CoA carboxylase (Tong, 2005; Polyak et al., 2012). Propionyl-CoA is produced upon degradation of branched chain amino acids (Ile, Thr, Val) and Met, odd-chain fatty acids and by the oxidation of the side chain of cholesterol (Kazakov et al., 2009; Wilburn et al., 2018). Propionyl-CoA can be converted to succinyl-CoA by the enzymes propionyl-CoA carboxylase (PCC), methylmalonyl-CoA epimerase (MCEE), and methylmalonyl-CoA mutase (MCM) (Haller et al., 2000). Succinyl-CoA is also an intermediate of the tricarboxylic acid (TCA)-cycle formed by oxidative decarboxylation of α-ketoglutarate by α-ketoglutarate dehydrogenase (KDH) (Thomas, 1974; Wolodko et al., 1986). Moreover, in many bacteria succinyl-CoA is formed in the reductive TCA cycle from succinate by succinyl-CoA synthetase (succinate-CoA ligase) in an ATP/GTP-dependent reaction (Jenkins and Weitzman, 1986). Crotonyl-CoA can be produced by acyl-CoA synthetase from the short-chain fatty acid crotonate. Moreover, crotonyl-CoA is formed during fermentation of butyrate, in fatty acid synthesis from β-hydroxybutyryl-CoA and also during degradation of the amino acids lysine and tryptophan. Some bacteria use the ethylmalonyl-CoA pathway for the assimilation of acetate. During the course of this pathway crotonyl-CoA is formed by condensation of two molecules acetyl-CoA catalyzed by β-ketothiolase and subsequent NADH-dependent reduction of the reaction product acetoacetyl-CoA to crotonyl-CoA by acetoacetyl-CoA reductase (Erb et al., 2009; Schink, 2009).

The presence and intracellular concentrations of these acyl-CoA molecules are direct indicators for the cellular metabolic state. A systematic evaluation of cellular concentrations of various acyl-CoAs in different growth phases and under different physiological conditions has not been performed in bacteria so far. This in combination with a thorough biochemical characterization of the enzymes is needed to judge the physiological importance of GNAT-catalyzed acyl-transfer. Some data on concentrations of acyl-CoA and CoA in bacteria is available. In E. coli, the concentrations of acetyl-CoA and malonyl-CoA were reported to fluctuate between 200–600 μM and 4–90 μM, respectively (Chohnan et al., 1997, 1998; Hou et al., 2015). Acetyl-CoA concentrations were highest during exponential growth phase and they declined during stationary growth phase. This shows that the acetyl-CoA level directly follows the available glucose levels. If other carbon sources are used, such as acetate and glycerol, the overall cellular acyl-CoA levels were strongly reduced (Takamura and Nomura, 1988). For the bacterial GNAT acetyltransferases K_M values of the GNAT domain for acetyl-CoA were reported that are in the micromolar range suggesting that the GNAT activities can directly and precisely be adjusted to alterations and availability of cellular acetyl-CoA concentrations and/or the cellular GNAT protein level (Nambi et al., 2010; Xu et al., 2014).

Studies in E. coli grown on different carbon sources showed that the overall acetyl-CoA level always exceeded the cellular malonyl-CoA levels. Furthermore, acetyl-CoA production increased strongly using various monosaccharides as carbon source such as D-glucose, D-mannose and D-fructose, while almost no acetyl-CoA production was induced using either succinate or acetate as carbon source. Under these conditions the cellular concentration of CoA exceeded the concentrations of acetyl-CoA/malonyl-CoA approximately 2-fold/20-fold (Takamura and Nomura, 1988; Chohnan et al., 1997, 1998). This ensures that the GNAT enzymes are not strongly active under these conditions. For the mammalian KATs, it was reported that the acetyl-CoA/CoA ratio is an important regulatory mechanism as affinities of KATs for acetyl-CoA and CoA are in the same order of magnitude. This suggests a possibility for a feedback regulatory mechanism of KAT activity by product inhibition due to increase in CoA concentration (Denisov and Sligar, 2012; Henry et al., 2015; Pietrocola et al., 2015). For exponentially growing E. coli cells intracellular concentrations of acetyl-CoA and CoA of 0.41 and 1.4 mM were reported (Bennett et al., 2009). This fluctuates dependent on the metabolic state and the nutrient availability. A similar regulatory mechanism likely exists for the activity of bacterial GNATs by the intracellular acetyl-CoA/CoA ratio rather than the concentration of acetyl-CoA alone.

Intracellular concentrations of propionyl-CoA were shown to be in the range of 40–200 μM in Haloferax mediterranii, but this might fluctuate dependent on the bacterial species and the physiological state (Hou et al., 2015).

It was shown that that under conditions in which intracellular acetyl-CoA levels are diminished, lysine succinylation becomes more prevalent showing that the availability of various acyl-CoA types can affect overall acylation with diverse acylations (Kosono et al., 2015; Mizuno et al., 2016). All acyl-CoA molecules are reactive thioesters that have the potential to modify lysine side chains either non-enzymatically or maybe also enzymatically. This needs further future investigation how diverse lysine acylations regulate protein function and how these protein acylations are regulated in bacteria.

For many mammalian KATs it was shown that the activities decreased with acyl-CoA chain length (Simithy et al., 2017). This means that acetyl-CoA is the preferred donor molecule for acetylation if it is present at least in equimolar concentrations in cells. If this is true also for the bacterial enzymes or if some bacterial GNAT enzymes have been evolved toward preferences for longer acyl-chain lengths needs to be evaluated in the future.

Several bacterial GNATs are regulated by binding of ligands such as acetyl-CoA, cAMP, amino acids and NADP⁺ to regulatory domains N- or C-terminal to the catalytic GNAT domain. as reported above. All of these ligands are themselves direct indicators for the cellular physiological and metabolic state (Figure 5A).

GNATs of type I and II use their N- or C-terminal NDP-forming acyl-CoA synthetase domain as a sensor for cellular ac(et)yl-CoA (Thao and Escalante-Semerena, 2012; de Diego Puente et al., 2015). For S. enterica SePat it was shown that it binds two molecules of ac(et)yl-CoA with the N-terminal regulatory and the C-terminal GNAT domain, respectively. Mutational analyses revealed that the N-terminal domain is essential for the catalytic activity in E. coli PatZ and SePat (Thao and Escalante-Semerena, 2011; de Diego Puente et al., 2015). For SePat it was shown that the N-terminal domain binds ac(et)yl-CoA with a nanomolar affinity (K_D: 290 nM), while binding to the GNAT domain was approximately one order of magnitude lower (K_D: 2.4 μM). This shows that the regulatory NDP-forming acyl-CoA synthetase domain in type I and type II GNATs might reflect a metabolic sensory domain that allows a tight control of the enzymatic activity avoiding a background activity under conditions of low cellular acetyl-CoA concentrations.

GNATs of type III are able to sense amino acids cysteine, arginine and asparagine, via their N-terminal ACT domains or cellular cAMP levels via their N-terminal cAMP-binding domains (Supplementary Figure 1A; Nambi et al., 2010; Shimada et al., 2011; Xu et al., 2011, 2014; Gancedo, 2013). Binding of these ligands was shown to stimulate their GNAT activity via an allosteric mechanism (Nambi et al., 2010; Xu et al., 2014). For the ACT-domain containing ACT-GNAT MaKat from Micromonospora aurantiaca (Micau_1670) affinities toward Cys and Arg were reported in the micromolar range (80–210 μM) similar to their reported intracellular cytosolic concentrations (E. coli 100–200 μM) (Park and Imlay, 2003; Caldara et al., 2008; Xu et al., 2014). The cyclic nucleotide cAMP is an important second messenger in mammalian cells and in bacteria. It is formed by adenylyl cyclase and can directly mediate signal transduction processes. In bacteria cAMP is generated by adenylate cyclase if cellular glucose uptake by the phosphotransferase system declines (Supplementary Figure 2). If the cell is energy deprived, cAMP accumulates in the bacterial cytosol (Saier, 1996a, b; Saier and Crasnier, 1996; Saier and Ramseier, 1996; Ye and Saier, 1996; Lin et al., 2004). For M. smegmatis intracellular concentrations of cAMP in the range of 100 μM–1 mM were reported (Dass et al., 2008). In glucose-fed E. coli cells the cAMP concentration is 35 μM, which increased to 146 μM with growth on acetate as carbon source (Bennett et al., 2009). The affinity of the cAMP-binding domain of the M. smegmatis cAMP-GNAT MSMEG_5458 was reported to be in the range of 100 nM suggesting complete saturation under these intracellular concentrations (Nambi et al., 2010; Lee et al., 2012). While binding of cAMP to MSMEG_5458 only resulted in a moderate, 3-fold increase in the GNAT activity for the M. tuberculosis cAMP-GNAT enzyme Rv0998 a strong increase in the acetyltransferase activity was shown upon cAMP-binding (Nambi et al., 2010, 2013). This cAMP accumulation results in activation of metabolic routes that allow utilization of alternative carbon sources (Shimada et al., 2011; Gancedo, 2013). Notably, under persistent conditions of glucose deprivation and starvation under which also the cellular ATP levels decrease drastically a decline of the cellular cAMP levels is observed as adenylyl cyclase uses ATP for production of cAMP (Nambi et al., 2010). As an example cAMP is used in bacteria to bind and to activate an important transcription factor, CAP/CRP, which modulates the expression of more than 100 genes (E. coli K12: 378 genes) (Bruckner and Titgemeyer, 2002; Shimada et al., 2011; Molina-Quiroz et al., 2018). Binding of cAMP to CAP/CRP induces a conformational change that allows to bind to the target DNA promotor sites which in turn recruits the RNA polymerase holoenzyme to initiate or repress transcription of target genes. CAP/CRP is a major regulator for metabolic adaptation to various carbon sources in bacteria. In E. coli it was shown that during growth on glucose upon entry into the stationary growth phase the transcription of the acetyltransferase pat gene was initiated by binding of cAMP to CAP/CRP. Alternatively, growth on acetate induces pat-expression in E. coli via the cAMP-CAP route. Generally, cAMP is the major regulator for regulation of carbon flux in glucose limited cultures in E. coli.

In contrast, some GNATs, such as cAMP-GNAT MtPat (Rv0998), are shown to be directly regulated allosterically by cAMP on the post-translational level. In both cases, activation of GNAT activity by rise in the cellular cAMP level results in inactivation of acetyl-CoA synthetase (Acs) activity through acetylation of a lysine residue in the C-terminus (E. coli K609. M. tuberculosis: K617, S. enterica: K609). Furthermore, for Salmonella enterica acetyl-CoA synthetase SeAcs it was shown that cAMP can act as direct competitive inhibitor by binding to its ATP/AMP pocket (Han X. et al., 2017). To this end, cAMP impairs Acs activity by applying three mechanisms: acs activity is reduced by inducing gnat expression via activation of cAMP-CAP (transcriptional regulation), directly by binding to and allosteric activation of cAMP-GNAT activity resulting to Acs acetylation (post-translational regulation) and its inactivation and by directly binding to the active site and competitively inhibiting Acs. These three mechanisms ultimately result in inhibition of Acs activity and therefore to a reduced cellular acetyl-CoA level. While for E. coli Acs, the K_M value toward acetate is reported to be 0.2 mM, the K_M value of SeAcs for acetate is in the millimolar range suggesting that Acs activity is important for generation of acetyl-CoA under conditions of high cellular acetate concentrations and low cAMP levels generated if glucose deprivation persists and cellular ATP level declines (Brown et al., 1977; Reger et al., 2007; Han X. et al., 2017).

Type III GNAT Mxan_3215 (MxPAT) was shown to be negatively regulated by binding to the coenzyme NADP⁺ (Supplementary Figures 1B, 2; Liu et al., 2015). The affinity of MxPAT to NADP⁺ was shown to be 2.9 μM in the same range of intracellular concentrations reported for E. coli (0.14–31.1 μM) suggesting that the activity of the NADP⁺-GNAT can be modified under physiological conditions by fluctuations in the intracellular NADP⁺ concentration (Bennett et al., 2009; Liu et al., 2015).

The type V GNATs Eis and Eis2 from M. tuberculosis and M. abscessus, respectively, were structurally characterized (Chen W. et al., 2011; Kim et al., 2012; Ung et al., 2019). Mycobacterium Eis GNATs were shown to act as aminoglycoside acetyltransferases and protein acetyltransferases acetylating proteins needed for host cell infection (Kim et al., 2012; Kumar et al., 2012; Houghton et al., 2013; Ung et al., 2019). The structures revealed that only the N-terminal GNAT domain crystallized with CoA or acetyl-CoA suggesting that either the affinity of the central GNAT domain for CoA and acetyl-CoA is very low or it is in an incompetent conformation for binding (Supplementary Figure 2; Chen W. et al., 2011; Houghton et al., 2013; Ung et al., 2019). To this end, most likely binding of acetyl-CoA/CoA to the central domain in Eis does not play a regulatory role under physiological conditions.

All sirtuins use an NAD⁺-dependent catalytic activity to either catalyze lysine deacylation or weak ADP-ribosylation (Supplementary Figure 2 and Figure 5A). The K_M values of sirtuins for NAD⁺ were reported to reside in the micromolar range (~30–900 μM) (Hirschey et al., 2011; Feldman et al., 2015; Olesen et al., 2018). For mammalian sirtuins the affinities for NAD⁺ were shown to depend on presence of the substrate, the identity of substrate and on the substrate concentration (Fischer et al., 2012). It is likely that a similar influence can also be observed for bacterial sirtuins and that K_M-values/affinities for NAD⁺ are similar in eukaryotic and prokaryotic sirtuins. The observed NAD⁺ K_M-values/affinities are within the range of intracellular NAD⁺ concentrations (E. coli: up to 2.6 mM; L. acidophilus: 0.2–5 mM) suggesting that the activities of bacterial enzymes are regulated by the availability of intracellular NAD⁺ (Bennett et al., 2009; Olesen et al., 2018). For the mammalian sirtuins, binding to NADH was shown to be substantially weaker and the intracellular concentrations of free NADH are much lower compared to NAD⁺ (Schmidt et al., 2004; Cambronne et al., 2016; Madsen et al., 2016; Anderson et al., 2017b). This was also confirmed in bacteria. E. coli and L. acidophilus showed an intracellular NADH concentration of 83 μM or it was almost undetectable, respectively (Bennett et al., 2009; Olesen et al., 2018). In analogy to the mammalian sirtuins, this suggests that the activity of bacterial sirtuins is regulated by the cellular NAD⁺ concentration rather than the NAD⁺/NADH ratio (Figure 5A; Schmidt et al., 2004; Madsen et al., 2016; Anderson et al., 2017b).

Mammalian sirtuins are mostly composed of the catalytic domain and additional N- and C-terminal extensions of various lengths. These additional regions were shown to affect sirtuin deacylase activity as autoregulatory domains, as binding sites for allosteric regulators such as small molecules or nucleic acids, as regions important for their subcellular localization or as regions of post-translational modifications regulating catalytic activity (North and Verdin, 2007; Han et al., 2008; Pandithage et al., 2008; Li et al., 2015; Tong et al., 2016, 2017; Huang et al., 2018; Shang et al., 2020). Most bacterial sirtuins do not contain extensive regions apart from their catalytic domain. However, CobB was shown to exist in two isoforms, a long (CobB_L) and an N-terminally by 37 residues shorter short isoform (CobB_S) (Tucker and Escalante-Semerena, 2010; Umehara et al., 2018). CobB_L was shown to be N-terminally acetylated by the acetyltransferase YiaC/NatA impairing CobB_L’s catalytic activity, shown to have a dual activity acetylating N-α- and N-ε-amino groups in proteins (Parks and Escalante-Semerena, 2020). The N-terminal region was shown to be important for binding to the intracellular signaling molecule c-di-GMP (Figure 5A and Supplementary Figure 2). Binding of c-di-GMP to CobB_L resulted in increased acetylation of known CobB substrates such as acetyl-CoA synthetase, CheY and NhoA (Xu et al., 2019). Binding constants were in the low micromolar range (K_D: 4.7 μM), which is in in the same range as intracellular c-diGMP concentrations in bacteria in the sub- to low micromolar range (Simm et al., 2004; Romling et al., 2013; Xu et al., 2019). Again, this shows that enzymatic activity can be modulated by dynamic changes in the cellular concentration of the regulator under physiological conditions.

Based on sequence, structure and function mammalian KDACs are classified into the classes I, II, and IV, while sirtuins build class III (Blasl et al., 2021). All mammalian KDACs contain a conserved catalytic domain of approximately 300–400 amino acids. Only KDAC6 contains two catalytic domains. Mammalian KDACs of class I (nuclear localization) are related to S. cerevisiae Rpd3 (reduced potassium dependency 3) and encompass KDAC1, 2, 3, and 8. Class II enzymes are related to yeast Hda1 (histone deacetylase 1). Class II is subdivided in subclass IIa including KDAC4, 5, 7, and 9 and subclass IIb with KDAC6 and KDAC10. Class IIb enzymes contain an N-terminal extension preceding the catalytic domain. For all members of class II it was shown that they shuttle between the nucleus and the cytosol. Class IV contains only a single enzyme, KDAC11, which was shown to localize to the nucleus. This classification was done based on structure, function and catalytic mechanism (de Ruijter et al., 2003). Recent mass-spectrometric data showed the presence of thousands of acetylation sites in all kingdoms of life, in proteins covering all essential cellular functions and in all cellular compartments (Choudhary et al., 2009, 2014; Lundby et al., 2012; Hansen et al., 2019). Mammalian KDACs are present in multiprotein-repressor complexes that determine their substrate specificity and the enzymatic activity. Several potent and selective mammalian KDAC-inhibitors are in clinical trials or are approved by the FDA (Food and Drug Administration) (Nebbioso et al., 2017; Lopez et al., 2018; Beyer et al., 2019). KDAC-inhibitors were shown to be neuroprotective in models of neurodegeneration (Ziemka-Nalecz et al., 2018). Classical KDAC inhibitors can be classified according to their chemical structures in hydroxamates, such as trichostatin A (TSA) or SAHA (vorinostat), short-chain fatty acids, such as butyrate, cyclic peptides or benzamides, such as MS-275 (etinostat) (Supplementary Figure 2; Yoshida et al., 2017; Zhang et al., 2018). Compounds were developed for classical KDACs that are very potent and selective with IC₅₀ values in the low nanomolar range (Li and Seto, 2016; Pasyukova and Vaiserman, 2017; Biran et al., 2018; Bourguet et al., 2018; Li et al., 2018; McIntyre et al., 2019). Hydroxamates, such as suberoylanilide hydroxamic acid (SAHA) and trichostatin A are pan inhibitors inhibiting almost all classical KDACs to a similar extend (Supplementary Figure 2). These inhibitors chelate the active-site Zn²⁺-ion, and replace the active site water molecule needed for nucleophilic attack of the electrophilic carbonyl carbon of the acetyl-group (Lombardi et al., 2011b; Yoshida et al., 2017; Zhang et al., 2018). Future studies are needed to show if bacterial KDACs are also inhibited using the inhibitors developed for mammalian enzymes or if these can be used in a drug-repurposing strategy.

During catalysis of classical KDACs a water molecule is activated by two histidine residues of a conserved tandem His-motif. One histidine (human KDAC8: H143; P. aeruginosa APAH: H144) acts as general base abstracting a proton from the catalytic water molecule thereby increasing its nucleophilicity for attack at the electrophilic carbonyl carbon of the acetyl-group (Figure 6B). The second histidine residue (human KDAC8: H142; P. aeruginosa APAH: H143) is important to orient and polarize the catalytic water molecule acting as electrostatic catalyst. Both histidine residues are polarized and oriented by two aspartate residues/one asparagine and one aspartate, respectively (human KDAC8: D183, D176; P. aeruginosa APAH: N186, D179) (Figure 6B). Another histidine is involved in coordination of the active site Zn²⁺-ion (human KDAC8: H180; P. aeruginosa APAH: H183). The active site Zn²⁺-ion coordinates and polarizes the attacking water molecule (Lombardi et al., 2011b). Furthermore, the Zn²⁺-ion, together with a Tyr (human KDAC8: Y306; P. aeruginosa APAH: Y313), which is replaced by a His in class IIa enzymes, increases the electrophilicity of the carbonyl carbon of the acetyl-group by contacting the carbonyl oxygen. During catalysis the Zn²⁺-ion and the Tyr stabilize the tetrahedral oxyanion intermediate. This is resolved by proton transfer of the histidine that initially acted as general base, now fulfilling the role as catalytic acid (Figure 6B). Specificity of the inhibitors is furthermore created by approaching sequence and structural differences between the KDAC-isoforms (Marmorstein, 2001; Marmorstein and Zhou, 2014; Li and Sun, 2019).

The knowledge on classical deacetylases in prokaryotes is limited and only a few classical deacetylases were identified and characterized in bacteria so far. In E. coli, no classical deacetylase is encoded. The Zn²⁺-dependent acetylpolyamine amidohydrolase AphA from Mycoplana ramosa was amongst the first classical KDACs identified (Sakurada et al., 1996). Activity studies revealed that M. ramosa AphA is effective on deacetylation of polyamines such as acetyl-spermidine, acetyl-spermine or acetyl-putrescin (Figure 1D). M. ramosa was shown to grow on medium with spermidine or putrescine as sole carbon source suggesting that AphA plays an important role during growth on mono- or diacetylated polyamines (Sakurada et al., 1996). Later structural data showed that M. ramosa AphA forms a dimer and that dimer formation is important for its catalytic activity. Moreover, a hydroxamate inhibitor was potent to inhibit M. ramosa catalytic activity showing a similar mode of action as shown for the mammalian KDACs (Lombardi et al., 2011a). Based on sequence and structure M. ramosa AphA was classified into class II KDACs (Lombardi et al., 2011a). In the same year the first mammalian KDAC, HDAC1/KDAC1, was isolated that had the activity to deacetylate histones (Taunton et al., 1996). The class I enzyme HDLP (Histone deacetylase-like protein) from the Gram-negative bacterium Aquifex aeolicus and the class IIb enzyme HdaH (Histone deacetylase-like amidohydrolase) from the Gram-negative Bordetella/Alcaligenes strain FB188 were amongst the first KDACs that were identified as protein deacetylases and structurally characterized (Finnin et al., 1999; Hildmann et al., 2004; Nielsen et al., 2005; Meyners et al., 2014; Mahindra et al., 2019). As all mammalian classical deacetylases all bacterial KDACs adopt an α/β-fold consisting of a central parallel β-sheet that is surrounded by α-helices (Figure 6A; Finnin et al., 1999; Somoza et al., 2004; Nielsen et al., 2005; Kramer et al., 2016b). All bacterial KDACs are, as their mammalian counterparts, metalloenzymes coordinating a Zn²⁺-ion that is essential for catalysis (Finnin et al., 1999; Nielsen et al., 2005; Kramer et al., 2016b). Bordetella/Alcaligenes FB188 HdaH was shown to possess activity as protein deacetylase with preference for basic residues directly neighboring to the acetylated lysine side chains (Hildmann et al., 2004; Nielsen et al., 2005). Both enzymes, Aquifex aeolicus HDLP and HdaH from Bordetella/Alcaligenes FB188 were potently inhibited by the hydroxamate mammalian KDAC inhibitors TSA and SAHA (Finnin et al., 1999; Nielsen et al., 2005). Three putative acetylpolyamine amidohydrolases (APAHs) were reported to be encoded in the human pathogen Pseudomonas aeruginosa, namely PA0321, PA1409, and PA3774 (Winsor et al., 2011; Kramer et al., 2016b). While PA0321 and PA1409 are active in deacetylating polyamines, such acetyl-putrescine and acetyl-cadaverine, PA3774 does not use polyamines as substrates (Kramer et al., 2016a). PA1409 was additionally capable in deacetylating N(1)-acetyl-spermidine and N(1)-acetyl-spermine (Kramer et al., 2016a). The K_M values of PA0321 and PA1409 for polyamines are 0.2–0.5 mM, which is approximately one order of magnitude lower than the reported intracellular polyamine concentrations suggesting that the enzyme shows full activity under physiological conditions (Patel et al., 2006; Kramer et al., 2016a). Polyamines were reported to be important for cell growth and cell proliferation in archaea, eukaryotes and bacteria (Michael, 2016). In bacteria polyamines affect biofilm formation and they play an important role in translation in eukaryotes and archaea (Patel et al., 2006; Michael, 2016). Apart from their role in biofilm formation, the role of polyamines in bacteria is only marginally understood (Michael, 2016). A possible role might be that polyamines form a buffer for non-enzymatic acetylation under physiological conditions that favor non-enzymatic, chemical acetylation of proteins such as high concentrations of reactive acyl-CoA thioesters. Thereby, polyamines would constitute a detoxification system to prevent systemic protein dysfunction due to unspecific, non-enzymatic acetylation. Notably, all three P. aeruginosa enzymes showed activity toward a fluorescent Boc-AcK-AMC substrate, which also opens the possibility that these enzymes are capable to also deacetylate lysine side chains that are located at the far termini of the proteins (Kramer et al., 2016a). P. aeruginosa enzyme HdaH (PA3774) can be classified into class IIb KDACs and it shows enzymatic activity as lysine deacetylase and has no activity toward acetylated polyamines (Kramer et al., 2016a, b). Due to the structural and sequence similarity of PA3774 to HdaH from Bordetella/Alcaligenes FB188 a similar substrate spectrum is assumed (Hildmann et al., 2004; Kramer et al., 2016a, b). Substrates for P. aeruginosa PA3774 were postulated such as the histone-like DNA-binding protein HU. However, this was not experimentally validated and further studies are needed to unravel the physiological substrates of P. aeruginosa PA3774 and Bordetella/Alcaligenes FB188 HdaH (Hildmann et al., 2004; Kramer et al., 2016a, b). The structure of PA3774 shows a special feature, an elongated surface loop (L1 loop) that mediates oligomerization (Figure 6A; Kramer et al., 2016b). A tetramer is formed consisting of two “head-to-head” dimers (Kramer et al., 2016b). A similar oligomerization was also observed in the structure of HdaH from Bordetella/Alcaligenes FB188 (Nielsen et al., 2005). This oligomeric state might affect substrate binding and thereby it might constitute an important mechanisms for the determination of substrate specificity (Kramer et al., 2016b). In contrast, the structure of the M. ramosa AphA polyamine deacetylase shows formation of a dimer. However, the dimer is formed via a loop insert not present neither in P. aeruginosa PA3774 nor in Bordetella/Alcaligenes FB188 HdaH resulting in a different dimer arrangement. Interestingly, this loop insert is also present in P. aeruginosa PA0321 and PA1409 suggesting that this is one molecular determinant of substrate specificity toward acetylated polyamines (Figure 6A; Kramer et al., 2016a, b). For Aeromonas hydrophila, a Gram-negative opportunistic human pathogen propagating extracellularly and surviving intracellularly in host phagocytes, the class II enzyme AcuC was shown to be needed for biofilm formation and virulence suggesting that targeting AcuC activity might be a promising strategy to treat bacterial infections (Jiang et al., 2017). For the classical KDACs that were structurally characterized all essential catalytic residues are conserved (Figure 6B).

So far, none classical KDAC of a Gram-positive bacterial species has been structurally characterized. B. subtilis encodes a classical KDAC, AcuC that is encoded as part of the acuABC operon (acu: acetoin-utilization) (Grundy et al., 1993; Gardner and Escalante-Semerena, 2009). The gene products of the acuABC-operon were initially thought to be important for growth and sporulation in acetoin and butanediol. AcuA was shown to be an type III GNAT and AcuC a Zn²⁺-dependent classical deacetylase deacetylating and activating acetyl-CoA synthetase (AcsA). The gene acsA encoding AcsA is reversely transcribed upstream of the acuABC operon (Gardner et al., 2006). The role of AcuB is not known but the genomic organization of acuB within the acuABC operon suggests that it plays a role on regulation of AcsA acetylation or deacetylation. A similar genomic organization is also present in S. aureus. However, the acuB gene is missing in S. aureus. AcuC from Gram-positive B. subtilis and S. aureus can be classified into the class I of mammalian KDACs (Leipe and Landsman, 1997; Thiagalingam et al., 2003). AcsA was shown to be acetylated by AcuA inhibiting AcsA and AcuC is the deacetylase reversing this acetylation thereby activating AcsA (Gardner et al., 2006). Apart from AcsA, only TufA was identified as additional substrate for AcuC so far, which was also shown to be deacetylated by SrtN in B. subtilis (Suzuki et al., 2019). For AcuA no further substrates were found so far not excluding that further substrates apart from AcsA exist.

As a summary, although some bacterial classical KDACs were structurally and functionally characterized, their physiological roles are only marginally understood. To this end, future studies should focus on investigation of their physiological roles including identification of substrates.

Beginning of 1960s lysine acetylation was discovered to occur on histones affecting RNA synthesis (Phillips, 1963; Allfrey and Mirsky, 1964; Allfrey et al., 1964). Afterward, it were only individual proteins identified as being lysine acetylated such as α-tubulin (L’Hernault and Rosenbaum, 1985). In the year 2000 the deacetylase SIR2 from S. cerevisiae was shown to be an NAD⁺-dependent lysine deacetylase with implications on the aging process (Kaeberlein et al., 1999; Imai et al., 2000). In the same period the first bacterial proteins were identified as being lysine acetylated, namely acetyl-CoA synthetase from S. enterica and the chemotaxis protein CheY from E. coli (Ramakrishnan et al., 1998; Barak and Eisenbach, 2001; Starai et al., 2002). The technological progress in mass-spectrometry enabled the identification of thousands of lysine ac(et)ylation sites in diverse bacterial species. This was due to the development of tools such improved mass-spectrometers concerning its dynamic range and sensitivity and the availability of specific antibodies recognizing anti-acetyl-L-lysine and other lysine acylations suited for immunoenrichment of ac(et)ylated peptides. However, as these antibodies are not unbiased regarding the sequences surrounding the acetyl-L-lysine often mixtures of available antibodies were used for immunoenrichment on the peptide level prior to LC-MS/MS analyses. This immunoenrichment step strongly improved the depth of the analyses represented by the number of acetylated peptides identified. Mass-spectrometric screens enabled the identification of thousands of lysine acetylation sites in bacteria. The first systemic mass-spectrometric screen performed on an E. coli proteome resulted in the identification of 125/138 acetylation sites in 85/91 proteins, respectively, predominantly involved in enzymes of the main metabolic pathways (Yu et al., 2008; Zhang et al., 2009). Today, more than 50 acetylomes were reported from diverse Gram-positive and Gram-negative bacterial strains and the number is continuously growing (Nakayasu et al., 2014, 2017; Christensen et al., 2019a, b). For most of these screens an immunoenrichment on the peptide level following proteolytic digest was performed with an anti-acetyl-lysine antibody. Notably, the fact that the number of acetylated proteins was strongly reduced in workflows without this immunoenrichment shows that the overall stoichiometry of the acetylation is low. Along this line, many of the sites identified in these workflows might not be of physiological significance as these are pure background noise present due to non-enzymatic acetylation by acetyl-phosphate and/or acetyl-CoA. It is one of the biggest challenges in the research field to unravel which of the many ace(et)ylation substrate sites are of real physiological importance. To this end, a huge study performed with 48 phylogenetically unrelated bacterial species analyzed the acetylomes without immunoenrichment and identified more than 9,000 acetylation sites (Nakayasu et al., 2017; Christensen et al., 2019a, b). It is likely that these sites detected without enrichment are important regulatory sites and in fact this study showed that many sites are conserved in enzymes of the main metabolic pathways (Nakayasu et al., 2017).

The workflows were improved to be able to determine the dynamics of lysine ac(et)ylation by chemical labeling on the peptide level as done in iTRAC or TMT labeling or by metabolic labeling such as done in SILAC (stable-isotope labeling by amino acids in cell culture), comparing different interventions (Ong and Mann, 2007; Choudhary et al., 2009; Sol et al., 2012; Hebert et al., 2013; Baeza et al., 2014; Carabetta and Cristea, 2017). There are several great reviews summarizing the mass-spectrometry screens performed on diverse bacterial species to assess dynamics of acetylomes comparing different interventions or physiological states (Carabetta and Cristea, 2017; Christensen et al., 2019a, b). A major technological breakthrough in acetylation research was the development of mass-spectrometric workflows that allow the determination of lysine ac(et)ylation stoichiometry on a systemic scale (Figure 8B; Baeza et al., 2014, 2015, 2016, 2020; Nakayasu et al., 2014; Weinert et al., 2015, 2017, 2018; Hansen et al., 2019; Lindahl et al., 2019). These are based on labeling of all accessible lysine side chains by isotopically labeled N-acetoxy-succinimide or acetic anhydride (Supplementary Figure 2) at the protein level prior to tryptic digest. Some workflows even combined SILAC and chemical acetylation or included internal standards to assess the dynamics of acetylation and obtain information on the stoichiometry of acetylation (Figure 8A; Weinert et al., 2015, 2017, 2018; Hansen et al., 2019; Lindahl et al., 2019). This allows to perform absolute quantification of the endogenous acetylation level and comparison of different states. Even time-resolved studies were performed in mammals to unravel the role of lysine acetyltransferases and the major cellular substrates and targets (Weinert et al., 2018; Baeza et al., 2020; Haws et al., 2020). The knowledge on the achieved acetylation stoichiometry is essential to assess the physiological importance of a specific lysine ac(et)ylation event. As an example, if a post-translational acetylation has a loss-of-function effect on a protein function, the modification must be present in rather high stoichiometry to be of physiological significance. Alternatively, if several enzymes of a metabolic pathway are affected by lysine acetylation to some extend it can accumulate to a significant outcome at the end of the pathway. The same is true for multi-subunit proteins, in which the acetylation in one subunit might affect the activity of the whole protein in a cooperative manner. If a modification creates a new functionality, in the sense of a gain-of-function, the stoichiometry does not need to reach high stoichiometries to be of significance. As an example, in a amplifying signal transduction cascade, such as present in chemotactic signaling in E. coli, affecting the activity of a minor proportion of one enzyme, such as activation of CheY by acetylation, might be sufficient to evoke a substantial response (Barak et al., 2006; Yan et al., 2008). The so far mostly for eukaryotic systems applied improved mass-spectrometric workflows that allow the systemic determination of ac(et)ylation stoichiometries combined with workflows that assess the ac(et)ylation dynamics comparing different interventions or physiological states, should be also applied to bacterial systems to narrow down which of the thousands of identified lysine ac(et)ylation sites are of physiological significance.

Our understanding on the mechanism how lysine acetylation mechanistically regulates protein function was mainly driven by the development of a system that allows to genetically encode acetyl-L-lysine in proteins—the genetic code expansion concept (GCEC) (Figure 8C). This is based on a synthetically evolved acetyl-lysyl-tRNA-synthetase (AcKRS)/tRNA_CUA (PylT) pair from Methanosarcina barkeri or Methanosarcina mazei originating on the pyrrolysyl-tRNA_CUA-synthetase (PylRS)/tRNA_CUA (PylT) pair. Methanogenic archaea of the genus Methanosarcina incorporate pyrrolysine as 22nd proteinogenic amino acid into enzymes involved in methane metabolism (Atkins and Gesteland, 2002; Hao et al., 2002). To this end, they encode PylRS, which charges the cognate tRNA_CUA, PylT, with pyrrolysine, which is incorporated into the enzymes as response to an amber stop codon (5′-UAG-3′). PylRS was evolved synthetically to charge the cognate PylT with acetyl-L-lysine instead of pyrrolysine (Neumann et al., 2008, 2009; Nguyen et al., 2009). As the PylRS/PylT pair from Methanosarcina is orthogonal in all model organisms this allows to incorporate acetyl-L-lysine as response to an amber stop codon into the protein of interest. Importantly, the system allows to obtain protein that is site-specifically and quantitatively lysine-acetylated and natively folded. We used the system in E. coli to produce proteins in yield and in a quality sufficient to perform biochemical and biophysical studies including structural analyses by X-ray crystallography (Figure 8C; Lammers et al., 2010; de Boor et al., 2015; Knyphausen et al., 2016a; Kuhlmann et al., 2016a, b; Lammers, 2018). We solved the first crystal structures of proteins carrying one or even two post-translationally relevant acetyl-lysines (Lammers et al., 2010; Kuhlmann et al., 2016a, b). Based on these studies we were able to derive mechanisms by which lysine acetylation regulate protein function (Lammers et al., 2010; de Boor et al., 2015; Knyphausen et al., 2016a; Kuhlmann et al., 2016a, b). Acetylation of the ε-amino group of lysine side chains neutralizes the positive charge at the lysine side chain, it alters the size of the amino acid, it affects the hydrophobicity of the side chain, it alters the surface complementarity and it also crosstalks with other post-translational modifications. We used the GCEC to produce site-specifically lysine-acetylated and natively-folded proteins as substrates for KDAC/sirtuin catalyzed deacetylation (de Boor et al., 2015; Knyphausen et al., 2016a; Lammers, 2018). This allows to analyze the molecular determinants of substrate specificity for the deacetylases. These studies revealed that the three-dimensional structure is next to the primary structure a major determinant for the substrate specificity (Knyphausen et al., 2016a; Lammers, 2018). Variants of AcKRS/PylT were developed that allow the incorporation of the acetyl-L-lysine analogs thioacetyl-lysine and trifluoroacetyllysine. As stated above these analogs are almost not deacylated by sirtuins representing valuable tools to block deacetylation increasing the lifetime of a lysine acetylation at a specific site for investigations in vivo or in vitro (Hancock et al., 2010; Venkat et al., 2017b). Notably, these analogs are better substrates compared to acetyl-L-lysine for some classical KDACs due to their increased electrophilicity at the carbonyl carbon of the thio-/trifluoroacetyl group (Lahm et al., 2007; Smith et al., 2008). Preparation of thioacetylated or trifluoroacetylated proteins in huge quantities and in high purity would allow to structurally characterize sirtuin-substrate complexes which was not successful so far. The available structural data is restricted on sirtuin complexes with acylated peptides (Spinck et al., 2020).

Recently, a novel prokaryotic class of deacetylases was suggested to be encoded in Gram-negative bacteria (Tu et al., 2015). The representative protein YcgC in E. coli was analyzed and shows no similarity in sequence and structure to any known deacetylases. It was suggested that YcgC exerts a catalytic mechanism involving a catalytic serine residue (Tu et al., 2015). Amongst the suggested substrates was the transcriptional regulator RutR involved in transcriptional regulation of pyrimidine metabolism. RutR was suggested to be quantitatively lysine acetylated on two lysine residues, K52 and K62, in the region of the helix-turn-helix motif important for DNA binding. Our laboratory applied the GCEC to produce K52- and K62-acetylated RutR proteins (RutR AcK52 and RutR AcK62) and used these proteins to assess YcgC and CobB catalyzed deacetylation. We showed that YcgC does not show any deacetylase activity toward acetylated RutR, while CobB is capable to deacetylate RutR (Kremer et al., 2018).

Moreover, the GCEC can be applied to assess the real consequences of lysine acetylation rather than performing studies with the Lys to Gln or Lys to Arg mutants, which are often applied for studies in vivo and in vitro to either mimic a lysine acetylation or to conserve the non-acetylated, positively charged state, respectively (Figure 1B; de Boor et al., 2015; Knyphausen et al., 2016b). These mutants were shown to be sometimes poor tools to study lysine acetylation. As an example, the structure of Gln does not resemble acetyl-L-lysine in size (Figure 1B). If the sole mechanism how lysine acetylation affects a protein function is charge neutralization, this might be reflected by mutation of Lys to Gln. However, if the lysine acetylation also acts via a steric contribution, a Gln is a poor mimic for acetylation (de Boor et al., 2015; Knyphausen et al., 2016b). Along that line, the mutation of Lys to Arg, which was initially used to preserve and assess the non-acetylated state, might reflect the steric contribution of lysine acetylation as Arg resembles acetyl-L-lysine more than Gln (Figure 1B). This has to be considered when analyzing the results obtained by these mutational studies. In a study analyzing the influence of acetylation of K180 (AcK180) on the DNA-binding capacity of the E. coli transcription factor RcsB, the mutants RcsB K180Q and RcsB K180R were analyzed (Thao et al., 2010). In this example, both RcsB K180Q and RcsB K180R almost completely abolished DNA-binding. This shows that RcsB AcK180 mechanistically impairs DNA-binding via charge neutralization (represented by K180Q) and via a steric contribution (represented by K180R) (Thao et al., 2010).

In another study the genetic code expansion was applied for incorporation of acetyl-L-lysine into E. coli isocitrate dehydrogenase (ICDH) at diverse sites. Notably, while some lysine acetylation events impaired the catalytic activity of ICDH some increased the catalytic activity in comparison to the non-acetylated enzyme, i.e., these show a gain-of-function. This activity stimulating effect was attributed to an potential affinity increasing effect toward the cofactor NADP⁺ and an influence on active site conformation (Venkat et al., 2018).

Other reports applying the GCEC for were performed in Corynebacterium glutamicum phosphoenylpyruvate carboxylase showing that acetylation at K653 impairs the enzymatic activity and that this acetylation is reversible (Nagano-Shoji et al., 2017). Moreover, genetically encoding acetyl-L-lysine results in the decrease in E. coli alanyl-tRNA-synthetase and tyrosyl-tRNA-synthetase activity (Venkat et al., 2017a; Umehara et al., 2018).

Further systems were developed that allow to genetically encode unnatural amino acids in proteins. The tyrosyl-tRNA-synthetase/tRNA_CUA from Methanocaldococcus jannaschii was evolved to allow incorporation of the photoactivatable crosslinkers p-benzoyl-L-phenylalanine and p-azido-L-phenylalanine proteins (Kauer et al., 1986; Chin et al., 2002a, b). This enables to photo-crosslink interaction partners and to stabilize otherwise transient interactions. This has been applied to human KDAC8 to identify novel substrates (Lopez et al., 2017). Next to photocrosslinkers, other genetic code expansion systems allow the site-specific incorporation of fluorescent amino acids, some of which are even photocageable, such as o-nitrobenzoyl-lysine, to follow protein dynamics in cells enabling light control of cellular processes using live cell fluorescence microscopy (Gautier et al., 2010, 2011; Lopez et al., 2017; Illes et al., 2021; Wolffgramm et al., 2021). Moreover, systems to genetically encode ubiquitin, phosphoserine and phosphothreonine were also developed that enables to study the impact of ubiquitination and phosphorylation on protein function (Chin, 2011, 2017; Park et al., 2011; Rogerson et al., 2015; Zhang M. S. et al., 2017; de la Torre and Chin, 2021). As stated above using mimetic mutations such as Ser/Thr to Glu/Asp, to study Ser/Thr phosphorylation can be misleading as described for mimetic mutations in studying lysine acetylation. The efficiency of multiple site incorporation of different amino acids in proteins was enhanced using different strategies. E. coli strains were developed that carry a genomic deletion of the gene encoding for release factor 1 (RF1), which recognizes the amber stop codon (5′-UAG-3′) thereby competing with the amber suppressor tRNA_CUA (Johnson et al., 2011). Moreover, E. coli strains were developed that carry whole genome deletion of amber stop codons avoiding incorporation of the unnatural amino acid at unwanted sites in endogenous proteins (Mukai et al., 2010). Often decoding the amber stop codon is no problem, as the amber stop codon is the least often used stop codon in E. coli. Moreover, the other stop codons opal (5′-UGA-3′) and ochre (5′-UAA-3′), rare sense codons or even quadruplet codons can be (re)assigned for unnatural amino acid incorporation (Neumann et al., 2010; Johnson et al., 2011; Heinemann et al., 2012; Wang et al., 2012; Mukai et al., 2015; Ho et al., 2016; Schmitt et al., 2018). Finally, even orthogonal ribosomes were created that are directed to an orthogonal messages allowing multisite incorporation of unnatural amino acids decoding several amber stop codons (Wang et al., 2007). The development of multiple orthogonal systems that allow the incorporation of different unnatural amino acids into proteins at different sites allows to study the crosstalk of different post-translational modifications. As an example, this could be applied to the E. coli chemotaxis regulator CheY, which is known to be regulated by phosphorylation and acetylation (Barak and Eisenbach, 2004). Our laboratory showed for the regulator RhoGDIα of the GTP-binding protein RhoA that lysine acetylation interferes with RhoGDIα SUMOylation (Kuhlmann et al., 2016a).

Future studies should involve the genetic code expansion concept to unravel how function of bacterial proteins and as a consequence also physiological process in bacteria are regulated by lysine acetylation. Most genetic code expansion systems, such as the Methanosarcina AcKRS/PylT system, were shown to be orthogonal in all model organisms tested and can be applied for studies in vitro and in vivo (Hancock et al., 2010; Greiss and Chin, 2011; Bianco et al., 2012; Elsasser et al., 2016; Han S. et al., 2017; Brown et al., 2018; Reille-Seroussi et al., 2019; Nikic-Spiegel, 2020). Application of genetic code expansion tools such as genetically encoded photoactivatable crosslinkers or fluorescent amino acids can be applied to bacteria to deepen our understanding on the physiological impact of lysine ac(et)ylation to control cellular processes.