Unmasking crucial residues in adipose triglyceride lipase for coactivation with comparative gene identification-58

Lipolysis is an essential metabolic process that releases unesterified fatty acids from neutral lipid stores to maintain energy homeostasis in living organisms. Adipose triglyceride lipase (ATGL) plays a key role in intracellular lipolysis and can be coactivated upon interaction with the protein comparative gene identification-58 (CGI-58). The underlying molecular mechanism of ATGL stimulation by CGI-58 is incompletely understood. Based on analysis of evolutionary conservation, we used site directed mutagenesis to study a C-terminally truncated variant and full-length mouse ATGL providing insights in the protein coactivation on a per-residue level. We identified the region from residues N209-N215 in ATGL as essential for coactivation by CGI-58. ATGL variants with amino acids exchanges in this region were still able to hydrolyze triacylglycerol at the basal level and to interact with CGI-58, yet could not be activated by CGI-58. Our studies also demonstrate that full-length mouse ATGL showed higher tolerance to specific single amino acid exchanges in the N209-N215 region upon CGI-58 coactivation compared to C-terminally truncated ATGL variants. The region is either directly involved in protein-protein interaction or essential for conformational changes required in the coactivation process. Three-dimensional models of the ATGL/CGI-58 complex with the artificial intelligence software AlphaFold demonstrated that a large surface area is involved in the protein-protein interaction. Mapping important amino acids for coactivation of both proteins, ATGL and CGI-58, onto the 3D model of the complex locates these essential amino acids at the predicted ATGL/CGI-58 interface thus strongly corroborating the significance of these residues in CGI-58–mediated coactivation of ATGL.

region were still able to hydrolyze triacylglycerol at the basal level and to interact with CGI-58, yet could not be activated by CGI-58. Our studies also demonstrate that full-length mouse ATGL showed higher tolerance to specific single amino acid exchanges in the N209-N215 region upon CGI-58 co-activation compared to C-terminally truncated ATGL variants. The region is either directly involved in protein-protein interaction or essential for conformational changes required in the co-activation process. Three-dimensional models of the ATGL/CGI-58 complex with the artificial intelligence software AlphaFold demonstrated that a large surface area is involved in the protein-protein interaction. Mapping important amino acids for coactivation of both proteins, ATGL and CGI-58, onto the 3D model of the complex locates these essential amino acids at the predicted ATGL/CGI-58 interface thus strongly corroborating the significance of these residues in CGI-58 mediated co-activation of ATGL.
The physiological relevance of ATGL was thoroughly studied in different genetic knockout and transgenic mouse models: Mice lacking ATGL exhibit increased TG deposition in many tissues and disrupted signaling pathways leading to disturbed energy homeostasis [21,22]. A comprehensive record of published ATGL mouse models and associated phenotypes was rigorously summarized elsewhere [23]. Human patients with mutations in the gene coding for ATGL suffer from TG accumulation in leukocytes and multiple tissues in addition to severe cardiomyopathy [22]. Interestingly, mutations in the human gene of the ATGL co-activator CGI-58 cause severe hepatic steatosis and systemic TG accumulation that is always associated with ichthyosis [7,24]. Analogously, global CGI-58 knockout mice suffer from a lethal skin permeability barrier defect due to impaired ω-O-acylceramide synthesis [21,[25][26][27].
Atomic resolution structures of both proteins, CGI-58 and ATGL, are not available yet. Mouse ATGL (mATGL, 486 amino acids) harbors a "patatin-like phospholipase domain (PNPLA)" within resides I10-K179 (InterPro IPR002641), which is name-giving for all PNPLA-family members [28]. The TG-hydrolytic activity of ATGL is catalyzed by a catalytic dyad, formed by S47 and D166, and residues G14-G19 forming the oxyanion hole [29]. Residues I10-G24 are suggested to be involved in TG binding [30]. In vitro studies showed increased lipolytic activity of the truncated variants of ATGL (truncated after D288 or L254 in mATGL) compared to the wild-type enzyme [5,10]. Nevertheless, the exact roles of the C-terminal half of ATGL (residues P260-C486) with respect to lipid droplet (LD) localization and autoregulatory function remain to be established.
Mouse CGI-58 (mCGI-58, 351 amino acids) is a member of the α/β-hydrolase-fold containing protein family comprising of an N-terminal region and an α/β-hydrolase core domain with a cap [31]. In contrast to other α/β-hydrolase protein family members, CGI-58 does not exhibit hydrolytic activity due to the lack of a catalytic nucleophile. Previously, we have shown that the N-terminal, Trp-rich region of CGI-58 is important for LD anchoring and largely disordered [32][33][34]. Removal of 30 N-terminal amino acids of mCGI-58 disrupted both its ability to localize to LDs and its ability to co-activate ATGL. However, this LD anchor by itself lacks the ability to activate ATGL, indicating that other regions of CGI-58 are necessary for ATGL coactivation [33]. Subsequent mutagenesis studies of CGI-58 (ABHD5) demonstrated a crucial function for residues R299, G328 and D334 of CGI-58 in ATGL co-activation [16,35].
In the current study, we identified evolutionary less conserved parts of ATGL by comparing mammalian ATGL with different phyla in the Kingdom Animalia. Importantly, ATGL variants at positions N209, I212 and N215 exhibited intact or marginally reduced basal activity, however drastically reduced activatability by CGI-58 providing evidence that these residues play a significant role in the co-activation process. Artificial intelligence-based modelling approaches for the three-dimensional (3D) structure of the ATGL/CGI-58 complex also predicted the region comprising N209, I212 and N215 to be involved in protein-protein interaction between enzyme and co-activator. Residues R299, G328 and D334 of CGI-58, are also located in the predicted ATGL/CGI-58 binding interface. In the absence of experimental complex structures, our data provide a good working model of the ATGL/CGI-58 complex.

Results: Evolutionary conservation provides an initial rationale for generating ATGL variants
Regulation of ATGL activity on a protein and activity level is very well studied with respect to the interaction of ATGL with CGI-58, G0S2 and HILPDA [10][11][12][13][14][15][16]21,22]. Interestingly, the regulatory proteins are not conserved in all species, e.g., the inhibitory protein G0S2 only exists in vertebrates while BLAST-searches of ATGL also reveal proteins with significant hits in non-vertebrates [15,36]. Therefore, we performed an unbiased computational screen to identify amino acids within the sequence of ATGL that mediate its interaction with regulatory proteins. We were primarily interested in the N-terminal half of ATGL since previous in vitro studies demonstrated that the truncated ATGL variants M1-L254 or M1-D288 can be activated by CGI-58 or inhibited by G0S2 and HILPDA [7,9,15,19].
We compared the amino acid sequences of ATGL from different species, within the subgroup of 'mammalia' and within a larger general group termed 'animalia' including different chordates (fish, reptiles, birds, mammals) insects, nematodes and mollusks. The analysis was carried out using the ConSurf software, a designated tool to identify functional regions in proteins by exploiting evolutionary data ( Figure 1A) [37]. We hypothesized that catalytically and structurally essential regions are highly conserved, whereas regulatory regions might be i) less conserved when comparing mammalian ATGL with ATGL of non-mammalian species and ii) surface exposed to enable protein-protein interaction.
As shown in Figure 1, the levels of conservation are quite different. Both groups, 'mammalia' and 'animalia' show large sequence variations in regions P3-K8, V57-C61, G96-T101, T158-Q160 and K179-N180. In line with our general hypothesis, the most conserved region even within 'animals' is in the central core and catalytic region of the protein, whereas surface exposed parts of α-helices show higher sequence variability ( Figure 1B). Based on our hypothesis (i) on sequence conservation, we identified regions F35-A40, V71-N89, V150-V165, F187-D197, H203-K229 and Y242 to match our criteria of conservation within 'mammalia', yet high diversity within 'animalia'. To further narrow down residues of interest, we tested these residues for criteria (ii) by plotting them on a 3D model of ATGL and analyzed the residues with respect to surface exposed residues. Accordingly, we decided to introduce the following residues for introduction of single-amino acid exchanges: L81A, L84A, Y151A, Y164A, F187A, S188A, I193A, L205A, N209A, I212A, I212S, N215A, L216A, Y220A, R221A, L226A, F227A, Y242A. These residues are predominately surface-exposed and cluster on an almost continuous surface area of the protein ( Figure 1C). The amino acid exchanges were introduces in the C-terminally truncated variant of mouse ATGL (mATGL288) [5,6,15,20,36].
We predominantly mutated aromatic and aliphatic amino acids residues since these sidechains are frequently involved in biological interactions, whereas small alanine residues typically contribute very little to protein-interactions. The exchange I212S alters the physicochemical properties more drastically by introducing a small, polar side-chain.  Figure 2B, 2D, 2F). To clearly show the difference between basal activity and activatability, we give the X-fold multiplier by which activity is increased upon addition of CGI-58. The multiplier is also insensitive to slight changes in the expression level of ATGL, since basal activity and activity upon CGI-58 stimulation are equally affected.
Amino acids F17 and at the active site serine S47 are key amino acids for ATGL activity and the corresponding exchanges F17A and S47A served as negative controls [7,38,39].

Figure 2.
TG hydrolase activity assays and expression control of ATGL variants from bacterial lysates. mATGL288 WT enzymatic activity in basal and mCGI-58 co-activated condition was used as positive control; the multiplier for x-fold increase in activity is indicated for each variant. A) Inactive variants of mATGL288, namely F17A, S47A, Y164A, F187A and I193A in comparison to WT mATGL288 under basal and mCGI-58 co-activated conditions based on enzymatic activity with zoom-in (right). B) Immunoblotting of the expression levels of inactive variants was used as expression control. C) Active variants of mATGL288 namely L81A, L84A, S188A, L205A, L216A, Y220A, R221A and F227A under basal and mCGI-58 co-activated conditions. D) Immunoblotting analysis of the active variants. E) Partially active variants with residual enzymatic activity of mATGL288 namely Y151A, L226A and Y242A; F) Immunoblotting of the expression levels of the partially active variants in comparison to WT mATGL288. Each TG hydrolase activity assay represents the assay of three technical replicates. At least 2 biological replicates were performed.
When we tested the N209A, I212A, I212S and N215A variants, we observed intact or only slightly reduced basal activity, but reduced activability by mCGI-58 (approximately 1.5 -4fold) when compared to activation of WT mATGL288 by mCGI-58 ( Figure 3A). To determine the half-maximal effective concentration (EC50) of mCGI-58 to activate ATGL, we performed dose-dependent activity measurements of mATGL288 WT, N209A, I212A, I212S and N215A with increasing concentrations of mCGI-58. The results revealed EC50 values in the range of 500 ± 50 nM for the N209A and I212A variants and 200 ± 20 nM for the I212S and N215A variants, in comparison to EC50 values of 178 ± 20 nM for WT mATGL288 ( Figure 3B).
Plotting single amino acid exchanges on a 3D model of mATGL: N209, I212 and N215 are located on a surface region on one face of the protein.
To locate the most crucial amino acid residues within the structure of ATGL, we generated a 3D model of mouse ATGL254 using AlphaFold ( Figure 4). It is important to note that the model confidence varies significantly between different regions of the enzyme [5,40]. Almost the entire PNPLA domain and two short additional strands of the central β-sheet are modeled with very high confidence, namely residues W8-S73, I94-L173 and T181-P195 (confidence score 'high', >90). Similarly, P231-N252 form a long α-helix on the surface of mATGL254 that is predicted with very high confidence. In contrast, residues Q196-E230, which also show largest sequence variability (see Figure 1A), are only predicted with 'low' to 'medium' confidence scores between 50 and 90. The residues N209, I212 and N215 are located within a region forming a short β-sheet on the surface of ATGL; amino acid exchanges of those residues resulted in retained basal activity yet substantial loss of activability by CGI-58 (Figures 3,4). Based on the 3D model of ATGL, this region forms a part of the substrate binding pocket that is distant from the catalytic site. We speculate that this region is not directly involved in the catalytic reaction at the scissile ester bond of the lipid, but plays a role in binding or release of the substrate or product ( Figure 4).
Substitution of the conserved aromatic residues F17, F187 and Y164 to alanine residues resulted in loss of ATGL-activity which could not be rescued by addition of CGI-58. Amino acid exchanges Y151A, L226A, Y242A resulted in ATGL-variants with reduced basal activity and reduced co-activation upon addition of CGI-58. In the 3D model of mATGL254 residues F17, Y151, F187, L226 are located in or at the entrance to the substrate binding pocket and might therefore not tolerate substitutions ( Figure 4). Y164 is positioned in a loop that helps in positioning of the catalytic residue D166 spatially close to S47. Thus, it might be essential for the formation of a principally functioning catalytic active site architecture within ATGL. Y242 is located on the surface exposed face of an α-helix. The complete loss of activity and coactivation upon introducing the change I193A might result from destabilization of ATGL due to loss of hydrophobic interactions (Figure 2 and Figure 4). To investigate whether our results with bacterially expressed mATGL288 can be recapitulated with mATGL variants in lysates from a eukaryotic expression system, we transfected the suspension-adapted human embryonic kidney cell line Expi293F with WT and mutated mATGL variants and tested cell lysates for TG-hydrolytic activity ( Figure 5). Upon mCGI-58 addition, truncated WT mATGL288 was activated 11-fold, whereas the mATGL288 variants harboring double or triple amino acid exchanges, were only stimulated up to 2-fold ( Figure 5A). All mATGL288 variants showed similar expression efficacy in transfected Expi293F cells ( Figure   5B). In the eukaryotic system, we also tested full-length mATGL to confirm the results obtained with the truncated mATGL288 enzyme. Addition of CGI-58 increased TG hydrolase activity of lysates containing WT full-length ATGL approximately 20-fold. The single amino acid exchange variants N209A, N212A and N215A were also activatable although to a lesser degree (8 and 14-fold). In contrast, double (N209A/N215A) or triple mutations (N209A/I212A/N215A) in full-length mATGL essentially lost the ability to be co-activated activated by mCGI-58 ( Figure 5A)) despite similar expression level ( Figure 5B).
After cell lysis, we employed affinity purification, to isolate mATGL288 and potential mATGL288/mCGI-58 protein-protein complexes. Lysates and column fractions were analyzed for complexed and uncomplexed mATGL288 and CGI-58 by Western blotting analyses using an anti-StrepII-antibody and anti-His-antibody for the detection of mATGL variants or mCGI-58, respectively ( Figure 6B  Using AlphaFold, we generated an in silico model of the ATGL/CGI-58 complex (Figure 7). For our analysis, we mapped the positions of N209, I212 and N215 of ATGL identified in this study and residues R228, G328, D334 of CGI-58 identified in previous work [16,35]. As seen in Figure  7A-7D, the interface between ATGL (truncated at residue 260 in this panel) and full-length mCGI-58 spans over large, yet continuous surface area of both proteins. I212 is central in a βstrand of a short two-stranded antiparallel β-sheet (E204-V207, T210-N213) at the far-end of the catalytic site in the predicted substrate binding pocket. N209 and N215 are located at the hinges of this β-strand. Substitution to of potential bidentate residues asparagine to alanine might change the flexibility of these hinges or be directly involved in changing protein-protein interaction (Figures 3, 4, 7). Importantly, data from experimental mutagenesis studies agreed very well with the calculated protein-protein interfaces. At the current stage of modelling, detailed analysis (e.g., with respect to inter-molecular H-bonding, formation of salt bridges) or fine-tuned modelling of the protein-protein interaction interface requires additional data, preferably from an experimentally determined structure ( Figure 7B). The analysis of the Cterminal region of ATGL is challenging, since large parts of the C-terminal half are modelled with very low, low and at best medium confidences scores. Accordingly, the C-terminal half might adopt different secondary structures or different spatial positions in the physiological environment (e.g., the full LDs decorated with different LD associated proteins, Figure 7C).
Next, we generated 3D models of truncated WT ATGL and the N209A/I212A/A215A ATGL variant. The triple variant exhibits large conformational rearrangements of region D197-L216 -comprising a long loop and a short β-sheet in WT ATGL -by forming a loop and a long α-helix ( Figure 7E). This region coincides with the region of high conservational variability ( Figure 1) and low to medium confidence scores for structure prediction. It is interesting to note, that previous 3D homology modeling of the WT sequence of ATGL had an α-helix predicted for N209-L226, while the PNPLA domain (resides I10-K179) was essentially identical [41,42].
When we modeled the N209A/I212A/A215A ATGL/CGI-58 complex, the overall complex looked similar, however with some small changes in the interface. The model confidence for AlphaFold-Multimer models is described as an intrinsic score pTM and interface score ipTM [43]. The truncated WT ATGL/CGI-58 model exhibited a combined pTM + ipTM of 0.78, while the triple variant truncated ATGL/CGI-58 model, which experienced significant conformational changes in the loop region at the interface, had a score of 0.68. These results suggest a weaker interaction between the two proteins. Nevertheless, the score remains relatively high, indicating a reasonably strong interaction. There is an overall breathing motion and shifting of the co-activator CGI-58 to accommodate the longer α-helix and the lack of the short β-sheet of the ATGL variant ( Figure 7F). Together, these predictions indicate conformational flexibility beyond the PNPLA-domain and awaits further experimental insights of ATGL by itself and in complex with the co-activator CGI-58. The 3D structure of mATGL (M1-N259 in (A), M1-C486 in (C)) is in cartoon representation, the annotated PNPLA-domain (I10-L178), which is predominantly predicted with very high confidence, is colored light blue; residues M1-N9 and K179-L254 are in gray. Active site residues (S47, D166) are depicted as green sticks. mCGI-58 (G18-D351) is in pea-green cartoon representation. The transparent surface of both proteins is depicted in gray and pea-green, to highlight the extensive predicted protein-protein interaction surface. N209, I212 and N215 of ATGL as well as amino acids R299, G328 and D334 of CGI-58 are highlighted in cyan. B) Close-up view of the predicted interaction surface with N209, I212 and N215 of ATGL and amino acids R299, G328 and D334 of CGI-58 in cyan stick representation. C) 3D model of the ATGL/CGI-58 complex including the C-terminal half (N256-C486, colored in orange) of ATGL that is difficult to model with high confidence. D) Surface representation of the 3D model of the truncated ATGL/CGI-58 complex. The αβ-hydrolase core of CGI-58 is colored pea-green as in A-C, yet the cap region of CGI-58 (P180-M279) is colored in yellow. E) Overlay of the 3D models of truncated WT ATGL with the N209A/I212A/N215A ATGL variant. Regions predominantly predicted with very high confidence, are colored light blue; whereas regions K179-L254 with lower confidence in the prediction are in gray and sand for WT ATGL and N209A/I212A/N215A ATGL, respectively. ATGL is in similar orientation as in Figures 1B, 1C and 7F. F) Overlay of the complexes of WT ATGL/CGI-58 and that of the variant N209A/I212A/A215A ATGL with CGI-58. WT ATGL is colored light blue and gray; N209A/I212A/A215A ATGL is also in light blue, yet residues K179-L254 are in sand; the side chains A209, I212A and A 215 are depicted as orange sticks. When CGI-58 is in complex with WT ATGL, it is colored as in (B), whereas CGI-58 in complex with N209A/I212A/N215A ATGL is colored in light green for the core and pink for the cap, respectively.

Discussion
Intracellular lipolysis is a crucial metabolic process in energy homeostasis and diligent balance of its regulation leads to metabolic equilibrium. CGI-58 is a crucial regulator of ATGL activity, but the mechanism by which CGI-58 co-activates ATGL remains elusive. It is not known if binding of mCGI-58 affects the conformation of ATGL, facilitates substrate presentation, or increases the lipolytic activity of ATGL by removing reaction products from the active sitequestions similar to the unknowns discussed for the G0S2-mediated inhibition mechanism of ATGL [31]. We employed mutagenesis studies to identify specific amino acids in ATGL which are required for its co-activation by CGI-58. The sites for amino acid exchanges were selected based on evolutionary conservation and surface exposure. For most experiments, we used a shortened version of mouse ATGL, mATGL288, in combination with mouse full-length CGI-58 (mCGI-58) expressed in a bacterial expression system. Doing so, we could identify amino acids exchanges that resulted in complete loss of ATGL activity and exchanges leading to similar basal activity and activability by CGI-58 as observed for WT ATGL. Importantly, we identified the region N209-N215 of mouse ATGL to play an essential role in mCGI-58-mediated coactivation of ATGL (Figure 3, Figure 4). N209, I212 and N215 were recognized as residues with low evolutionary conservation ( Figure 1) and they turned out to be crucial residues for coactivation based on in vitro assays (Figure 3). AlphaFold modeling revealed that these residues were also central for the mATGL288/mCGI-58 interaction in the interface of protein contacts ( Figure 7). The findings suggest that the N209-N215 region is not directly involved in executing the hydrolysis reaction, but mediates CGI-58 co-activation of ATGL.
The comparison of ATGL variants from bacterial and mammalian expression system demonstrates that post-translational modifications are not required for the co-activation of ATGL by mCGI-58 ( Figure 5). Furthermore, the comparison of full-length versus C-terminally truncated variants indicate that the full-length variants are slightly more tolerant towards single amino-acid exchanges. This is very similar to our previous observations on full length and truncated ATGL inhibition by variants of G0S2 [36]. The 254 N-terminal amino acids of mouse ATGL comprise the minimal domain that can be activated by mCGI-58 and inhibited by G0S2 [10]. Supposedly, additional regions within the C-terminal half of ATGL together with other factors either affect the protein-protein interaction, or provoke conformational changes in the lipase, the co-activator or the LD-associated TG substrates. ATGL variants associated with neutral lipid storage disease (NLSDM) are either enzymatically inactive proteins localizing to LDs or active TG hydrolases lacking LD localization [44]. Deletion of 220 amino acids from the C-terminus of human ATGL increases its interaction and activation by CGI-58 in vitro, in spite of defective LD localization in vivo in cultured cells [44]. This finding indicates that the C-terminal region of ATGL is required for its targeting to LDs and plays an important (auto)regulatory role that requires further characterization.
When we modelled the structure of the ATGL/CGI-58 complex, we found that large surfaces of both proteins, ATGL and CGI-58, are predicted to be involved in protein-protein contacts.
Plotting N209, I212 and N215 on our 3D model of ATGL places these residues on an extension from the highly conserved PNPLA (I10-K179) domain on one face of the protein. The predicted interface of ATGL is completely in line with the herein presented and previous experimental results. Furthermore, the predicted interface region of CGI-58 matches well with previously published data: The Trp-rich N-terminal region is not predicted to be involved in the interface, which is in agreement with its annotated role in serving as LD localization anchor [33]. Both, the α/β-hydrolase core domain and the cap region (P180-M279, colored yellow in Figure 7D, 7F) of CGI-58 are predicted to be involved in formation of a large protein-protein interaction surface. The extent of the surface might also explain, why the single and even triple amino acid exchanges did not totally abolish the interaction in co-expression and co-purification assays ( Figure 6). In the modeled ATGL/CGI-58 complex, the far end of the substrate binding cavity involving N209-N215 seamlessly merges into the pocket between the cap and the core of CGI-58. Previous studies had also demonstrated that CGI-58 interacts with fatty acid binding protein 4 (FABP4), which further promotes ATGL-mediated lipolysis and indicates the existence of an ATGL/CGI-58/FABP4 complex [45]. In the model of the binary ATGL/CGI-58 complex, a direct transfer of hydrophobic ligands (substrates or products) from ATGL to CGI-58 appears possible (Figure 7). AlphaFold-modeling of the binary CGI-58/FABP4 complex predicts that the helical portal region of FABP4 interacts with the cap region of CGI-58. This region of FABP4 has been shown experimentally to be involved in the FABP4/CGI-58 interaction using biochemical and biophysical methods [45].
In the absence of experimental 3D structures for ATGL and ATGL/CGI-58 complexes, our results represent a significant advance in our understanding of the protein-protein complex formed by ATGL/CGI-58 and provides novel insights into ATGL co-activation.

Materials
If not stated otherwise, chemicals were obtained from Merck (Darmstadt, Germany) or Carl Roth GmbH (Karlsruhe, Germany); columns for protein purification were obtained from Cytiva (formerly GE Healthcare Life Sciences (Uppsala, Sweden)). [9,

Site-directed mutagenesis of bacterial expression vectors
The WT variants coding for Mus musculus ATGL (UniProt Accession: Q8BJ56) in two different expression vectors namely pST44 and pcDNA4/HisMaxC (Thermo Fisher Scientific, Waltham, USA) were described in [46]. The primers listed in Table 1 and 2 were designed to introduce point mutations in the WT variants using the Q5® site-directed mutagenesis kit (New England BioLabs, Ipswich, USA). All variants were verified by Sanger sequencing (Microsynth, Balgach, Switzerland).
Site-directed mutagenesis of eukaryotic expression vectors

Bacterial expression of recombinant mCGI-58, ATGL-proteins, mATGL/mCGI-58 complexes and preparation of bacterial cell extracts
Expression of mATGL288 and mCGI-58 have been described before [5]. Expression of mATGL288 protein and single amino acids exchange variants were performed in E.
coli ArcticExpress (DE3) cells (Agilent Technologies, Santa Clara, CA, USA) similar as previously described [5]. Expression of the complexes mATGL288/mCGI-58, mATGL_N209A/mCGI-58, stands for the cleared cell lysate which was loaded onto the column and "wash" is the last wash fraction taken before the elution.

Protein expression in Expi293F TM cells and preparation of cell-extracts for TGH assays
His-mATGL and its point mutants were recombinantly expressed in Expi293F TM cells (Thermo Laboratories, Hercules, USA) using bovine serum albumin (BSA) as standard.

Purification of recombinant mouse mCGI-58
Purification of His6-smt3-mCGI-58 was performed via immobilized metal ion affinity chromatography as previously described [5]. The concentration of the purified protein was determined via absorption at 280 nm.

TG Hydrolase Assay
TG hydrolase activity assays was performed with some minor modifications as described elsewhere [5,10,36]

Modelling of ATGL and the ATGL/CGI-58 complex using AlphaFold
ATGL and hetero multimers of ATGL with CGI-58 (ABHD5) were calculated using AlphaFold multimer. The predictions were based on either full-length protein from mouse (ATGL M1-C486; CGI-58 M1-351) or a truncated form of ATGL with a C-terminal truncation (ATGL M1-D264). Despite the differences, both predictions resulted in the identification of the same interaction interface. Furthermore, the same result was obtained when both combinations were predicted using AlphaFold monomer mode with an unstructured 50x glycine linker between the two different proteins. Predictions were either calculated with AlphaFold v2.1.2 full installation on a local workstation with following specifications: Nvidia GeForce RTX 3090 24 GB, AMD Ryzen Thread Ripper 3975WX using 48 cores and 192 GB RAM or with AlphaFold v2.3.1 and local ColabFold v1.5.2 [48] installation using a workstation with Nvidia GeForce RTX 3090 24 GB, AMD Ryzen 9 5900X using 12 cores and 64 GB RAM.