The ubiquitin-like modifier FAT10 inhibits retinal PDE6 activity and mediates its proteasomal degradation

The retina-specific chaperone AIPL1 is essential for the correct assembly of phosphodiesterase 6 (PDE6), which is a pivotal effector enzyme for phototransduction and vision because it hydrolyzes cGMP. AIPL1 interacts with the cytokine-inducible ubiquitin-like modifier FAT10 that gets covalently conjugated to hundreds of proteins and targets its conjugation substrates for proteasomal degradation, but whether FAT10 affects PDE6 function or turnover is unknown. Here, we show that FAT10 mRNA is expressed in human retina and identify rod PDE6 as a retina-specific substrate of FAT10 conjugation. We found that AIPL1 stabilizes the FAT10 monomer as well as the PDE6-FAT10 conjugate. Additionally, we elucidated the functional consequences of PDE6 FAT10ylation. On the one hand, we demonstrate that FAT10 targets PDE6 for proteasomal degradation by formation of a covalent isopeptide linkage. On the other hand, FAT10 inhibits PDE6 cGMP hydrolyzing activity by non-covalently interacting with the PDE6 GAFa and catalytic domains. Therefore, FAT10 may contribute to loss of PDE6 and, as a consequence, degeneration of retinal cells in eye diseases linked to inflammation and inherited blindness causing mutations in AIPL1.


Introduction
The process of vision is very complex and needs to be tightly controlled. Since retinal cells, especially the rod and cone photoreceptors, have an unusually high demand for the synthesis and thus folding of proteins involved in phototransduction, a specialized chaperone machinery is needed to guarantee retinal proteostasis (1)(2)(3)(4). One member of this machinery is the aryl hydrocarbon interacting protein-like 1 (AIPL1), which together with Hsp70 and Hsp90 forms a chaperone heterocomplex (5). AIPL1 is exclusively expressed in developing photoreceptors as well as in adult rods and the pineal gland (6)(7)(8). Its importance becomes apparent as even point mutations within the Aipl1 gene, such as A197P or C239R induce an inherited retinal degenerative disease called Leber congenital amaurosis (LCA), which is characterized by the loss of vision during the first months of life (6,9).
Within photoreceptors, AIPL1 is important for chaperoning the retinal phosphodiesterase 6 (PDE6) (3). AIPL1 was not only shown to interact with PDE6 but to be essential for the proper assembly of PDE6 subunits into a holoenzyme (10). Within rods and cones, PDE6 differs in the composition of the catalytic core. While rod PDE6 possesses a heterodimeric catalytic core consisting of a α-and a β-subunit, cone PDE6 is composed of two catalytic α`subunits. Nonetheless, both interact with two photoreceptor specific inhibitory γ-subunits (11)(12)(13). Just like other PDE superfamily members, the PDE6 catalytic subunits harbor two N-terminal regulatory GAF domains and a C-terminal catalytic domain (14)(15)(16). PDE6 is one of the key enzymes of phototransduction. Briefly, after activation of a visual pigment, the G protein transducin is activated leading to the interaction of the activated transducin α-subunit with the PDE6 γ-subunits. This leads to an increased and accelerated PDE6-mediated cGMP hydrolysis resulting in a drop of intracellular cGMP levels. Subsequently cGMP-gated ion channels are closed and the photoreceptor membrane is hyperpolarized resulting in a synaptic transmission of the signal (2,17,18). This whole phototransduction process takes place in the outer segment of photoreceptors where PDE6 is anchored in membrane disks via posttranslational prenylation of the catalytic subunits (19,20).
We have recently published another interaction partner of AIPL1, human leukocyte antigen (HLA)-F adjacent transcript 10 (FAT10) protein (21). FAT10 belongs to the family of ubiquitin-like modifiers and is, in contrast to the ubiquitously expressed ubiquitin, constitutively expressed only in organs and unique cell types of the immune system (22)(23)(24)(25). In other cell types, FAT10 is expressed upon synergistic induction by the pro-inflammatory cytokines tumor necrosis factor (TNF)-α and interferon (IFN)-γ (26,27). The 18 kDa FAT10 protein consists of two ubiquitin-like domains (UBD) bearing the typical ubiquitin β-grasp fold, while the C-terminal domain contains the diglycine motif important for covalent attachment to substrates (28,29). This conjugation is mediated by an enzyme cascade consisting of the E1 activating enzyme UBA6 (30)(31)(32), the E2 conjugating enzyme USE1 (26,33) and putative E3 ligases. FAT10ylation of conjugation substrate proteins such as USE1 itself (26), the autophagy receptor p62 (34) or the ubiquitin activating enzyme UBE1 (35), targets them directly for proteasomal degradation. Of note, this is independent of additional ubiquitin attachment or the activity of the segregase valosin-containing protein (VCP) (28,(36)(37)(38). Next to covalent modification of substrates, FAT10 can also interact with proteins in a noncovalent manner resulting for example in an altered functionality of the interaction partner, as recently shown for the deubiquitinating enzyme OTUB1 (39) or for the SUMO E1 activating enzyme AOS1/UBA2 (40).
In this study, we examined the specific function of FAT10 in photoreceptor cells inspired by the recently identified interaction between AIPL1 and FAT10. We present evidence that FAT10 expression is inducible in human retinal cells and we identify PDE6 as the first retina-specific FAT10 conjugation substrate. By investigating the impact of AIPL1 on FAT10 and the FAT10-PDE6β conjugate, we show that both accumulate in the presence of AIPL1. Finally, we examined the functional consequences and show that the modification of PDE6 by covalent attachment of FAT10 targets PDE6 for proteasomal degradation while non-covalent interaction with FAT10 downregulates the enzymatic activity of PDE6.
As AIPL1 wild type had been shown to interact with FAT10 in a direct manner (21), we investigated whether the pathogenic mutants AIPL1-A197P and AIPL1-C239R directly interact with FAT10 as well. In vitro Co-IP of recombinant proteins revealed that all three recombinant AIPL1 variants were able to directly interact with recombinant FAT10 (Fig.  1B, lanes 3-5). As FAT10 is a ubiquitin-like modifier containing typical ubiquitin-like folds (28), we tested whether AIPL1 interacts with ubiquitin as well. However, in vitro Co-IP studies using recombinant His-tagged ubiquitin and HA-tagged AIPL1, AIPL1-A197P and AIPL1-C239R showed no interaction of the chaperone with ubiquitin ( Fig. S1C), underlining a specific interaction with FAT10.
To confine the interaction sites in more detail, we tested which FAT10 domain is involved in the interaction with AIPL1 (Fig. 1C). Either HA-GFP-tagged full length FAT10 (FAT10-GG), the single FAT10 N-terminal (FAT10-N) or C-terminal UBD (FAT10-C) were expressed together with FLAG-tagged AIPL1 in HEK293 cells, as indicated. Again, the interaction between AIPL1 and full length FAT10 was detectable (Fig. 1C, IP: FLAG IB: HA or GFP, lane 6). Interestingly, AIPL1 also interacted with both, the N-and the C-terminal FAT10 UBD (Fig. 1C, IP: FLAG IB: HA or GFP, lanes 7 and 8), suggesting that both UBDs are used for the interaction. In the negative control, in contrast, no interaction between AIPL1 and GFP was detectable (Fig. 1C, IP: FLAG IB: GFP, lane 10). AIPL1 possesses three TPR motifs that are described to be important for protein-protein interactions (5,41). To test whether this holds true for FAT10 as well, Co-IP experiments with untagged FAT10 and FLAG-tagged AIPL1 truncation variants were performed ( Fig. 1D  and 1E). Indeed, FAT10 interacted mainly with the TPR motifs since an interaction was detected only for those AIPL1 truncations still expressing all three TPR motifs. In detail, FAT10 interacted with AIPL1 wild type (Fig.  1D, lane 8) as well as with AIPL1-S328X (expression of FKBP and TPR motifs; Fig. 1D, lane 10), AIPL1 TPR (expression of only the TPR motifs; Fig. 1D, lane 11) and slightly with AIPL1 TPR+PRD (expression of the TPR motifs and the C-terminal proline rich domain, Fig. 1D, lane 12). No interaction was observed between FAT10 and the AIPL1-Q163X only expressing the FKBP domain of AIPL1 (Fig.  1D, lane 9). However, the most prominent interaction was detected with AIPL1 wild type (Fig. 1D, lane 8) suggesting that proper folding of AIPL1 is important for the interaction with FAT10.
Taken together, our data show that not only AIPL1 wild type interacts specifically with FAT10 in a direct manner but also pathogenic mutants such as AIPL1-A197P and AIPL1-C239R. Moreover, the data demonstrate that AIPL1 interacts via its TPR domains with both FAT10 UBDs (Fig. 1E).

FAT10 mRNA is expressed in human retina
So far, the direct interaction between the ubiquitin-like modifier FAT10 and the retinal chaperone AIPL1 was shown in vitro and in cellulo (Fig. 1). In general, AIPL1 is expressed only in developing rods and cones, mature rods and the pineal gland (6)(7)(8). Contrary to this, FAT10 mRNA and protein is described to be constitutively expressed in organs of the immune system (25), while its expression has not yet been investigated in retinal structures. As FAT10 expression is inducible in several tissues and cell lines by the pro-inflammatory cytokines TNF-α and IFN-γ, we first aimed to test whether FAT10 mRNA expression can also be induced in the retinoblastoma cell line Weri-Rb1 ( Fig. 2A). As control, FAT10 mRNA was markedly increased after stimulating HEK293 cells with TNF-α and IFN-γ for 24 h, while no mRNA expression was found in untreated cells ( Fig. 2A and Ref (34)). The same outcome was detected in Weri-Rb1 cells ( Fig. 2A). Remarkably, the upregulation of FAT10 mRNA after induction with TNF-α and IFN-γ was even higher than in HEK293 cells.
Showing that FAT10 mRNA expression is inducible in a retinoblastoma cell line led to the suggestion that FAT10 might also be expressed in human retina. To test this, quantitative real time RT-PCR was performed with three human retina samples. As control, AIPL1 mRNA expression was prevalent in all three retina samples but not in HEK293 cells (Fig. 2B). Next, FAT10 mRNA expression was measured in non-induced HEK293 cells, all three donor retina and in a purchased retinal RNA sample (Clontech; named hereafter donor 4) (Fig. 2C). While FAT10 mRNA was not expressed in unstimulated HEK293 cells, it was clearly expressed in three of the four retina samples (Fig. 2C, donors 1, 2 and 4), indicating that FAT10 mRNA is expressed in human retina. Finally, we also checked mRNA levels of the FAT10-specific E1 activating enzyme UBA6 as well as of the E2 conjugating enzyme USE1 ( Fig. 2D and 2E). Hereby, mRNA expression in retina samples was again compared to HEK293 cells that are known to express both enzymes (26). We could show that both, UBA6 and USE1, were also expressed in human retina although the expression levels fluctuated probably due to donor variations. Overall, these findings lead to the suggestion that FAT10 is not only expressed but might also be activated and conjugated in the human retina.

The rod phosphodiesterase 6 is a retinaspecific FAT10 conjugation substrate
Having shown that FAT10 mRNA is expressed in human retina and that FAT10 protein interacts with the retina-specific chaperone AIPL1, we searched for a retina-specific FAT10 conjugation substrate. Hereby we focused on a known AIPL1 client, the rod phosphodiesterase 6 (PDE6) which is a heterotetramer consisting of one α-subunit (PDE6α), one β-subunit (PDE6β) and two inhibitory γ-subunits (PDE6γ) (3,10,12,16). In order to investigate rod PDE6 as a possible FAT10ylation target we amplified the single rod PDE6 subunits out of human retina cDNA and cloned them into expression vectors, containing different peptide tags (42). A possible FAT10ylation of the single subunits was tested by overexpressing the respective subunit together with two different FAT10 variants, His-3xFLAG-FAT10 (further referred to as FLAG-FAT10) and the conjugation incompetent variant His-3xFLAG-FAT10-AV (FLAG-FAT10-AV) characterized by a Cterminal AV motif replacing the wild type GG. The experiments were performed under reducing conditions to visualize both, covalent and non-covalent interactions ( Fig. 3B and S2) as well as under denaturing conditions (Fig. 3A, C, D) to further confirm isopeptide formation between FAT10 and the single PDE6 subunits.
After immunoprecipitation of the FAT10 variants, western blot analysis showed that a small amount of His-PDE6α was covalently modified with FAT10 (Fig. 3A, lane 5). In contrast, the FAT10-AV mutant was not conjugated to PDE6α (Fig. 3A, lane 6). A possible non-covalent interaction with His-PDE6α was detectable for both FAT10 variants (Fig. S2A, top panel, lanes [5][6][7][8] which was independent of the C-terminal diglycine motif of FAT10. Of note, during inhibition of the proteasome with MG132 the amount of PDE6α-FAT10 conjugate increased (Fig. S2A, lane 6) indicating that FAT10ylation of PDE6α might lead to its proteasomal degradation. Accordingly, the other catalytic subunit PDE6β was tested as possible FAT10 conjugation substrate. An isopeptide-linked PDE6β-FAT10 conjugate was detected under reducing (Fig. 3B, lanes 5 and 6) as well as under denaturing conditions (Fig. 3C, lane 5). Additionally, a non-covalent interaction occurred between both FAT10 variants and PDE6β (Fig. 3B, top panel, lanes [5][6][7][8]. FAT10ylation of PDE6β was likewise dependent on the C-terminal diglycine motif since it was not detectable with the FAT10-AV mutant (Fig. 3B, lanes 7 and 8 and Fig. 3C, lane 6). Inhibition of the 26S proteasome led to only a slight increase in PDE6β-FAT10 conjugate amount (Fig. 3B, lane 6), consistent with a proteasomal degradation of the conjugate. Performing the same experiment using a low percentage SDS-PAGE revealed that PDE6β did not only become FAT10ylated with one FAT10 moiety but seemed to be oligo mono-FAT10ylated with at least two FAT10 molecules (Fig. S2B, lanes 5-8). Since no difference was observed between FAT10 WT and a FAT10 variant lacking all internal lysine residues, FAT10 K0, it can be presumed that several single FAT10 proteins were covalently conjugated to the subunit and no FAT10 chain was built (Fig. S2B, lanes 5 and 6 versus 7 and 8). As already seen in previous experiments, proteasomal inhibition with MG132 led to an accumulation of the conjugate (Fig. S2B, top panel). Completing this study for all PDE6 components, we investigated a possible FAT10ylation of the inhibitory PDE6γ subunit (Fig. 3C). In line with the catalytic subunits PDE6αβ, a covalent FAT10ylation of PDE6γ was detectable under denaturing (Fig. 3D, lane 5) and under reducing conditions (Fig. S2C, lanes 5 and 6). This was likewise dependent on the C-terminal diglycine motif (Fig. 3C, lanes 7 and 8 and Fig. S2C lanes 7 and 8). No clear statement can be drawn concerning a possible non-covalent interaction between PDE6γ and FAT10, as His-3xFLAG-tagged FAT10 showed a high unspecific background binding to the myc-beads ( Fig. 3C and S2C).
To investigate whether PDE6β becomes FAT10ylated not only in cellulo but also under in vitro conditions, recombinant FLAG-UBA6, His-USE1, FAT10, FAT10-AV and myc-PDE6β were incubated at 30°C followed by western blot analysis (Fig. 3E). With this approach a PDE6β-FAT10 conjugate was detected in presence of UBA6 and USE1 (Fig.  3E, lane 6), showing that PDE6β is also covalently modified with FAT10 in vitro. No FAT10ylation of PDE6β was observable with the FAT10-AV mutant (Fig. 3E, lane 7). Moreover, an oligo mono-FAT10ylation of the recombinant myc-PDE6β with two FAT10 moieties could be detected, which is in line with the in cellulo data (Fig. S2B). Interestingly PDE6β was also FAT10ylated in the absence of the E2 conjugating enzyme USE1 (Fig. 3E, lane 5) indicating that at least in vitro UBA6 alone was sufficient to mediate the FAT10 transfer onto PDE6β as recently published by our group for other substrates (39,40).
Taken together, our data identify rod PDE6 as a retina-specific FAT10 conjugation substrate. Both rod PDE6 catalytic subunits, PDE6α andβ as well as the inhibitory PDE6γ subunit become covalently modified with FAT10. Additionally, PDE6α and PDE6β interact noncovalently with FAT10, as well.

The PDE6β subunit is also FAT10ylated when incorporated in the holoenzyme
So far, only FAT10ylation of single PDE6 subunits was investigated raising the question, whether a properly formed PDE6 holoenzyme might be modified with FAT10 as well. Therefore, we overexpressed all three rod PDE6 subunits together with either FAT10 WT or the FAT10-AV mutant in HEK293 cells. In accordance with our previous publication where co-localization of the ectopically expressed PDE6 subunits had been shown by confocal microscopy (42), the interaction of the single subunits was confirmed by Co-IP experiments pointing to a formation of the PDE6 holoenzyme upon ectopic expression of the subunits in HEK293 cells ( Fig. S3A and S3B).
Combined immunoprecipitation and western blot analysis under reducing conditions (4% 2-ME) revealed, that PDE6β, although interacting with the other subunits and being eventually incorporated in the holoenzyme, was indeed FAT10ylated (Fig. 4A, lane 5). As already seen before, no signal was obtained with the FAT10-AV mutant (Fig. 4A, lane 6). However, both FAT10 variants were still able to noncovalently interact with PDEβ ( Fig. 4A, lanes 5 and 6). This led to the suggestion, that incorporation of the β-subunit into a holoenzyme complex or interaction of PDE6β with other subunits does not prevent its covalent modification with FAT10. Nonetheless, it cannot be excluded that only free PDE6β was FAT10ylated in our experimental setup. Unfortunately, no antibody for the immunoprecipitation of human PDE6 holoenzyme is available and also recombinant purification of the fully assembled human enzyme has not been established so far. We circumvented these issues by the usage of mouse PDE6 enzyme since mouse and human PDE6β share 93% sequence similarity at amino acid level. C57BL/6 mouse retina was prepared, lysed and PDE6 was immunoprecipitated with ROS-1 antibody that is described to only recognize fully assembled mouse PDE6 (43). The immunoprecipitated PDE6 was further used for in vitro FAT10ylation assays using recombinant human FAT10, UBA6 and USE1 (50% input of the recombinant proteins used is shown in Fig. S3C). Thereby we could show that PDE6β was clearly FAT10ylated with FAT10 wild type but not with FAT10-AV ( Fig.  4B, lanes 6-8). As already shown above, UBA6 alone was sufficient for PDE6β FAT10ylation (Fig. 4B, lane 6). Similar amounts of FAT10 and FAT10-AV were used while only FAT10 was activated by UBA6 and transferred onto USE1 (Fig. 4B, supernatant).
These data confirm that the catalytic β-subunit of PDE6 becomes FAT10ylated despite its incorporation into the holoenzyme.

AIPL1 stabilizes FAT10 and the PDE6β-FAT10 conjugate
As PDE6 is a client of the chaperone AIPL1, we investigated whether AIPL1 has an impact on the FAT10ylation of this enzyme. During their studies, Bett et al. had reported that FAT10 accumulates in the presence of AIPL1 (21). Therefore, a cycloheximide (CHX) chase experiment to monitor stability of monomeric FAT10 protein in the absence or presence of AIPL1 WT or AIPL1-A197P was performed (Fig. 5A). Treatment of cells with CHX for the indicated time periods inhibited protein de novo synthesis giving the possibility to track the degradation of proteins. In line with the reported observation, a slower degradation rate for monomeric FAT10 was observed when AIPL1 WT was co-expressed ( Fig. 5A, lanes 4-6 versus lanes [1][2][3]. Interestingly, the pathogenic mutant AIPL1-A197P likewise stabilized FAT10 (Fig. 5A, lanes 7-9 versus lanes [1][2][3] indicating that the direct interaction with FAT10 ( Fig. 1) is sufficient for its stabilization.
In a next step, the formation of the PDE6β-FAT10 conjugate was investigated in the presence of different AIPL1 variants (Fig. 5B). Again, an accumulation of FAT10 in presence of all AIPL1 variants was detectable (Fig. 5B, lysate IB: FLAG, lanes [8][9]. Interestingly, the amount of PDE6β also increased in the presence of all AIPL1 variants (Fig. 5B, lysate IB: HA, lanes [8][9][10]. In line with this, more PDE6β-FAT10 conjugate and especially an increased non-covalent interaction between FAT10 and PDE6β was observed in the presence of AIPL1, AIPL1-A197P and AIPL1-C239R (Fig. 5B, top panel, lanes [7][8][9][10]. This led to the suggestion that AIPL1, FAT10 and PDE6β might build a trimeric complex based on the interaction between FAT10 and AIPL1 resulting in a stabilization of the modifier and in consequence of the FAT10 conjugation substrate. In addition, we could show that the stabilization of the conjugate was dependent on the amount of AIPL1 (Fig. 5C). Upon transfection of cells with fixed plasmid amounts for HA-PDE6β and FLAG-FAT10 but increasing amounts of plasmid DNA encoding for AIPL1, more covalent and non-covalent interaction between PDE6β and FAT10 was detectable correlating with the increasing amounts of AIPL1 (Fig. 5C, top panel). Consistent with that, amounts of FAT10 and PDE6β were elevated upon stronger expression of AIPL1 (Fig. 5C, lysate IB: HA and FLAG, lane 9 versus 5). The same results were obtained when expressing the AIPL1 mutant AIPL1-A197P (Fig. 5D).
In summary we showed that the degradation rate of monomeric FAT10 was slowed down in the presence of AIPL1 WT or AIPL1-A197P resulting in an increased PDE6β-FAT10 conjugate formation that was dependent on the concentration of AIPL1.

FAT10 targets PDE6 for proteasomal degradation
FAT10ylation of substrates targets them for degradation by the 26S proteasome (36,38). We wanted to know whether this holds true for PDE6β since we saw a slight increase in PDE6β-FAT10 conjugate upon proteasomal inhibition (Fig. 3B). To verify proteasomal degradation of the PDE6β-FAT10 conjugate, we performed CHX chase experiments with HEK293 cells transiently expressing HA-PDE6β and FLAG-FAT10 (Fig. 6A). As a control, MG132 was added to block the catalytic activity of the 26S proteasome (Fig.  6A, lane 4). Moreover, endosomal acidification and maturation of autophagic vacuoles were inhibited with Chloroquine or Bafilomycin A1, respectively (Fig. 6A, lanes 5 and 6). In line with previous data ((26) and Fig. 5A), monomeric FAT10 was almost completely degraded within 5 hours of chase while its degradation was rescued upon proteasome inhibition (Fig. 6A, lysate IB:FLAG, lanes 3 vs 4). Likewise, the PDE6β-FAT10 conjugate was degraded over time and rescued by proteasomal inhibition although the amount of rescued protein was not quite as high as for monomeric FAT10 (Fig. 6A, top panel). As monomeric FAT10 itself is degraded over time, only small amounts could be additionally conjugated to PDE6β during the time of CHX treatment. Considering that no FAT10-specific deconjugating enzyme has been found and that covalent conjugation of FAT10 targets the bulk of its substrates to proteasomal degradation at the same pace as monomeric FAT10 (36,38), the observed decline of PDE6β-FAT10 conjugate over time is very likely due to proteasomal degradation. As neither treatment of cells with Chloroquine nor Bafilomycin A1 affected the degradation, an involvement of lysosomal degradation or autophagy could be excluded (Fig. 6A, top panel, lanes 5 and 6 and densitometric analysis in Fig.  6B). Densitometric analysis of the ECL signals of the conjugate normalized to the levels of β-actin in the respective lanes revealed that approximately 50% of the conjugate was degraded after 2.5 hours of CHX treatment (Fig. 6B).
Altogether, FAT10ylation of PDE6β targets it for proteasomal degradation as already documented for other FAT10 substrates (35).

Non-covalent interacting FAT10 inhibits PDE6 activity
To gain insights whether FAT10 has an impact on the functionality of PDE6 in addition to targeting it for proteasomal degradation, we first elucidated the FAT10ylation site within PDE6β. To this aim we co-expressed FLAG-FAT10 together with truncated PDE6β mutants ( Fig. 7A and B). Either the N-terminal GAFa domain alone (HA-PDE6β GAFa), both GAF domains (HA-PDE6β GAFa+b) or only the Cterminal catalytic PDEase domain (HA-PDE6β PDEase) were expressed. Full length PDE6β was used as control. Co-IP of the HA-tagged PDE6β truncations and subsequent western blot analysis showed that FAT10 interacted with all expressed domains in a non-covalent manner (Fig. 7B, lane 7-10). As expected, full length PDE6β was strongly FAT10ylated (Fig. 7B,  lane 10). In addition, a covalent modification was observed for the GAFa domain as well as for the PDEase domain (Fig. 7B, lanes 7 and 9). In the case of GAFa+b, a single band was detected at around 75 kDa representing most likely mono-FAT0ylation in the GAFa domain since it could be shown that this domain is even As control, repetition of the experiment using FAT10-AV instead of FAT10 wild type showed no covalent isopeptide linkages between FAT10 and different PDE6β versions while non-covalent interactions were not affected (Fig. S4A). This is in line with our finding of oligo mono-FAT10ylation of the PDE6β subunit (Fig. S2).
Within the PDE6β subunit the GAFa domain is important for cGMP binding while the PDEase domain possesses the catalytic activity for hydrolysis of cGMP. Moreover, the inhibitory PDE6γ subunits bind to these two domains (14,44). As we saw that FAT10 is able to interact with these two important domains in a covalent but also a non-covalent manner, we investigated a possible influence of FAT10 on the activity of PDE6 using a cGMP hydrolysis assay. Briefly, PDE6 activity was measured by using a defined amount of cGMP as a substrate. After hydrolysis of cGMP by PDE6, a supplemented 5`-nucleotidase enzymatically cleaves 5`-GMP into the nucleoside and phosphate, which in turn can be detected in a colorimetric assay by measuring the absorbance at 620 nm. To ensure a high amount of functional enzyme in the assay, lysate of mouse retina served as source for PDE6 holoenzyme while recombinant human FAT10 was used to test a possible influence of the modifier onto the PDE6 activity. Retinae of five C57BL/6 mice were prepared and lysed in a phosphate-free lysis buffer and protein amount was comparable for all prepared retinae as seen in a colloidal coomassie staining of whole retina lysate (Fig.  S4B). A control western blot analysis using the mouse retinal lysates showed similar PDE6β expression levels (Fig. S4C), indicating that comparable amounts of PDE6 were used in the assay.
After the reaction and incubation with the reagent reacting with free phosphate for 30 min, the absorbance at 620 nm was measured and amounts of produced 5`-GMP were calculated (Fig. 7C). Recombinant enzyme of PDE class I served as a positive control and revealed that the enzyme was active and able to hydrolyse cGMP (Fig. 7C, PDE) confirming the functional experimental setup. To exclude that neither impurities in the FAT10 or FAT10-AV preparations nor the FAT10 storage buffer itself influenced the system, cGMP was incubated only with recombinant FAT10, FAT10-AV or buffer. As no enhanced amounts of 5`-GMP could be detected, it can be concluded that the proteins themselves did not influence cGMP turnover (Fig. 7C, FAT10, FAT10-AV, FAT10 buffer). In addition, background signal within the lysates was measured and revealed that the phosphate-free lysis and subsequent desalting was successful as no background was detectable (Fig. 7C, lysate w/o substrate). Incubation of mouse retina lysate with cGMP resulted in an increased absorbance and accordingly increased amounts of 5`-GMP indicating that mouse PDE6 successfully hydrolysed its substrate (Fig. 7C, lysate). Strikingly, the addition of increasing amounts of FAT10 significantly decreased cGMP turnover in a dose-dependent manner (Fig. 7C, lysate + FAT10 (+) and (++)). Thus, FAT10 has a suppressive impact on PDE6 activity. To check whether the inhibition is dependent on the covalent FAT10ylation of PDE6, we additionally used comparable amounts of the conjugation defective mutant FAT10-AV. Interestingly, FAT10-AV was as potent as FAT10 wild type and significantly reduced PDE6 activity (Fig. 7C, lysate + FAT10-AV (+) and (++)). This showed that the non-covalent interaction between FAT10 and PDE6 was sufficient for PDE6 impairment. When measuring the absorbance starting from the beginning of the colorimetric reaction over a time period of 30 min, the same result was obtained (Fig. 7D). While a strong PDE6 activity was measurable in absence of FAT10 variants or in presence of FAT10 buffer, a strong decrease was observed with FAT10 or FAT10-AV (Fig. 7D, lysate + FAT10/-AV (+)). This decrease was even more prominent with higher amounts of FAT10 (Fig. 7D, FAT10/-AV (++)).
Taking all results together, we have shown that FAT10 mRNA is expressed in human retina and we suggest that its expression is inducible in retina under inflammatory conditions. Upregulation of FAT10 enables the covalent modification of the enzyme PDE6 whereby PDE6-FAT10 conjugates are stabilized by AIPL1. FAT10 protein expression has two different functional consequences for PDE6. While covalent modification targets PDE6 to proteasomal degradation, the non-covalent interaction of FAT10 with PDE6 downregulates PDE6 function thus resulting in a reduction of cGMP hydrolysis.

Discussion
The retinal chaperone AIPL1 is very important for the maintenance of retinal phototransduction in photoreceptors and a loss of AIPL1 leads to rapid degeneration of rods and cones (45). Already single point mutations in the Aipl1 gene cause the retinal dystrophy LCA resulting in the degeneration of photoreceptors (6). Recently, a connection between AIPL1 and the ubiquitinlike modifier FAT10 was identified (21). Considering that AIPL1 is only expressed in photoreceptor cells and the pineal gland (6-8), we investigated a possible role of FAT10 in photoreceptor cells. Our data revealed that FAT10 mRNA is indeed expressed in retina and that the rod PDE6 protein is a target of FAT10ylation as well as a non-covalent interaction partner of the FAT10 protein, leading either to proteasomal degradation of PDE6 or to a decreased PDE6 cGMP hydrolysis activity.
In our previous publication we had shown that AIPL1 interacts with FAT10 in a non-covalent manner (21). We now confirmed these findings and characterized this interaction further showing that AIPL1 binds to both UBLdomains of FAT10 (Fig. 1C). This is in contrast to other FAT10 interaction partners such as NUB1L, which was shown to bind only to the N-terminal UBL domain of FAT10 (46). In line with other AIPL1 interaction partners, FAT10 was also interacting with the AIPL1 TPR motifs, that are already described to be important for protein-protein interactions (Fig.  1D) (41,47). Interestingly, neither point mutations within the TPR domain (A197P, C239R) nor mutations in any other domain of AIPL1 (e.g. R38C in FKBP, P376S in prolinerich domain) influenced the interaction with FAT10 ( Fig. 1 and Fig. S1A and S1B). As AIPL1 binds both FAT10 UBD domains (Fig.  1C), it seems likely that FAT10 binds to multiple sites within the TPR region diminishing the effect of a single point mutation. Interestingly, no interaction of AIPL1 with ubiquitin was observed in vitro (Fig. S1C). This might be explained by the fact that the surface charge distributions of ubiquitin and FAT10 differ entirely from each other (28). Hereby, the hydrophobic patch in ubiquitin (L8, I44 and V70) that serves in most cases for ubiquitin-protein interactions is neither conserved in the FAT10-N nor the FAT10-C domain. As we could show that AIPL1 is able to interact with both, N and C-terminal UBDs of FAT10, one could further speculate that the single UBD of ubiquitin is not sufficient for a stable interaction with AIPL1.
So far, FAT10 expression in retinal cells has not been investigated. Here we provide data, that FAT10 mRNA expression is inducible in the retinoblastoma cell line Weri-Rb1 upon cytokine treatment with TNFα and IFNγ ( Fig.  2A). We further show FAT10 mRNA expression in human retina samples (Fig. 2C).
Here, one has to keep in mind that the retina is a multi-layer structure including different cell types. However, the precise cell type expressing FAT10 could not be determined due to the lack of a suitable antibody reactive against mouse FAT10 that would be sensitive and specific enough to visualize endogenous mouse FAT10 in immunostainings of mouse retina. Likewise difficulties in obtaining human retina samples hindered us from performing additional biochemical interaction studies for human AIPL1 and FAT10. However, since FAT10 mRNA expression was not detectable in every donor sample and as it can be excluded that donors of the retinal tissues suffered from a known retinopathy, it can be suggested that FAT10 might not be constitutively expressed but rather is upregulated as a consequence of inflammation. Therefore, it cannot be excluded that FAT10 mRNA and protein expression is also induced during the pathogenesis of retinopathies such as LCA. Increased levels of cGMP caused by the inhibition of PDE6 with the drug zaprinast in porcine retinal explants resulted in the production of TNFα and IL-6 (48), a cytokine combination which was shown to induce FAT10 expression in HepG2 cells (49). This leads to the hypothesis that FAT10 might be expressed in the retina of LCA patients due to unproductive PDE6 assembly caused by pathogenic AIPL1 mutants which in turn cause an inflammatory environment enabling FAT10 induction that further aggravates the situation.
We have identified the rod PDE6 as the first retina-specific FAT10 conjugation substrate (Fig. 3) targeting it to proteasomal degradation (Fig. 6) as already seen for other FAT10 substrates such as the ubiquitin E1 activating enzyme UBE1, the transcription factor JunB or the ubiquitin deconjugating enzyme OTUB1 (35,39,40). As AIPL1 is a chaperone for PDE6 (3), we investigated whether AIPL1 might influence PDE6 FAT10ylation. Indeed, we showed that both, AIPL1 WT and AIPL1-A197P, stabilized monomeric FAT10 and thus also the PDE6β-FAT10 conjugate (Fig. 5). In contrast, no significant differences between AIPL1 WT and pathogenic AIPL1 mutants causing LCA were observed concerning their interaction with FAT10 or their ability to stabilize FAT10ylated PDE6. It should be kept in mind, however, that FAT10 interacts strongly with NUB1, which is highly expressed in the brain and accelerates the degradation of FAT10 and FAT10 conjugates by the 26S proteasome (46,50). Since NUB1 was shown to bind to wild type AIPL1 but not to LCA causing variants of AIPL1 (47) it is quite possible that binding of NUB1 to AIPL1 is required to more effectively sequester FAT10 and keep it from inhibiting PDE6 or from mediating its degradation. This possibility is currently experimentally addressed in our laboratory. Importantly, AIPL1 and PDE6 are exclusively expressed in rod and cone photoreceptors within the retina (3,7,8). Therefore, the identification of AIPL1 as an interaction partner and PDE6 as a substrate of FAT10 conjugation led to the suggestion that FAT10 itself is expressed by the photoreceptors functioning in the maintenance of proteostasis during phototransduction.
So far, it is not clear where FAT10 is localized within photoreceptors. Since AIPL1 is expressed in the inner segment and is not readily detected in the outer segment (7), one possibility would be that it captures FAT10 and hinders its movement to the outer segment. Within this study, an inhibitory effect of FAT10 onto the PDE6 activity was found (Fig. 7). Although whole retina lysate was used for the investigation, one can assume that FAT10 specifically inhibited PDE6 activity as to our knowledge PDE6 is the only cGMP-specific class of PDE expressed in photoreceptors of the retina (51). Showing that FAT10 influences PDE6 activity by its non-covalent interaction resulting in a decreased cGMP hydrolysis (Fig.  7), hindrance of FAT10 translocation into the outer segment would help to maintain PDE6 function within the already exported PDE6 in the disc membranes. Additionally, we saw an accumulation of PDE6β-FAT10 conjugate in the presence of AIPL1 suggesting that a trimeric complex of PDE6, FAT10 and AIPL1 is formed. Thereby FAT10ylated PDE6 might be retained in the inner segment. Thus, eventually reduced amounts of PDE6 in the outer segment due to a lack of renewal or to a FAT10dependent inhibition of PDE6 function could cause visual impairment during inflammation as e.g. in chorioretinitis or during retinopathies. Next to its localization, several possibilities how FAT10 can inhibit PDE6 activity in a noncovalent manner appear imaginable. Structural studies revealed that PDE6 undergoes a conformational change upon activation (14,16). Therefore, one could speculate that FAT10 blocks this conformational change of PDE6. It was further shown that two transducin αsubunits binding to the catalytic domain of the catalytic subunits are necessary for PDE6 activation (52). As FAT10 is binding to the catalytic domain in a covalent and non-covalent manner (Fig. 7), it may block these binding sites. Finally, the cGMP substrate as well as FAT10 are both binding to the GAFa domain opening the possibility of competition for binding.
Interestingly, this is another example showing that the non-covalent interaction of FAT10 with a substrate protein has a strong influence on substrate functionality. While the activity of the deconjugating enzyme OTUB1 is enhanced by the interaction with FAT10 (39), the activation of the ubiquitin-like modifier SUMO is blocked when FAT10 interacts with the SUMO E1 activating enzyme AOS1/UBA2 (40). In the present study, FAT10 also has a negative impact onto PDE6 activity as non-covalent interaction with FAT10 seems to inhibit the capability of PDE6 for cGMP hydrolysis (Fig. 7). Overall, these examples imply that FAT10 can have major impacts on substrate proteins besides targeting them to proteasomal degradation.
Taken together, we have identified the rod PDE6 as a target protein for FAT10ylation as well as a non-covalent interaction partner of FAT10 and show that FAT10 downregulates the activity by guiding PDE6 to proteasomal degradation as well as by inhibiting its activity. Since PDE6 is a very important protein for the transmission of vision, upcoming studies using different mouse models will help to elucidate the role of FAT10 in more detail and may also open new possibilities for the development of treatment options for certain retinopathies or inflammatory eye diseases, adversely affecting vision or even causing blindness.

Cell lines, mice and human tissue
HEK293 cells were originally obtained from ATCC (Manassas, VA, USA) and cultivated in Iscove`s modified Dulbecco`s Medium (IMDM) (Pan Biotech) supplemented with 10% fetal calf serum (FCS; GIBCO), 1% stable glutamine (100x, 200 mM, Biowest) and 1% penicillin/streptomycin (100x, Pan Biotech). Weri-Rb1 cells (ATCC, Manassas, VA, USA) were cultivated in RPMI 1640 medium (Pan Biotech) supplemented with 10% FCS (GIBCO) and 1% penicillin/streptomycin (100x, Pan Biotech). Cells were routinely tested to be negative for mycoplasma contamination using the MycoAlert TM Mycoplasma Detection Kit (Lonza). C57BL/6 mice (H-2b) were originally purchased from Charles River, Germany. Mice were kept in a specific pathogen-free facility at the University of Konstanz and for all experiments, 6-8 week old mice were used. Sacrifice and organ retrieval was approved by the Review Board of Regierungspräsidium Freiburg. Human retina tissue was provided by the Moorfields Lions Eye Bank (Moorfields Eye Hospital, London EC1V 2PD, UK) and the study was approved by the Moorfields Biobank Internal Ethics Committee. All of the experiments were undertaken with the understanding and written informed consent of each subject.
GST was expressed in E. coli BL21(DE3) using the pGEX-4T-3 expression construct and bacteria were grown in modified LB medium at 37°C to an OD600 of 0.6. Protein expression was induced for 5 h at 20°C by the addition of 0.1 mM IPTG. Cells were harvested and lysed as described above and GST was pulled down for 2 h at 8°C by using GSH-beads (Sigma-Aldrich). Elution was performed for 30 min at 8°C in 50 mM Tris pH 8.0 supplemented with 5 mM GSH (Fluka). Free GSH was removed using PD10 columns (GE Healthcare) and protein was stored in 50 mM Tris pH 8.0 at -80°C. For all purifications of recombinant proteins, purity and concentrations were confirmed by BCA assays (Thermo Scientific) and Coomassie stained SDS gels.
Bacteria were harvested by centrifugation (8000 x g, 8°C, 15 min) and mechanically lysed with at least two cycles at 2.5 kbar in a cell disrupter (Constant Cell Disruptor TS, Constant Systems Ltd.) in binding buffer (20 mM Tris pH 7.6, 150 mM NaCl, 20 mM imidazole supplemented with 1x protease inhibitor mix (cOmplete TM , Mini, EDTA-free Protease Inhibitor Cocktail; Roche). Cleared lysates were used for Ni pulldown with Ni Superflow60 resin (Takara). His-tagged proteins were eluted by the addition of buffer containing 500 mM imidazole. Buffer exchange for 50 mM Tris-HCl pH 8.0 and Ulp1 digest were performed by dialysis overnight at 8°C. After removal of the 6His-SUMO tag and His-Ulp1 by Ni pulldown, HA-tagged AIPL1 variants were applied to size exclusion chromatography using a 16/600 75 column (GE Healthcare). Purified proteins were stored in 50 mM Tris-HCl pH 8.0, supplemented with 10% glycerol at -80°C. Protein purity and concentration were confirmed by BCA assays (Thermo Scientific), dot blots with anti-HA peroxidase-conjugated antibody (Sigma) and colloidal Coomassie stained SDS gels.
The expression of the pSUMO-myc-PDE6β construct in E.coli BL21(DE3) RIPL bacteria was induced with 0.4 mM IPTG at an OD600 of 0.5-0.7. Bacteria were grown overnight at 21°C and harvested via centrifugation before cell lysis in 5 ml/g binding buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 20 mM imidazole, 1 mM TCEP) using a cell disruptor (Constant Cell Disruptor TS, Constant Systems Ltd.) with at least two cycles at 2.5 kbar. Capture Ni 2+ affinity chromatography was performed with HisTrap FF 5 ml columns using the AektaPure system (both GE Healthcare). 6His-SUMOmyc-PDE6β was eluted with elution buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 500 mM imidazole, 1 mM TCEP) and the Ulp1 digest was performed overnight at 8°C after buffer exchange with HiPrep 26/10 Desalting columns (GE Healthcare). After a second affinity chromatography with HisTrap FF 5 ml columns and the AektaPure system (both GE Healthcare), the 6His-SUMO as well as the Ulp1-6His was separated from myc-PDE6β. The final myc-PDE6β was stored in storage buffer (20 mM Tris-Hcl pH7.5, 50 mM NaCl, 10 mM MgCl2, 1 mM TCEP) at 4°C and immediately used for the in vitro experiments.

RT-qPCR for human retina samples
Three human retina tissues were provided by the Moorfields Lions Eye Bank (Moorfields Eye Hospital, London EC1V 2PD, UK). The study was approved by the Moorfields Biobank Internal Ethics Committee. Sex and age characteristics of the donors were: donor 1 (male, 36 years old), donor 2 (female, 37 years old), donor 3 (male, 81 years old). A fourth sample of retina RNA was purchased from Clontech and was described to originate from a Caucasian child. All of the experiments were undertaken with the understanding and written informed consent of each subject and the study methodologies conformed to the standards set by the Declaration of Helsinki. After removal, retina were stored in RNAlater until RNA was extracted using RNeasy Mini Kit (Qiagen) in combination with QIAshredder (Qiagen) according to the manufacturer`s protocol. RNA was reverse-transcribed into cDNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) according to the manual. As control, endogenous FAT10 expression was induced by treating HEK293 cells and Weri-Rb1 cells with the proinflammatory cytokines TNFα (600U/ml) and interferon IFNγ (300U/ml) (both from Peprotech) for 24 h. Quantitative PCR was performed with the 7900 HT Fast Real Time PCR instrument (Applied Biosystems) using primers for human FAT10 (FAT10 fwd: CTGTCTCTGGTTTCTGGCCC; FAT10 rev: GGAAGCATTGGGAGCCATCT; (26)), human AIPL1 (Hs_AIPL1_1_SG; Qiagen), UBA6 (Hs_UBA6_1_SG; Qiagen), USE1 (Hs_FLJ13855_1_SG; Qiagen) and GAPDH (Hs_GAPDH_1_SG; Qiagen).

Quantification and statistical analysis
For all figures, error bars show mean ± SEM. Densitometric analysis was performed by calculating the respective ECL signal with Image Lab 4.1 software. Statistical analysis for cGMP hydrolysis assay was performed using one-way ANOVA with Dunnett correction with GraphPad Prism 6 (GraphPad Software). Differences were considered as significant for p values of *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.   Immunoprecipitations were performed using anti-FLAG M2 affinity gel, anti-HA agarose or anti-c-myc affinity gel. SDS-PAGE under reducing conditions with 4-12% gradient gels (NuPage) was followed by western blot analysis using the respective antibodies. β-actin served as loading control. Shown are single experiments out of three experiments each with similar outcomes. B, The experiment was performed as described in C, but under reducing conditions (4% 2-ME in the gel sample buffer). E, In vitro FAT10ylation of PDE6β. Recombinant proteins were incubated for 60 min at 30°C in an ATP containing in vitro buffer. The reaction was stopped with 5 x SDS gel sample buffer containing 4% 2-ME followed by SDS-PAGE and western blot analysis. The exact recombinant protein amounts can be found in the methods section. Shown is one experiment out of two (A and D) or three (B, C, E) experiments with similar outcomes.

Figure 4. The PDE6β subunit is also FAT10ylated when incorporated in the holoenzyme. A, PDE6
FAT10ylation under overexpression conditions in HEK293 cells. Cells were transiently transfected with expression constructs for all rod PDE6 subunits, FLAG-FAT10 or FLAG-FAT10-AV. An immunoprecipitation against HA-PDE6β was performed and samples were separated on 4-12% gradient gels (NuPage) followed by western blot analysis. Unspecific background is marked with an asterisk. B, Conjugation of FAT10 to mouse PDE6 holoenzyme. Retina from C57BL/6 mice were prepared and lysed in RIPA buffer. Mouse PDE6 holoenzyme was immunoprecipitated overnight with ROS-1 antibody coupled to protein A sepharose. Mouse IgG2a was used as isotype control for the IP. Immunoprecipitated PDE6 was used in an in vitro FAT10ylation assay with recombinant proteins, as indicated. Reactions were incubated at 30°C for 30 min. Supernatant samples were analyzed under nonreducing conditions and IP samples were analyzed under reducing conditions using 4-12% gradient gels (NuPage) and the respective antibodies. The recombinant protein amounts can be found in the methods section. Shown is one experiment out of three experiments with similar outcomes, respectively. showing the degradation rate of monomeric FAT10 analyzed in HEK293 cells co-expressing myctagged AIPL1 or AIPL1-A197P. Before harvesting, cells were treated for 2.5 or 5 h with 50 µg/ml CHX to inhibit protein de novo synthesis. One experiment out of three experiments with similar outcomes is shown. B, HEK293 cells were transiently transfected with expression constructs for HA-PDE6β, FLAG-FAT10, myc-AIPL1, myc-AIPL1-A197P and myc-AIPL1-C239R, as indicated. After cell lysis, HA-PDE6β was immunoprecipitated with anti-HA agarose and proteins were separated under reducing conditions using 4-12% gradient gels (NuPage). FAT10ylation of PDE6β was visualized by western blot analysis. Shown is one experiment out of three experiments with similar outcomes. C and D, Investigation of PDE6β FAT10ylation was performed as described in (B) but with increasing amounts of expression constructs (0.5 -5 µg DNA) for untagged AIPL1 WT (C) or AIPL1-A197P (D). One experiment out of three (AIPL1 WT; C) or out of two (AIPL1-A197P; D) experiments with similar outcomes is shown. Figure 6. FAT10 targets PDE6 for proteasomal degradation. A, The degradation rate of PDE6β-FAT10 conjugate in transiently transfected HEK293 cells, 24 h after transfection, was monitored by treating the cells with CHX (50 µg/ml; 2.5 or 5 h). When indicated, cells were additionally treated for 6 h prior to harvesting with MG132 (10 µM) to block proteasomal degradation or with Bafilomycin A1 (0.2 µM) or Chloroquine (100 µM) to interfere with lysosomal degradation. Immunoprecipitation was performed using anti-HA agarose and proteins were separated under reducing conditions. One representative experiment out of four experiments with similar outcomes is shown. B, Densitometric analysis of ECL signals of CHX chase experiments. ECL signals of PDE6β-FAT10 conjugate were normalized to the ECL signals of β-actin. Values of untreated cells were set to unity. Shown is the mean ± SEM of four experiments with similar outcomes (n=4). Shown is one out of three experiments with similar outcomes. C, cGMP hydrolysis assay in presence or absence of FAT10 using mouse PDE6. Retina from five C57BL/6 mice were prepared, lysed in phosphate-free lysis buffer and desalted using PD Mini Trap G-25 columns. Production of 5`-GMP was measured by subsequent enzymatic cleavage of 5`-GMP into the nucleoside and phosphate via a 5`-nucleotidase. The amount of released phosphate was quantified with a colorimetric reaction in a modified Malachite Green assay by measuring the absorbance at 620nm. As inhibitors, recombinant FAT10 ((+) = 1.67 µg, (++) = 5 µg) or FAT10-AV ((+) = 1.5 µg, (++) = 4.5 µg) were added. FAT10 storage buffer served as a control. Shown is the mean ± SEM of five experiments (n=5). Significance was calculated using one-way ANOVA with Dunnett correction. A p values of p < 0.0001 (****) was considered to be highly statistically significant. D, Absorbance was measured during the 30 minutes of colorimetric reaction for the experiment described in (C). Shown is the mean ± SEM of five experiments (n=5).