PAM forms an atypical SCF ubiquitin ligase complex that ubiquitinates and degrades NMNAT2

PHR (PAM/Highwire/RPM-1) proteins are conserved RING E3 ubiquitin ligases that function in developmental processes, such as axon termination and synapse formation, as well as axon degeneration. At present, our understanding of how PHR proteins form ubiquitin ligase complexes remains incomplete. Although genetic studies indicate NMNAT2 is an important mediator of PHR protein function in axon degeneration, it remains unknown how PHR proteins inhibit NMNAT2. Here, we decipher the biochemical basis for how the human PHR protein PAM, also called MYCBP2, forms a noncanonical Skp/Cullin/F-box (SCF) complex that contains the F-box protein FBXO45 and SKP1 but lacks CUL1. We show FBXO45 does not simply function in substrate recognition but is important for assembly of the PAM/FBXO45/SKP1 complex. Interestingly, we demonstrate a novel role for SKP1 as an auxiliary component of the target recognition module that enhances binding of FBXO45 to NMNAT2. Finally, we provide biochemical evidence that PAM polyubiquitinates NMNAT2 and regulates NMNAT2 protein stability and degradation by the proteasome.

Post-translational modification of proteins by ubiquitination is a prominent mechanism for regulating protein trafficking, localization, stability, and activation. Ubiquitination plays an important role in cell cycle control (1,2) and oncogenesis (3,4). In the nervous system, ubiquitination impacts both neuronal development and function (5,6) and is implicated in neurodegenerative disease (7). Proteins can be monoubiquitinated by the addition of a single ubiquitin molecule or polyubiquitinated with multiple ubiquitin species. Polyubiquitination occurs from successive reactions by a series of ubiquitin ligases called E1, E2, and E3. One outcome of polyubiquitination is degradation by the 26S proteasome.
The Pam/Highwire/RPM-1 (PHR) proteins, which include human PAM/MYCBP2, mouse Phr1, Drosophila Highwire, and Caenorhabditis elegans RPM-1, have been suggested to form atypical SCF complexes that lack a Cullin (17). PHR proteins are RING domain E3 ligases that play conserved functions in the developing nervous system where they regulate axon termination, axon guidance, and synapse formation (18 -26). In mature neurons, PHR proteins take on an important role in axon degeneration following injury (27,28). Outside of the nervous system, PHR proteins are implicated in oncogenesis (29 -34). Despite growing recognition of the functional importance of PHR proteins, relatively little is known about how components of atypical PHR ubiquitin ligase complexes assemble. Previous work using model organisms and cell-based biochemistry identified a conserved F-box protein, called FBXO45 in mammals or FSN-1 in C. elegans, that binds PHR ubiquitin ligases (35)(36)(37). At present, we have the best understanding of how C. elegans RPM-1 forms a complex with FSN-1. Biochemistry with 293 cells and in vivo biochemistry from the worm nervous system indicate that RPM-1 contains multiple FSN-1binding domains that interact with FSN-1 (38,39). It remains unknown whether a similar biochemical mechanism underpins assembly of ubiquitin ligase complexes formed by other PHR proteins, including human PAM. Furthermore, we do not know whether PHR proteins bind directly or indirectly to F-box proteins.
Proteomics and genetics in flies, as well as biochemistry in 293 cells, have suggested PHR proteins can form noncanonical SCF complexes, which contain SKP1 but lack a Cullin (37,40). Although these studies provided evidence that individual components can interact, biochemistry showing that FBXO45 and SKP1, but not CUL1, assemble into a complex with PAM remains notably absent. We also know nothing about the biochemical function of SKP1 in PHR ubiquitin ligase complexes. Is SKP1 an adaptor that mediates FBXO45 binding to PAM? Does SKP1 influence FBXO45 binding to substrates?
Studies with mouse Phr1 and fly Highwire indicate NMNAT2 mediates PHR protein effects on axon degeneration (17,27,28). NMNAT2 is part of the nicotinamide mononucleotide adenylyltransferase protein family that catalyzes formation of NAD ϩ (41)(42)(43). NAD is an essential coenzyme for several biological process, such as cell death, energy metabolism, and calcium mobilization. NMNAT2 is highly expressed in the brain, and along with axon protection following injury (44 -49), it is implicated in neurodegenerative disease (50 -53). Although functional genetics indicate PHR proteins can inhibit NMNAT2, the biochemical mechanism by which this occurs remains untested. For example, we lack evidence that PHR proteins interact biochemically with NMNAT2. Furthermore, it is unknown whether PHR proteins polyubiquitinate NMNAT2 and if so whether ubiquitination results in NMNAT2 degradation by the proteasome.
In this study, we comprehensively evaluate the biochemistry that underpins assembly of the PAM ubiquitin ligase complex. We demonstrate that PAM assembles into a noncanonical SCF complex that contains FBXO45 and SKP1 but not CUL1. Importantly, our results indicate the biochemical mechanism behind this is a direct interaction between FBXO45 and PAM, which facilitates recruitment of SKP1 into the ligase complex. This precludes the need for a Cullin scaffold.
Interestingly, we uncover a novel role for SKP1 in this atypical SCF complex. SKP1 acts as part of the target recognition module with FBXO45 rather than acting as an adaptor to mediate formation of the PAM/FBXO45/SKP1 complex. Thus, SKP1 plays a very different role in the noncanonical PAM/ FBXO45/SKP1 complex compared with traditional SCF complexes formed by the Rbx1 E3 ligase.
Increased insight into the biochemistry underpinning assembly of the PAM/FBXO45/SKP1 ubiquitin ligase complex prompted us to dissect the biochemical relationship between PAM and NMNAT2. Our results indicate NMNAT2 binds to the PAM/FBXO45/SKP1 complex via FBXO45. PAM polyubiquitinates NMNAT2 thereby regulating NMNAT2 protein stability and degradation by the proteasome. Importantly, experiments with PAM overexpression and endogenous PAM knockout cell lines indicate that PAM E3 ubiquitin ligase activity regulates NMNAT2 protein turnover. Our findings have interesting implications, given the importance of PHR proteins and NMNAT2 in axon degeneration, and growing links between NMNAT2 and neurodegenerative disease.

FBXO45 binds directly to PAM via a conserved mechanism
Previous biochemistry using HEK 293 cells and in vivo biochemistry from transgenic animals showed that C. elegans

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
RPM-1 binds the F-box protein FSN-1 via three domains in RPM-1 called FBD1, FBD2, and FBD3 (38,39). We tested whether human PAM/MYCBP2 relies upon a similar conserved mechanism to bind the human F-box protein FBXO45. To do so, we began by generating three fragments of PAM, and we tested their binding to FBXO45 in HEK 293 cells (Fig. 1A).
To minimize the possibility that we might miss FBXO45 interaction sites, we generated PAM fragments that overlapped with one another. As shown in Fig. 1B, only the central fragment of PAM (GFP-PAM Central) coprecipitated with FLAG-tagged FBXO45 (FLAG-FBXO45). To further map this interaction, we transfected 293 cells with FLAG-FBXO45 and four smaller domains within the central PAM fragment, termed D4 to D7 (Fig. 1A). We only observed an interaction between PAM D5 and FBXO45 (Fig. 1C).
Interestingly, D5 contains the conserved region of PAM called FBD1, which was previously shown to be sufficient for binding to C. elegans FSN-1 (Fig. 1A) (38). This prompted us to further determine whether the FBD1 of PAM was sufficient for binding to FBXO45. Coimmunoprecipitation (coIP) experiments with several smaller constructs derived from PAM D5 showed only GFP-FBD1 coprecipitated with FLAG-FBXO45 (Fig. 1D). Thus, FBD1 of PAM is capable of binding FBXO45.
Next, we wanted to determine whether the interaction between PAM and FBXO45 was direct. We expressed GST-PAM D5, GST, His 6 -FBXO45, or His 6 in Escherichia coli.
Lysates from E. coli expressing GST-PAM D5 or the negative control GST were incubated with glutathione (GT) beads. E. coli lysates expressing His-FBXO45 or His alone were incubated with GT beads bound with GST-PAM D5 or GST. Pulldown of His-FBXO45 was only observed when GST-PAM D5 was used as bait (Fig. 1E). As a control, lysates expressing His-FBXO45 or His alone were also precipitated with nickel beads and immunoblotted with anti-His antibody. This ensured equal amounts of His-FBXO45 were applied to GT beads bound with GST-PAM D5 or GST (Fig. 1E). Given that homologs of PAM and FBXO45 are not present in E. coli, these results show that PAM D5 is sufficient for direct binding to FBXO45 via the FBD1 domain (Fig. 1F).

RPM-1 and FSN-1 interact in subcellular axonal compartments in vivo
Our results here show that the FBD1 domain of PAM binds directly to FBXO45 (Fig. 1). Similarly, prior work showed the PAM ortholog RPM-1 utilizes FBD1 to bind the F-box protein FSN-1 (38). Because these findings suggest a conserved, direct Scatterplots show mean fluorescence lifetime of RPM-1::GFP at axon termination sites of indicated genotypes. RPM-1::GFP lifetime is significantly reduced when mCherry::FSN-1 is coexpressed compared with RPM-1::GFP alone. RPM-1::GFP lifetime is not reduced by coexpression of mCherry cell fill. Data are shown for nine animals (n ϭ 9) from two independent experiments for all genotypes. Significance was determined using Student's t test, and error bars represent standard error of mean. ***, p Յ 0.001; ns ϭ not significant.

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
interaction occurs between PHR proteins and F-box proteins, we wanted to explore where this interaction occurs in neurons in vivo. Therefore, we turned to in vivo two-photon FLIM-FRET (2pFLIM) using C. elegans RPM-1 and FSN-1.
Having shown RPM-1 and FSN-1 colocalize in different subcellular compartments, we performed 2pFLIM using these reagents (Fig. 2D). Fluorescence lifetime imaging is a method of quantifying FRET, which provides many advantages, including enhanced sensitivity over ratiometric imaging or other intensity-based FRET measurements (55,56). This approach allows us to test whether both proteins are in close molecular proximity. Importantly, FLIM only measures the fluorescence lifetime of the donor, which becomes shortened when the fluorophore undergoes FRET. 2pFLIM revealed that the fluorescence lifetime of GFP (donor) fused to RPM-1 was shorter when mCherry::FSN-1 was present at the axon termination site of mechanosensory neurons (Fig. 2E). In contrast, RPM-1::GFP fluorescence lifetime was not affected by expression of mCherry alone (Fig. 2E). Quantitation of 2pFLIM showed that RPM-1::GFP fluorescence lifetime was significantly reduced in the presence of mCherry::FSN-1 but not mCherry (Fig. 2F). These results show for the first time that RPM-1 physically interacts with FSN-1 at axon termination sites in vivo and is consistent with RPM-1 binding directly to FSN-1.

Conserved amino acids in FBD1 are sufficient for PAM binding to FBXO45
To further evaluate the mechanism by which the FBD1 domain of PAM mediates direct binding to FBXO45, we performed coIP with several FBD1 point mutants. Fig. 3A highlights several conserved motifs in FBD1 that we tested by alanine screening. Conserved amino acids in either individual motifs or residues across a combination of different motifs were mutated to alanine (Fig. 3A). GFP-tagged FBD1 mutants were cotransfected with FLAG-FBXO45 in 293 cells and tested for binding (Fig. 3B). GFP-FBD1 alanine mutants were expressed at similar levels, and FBXO45 coIP was reduced when three different motifs were mutated in FBD1, GRR (G2404A/ R2406A/R2408A), FE (F2429A/E2432A), and WCL (W2462A/ C2463A/L2464A) (Fig. 3B). Quantitation showed FBXO45

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
binding to FBD1 GRR and WCL alanine mutants was significantly reduced compared with WT FBD1 (Fig. 3C).
Collectively, these observations support two conclusions. 1) Several conserved motifs in the FBD1 domain of PAM are necessary for binding to FBXO45. 2) The NDD motif of FBD1 restricts binding to FBXO45.

FBXO45 recruits SKP1 into a complex with PAM
Previous biochemical and proteomic experiments using cultured cells showed SKP1 binds to FBXO45 (37,57). Affinity purification proteomics with Highwire in flies also identified the SKP1 ortholog SkpA (40). These observations invoke several interesting questions. How is SKP1 recruited to PHR ubiquitin ligase complexes? Does SKP1 function as an adaptor that helps stabilize the interaction between FBXO45 and PAM? Can SKP1 influence target recognition?
To answer these questions, we began by testing whether SKP1 binds PAM. HEK 293 cells were cotransfected with HA-SKP1 and GFP-PAM fragments (N-term, Central, and C-term, Fig. 4A). We did not observe coprecipitation of HA-SKP1 with any GFP-PAM fragments (Fig. 4C). Consistent with prior work (37,57), HA-SKP1 did coIP with GFP-FBXO45 (Fig. 4C). These results indicate FBXO45 and SKP1 interact, and this is likely to be independent of PAM.
Because SKP1 binds FBXO45 but not PAM, we tested whether FBXO45 can recruit SKP1 into a complex with PAM. To do so, we cotransfected different combinations of GFP-PAM D5, FLAG-FBXO45, and HA-SKP1, and we analyzed binding by coIP. Although GFP-PAM D5 and HA-SKP1 did not coprecipitate, we did observe binding of these two proteins when FLAG-FBXO45 was included in transfections (Fig. 4D). Similar results occurred with full-length PAM, as we only observed coprecipitation of HA-SKP1 with GFP-PAM when FLAG-FBXO45 was coexpressed (Fig. 4E).
Next, we mapped the interaction between FBXO45 and SKP1. FBXO45 is highly conserved across species (Fig. S1) and contains three domains: an F-box domain, a conserved domain (CD) of unknown function, and a SPRY domain ( Fig. 4B and Fig.  S1). We generated three FBXO45 constructs: the F-box domain alone, the CD domain alone, and a construct containing both CD and SPRY domains (Fig. 4B). The SPRY domain alone was not generated because it was problematic to construct. We cotransfected GFP-tagged FBXO45 domains or full-length GFP-FBXO45 with HA-SKP1 in 293 cells, and evaluated bind-

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
ing by coIP. We only observed an interaction between SKP1 and the F-box domain of FBXO45. This indicates the F-box domain is sufficient for FBXO45 to bind SKP1 (Fig. 4I). Previous work by Nakayama and colleagues (37) showed that the SPRY domain of FBXO45 mediates binding to PAM. Our results taken with this prior observation demonstrate that FBXO45 uses different biochemical mechanisms to bind SKP1 and PAM. Our results now provide an explanation for why FBXO45 is sufficient to recruit SKP1 into a complex with PAM (Fig. 4I). This provides new and more comprehensive insight into how human PAM forms a noncanonical SCF complex with SKP1 and FBXO45.

CUL1 is not part of the PAM/FBXO45/SKP1 complex
In traditional SCF complexes formed by the Rbx1 E3 ligase, the Cullin scaffolding protein binds directly to Rbx1 (8, 10). Skp1 then acts as an adaptor between the F-box protein and the Cullin. In contrast to traditional SCF complexes, we showed FBXO45 binds directly to PAM (Fig. 1), and SKP1 is recruited by FBXO45 into a complex with PAM (Fig. 4, C-F). These results are consistent with the lack of the Cullin CUL1 in this complex. To directly test this, we cotransfected GFP-PAM and MYC-CUL1 into 293 cells. We did not observe coIP of PAM with CUL1, but PAM did coIP with our positive control FBXO45, which was expressed at similar levels to CUL1 ( Fig.  4G; Fig. S2). We then determined whether CUL1 and FBXO45 interact by coexpressing MYC-CUL1 and FLAG-FBXO45. We observed no coIP of FLAG-FBXO45 with MYC-CUL1 (Fig. 4H). CoIP of FLAG-FBXO45 with MYC-SKP1 was used as a positive control for binding (Fig. 4H). These results confirm that CUL1 fails to bind PAM or FBXO45 and does not serve as a scaffold for the atypical SCF complex formed by PAM, FBXO45, and SKP1 (Fig. 4I).

PAM polyubiquitination of NMNAT2 results in proteasomemediated degradation
Genetic and cellular experiments in flies and rodents have shown PAM inhibits NMNAT2, but the biochemical mechanism of how this occurs remains untested. We hypothesized that PAM ubiquitinates NMNAT2, which results in degradation by the proteasome.
To test this hypothesis, we initially determined whether the PAM ubiquitin ligase complex binds NMNAT2. We cotransfected 293 cells with NMNAT2-MYC and GFP-PAM D5 in the presence or absence of FLAG-FBXO45. Consistent with a prior observation (28), we observed robust coIP between FLAG-FBXO45 and NMNAT2-MYC (Fig. 5A). When GFP-PAM D5 and NMNAT2-MYC were expressed together, we detected a low level of coIP, which is likely to be mediated by endogenous FBXO45 (Fig. 5A). Alternatively, PAM could interact at low levels with NMNAT2 in the absence of FBXO45. More importantly, binding of PAM D5 to NMNAT2 was much stronger when FLAG-FBXO45 was coexpressed (Fig. 5A). These results indicate FBXO45 recruits NMNAT2 into a protein complex with PAM (Fig. 5F).
We then evaluated whether PAM can stimulate ubiquitination of NMNAT2. 293 cells were cotransfected with NMNAT2-MYC and HA-tagged ubiquitin (HA-Ub) with or without full-length GFP-PAM. As shown in Fig. 5B, PAM coexpression triggered mild increases in polyubiquitination of NMNAT2-MYC (size Ͼ171 kDa). When the proteasome was inhibited with MG132, PAM caused further large increases in How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles polyubiquitination of NMNAT2 (Fig. 5B). We note in this experiment that it is likely PAM utilizes endogenous FBXO45 to recruit and ubiquitinate transfected NMNAT2-MYC. These results demonstrate that PAM can stimulate polyubiquitination of NMNAT2, and ubiquitination is enhanced when the 26S proteasome is inhibited.
Because PAM polyubiquitinates NMNAT2 in a proteasomedependent manner, we evaluated how proteasome activity and PAM affect the stability of NMNAT2. Initially, we evaluated whether NMNAT2 is degraded by the proteasome. 293 cells were transfected with NMNAT2-MYC and treated for 3 h with DMSO, the protein synthesis inhibitor cycloheximide (CHX), or a combination of CHX and MG132. As shown in Fig. 5C, NMNAT2-MYC levels are reduced following cycloheximide treatment due to NMNAT2 protein turnover in the absence of new protein synthesis. When MG132 is used to inhibit the proteasome, NMNAT2 turnover was partially blocked. Thus, NMNAT2 protein turnover occurs, in part, via the proteasome.
Because PAM is expressed in 293 cells (37), we examined how NMNAT2 protein stability is affected by endogenous PAM. To do so, we generated two independent 293 cell lines in which endogenous PAM was knocked out using CRISPR/Cas9 gene editing and nonhomologous end joining (58). To validate the efficiency of CRISPR with two different PAM single guide RNAs (sgRNAs), we cotransfected individual plasmids that express a single PAM sgRNA and Cas9, as well as a plasmid expressing GFP-PAM. Following puromycin selection for CRISPR editing, which induces deletions via nonhomologous end joining, we observed a decrease in GFP-PAM levels (Fig. S3). This did not occur when a nonspecific sgRNA was used (Fig. S3). This showed that PAM sgRNAs successfully recognize the PAM sequence, allow CRISPR cleavage of the plasmid encoding GFP-PAM, and result in reduced GFP-PAM expression.
We used these validated PAM sgRNAs to generate two independent PAM knockout cell lines. NMNAT2-MYC was transfected into PAM knockout cells, and we tested the level of NMNAT2 protein turnover when new NMNAT2 synthesis was blocked by cycloheximide. We observed that NMNAT2 turnover was reduced in both PAM knockout lines compare with WT HEK 293 in which endogenous PAM is expressed normally (Fig. 5D). This result demonstrates that endogenous PAM regulates turnover of NMNAT2.
To further test whether PAM ubiquitin ligase activity affects NMNAT2 turnover, we coexpressed NMNAT2-MYC with WT PAM or a PAM ligase-dead (LD) point mutant, which lacks catalytic RING activity. Interestingly, NMNAT2 turnover was not as rapid in the presence of PAM LD compared with WT PAM (Fig. 5E). This is consistent with PAM ubiquitin ligase activity affecting NMNAT2 protein stability.
Taken as a whole, our results demonstrate that the PAM ubiquitin ligase complex can bind and polyubiquitinate NMNAT2 (Fig. 5F). Polyubiquitination of NMNAT2 by PAM results in NMNAT2 degradation by the proteasome and NMNAT2 turnover. Finally, our results provide the first evidence that endogenous human PAM regulates NMNAT2 protein stability in human cells.

SKP1 increases FBXO45 binding to NMNAT2
Our observation that FBXO45 recruits SKP1 into a complex with PAM (Fig. 4) indicates SKP1 does not play its traditional role as an adaptor protein in the noncanonical PAM/FBXO45/ SKP1 complex. As a first step in deciphering the function of SKP1 in the PAM/FBXO45/SKP1 complex, we assessed whether SKP1 binds to NMNAT2, which we showed is a ubiquitination substrate for this ubiquitin ligase complex (Fig. 5).
We coexpressed GFP-SKP1 with NMNAT2-MYC in 293 cells and examined binding by coIP. As shown in Fig. 6A, SKP1 did not coprecipitate with NMNAT2. However, we did observe robust coIP between GFP-FBXO45 and NMNAT2-MYC (Fig.  6A). These results demonstrate that FBXO45, but not SKP1, is sufficient to bind NMNAT2.
There are a few possible explanations for our observation. 1) SKP1 does not play a role in target recognition. 2) SKP1 can modify the level of binding between FBXO45 and NMNAT2. 3) SKP1 is target-specific and might recognize other targets of PAM, such as DLK/MAP3K12 (59,60). To differentiate between these possibilities, we examined protein interactions when FLAG-FBXO45, HA-SKP1, and NMNAT2-MYC were coexpressed. Interestingly, when FBXO45 was present, we detected coprecipitation of SKP1 with NMNAT2 (Fig. 6B). This demonstrates that FBXO45 can form a complex with SKP1 and NMNAT2.
Next, we tested whether SKP1 influences the interaction between FBXO45 and NMNAT2. As shown in Fig. 6C, FLAG-FBXO45 binding to NMNAT2-MYC is greatly increased when

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
HA-SKP1 is present. Quantitation confirmed that FBXO45 binding to NMNAT2 is significantly increased severalfold when SKP1 is coexpressed compared with when it is absent (Fig. 6D).
These observations suggest that in the noncanonical PAM/ FBXO45/SKP1 complex SKP1 plays a novel function as an auxiliary component of the substrate recognition module and facilitates FBXO45 binding to NMNAT2 (Fig. 6E). This contrasts with traditional SCF complexes where SKP1 functions as an adaptor between the Cullin and F-box protein. However, we also note that the level of FBXO45 was increased slightly with coexpression of SKP1 (Fig. 6C, 2nd panel from bottom). Thus, we cannot rule out that NMNAT2 binding to FBXO45 is increasing when SKP1 is coexpressed because SKP1 stabilizes FBXO45 expression.

Discussion
In this study, we show that the human PHR protein PAM assembles into a noncanonical SCF complex that contains FBXO45 and SKP1 but lacks CUL1 (Fig. 7). The SCF complex formed by PAM, FBXO45, and SKP1 displays further atypical characteristics, as SKP1 does not act as an adaptor protein in complex assembly but rather is an auxiliary subunit that enhances substrate binding to the F-box protein FBXO45. Furthermore, we provide the first biochemical evidence that PAM polyubiquitinates NMNAT2 thereby triggering NMNAT2 degradation by the proteasome.

Biochemistry underpinning assembly of the noncanonical PAM/FBXO45/SKP1 ubiquitin ligase complex
Results from previous studies suggested that SCF ubiquitin ligase complexes formed by PHR proteins, including human PAM and fly Highwire, do not require Cullin as a scaffold protein (37,40).
Our results confirm that PAM forms an atypical SCF complex that contains the F-box protein FBXO45 and SKP1 but lacks CUL1 (Fig. 7A). We performed extensive biochemical analysis that now provides a biochemical mechanism for how this noncanonical SCF complex assembles. Our results show that the F-box protein FBXO45 binds directly to PAM, the RING E3 ubiquitin ligase (Fig. 1). This interaction is mediated by the FBD1 domain of PAM, which utilizes at least three conserved motifs, GRR, WCL, and FE, to bind FBXO45 (Fig. 3, B  and C). Previous work showed these same conserved motifs mediate binding of the C. elegans PHR protein RPM-1 to the F-box protein FSN-1 (38). Thus, the FBD1 domain represents a conserved biochemical mechanism by which PHR proteins interact with F-box proteins. Interestingly, one motif, NDD, appears to limit binding to FBXO45 and could be important for releasing ubiquitination substrates (Fig. 3, B and D).
Because human PAM relies upon a conserved mechanism to bind directly to FBXO45, we explored where the interaction between orthologous proteins in C. elegans occurs in neurons in vivo. Our results with two photon FLIM-FRET indicate that RPM-1 and FSN-1 physically interact at the axon termination site of C. elegans mechanosensory neurons (Fig. 2). This observation expands upon prior work showing RPM-1 and FSN-1 form a ubiquitin ligase complex that regulates axon termination (21,38,39). Our findings now provide the first in vivo biochemical evidence that RPM-1/FSN-1 ubiquitin ligase activity is likely to occur locally at axon termination sites.
When taken together, previous work (37) and results from our study (Fig. 4, G and H) show convincingly that the CUL1 scaffolding protein is not present in the atypical SCF complex formed by PAM, FBXO45, and SKP1. Several of our results now provide a molecular explanation for why the CUL1 scaffold is absent. First, PAM binds directly to FBXO45 (Fig. 1), which removes the necessity of CUL1 to facilitate binding to the F-box protein. Second, FBXO45 binds SKP1 via its F-box domain and recruits SKP1 into a complex with PAM (Fig. 4, D-F). Thus, another role CUL1 normally plays in canonical SCF complexes, binding to SKP1, is facilitated by the F-box protein FBXO45.

How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles
Finally, because FBD1 is located near the middle of the extremely large PAM protein, it is plausible PAM has sufficient distance between its catalytic RING domain and FBXO45 to ubiquitinate targets without a CUL1 scaffold (Fig. 7, compare A  and B). Consistent with this model, PAM (4664 amino acids) is extremely large compared with CUL1 (776 amino acids), which forms a canonical SCF complex with the E3 ligase Rbx1 (108 amino acids) (8).
Interestingly, we found that SKP1 plays a novel, very different role in the noncanonical SCF complex formed by PAM compared with traditional SCF complexes containing Rbx1. In traditional SCF complexes, SKP1 functions as an adaptor protein to facilitate binding between the F-box protein and the CUL1 scaffold. In contrast, in the PAM ubiquitin ligase complex, SKP1 interacts only with the F-box protein FBXO45 and facilitates increased binding of FBXO45 to the ubiquitination substrate NMNAT2. This suggests that in the atypical PAM/ FBXO45/SKP1 SCF complex, SKP1 is likely to function as an auxiliary protein in the target recognition module.

PAM/FBXO45/SKP1 complex polyubiquitinates NMNAT2, targeting it for degradation
Previous knockdown and loss of function genetic experiments in fly and rodent neurons have shown that PHR proteins, such as Drosophila Highwire and mouse Phr1, regulate axon degeneration by inhibiting Nmnat2 (27,28,61). These findings implied a model in which Nmnat2 ubiquitination by Phr1 results in Nmnat2 degradation.
To date, biochemistry supporting this model has remained notably absent (43). We address this gap in our knowledge via a range of biochemical findings with human PAM, FBXO45, and NMNAT2. 1) We show FBXO45 binds to NMNAT2, and this interaction is increased in the presence of SKP1 (Fig. 6). 2) FBXO45 recruits NMNAT2 into a complex with PAM (Fig.  5A). 3) NMNAT2 is turned over by proteasomal degradation (Fig. 5C). 4) PAM polyubiquitinates NMNAT2, and polyubiquitination is increased when the proteasome is inhibited (Fig.  5B). 5) CRISPR knockout of endogenous PAM decreases the turnover rate of NMNAT2 (Fig. 5D). 6) Finally, catalytically inactive PAM shows reduced ability to trigger turnover of NMNAT2 compared with WT PAM (Fig. 5E). Collectively, these results support the conclusion that PAM polyubiquitinates NMNAT2, which leads to proteasomal degradation and NMNAT2 turnover (Fig. 7A). Our findings now provide a biochemical explanation for how PHR proteins inhibit NMNAT2 during axon degeneration.
An intriguing recent study, published during revision of this manuscript, showed that PAM is an unusual RING E3 ligase that can ubiquitinate threonine and serine residues, as well as lysine (62). Although our results indicate that PAM polyubiquitination of NMNAT2 (which would occur on lysine residues) leads to proteasomal degradation, it is entirely plausible that PAM monoubiquitination of NMNAT2 on threonine or serine residues could act as a second molecular mechanism to restrain NMNAT2. If PAM indeed mediates both these types of ubiquitin modifications of NMNAT2, it is plausible PAM could regulate both local activity of NMNAT2 via monoubiquitination and NMNAT2 protein turnover via polyubiquitination. Impor-tantly, prior work in worms looking at axon and synapse development showed that the PHR protein RPM-1 utilizes two molecular mechanisms to inhibit the DLK-1 MAP3K: recruitment of the PPM-2 phosphatase to dephosphorylate DLK-1, and polyubiquitination and degradation of DLK-1 (59,63). It is an interesting possibility that PAM might also employ multiple molecular mechanisms, i.e. different types of ubiquitination, to regulate NMNAT2 degradation and activity during axon degeneration. If this occurs, PHR proteins could potentially exercise sophisticated temporal and spatial control over NMNAT2 in neurons following injury or during neurodegenerative disease.

Biochemistry
For all experiments, HEK293T cells were transfected using Lipofectamine 3000 (Invitrogen). Transfections were performed with plasmids of interest and pBluescript as needed to reach an equal amount of total DNA per transfection. Detailed descriptions of DNA amounts transfected for each experiment can be found in Table S1. 22-26 h after transfection, cells were lysed with 0.1% Nonidet P-40 buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 10% glyc-How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles erol, 1 mM DTT, and 1ϫ Pierce HALT protease inhibitor mixture or EDTA-free protease inhibitor tablet, Roche Applied Science). 0.25 to 1 mg of total protein lysate was used for coIPs. For coIP, lysates were incubated with one of the following primary antibodies for 30 min: 1.5 l (1.6 g) of M2 anti-FLAG (mouse monoclonal, Sigma); 1.5-3 l (0.6 g) of 3E6 anti-GFP (mouse monoclonal, MP Biomedicals); 1.5-3 l (0.75 g) of anti-HA (rabbit polyclonal, Invitrogen 715500); or 1.5 l (4 g) of 9E10 anti-MYC (mouse monoclonal, Sigma). Antibody complexes were immunoprecipitated by applying 10 l of protein G-agarose (Roche Applied Science) for 4 h at 4°C. Precipitates were boiled in Laemmli sample buffer containing ␤-mercaptoethanol (Bio-Rad) and run on 4 -12% BisTris gels (Invitrogen). An exception to this was full-length PAM, which was run on a 3-8% NuPAGE Tris acetate gel. 20 -25% of IPs were run for input blots and 40 -75% of IPs were run for coIP blots. Gels were transferred to PVDF membranes in Tris acetate transfer buffer (1 h at 100 V or 16 h at 30 V for large molecular weight proteins) and immunoblotted. The following primary antibodies were diluted 1:1000 in 5% nonfat milk in TBST and applied to blots overnight at 4°C: mixture of mouse monoclonal anti-GFP antibodies (Roche Applied Science); rabbit polyclonal anti-FLAG (Cell Signaling); mouse monoclonal 9B11 anti-MYC (Sigma); or mouse monoclonal 6E2 anti-HA antibody (Cell Signaling). Blots were visualized with HRP-conjugated anti-mouse (GE Healthcare or Veriblot, Abcam), anti-rabbit (ThermoFisher Scientific), or light chain reactive anti-mouse (Millipore) secondary antibodies. Secondaries were detected using enhanced chemiluminescent reagent (1:5 dilution of Supersignal West Femto (ThermoFisher Scientific) in TBS), and visualized with X-ray film.
Quantitation of immunoblots was done using Fiji software from NIH image (http://rsb.info.nih.gov/ij/). Quantitation was done from three or more replicates for a given condition run on the same gel. Band intensity (values in arbitrary units) of coprecipitating proteins was normalized to the intensity of bands corresponding to the target of the IP. Statistical significance was determined using a Student's t test.

Turnover of NMNAT2 by PAM
HEK 293T cells were transfected with 1 g of NMNAT2-MYC and 7 g of GFP-PAM WT or GFP-PAM LD. 20 h after transfection, cells were treated for the indicated length of time with DMSO or CHX (10 g/ml). Following drug treatment, cells were lysed with 1% Triton RIPA buffer. For immunoblots, 30 g of WCL was run on gels. ERK-1 was used as a nontransfected loading control and was used to re-probe blots quenched with sodium azide.

Establishment of PAM knockout cell lines
The two PAM sgRNA CRISPR plasmids were generated by GenScript and designed to target the PAM start site and generate deletions via nonhomologous end joining (58). 5 g of PAM sgRNA pLentiCRISPR version 2 were transfected into HEK293T cells. 24 h after transfection, 2.5 g/ml of puromycin was added to plates. Cells were cultured with puromycin for 6 days, and experiments were performed 3-4 days after drug selection.

Ubiquitination of NMNAT2
HEK 293T cells were transfected with 0.5 g of HAubiquitin (pBG-325), 0.5-1.5 g of NMNAT2-MYC or control MYC plasmid, and 1.5 g of GFP-PAM or GFP control plasmid. 22-24 h after transfection, cells were treated for 1 h with DMSO or MG132 (50 M) and lysed using 1% Triton RIPA buffer. To detect ubiquitination of NMNAT2, 0.5 mg of total protein lysate was used for anti-MYC IPs, which were run on a 3-8% Tris acetate gradient gel (Invitrogen) and then blotted for HA-ubiquitin.
Expression of GFP-PAM, NMNAT2-MYC, and HA-Ub was assessed by immunoblotting 12.5-25 g of WCL. ERK-1 was used as a nontransfected loading control, and it was used to re-probe blots quenched with sodium azide.

Pulldown assays
Constructs were expressed in BL21-CodonPlus (DE3)-RILP cells (Agilent technology). 100 ml of LB media ϩ antibiotic was inoculated from overnight cultures and grown to 0.6 -0.8 an A 600 . Recombinant protein expression was induced with isopropyl 1-thio-␤-D-galactopyranoside (0.05 mM) at 16°C for 20 -25 h. Bacterial lysates were prepared using two different buffers: 1) His binding buffer (300 mM NaCl, 20 mM sodium phosphate, pH 7.4, 40 mM imidazole, 10% glycerol, 1 mM DTT, and 1ϫ protease inhibitor mixture (Pierce Halt)), or 2) GST binding buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 10% glycerol, 1 mM DTT and 1ϫ protease inhibitor mixture). Lysates were generated by sonication. GSH beads and nickel beads were washed (six times) with PBS or His-binding buffer, respectively. GST-PAM (1.5-3 mg) or GST (0.15-0.3 mg) lysates were incubated with the GSH beads for 2 h at 4°C. Beads with bound protein were washed (three times) with binding buffer and incubated for 2 h at 4°C with lysate expressing His 6 -FBXO45 (1.5 mg) or His 6 . Subsequently, beads were washed (six times) with binding buffer, boiled in sample buffer, and run on stainfree gel (Mini-PROTEAN TGX Stain-Free TM precast gels). Proteins were transferred to PVDF membranes using Tris acetate transfer buffer (1 h at 100 V). GST-tagged proteins were immunoblotted with Z-5 anti-GST polyclonal mouse antibody (Santa Cruz Biotechnology). His-tagged proteins were immunoblotted with an H8 anti-His monoclonal mouse antibody (ThermoFisher Scientific). Appropriate secondary How the PAM/FBXO45/SKP1 ubiquitin ligase complex assembles antibodies were applied and detected using Supersignal West Pico ECL (ThermoFisher Scientific) and visualized with X-ray film.
For confocal microscopy, young adults were immobilized on 2% agarose pads in levamisole (5 M). Animals were imaged on a Leica SP8 confocal microscope under ϫ25 objective and ϫ4 scan zoom factor. Leica Application Suite (LAS) software defined Z-stacks (0.8 -1.0-m intervals), which are shown as maximum intensity projections.

Two-photon FLIM
The interaction between RPM-1::GFP and mCherry::FSN-1 at axon termination sites of PLM mechanosensory neurons was measured using 2pFLIM. Briefly, mEGFP and mCherry were excited with a Ti:sapphire laser (Chameleon, Coherent) at a wavelength of 920 nm and a power of 1.4 -1.7 milliwatts. Fluorescence lifetime images were obtained using a time-correlated single photon counting board (SPC-150; Becker and Hickl) controlled with custom software in Matlab (55). Fluorescence lifetime images were collected by 128 ϫ 128 pixels acquired at 2 ms/line and averaged over 24 frames. The fluorescence lifetime of mEGFP was measured by curve fitting as described previously (65).