A high-coverage shRNA screen identifies TMEM129 as an E3 ligase involved in ER-associated protein degradation

Misfolded ER proteins are retrotranslocated into the cytosol for degradation via the ubiquitin–proteasome system. The human cytomegalovirus protein US11 exploits this ER-associated protein degradation (ERAD) pathway to downregulate HLA class I molecules in virus-infected cells, thereby evading elimination by cytotoxic T-lymphocytes. US11-mediated degradation of HLA class I has been instrumental in the identification of key components of mammalian ERAD, including Derlin-1, p97, VIMP and SEL1L. Despite this, the process governing retrotranslocation of the substrate is still poorly understood. Here using a high-coverage genome-wide shRNA library, we identify the uncharacterized protein TMEM129 and the ubiquitin-conjugating E2 enzyme UBE2J2 to be essential for US11-mediated HLA class I downregulation. TMEM129 is an unconventional C4C4-type RING finger E3 ubiquitin ligase that resides within a complex containing various other ERAD components, including Derlin-1, Derlin-2, VIMP and p97, indicating that TMEM129 is an integral part of the ER-resident dislocation complex mediating US11-induced HLA class I degradation.

I n the ER, newly synthesized proteins undergo a quality check by chaperones, which attempt to induce proper folding 1 . Failure to do so may result in protein accumulation and aggregation in the ER, thereby compromising protein and cellular homeostasis 2,3 . To prevent this, terminally misfolded luminal and transmembrane ER proteins are targeted for ubiquitin-dependent degradation by the proteasome in the cytosol 4 , a process called ER-associated protein degradation (ERAD) 5,6 . This process depends on retrograde transport, or dislocation, of proteins into the cytosol via a reaction that is facilitated by a multiprotein complex that combines several functions essential for ERAD, such as recognition, guidance, ubiquitination, dislocation and deglycosylation of the substrate 7,8 . ERAD is not only used for removal of misfolded proteins but also for physiologically regulated proteolysis of ER-resident proteins 9,10 .
Human cytomegalovirus (HCMV) encodes several proteins that impair the HLA class I antigen presentation pathway 11 , thus avoiding detection of infected cells by cytotoxic T lymphocytes 12 . In particular, HCMV US11 exploits ERAD to induce rapid dislocation of newly synthesized HLA class I heavy chains (HCs) from the ER into the cytosol, where the HCs are subsequently degraded via the ubiquitin-proteasome system 13 . The ER luminal domain of US11 is required for interaction with HLA class I, while the transmembrane domain of US11 is essential for interaction with Derlin-1 (refs [14][15][16]. In this way, US11 recruits the HLA class I molecule to the dislocation complex, which besides Derlin-1 (refs 14,17) contains VIMP 17 , the AAA ATPase p97 (refs 17,18), Derlin-2 (ref. 19) and SEL1L 19 . HLA class I is then ubiquitinated, dislocated and subsequently directed towards the proteasome for degradation 13,[17][18][19][20][21] .
The US11-mediated degradation of HLA class I has been instrumental in the identification of key components of mammalian ERAD, such as Derlin-1 (refs 15,17), p97 (ref. 18), VIMP 17 and SEL1L 22 . However, a complete understanding of the dislocation complex and its modes of action is currently lacking. Most known pathways of ERAD rely on multispanning transmembrane E3 ubiquitin ligases containing a cytosolic RING domain 23 , such as the yeast Hrd1p 24 and Doa10 (ref. 25), and the mammalian HRD1 (mammalian homologue of yeast Hrd1p) 26 , AMFR/gp78 (ref. 27), TEB4 (mammalian homologue of yeast Doa10) 28 and TRC8 (ref. 29). The RING domain of the E3 ubiquitin ligase forms a docking site for an E2 ubiquitinconjugating enzyme, which in turn catalyses polyubiquitination of target substrates 30 . In the context of US11-induced HLA class I degradation, the E3 ubiquitin ligase responsible has remained elusive.
Functional genomics in mammalian cells has been greatly aided by the use of silencing techniques to study loss-of-function phenotypes. Long-term silencing and analysis of non-transfectable cell lines can be achieved by genomic integration of shorthairpin RNAs (shRNAs) through the use of viral vectors for delivery 31 . As with other RNAi-based approaches, the shRNA utility is limited by the low efficacy of many shRNAs, resulting in high false-negative rates, and by off-target effects, leading to high false-positive rates 32 . To overcome these issues, we recently reported on the use of pooled ultracomplex shRNA libraries where each gene is targeted by many different shRNA sequences 33,34 . When combined with deep-sequencing-based readouts 33,35 , such pooled shRNA library screens allow accurate massive multiplexing in a controlled, identical environment for all cells.
Here, we construct a novel high-complexity shRNA library targeting all known human protein-encoding genes and subsequently perform a genome-wide screen to identify novel proteins that are essential for US11-mediated HLA class I degradation. Besides known players, we identify the previously uncharacterized TMEM129 and the ubiquitin-conjugating E2 enzyme UBE2J2 as essential components for US11-mediated HLA class I downregulation. We demonstrate that TMEM129 is a C4C4-type RING E3 ubiquitin ligase, of which its RING domain is essential for US11-mediated HLA class I downregulation. Coimmunoprecipitation studies indicate that TMEM129 is present in US11-containing dislocation complexes. Besides the previously reported UBE2K, also UBE2J2 is essential for US11-mediated HLA class I downregulation and it is part of TMEM129containing dislocation complexes. Notably, TMEM129 is associated with various dislocation components independently of US11, suggesting that TMEM129 might be involved in ERAD of proteins outside the context of HCMV US11-mediated HLA class I degradation.

Results
An shRNA screen identifies TMEM129 as a player in HLA-I ERAD. We have previously established a screening platform in which pooled shRNA library screens were coupled to deepsequencing as a means to perform large RNAi screens in a single population of cells 33,34,36 . Here, we constructed a new pooled lentiviral shRNA library targeting each annotated human proteincoding gene with B30 independent shRNAs (see Methods section for details). We exploited this new resource to identify proteins that are essential for US11-mediated HLA class I downregulation in human cells. To this end, U937 monocytic cells were generated expressing a chimaeric HLA-A2 molecule that was N-terminally tagged with eGFP and a Myc-epitope (eGFP-Myc-HLA-A2). In this way, relative total HLA class I expression can be measured by assessing the eGFP signal and surface display can be visualized by flow cytometry using a Myc-specific antibody. As expected, introduction of the HCMV US11 protein into these cells induced potent downregulation of the chimaeric HLA-A2 molecule as well as that of endogenously expressed HLA-A3 ( Supplementary Fig. 1a).
A single population of U937 cells co-expressing eGFP-Myc-HLA-A2 and US11 was infected with the pooled shRNA library and the cells were divided into eGFP-bright and -dim populations by FACS sorting. Subsequently, we assessed which shRNAs were enriched in the eGFP-bright population as compared with the eGFP-dim population and performed statistical analysis on these ( Fig. 1a). Two independent screens were executed and the overlap of the top 100 enriched genes from both screens is presented as our primary hit list in Table 1. GO term enrichment analysis of the hits showed a strong enrichment for genes involved in Golgi vesicle transport and proteasome-dependent catabolic processes (Fig. 1b). Indeed, network analysis of the hits revealed two clear clusters comprising most members of the coat protein complex involved in retrograde transport of proteins from the Golgi complex back to the rough ER 37 (Fig. 1c) and numerous proteasome subunits (Fig. 1d). Since it is known that US11-mediated ERAD of HLA class I relies on ubiquitindependent proteasomal degradation of the HLA class I substrate 13 , we expected to identify proteasome subunits in our screen. Additionally, the valosin-containing protein (VCP, commonly referred to as p97) was a clear hit. p97 is an AAA ATPase responsible for extracting and shuttling dislocated substrates from the ER environment towards the proteasome for degradation 18 . Besides these, there were numerous genes involved in ubiquitin biology that were clearly enriched in both screens but which did not meet our strict cutoff requirements; these genes included ubiquitin B (UBB), ubiquitin C (UBC), Ubiquitin-Activating Enzyme E1 (UBA1), the E2-conjugating enzyme UBE2J2, the p97-interacting proteins UFD1L and NPLOC4, as well as multiple additional proteasome subunits.
TMEM129 is essential for HLA class I dislocation by US11. We focused our validation analysis on the uncharacterized protein TMEM129. Depletion of TMEM129 by using independent shRNAs induced potent rescue of both the chimaeric HLA-A2 and the endogenous HLA-A3 expression in U937 cells coexpressing US11 ( Fig. 2a; Supplementary Fig. 1b). Consistent with the flow cytometry data, immunoblot analysis for HLA class I expression revealed increased protein levels in the presence of TMEM129 shRNAs ( Fig. 2b; Supplementary Fig. 1c). Intriguingly, US11 protein levels were upregulated upon TMEM129 depletion as well. The impaired HLA class I downregulation in TMEM129-depleted cells was restored upon overexpression of TMEM129 (Fig. 2c, upper panels), which excludes the possibility that HLA class I rescue was caused by off-targeting effects of the used shRNAs. Notably, overexpression of TMEM129 even enhanced HLA class I downregulation in both TMEM129-and mock-depleted cells (Fig. 2c, lower panels). The efficiency of shRNA-mediated TMEM129 depletion is depicted in Fig. 2d. Similar results were obtained for additional TMEM129-targeting shRNAs ( Supplementary Fig. 1b-d).
Besides validating TMEM129 by using RNAi, we employed the recently established CRISPR/Cas genome-engineering system to generate TMEM129-null cells. Expression of a TMEM129targeting guideRNA (gRNA) induced potent rescue of both chimaeric and endogenous HLA class I molecules to levels comparable to those in US11-negative cells ( Fig. 2e and Supplementary Fig. 2a,b). Loss of the TMEM129 protein was confirmed by TMEM129 immunoprecipitation experiments ( Supplementary Fig. 9d). Subsequent reconstitution of TMEM129 restored US11-mediated HLA class I downregulation in TMEM129-null cells, and even reduced HLA class I protein levels to almost undetectable levels ( Fig. 2e and Supplementary  Fig. 2b, lower panels).
US11 induces rapid dislocation of HLA class I molecules from the ER to the cytosol, where HLA class I is subsequently degraded by the proteasome (Supplementary Fig. 3) 13 . Upon dislocation into the cytosol, HLA class I HCs are deglycosylated, and these species can be visualized upon proteasome inhibition 13 . We performed pulse chase experiments to determine whether TMEM129 depletion impairs HLA class I HC dislocation and/or its subsequent degradation by the proteasome. As expected, TMEM129 depletion greatly stabilized HLA class I in the presence of US11 (Fig. 2f). Upon proteasome inhibition, we observed accumulation of deglycosylated HCs in control cells, whereas these species were completely absent in cells depleted for TMEM129 (Fig. 2g). Results were comparable for eGFP-Myc-HLA-A2 and endogenous HLA class I HCs (compare upper and lower panels in Fig. 2f,g). These data show that TMEM129 is critically involved in US11-mediated HLA class I dislocation and degradation.
TMEM129 is a novel E3 ubiquitin ligase. TMEM129 is an uncharacterized transmembrane protein of 362 amino acids that is predicted to have 3 transmembrane domains (TOPCONS; Fig. 3a), and is localized to the ER (Fig. 3b) Expression of the full-length TMEM129 enhanced HLA class I downregulation (Fig. 3e, upper panels), whereas expression of RING-less TMEM129 in U937 eGFP-Myc-HLA-A2 US11 cells resulted in a dominant-negative phenotype, that is, rescue of surface HLA-A3, and total and surface eGFP-Myc-HLA-A2 (Fig. 3e, lower panels). These results were confirmed by immunoblotting experiments (Fig. 3f), in which increased levels of HLA class I HCs were observed in the presence of RING-less TMEM129. Again, increased US11 protein levels were observed, this time upon RING deletion. The RING-less TMEM129 data recapitulate our findings obtained by TMEM129 depletion via shRNAs and CRISPR gRNAs. In conclusion, these findings identify TMEM129 as an E3 ubiquitin ligase with a C4C4 RING domain, removal of which results in a dominant-negative phenotype rescuing HLA class I from US11-mediated dislocation.
UBE2J2 is essential for US11-mediated HLA class I ERAD.
Ubiquitination of a target substrate is often facilitated by more than one E2 ubiquitin-conjugating enzyme 38 . Currently, the identity of one E2 ubiquitin-conjugating enzyme contributing to US11-mediated HLA class I downregulation is known, namely UBE2K/E2-25K (ref. 39). The combined results of our two genome-wide shRNA screens did not identify any additional candidates (Table 1). However, we did observe a clear enrichment for the E2 UBE2J2. We therefore tested UBE2J2-targeting  shRNAs, which indeed rescued endogenous and chimaeric HLA class I molecules in US11-expressing cells (Fig. 5a, upper two panels). This rescue was completely abolished upon co-expression of an UBE2J2 cDNA, showing that the observed phenotype was caused by on-target effects of the UBE2J2-targeting shRNAs. These results were confirmed by the generation of UBE2J2-null cells using the CRISPR/Cas system (Fig. 5b, upper panel). ShRNAs (Fig. 5a, lower two panels) and CRISPR gRNAs (Fig. 5b, third row) targeting the homologous UBE2J1 did not have any effect on HLA class I expression. Expression of a dominant-negative UBE2J2 (C94S) increased HLA class I expression in US11expressing cells (Fig. 5c). Co-immunoprecipitation experiments showed that UBE2J2 is in the same complex as the E3 ubiquitin ligase TMEM129 (Fig. 5d). RING removal of TMEM129 does not affect association of UBE2J2 with the complex. Intriguingly, UBE2K-null cells showed a clear rescue of HLA class I molecules (Fig. 5b, second row), suggesting that both UBE2J2 and UBE2K are essential for HLA class I downregulation by US11. We next assessed whether TMEM129 is part of ER-resident dislocation complexes independent of US11, which might indicate that TMEM129 plays a role in general ERAD. Co-immunoprecipitation experiments were performed using C-terminal Strep-tag II-and FLAG-tagged TMEM129 and TMEM129-DRING. Both proteins were found to be associated with Derlin-1, Derlin-2, VIMP, p97, SEL1L and HRD1 to similar levels in both US11-expressing and US11-negative cells (Fig. 6a). In the absence of US11, no or very low levels of HLA class I and eGFP-Myc-HLA-A2 associated with TMEM129, despite the presence of considerable amounts of HLA class I molecules in these cells (compare upper two panels, lanes 2 and 8, and lanes 3 and 9). In the presence of US11, again increased levels of HLA class I were found in association with TMEM129DRING as compared with wild-type TMEM129 (compare lanes 5 and 6, upper two panels). As HRD1 and SEL1L co-precipitated together with US11 ( Fig. 4) and TMEM129 (Fig. 6a), we compared overexpressed TMEM129 and HRD1 in their ability to associate with US11, Derlin-1 and SEL1L (Fig. 6b). Immunoprecipitation of TMEM129 recovered higher quantities of US11 and Derlin-1 compared with immunoprecipitation of HRD1-FLAG (compare Fig. 6b, lanes 5 and 6). In addition, immunoprecipitation of HRD1 recovered SEL1L (lane 6), which was not observed upon localizes to the ER. MelJuSo cells stably co-expressing TMEM129-eGFP with either mCherry (cytosolic and nuclear) or mCherry-KDEL (ER-localized) were subjected to fluorescent confocal microscopy to assess colocalization of TMEM129-eGFP with either marker. White bars represent 10 mm. (c) TMEM129 is an E3 ubiquitin ligase. In vitro ubiquitination assays were performed by using purified ubiquitin, E1 enzyme UBA1, E2 enzyme UBE2D3 and E3 enzyme MuRF1 (ctrl E3), immunoprecipitated TMEM129-FLAG-ST2 (T129) or TMEM129DRING-FLAG-ST2 (DRING), in the absence or presence of ATP. Purified S5a was used as a substrate. Immunoblot analysis was performed using an anti-S5a antibody to visualize poly-ubiquitinated S5a. Addition of the tags did not interfere with TMEM129 function (see Supplementary Fig. 10c). (d) The experiment was performed as in Fig. 3c, although a specific substrate was omitted from the reaction. Immunoblot analysis was performed using the anti-ubiquitin P4D1 mAb to visualize polyubiquitin (Ub n ). (e) The TMEM129 RING domain is essential for US11-mediated HLA class I downregulation. Flow cytometry analysis of endogenous surface HLA-A3, and surface (Myc) and total (eGFP) eGFP-Myc-HLA-A2 in U937 eGFP-Myc-HLA-A2 cells expressing US11 and co-expressing either TMEM129 (T129), TMEM129DRING (DRING) or an empty vector (ctrl). (f) Same cells as in Fig. 3e, now analysed by immunoblotting for the indicated proteins. TMEM129-FLAG retained its ability to enhance HLA class I downregulation (Supplementary Fig. 10a).
TMEM129 pull-down (lane 5). The same results were observed when US11 was absent (lanes 1-3). Thus, TMEM129, and not HRD1, preferentially associates with US11 and Derlin-1. Moreover, SEL1L preferentially interacts with HRD1 and not with TMEM129. These results demonstrate that TMEM129 is an integral constituent of the ER-resident dislocation complexes, suggesting that this E3 ubiquitin ligase might play a role in physiological ERAD, independently of US11. A model of the US11 dislocation complex is depicted in Fig. 6c.

Discussion
HCMV US11-mediated degradation of HLA class I molecules serves as a paradigm for ERAD and has facilitated the identification of multiple essential components of this important degradation pathway. In this study, we identified additional essential players of US11-mediated HLA class I downregulation by performing a pooled genome-wide shRNA library screen in human cells expressing the US11 protein. To this end, we constructed a new high-complexity genome-wide shRNA library consisting of B550,000 different shRNAs. We assessed whether we could successfully conduct a genome-wide screen using the entire library in a single population of cells. This approach proved successful, as we identified known players, such as p97 and a large compendium of proteasome subunits. Besides this, we identified multiple new genes potentially involved, such as subunits of the COPI retrograde trafficking machinery and the uncharacterized TMEM129. We chose a strict cutoff for our validation experiments and only allowed the overlap of the top 100 enriched genes from two independent screens in our hit list. This screen is the first to show that ultracomplex libraries with over 550,000 different shRNAs can be successfully applied in genetic screens in a single population of cells using deep-sequencing as a read-out for library complexity and dynamics.
Our hit list contained multiple genes of the COPI retrograde transport system. In agreement with this, we observed a moderate rescue of eGFP-Myc-HLA-A2 expression in COPI-depleted cells co-expressing the chimaeric molecule and US11 ( Supplementary  Fig. 4a). Since immature HLA class I molecules that escape to the Golgi complex are efficiently recruited back to the ER in a COPI-dependent manner 40 , depletion of COPI subunits would theoretically cause accumulation of these HCs in the Golgi, which could protect these from US11-induced ERAD initiated in the ER. Indeed, we observed accumulation of both endogenous and chimaeric HLA class I molecules in a post-ER compartment upon COPI depletion, as EndoH-resistant HCs became apparent in immunoblot analysis (Supplementary Fig. 4b). Additionally, confocal studies clearly demonstrated a relocalization of chimaeric HLA class I molecules to the Golgi upon COPI depletion ( Supplementary Fig. 4c). Rescue of HLA class I in COPI-depleted cells could be related to the inability of ERlocalized US11 to initiate degradation of Golgi-localized HCs, or by a lack of ERAD components in the ER due to an impaired retrieval of membrane proteins from the Golgi to  Fig. 10c, left panels). Immunoprecipitated proteins were eluted using d-Desthiobiotin, after which immunoblot analysis was performed for the proteins indicated.
the ER upon blockage of the retrograde transport system. We therefore believe that the COPI complex is not specifically involved in US11-mediated HLA class I ERAD, but that the identification of these subunits in the screen is likely caused by an indirect effect.
Our hit list contained the previously uncharacterized protein TMEM129. Subsequent validation experiments, including shRNA-mediated knockdown, CRISPR/Cas-mediated knockout and overexpression of TMEM129, showed that TMEM129 is crucial for US11-mediated HLA class I downregulation. Sequence analysis revealed that TMEM129 contains a putative non-classical RING domain. Whereas a conventional RING domain contains three zinc-coordinating cysteine pairs and one cysteine-histidine pair, the RING domain of TMEM129 possesses four zinccoordinating cysteine pairs (C4C4) that are fully conserved among a wide range of TMEM129 orthologs ( Supplementary  Fig. 5). Thus far, only a single mammalian E3 ubiquitin ligase (CNOT4) has been described to contain a C4C4 RING domain 41 . We show that TMEM129 is a bona fide E3 ubiquitin ligase as it possesses such activity in vitro, and removal of the RING domain completely abrogated this activity. In line with this, expression of RING-less TMEM129 in US11-expressing cells caused a dominant-negative phenotype of increased HLA class I levels.
Not only HLA class I levels were found to be elevated but also US11 protein levels were increased upon expression of RING-less TMEM129, an observation also made upon shRNA-mediated depletion of TMEM129. However, when TMEM129 was overexpressed in US11-expressing cells, US11 levels remained unchanged, whereas HLA class I levels decreased to almost undetectable levels. Thus, while we cannot rule out that US11 is a substrate for TMEM129 to some extent, we show that TMEM129 primarily mediates dislocation of HLA class I molecules.
Immunoprecipitation of US11 and TMEM129 demonstrated that TMEM129 is an integral part of the US11 dislocation complex. Depletion of TMEM129 and expression of RING-less TMEM129 appear to lock the dislocation complex in a dislocation-incompetent state, indicated by high levels of HLA class I in complex with US11, while normally, association of HLA class I with US11 is transient 15 . The presence of TMEM129 in ERAD complexes independently of US11 suggests that the protein may also be involved in dislocation of other cellular substrates. Among these substrates may be cellular proteins destined for degradation in the context of ER quality control and/ or alleviation of ER stress through the unfolded protein response (UPR). CRISPR/Cas-mediated TMEM129 knockout results in a slight induction of the UPR as assessed by XBP-1 splicing analysis ( Supplementary Fig. 6), suggesting that TMEM129 might play a role in this process.
The identification of the E3 ubiquitin ligase crucial for US11mediated HLA class I downregulation has been a subject of intense investigation. Currently, only a few E3 ubiquitin ligases involved in human ERAD are known, of which HRD1 is the most studied. However, previous attempts to link HRD1 to US11mediated HLA class I degradation have not been successful, as HRD1 depletion studies did not rescue HLA class I from US11mediated degradation 42 . In line with this, we did not observe increased US11-mediated HLA class I downregulation upon HRD1 overexpression (Supplementary Fig. 7a), nor did we see rescue of HLA class I from US11-mediated degradation upon either shRNA-mediated HRD1 depletion or gRNA-mediated HRD1 gene disruption ( Supplementary Fig. 7c,e respectively). Nevertheless, HRD1 was present in US11 and TMEM129containing dislocation complexes (Figs 4 and 6a). A subsequent immunoprecipitation experiment to compare the binding of HRD1 and TMEM129 to other dislocation components revealed that US11 and Derlin-1 associated more strongly with TMEM129, compared with HRD1 (Fig. 6b). In this context, SEL1L was preferentially associated with HRD1, and not with TMEM129, even though both E3 ligases were immunoprecipitated to similar quantities. Hence, SEL1L may thus be recruited to the dislocation complex through its association with HRD1, which is also present in the complex, albeit in small amounts. We therefore propose that TMEM129 is the E3 ubiquitin ligase that is essential for US11-mediated HLA class I degradation.
The fact that TMEM129 has escaped identification thus far 43 is likely due to its low endogenous expression levels, which complicates its identification via standard biochemical approaches. Additionally, TMEM129 was a previously uncharacterized protein, which prevented including TMEM129 as a candidate in focused E3 ubiquitin ligase screening approaches by, for example RNAi. Our strategy involved the use of an unbiased genome-wide shRNA library screen in which all protein-encoding genes were included, thereby eliminating these hurdles and illustrating the advantages of genome-wide pooled shRNA library screens.
E3 ubiquitin ligases collaborate with E2 enzymes to facilitate polyubiquitination of a target substrate 30 . Previously, the E2 UBE2J1 was implicated to be involved in US11-mediated HLA class I dislocation 43 . Unexpectedly, not only the dominantnegative UBE2J1 but also the wild-type E2 protein was found to impair US11-mediated HLA class I dislocation 43 . We now show that UBE2J2, and not UBE2J1, is essential for HLA class I degradation by US11, since depletion or knockout of UBE2J2, but not UBE2J1, causes a dramatic rescue of HLA class I from US11mediated degradation. Considering that mammalian UBE2J1 and UBE2J2 are highly homologous 44 , UBE2J1 might compete with UBE2J2 for binding to certain E3 ubiquitin ligases; when present in excess, UBE2J1 might prevent UBE2J2 from binding, which could explain the earlier observations. Additionally, co-immunoprecipitation experiments confirmed that UBE2J2 is present in complexes with TMEM129. Also, the E2 UBE2K/E2-25K has been reported to facilitate ubiquitination of HLA class I in permeabilized US11-expressing cells 39 . In this experimental setup however, membrane-anchored E2 enzymes (for example, UBE2J2) remained uninvestigated since the usurped semipermeabilized system only allowed removal and addition of cytosolic E2 enzymes. Using CRISPR/Cas-mediated gene disruption, we show that UBE2K is also essential for US11mediated HLA class I downregulation, thereby confirming the previous findings for UBE2K. UBE2J2-and UBE2K-null cells expressing US11 both show high levels of rescued HLA class I, which indicates that these two E2 enzymes are not interchangeable and catalyse two different ubiquitination steps essential for the dislocation process.
The processivity of polyubiquitin formation is often orchestrated by at least two different E2 enzymes, one of which is responsible for ubiquitin chain initiation, while the other E2 promotes ubiquitin chain elongation 38 . Mono-ubiquitination of HLA class I is not sufficient to induce US11-mediated dislocation; completion of this reaction requires Lys48-linked polyubiquitination 21,45 . While UBE2K is known to catalyse Lys48-linked ubiquitin chain elongation, and not chain initiation 46 , it is unknown what type of ubiquitin chain formation is catalysed by UBE2J2. However, in the context of US11-mediated HLA class I degradation, the membrane-bound UBE2J2 might promote mono-ubiquitination of HLA class I, thereby enabling the cytosolic UBE2K to catalyse Lys48-linked ubiquitin chain elongation. Removal of either UBE2J2 or UBE2K would prevent polyubiquitination of HLA class I and thereby inhibit its dislocation. This hypothesis would also be in line with the results previously obtained for UBE2K using the semipermeabilized cells 39 ; in this system, the membrane-bound UBE2J2, which was not removed, would enable the supplied UBE2K to catalyse ubiquitin chain elongation.
Besides US11, the HCMV-encoded protein US2 also mediates dislocation of HLA class I 48  HCs is accelerated via depletion of b 2 m, yet another E3 ubiquitin ligase, HRD1, catalyses the degradation of HLA class I HCs 42,47 . It is notable that the dislocation of a single degradation substrate may be catalysed by at least three different E3 ubiquitin ligases, depending on its context. This observation testifies to the versatility of ER-associated protein degradation and underscores the essential role that E3 ubiquitin ligases play in regulating the dislocation of substrates into the cytosol for degradation by the ubiquitin-proteasome system.

Methods
Cell culture and lentiviral infection. U937 human monocytic cells, 293T human embryonic kidney cells and MelJuSo (MJS) human melanoma cells were obtained from ATCC (American Type Culture Collection) and grown in RPMI medium supplemented with glutamine, penicillin/streptomycin and 10% FCS. For individual gene infections using lentiviruses, virus was produced in 24-well plates using standard lentiviral production protocols, third-generation packaging vectors, and 50 ul viral supernatant adjusted to 8 mg ml À 1 polybrene was used to infect B20,000 cells by spin infection at 1,000g for 2 h at 33°C. For confocal colocalization studies, we used the same dual promoter lentiviral vectors described above in which we either cloned TMEM129 C-terminally fused to a FLAG-epitope and eGFP (TMEM129-eGFP), mCherry alone (mCherry), mCherry N-terminally tagged to amino acids 1-60 of the human galactosyltransferase (GalT-mCherry), or mCherry fused to a C-terminal BiP-leader and N-terminal KDEL-sequence (mCherry-KDEL). RING-less TMEM129 was generated by removing amino acids 285-362. Dominant-negative UBE2J2 was generated by mutating the cysteine of the active site at position 94 into a serine, as described previously 44 . When indicated, we used the following (epitope) tags: HA-tag (YPYDVPDYA), Myc-tag (EQKLISEEDL), FLAG-tag (DYKDDDDK) and triple Strep-Tag II (ST2; WSHPQFEKGSWSHPQ FEKGSWSHPQFEKGS). For CRISPR/Cas experiments, we obtained vectors from the Church lab 49 via Addgene (Addgene plasmid 41824: gRNA_Cloning Vector, and Addgene plasmid 41815: Cas9 expression vector). Additionally, we constructed a selectable lentiviral CRISPR/Cas vector (which will be described elsewhere, manuscript in preparation) to facilitate efficient and selectable expression of Cas9 and gRNAs in target cells: briefly, the lentiviral pSicoR vector (Jacks Lab, MIT) was altered to express a nuclear-localized Cas9 gene that was N-terminally fused to puroR and a T2A sequence. Additionally, the region immediately downstream of the U6 promoter was replaced by a cassette consisting of two unique restriction sites (BsmBI) to allow cloning of gRNA target sites followed by the gRNA scaffold and a terminator consisting of 5 T-residues.
Pooled shRNA libraries. We designed shRNAs targeting all protein-encoding transcripts in human cells using the shRNA retriever programme 50 , which generates shRNAs with 22-nucleotide guide strands. We designed oligos consisting of the shRNA hairpin (fully complementary 22-bp arms, and a 9-bp loop) surrounded by 3 0 and 5 0 primer-binding sites. The shRNAs include an initial G before the sense arm, for correct expression by the U6 promoter, a complementary C after the antisense arm, and a TTTTT (UUUUU in the RNA sequence) terminator to end transcription immediately after the hairpin. Oligonucleotides encoding shRNAs in a sub-library of 55,000 sequences were synthesized by Agilent, and used to generate shRNA libraries as previously described 33 . Twelve sublibraries were generated where genes were organized into functionally related groups, using GO annotation, curated localization and data from various proteomic surveys of organelles. ShRNAs were cloned into our vector MP-177, for which a detailed map will be provided on request. This vector was derived from the lentiviral pSicoR vector (Jacks Lab, MIT) in which the U6 promoter was altered to allow for sticky cloning of shRNAs in between a BstXI and XhoI site, and in which the CMV promoter was used to drive expression a cassette encoding a puromycin resistance marker, the ribosomal skipping peptide T2A and mCherry. Library quality was evaluated by sequencing 96 colonies from each sub-library, the percentage of perfect sequences ranged from 58 to 76% with an average of 66%. The coverage of each library ranged from 31-to 128-fold with an average of 50-fold, as assessed by bacterial colony counts.
Genome-wide shRNA screen. U937 monocytic cells co-expressing eGFP-Myc-HLA-A2 and US11 were grown in RPMI medium supplemented with glutamine, penicillin-streptomycin and 10% FCS. Virus was produced in 15-cm plates of 293T cells using standard lentivirus production protocols. For each screen, B160 Â 10 6 cells were infected with the entire genome-wide shRNA library in 320 ml virus supernatant supplemented with 8 mg ml À 1 polybrene, divided into wells of a sixwell plate and spin-infected at 1,000g for 2 h at 33°C to reach a target infection of B80%. Six days after infection, B800 Â 10 6 cells were harvested and subjected to cell sorting via a two-step sort-protocol using a Becton Dickinson Influx cell sorter. First, eGFP-upregulated cells were sorted using an 'enrichment-protocol', which allowed for high-speed cell sorting of the entire population of cells in a short timeframe. Next, the selected cells were sorted for purity selecting mCherrypositive cells (cells transduced with an shRNA lentivector) and eGFP-bright cells. The top 0.75% of eGFP bright cells was selected, which resulted in a total of B2.8 Â 10 6 cells per screen. As control, B115 Â 10 6 mCherry-positive, eGFP dim cells were sorted. Subsequently, genomic DNA was isolated from the selected cells using standard phenol-chloroform extractions. Next, the lentiviral shRNA inserts were PCR-amplified for 23 cycles using primers AATGATACGGCGACCACC GACACTCTTTCCCacaaaaggaaactcaccctaac and CAAGCAGAAGACGGCATAC GAgcggtaatacggttatccacg and Phusion polymerase (NEB) in the presence of buffer GC supplemented with DMSO. These primers contain Illumina adapter sequences (displayed in capital letters) that allow direct loading on an Illumina HiSeq2000 sequencer. As input we performed a PCR on genomic DNA equivalent to B2.8 Â 10 6 cells for the selected eGFP bright cells, and 20 Â 10 6 cells for the eGFP dim control cells. This resulted in B35 and 250 50-ul PCR reactions for the selected and control cells, respectively. The PCR products were pooled and purified/concentrated using a PCR purification kit (Qiagen), and subsequently loaded on a 20% polyacrylamide gel in 0.5 Â TBE. Bands of the correct size were excised, electro-eluted, purified by phenol-chloroform extraction and subsequently quantified using a Nanodrop quantification device (Nanodrop, Rockland, DE, USA) and an Agilent bioanalyzer (Agilent Technologies, Palo Alto, CA, USA). Deep sequencing was carried out on an Illumina HiSeq2000 (performed by BGI, Hong Kong) using the sequencing primer 5 0 -GAGACTATAAGTATCCCTTGGA-GAACCACCTTGTTGG-3 0 , in which the guide strands of the shRNA-encoding constructs were detected. Sequences were aligned to the known library sequences using Bowtie 51 and the counts per shRNA were calculated. On average we obtained B78 Â 10 6 aligned sequences per sample, which ranged from 56 Â 10 6 cells to 106 Â 10 6 reads per sample. To identify hit genes in the genome-wide screen, we Figure 6 | TMEM129 is present in ERAD complexes in the absence of US11. (a) C-terminally Strep-tag-II-and FLAG-tagged TMEM129 or TMEM129DRING were immunoprecipitated using StrepTactin beads from 1.0% digitonin lysates of US11-negative (lanes 1-3) and US11-positive (lanes 4-6) U937 eGFP-Myc-HLA-A2 cells. The cells were co-expressing a control vector (ctrl, lanes 1 and 4), TMEM129-FLAG-ST2 (T129, lanes 2 and 5), or TMEM129DRING-FLAG-ST2 (DRING, lanes 3 and 6). Immunoprecipitated proteins were eluted using d-Desthiobiotin, after which immunoblot analysis was performed for the proteins indicated. The right panels (lanes 7-12) indicate loading controls to analyse input of the indicated proteins before immunoprecipitation. The C-terminally Strep-tag-II-and FLAG-tagged TMEM129 construct retained its ability to enhance US11-mediated HLA class I downregulation ( Supplementary Fig. 10c, left panels), whereas the C-terminally tagged TMEM129DRING retained its dominant-negative phenotype ( Supplementary Fig. 10c, right panels). The HRD1 antibody could not detect HRD1 in digitonin cell lysates, indicated by the asterisk. (b) C-terminally FLAGtagged TMEM129 and HRD1 were immunoprecipitated using FLAG-M2-coupled beads from 1.0% digitonin lysates of US11-negative (lanes 1-3) and US11positive (lanes 4-6) U937 eGFP-Myc-HLA-A2 cells. Immunoprecipitated proteins were eluted using FLAG peptides, after which immunoblot analysis was performed for the proteins indicated. (c) Schematic overview of the US11 dislocation complex. The depicted location of the proteins is not accurate. calculated a P-value for each gene in our library using a Mann-Whitney U-test as described previously 34 . The genome-wide screen was carried out in two independent replicates and the overlap between the top 100 genes with the lowest P-values from both screens was selected as hit list. These hits were assessed for enrichment of GO terms using DAVID 52, 53 and network analysis was performed using Ingenuity pathway analysis (Ingenuity Systems, Redwood City, CA, USA).
Generation of CRISPR/Cas-mediated knockout cells. U937 cells co-expressing eGFP-Myc-HLA-A2 and US11 were transfected twice on subsequent days using the Neon Transfection System (Life technologies, Breda, The Netherlands) with a Cas9-encoding vector and a gRNA expression vector targeting human TMEM129. Typically, we obtained B10% knockout cells as determined by flow-cytometry assessment of upregulated total eGFP-Myc-HLA-A2 (eGFP) levels. Cells were sorted by flow cytometry (FACSAriaII, BD Biosciences) and single-cell cloned by limiting dilution.
Later on, we moved to a lentiviral CRISPR/Cas system (see 'Plasmids' section) in which a single lentiviral vector co-expresses a Cas9, puroR and a gRNA sequence. Virus was produced and cells were infected as described in the 'Cell Culture and Lentiviral Infection' section. Two days post-infection (dpi), infected cells were selected by using 2 mg ml À 1 puromycin and allowed to recover. Typically, we observed 30-90% of selected cells to display a full knockout phenotype as assessed by flow cytometry. See also Supplementary Table 2 for gRNA sequences used in this study and Supplementary Fig. 8 for a visualization of the genomic target sites of these gRNAs.
Fluorescence confocal microscopy. MelJuSo cells were grown overnight on 12 mm circular glass coverslips. Attached cells were washed with PBS supplemented with 0.5 mM MgCl 2 and 1 mM CaCl 2 (PBS þ þ ) and fixed in 3% paraformaldehyde for 15 min at room temperature. Coverslips were washed again in PBS þ þ and embedded on microscope slides with Mowiol 4-88 (Carl Roth, Germany). Slides were air-dried overnight at room temperature and imaged using a Leica TCS SP5 confocal microscope equipped with a HCX PL APO CS Â 63/1.40-0.60 OIL objective (Leica Microsystems, the Netherlands). eGFP and mCherry fluorescent signals were detected with PMTs set at the appropriate bandwidth after excitation using the 488 nm argon laser for eGFP and the 543 nm helium neon laser for mCherry. Images were processed and analysed using Leica SP5 software.
UPR assessment. Cells were either incubated with Thapsigargin (50 mM) or DMSO. After 6 h of incubation, cells were harvested and mRNA was isolated using standard Trizol RNA isolation procedures. Possible genomic DNA contamination was removed using a TURBO DNase kit (Life Technologies) according to the manufacturer's protocol. Using oligo dT primers, mRNA was specifically reverse transcribed into cDNA (SuperScriptIII, Invitrogen) according to the manufacturer's protocol. GAPDH and XBP-1 were then amplified using specific primers (Supplementary Table 3) and DreamTaq Green (Fermentas). PCR products were separated in SYBRsafe (Life Technologies)-containing 2.5% agarose gels.