UPF3A and UPF3B are redundant and modular activators of nonsense-mediated mRNA decay in human cells

The paralogous human proteins UPF3A and UPF3B are involved in recognizing mRNAs targeted by nonsense-mediated mRNA decay (NMD). While UPF3B has been demonstrated to support NMD, contradicting reports describe UPF3A either as an NMD activator or inhibitor. Here, we present a comprehensive functional analysis of UPF3A and UPF3B in human cells using combinatory experimental approaches. Overexpression or knockout of UPF3A as well as knockout of UPF3B did not detectably change global NMD activity. In contrast, the co-depletion of UPF3A and UPF3B resulted in a marked NMD inhibition and a transcriptome-wide upregulation of NMD substrates, demonstrating a functional redundancy between both NMD factors. Although current models assume that UPF3 bridges NMD-activating exon-junction complexes (EJC) to the NMD factor UPF2, UPF3B exhibited normal NMD activity in rescue experiments when UPF2 or EJC binding was impaired. Further rescue experiments revealed partially redundant functions of UPF3B domains in supporting NMD, involving both UPF2 and EJC interaction sites and the central region of UPF3. Collectively, UPF3A and UPF3B serve as fault-tolerant NMD activators in human cells.


Precisely regulated expression of correct gene products is indispensable for eukaryotic life. 44
This is underlined by the existence of several quality control mechanisms for gene expression, 45 one of which is the nonsense-mediated mRNA decay (NMD). NMD is primarily known for its 46 ability to eliminate mature mRNAs that contain a premature termination codon (PTC). Thereby, 47 rather indicate opposing functions of the two UPF3 paralogs with UPF3A being an antagonist 123 of UPF3B and broadly acting as an NMD inhibitor. 124 In this study, we resolved the controversy about the functions of UPF3A and UPF3B in the 125 NMD pathway using different UPF3 overexpression and knockout (KO) HEK293 cell lines. We 126 found that neither overexpression nor genomic KO of UPF3A resulted in substantial changes 127 of NMD activity or global alterations of the transcriptome. In UPF3B KO cells UPF3A protein 128 levels were upregulated, but NMD activity was maintained at almost normal level. In contrast, 129 the co-depletion of both UPF3 paralogs resulted in a marked NMD inhibition and a global 130 upregulation of PTC-containing transcripts. Moreover, rescue experiments revealed that UPF3 131 proteins have additional functions besides bridging the EJC and the NMD machinery. Taken 132 together, our data support a model of human NMD, in which UPF3A and UPF3B can replace 133 each other and therefore perform redundant functions. 134 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint

135
UPF3A overexpression or knockout does not affect NMD efficiency 136 Prior work using different mammalian models and various experimental approaches reached 137 different conclusions regarding the function of UPF3A in NMD (Fig 1A). Therefore, we set out 138 to re-examine the role of UPF3A in human cells by specifically manipulating its expression 139 levels. Under regular conditions UPF3A is barely present in cultured cells, presumably due to 140 its lower binding affinity to the stabilizing interaction partner UPF2 compared to UPF3B, 141 resulting in a rapid turnover of "free" UPF3A (Chan et al., 2009). We hypothesized that 142 increasing the abundance of UPF3A should lead to the stabilization of NMD targets if UPF3A 143 is an NMD inhibitor. To test this hypothesis, we generated Flp-In T-REx 293 (HEK 293) cells 144 inducibly overexpressing FLAG-tagged wildtype UPF3A to high protein levels ( Fig 1B). Global 145 analysis of the transcriptome using RNA-seq ( Fig EV1A and Datasets EV1-EV3) revealed, 146 except for UPF3A itself, no significant differential gene expression (DGE), differential transcript 147 usage (DTU) or alternative splicing (AS) events upon UPF3A overexpression compared to 148 control conditions (Figs 1C and D). Using these RNA-seq data, we analyzed NMD targets that 149 were previously described to be strongly upregulated in UPF3A overexpressing HeLa cells 150 (Shum et al., 2016). The DGE analysis and visualization of the respective read coverage 151 showed no substantial effects in our setup (Figs 1E and EV1B-E). Furthermore, quantification 152 of differential transcript usage via IsoformSwitchAnalyzeR (Vitting-Seerup & Sandelin, 2019) 153 could neither detect any differences in the global isoform fraction distribution, nor an 154 accumulation of PTC-containing transcripts ( Fig EV1F). Collectively, these analyses indicated 155 that UPF3A overexpression in HEK 293 cells does not negatively affect gene expression in 156 general or NMD in particular. C Fraction of expressed genes (genes with non-zero counts in DESeq2) were calculated which exhibit individual or combinations of differential gene expression (DGE), differential transcript usage (DTU) and/or alternative splicing (AS) events in WT cells overexpressing UPF3A using the respective computational analysis (cutoffs are indicated). AS and DTU events were collapsed on the gene level. For DGE, p-values were calculated by DESeq2 using a two-sided Wald test and corrected for multiple testing using the Benjamini-Hochberg method. For DTU, p-values were calculated by IsoformSwitchAnalyzeR (ISAR) using a DEXSeqbased test and corrected for multiple testing using the Benjamini-Hochberg method. For AS, p-values were calculated by LeafCutter using an asymptotic Chi-squared distribution and corrected for multiple testing using the Benjamini-Hochberg method. D Volcano plot showing the differential gene expression analyses from the RNA-Seq dataset of WT cells overexpressing UPF3A. The log2 fold change is plotted against the -log10 adjusted p-value (adj. p-value). Pvalues were calculated by DESeq2 using a two-sided Wald test and corrected for multiple testing using the Benjamini-Hochberg method. OE = overexpression. E Read coverage of SMG5 from WT HEK 293 RNA-seq data with or without induced UPF3A overexpression shown as Integrative Genomics Viewer (IGV) snapshot. Differential gene expression (from DESeq2) is indicated as Log2 fold change (Log2FC) on the right. Schematic representation of the protein coding transcript below. (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint Next, we approached the question of UPF3A function in the opposite way by generating 159 UPF3A knockout (KO) HEK 293 cell lines. Using CRISPR-Cas9 genome editing we isolated 160 three clones that lacked the UPF3A-specific band on the Western blot even after 161 downregulation of UPF3B (Fig 2A). Two clones (14 and 20) were characterized in detail. 162 In both cell lines, the UPF3A genomic locus contained insertions and/or deletions causing 163 frame-shifts and eventually PTCs (Figs EV2A-C Fig 2B). However, this effect was not rescued 172 by the (over)expression of transgenic UPF3A, indicating that it is not caused by the lack of 173 UPF3A but rather represents random variations in gene expression or clonal effects (Figs 174 EV2D-E). To get a complete overview of the effects of the UPF3A KO, we performed RNA-seq 175 of two UPF3A KO cell lines with or without an additional UPF3B knockdown (KD; Fig EV2F  176 and Datasets EV1-EV3). Initially, we focused on the UPF3A KO cell lines without KDs, for 177 which the global transcriptome analysis revealed that about 4-9 % of the expressed genes are 178 altered (Figs 2C-E). The observation that in the absence of UPF3A more genes were 179 downregulated than upregulated (879 vs. 676) could be an indicator for UPF3A NMD inhibiting 180 properties ( Fig 2F). However, the majority of genes with altered expression were clone specific 181 and only 110 genes showed downregulation in both UPF3A KO cell lines (Fig 2F). Investigation 182 of selected targets that were significantly up-or downregulated in both clones revealed that 183 the changes were not rescued after UPF3A overexpression, suggesting that they are UPF3A-184 independent (Figs EV2D-E). Another indication that UPF3A depletion does not generally affect 185 NMD efficiency came from the DTU analysis. 186 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021.

10
. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint Although clone 14 showed a minor downregulation of PTC-containing transcripts, which could 188 indicate more active NMD, this effect was not reproducible in the second clone ( Fig EV2G). isoforms of the NMD-targets RSRC2 and SRSF2 was observed ( Fig 3B). This indicates that 205 either UPF3B is not essential for NMD or that another protein is able to compensate for its 206 loss. The most obvious candidate for this function is its own paralog UPF3A, which was also 207 suggested previously to functionally replace UPF3B in NMD. Indeed, knocking down UPF3A 208 in the UPF3B KO cells resulted in the increase of NMD-sensitive RSRC2 and SRSF2 isoforms 209 ( Fig 3B). Of note, the combination of the UPF3B KO with UPF3A KD showed stronger effects 210 than the previously analyzed UPF3A KO plus UPF3B KD. This is probably caused by the lower 211 KD efficiency of the UPF3B siRNAs which can be observed by comparing the respective 212 protein levels (Fig 2A vs. Fig 3A). We suspect that the remaining UPF3B levels after siRNA-213 mediated UPF3B KD still support NMD.    qPCR how strongly the dKO affected NMD (Fig 4C). For all three tested genes, the expression 237 of the NMD-sensitive isoform was further increased compared to the previously used 238 combination of UPF3B KO with additional UPF3A KD. Of note, the NMD inhibitory effect 239 observed in the dKOs became more pronounced after UPF3B siRNA transfection, suggesting 240 that low levels of residual UPF3B protein were still present in the dKO cells ( Fig EV4C). 241 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint The expression levels of endogenous NMD substrates could be influenced by transcription 243 rates or other indirect effects, which could lead to over-or underestimating NMD inhibition. 244 Therefore, we investigated the NMD efficiency in the dKO cells using the well-established β-245 globin NMD reporter. To this end, we stably integrated β-globin WT or PTC39 constructs in 246 WT and UPF3 dKO cells (Fig 4D). These reporters also contained XRN1-resistant sequences 247 (xrRNAs) in their 3' UTRs, which allowed us to analyze not only the degradation of the full-248 length reporter mRNA but also to quantify decay intermediates (called xrFrag) ( To establish transcriptome-wide insights into UPF3A and UPF3B function, we carried out RNA-257 seq for both dKO clones, which was combined with and without UPF3B KD treatment to 258 eliminate as many of the potentially present remaining UPF3B proteins ( Fig EV5A and  259 Datasets EV1-EV3). Differential gene expression analysis showed that nearly three times as 260 many genes were upregulated than downregulated in both dKO cells (Figs 5A and EV5B). This 261 is consistent with the redundant role of UPF3B and UPF3A as supporting NMD factors. The 262 considerable overlap between both clones also suggests that we identified high-confidence 263 UPF3 targets. Furthermore, 890 of these gene were also significantly upregulated in previously 264 generated SMG7 KO plus SMG6 KD data ( that these are universal NMD-targets and not specific to a certain branch of the NMD pathway. 266 In addition to DGE, subsets of the expressed genes showed changes in alternative splicing 267 or/and differential transcript usage ( Fig 5B). In total, 14-16 % of the global transcriptome 268 showed single or combined changes (DGE, DTU and/or AS) and up to 20% when the cells 269 were treated with an additional UPF3B KD. 270 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021.

16
. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint In agreement with NMD inhibition in the dKOs, we saw that many transcripts containing a PTC 272 were up-regulated, while the corresponding transcripts without a PTC were down-regulated. 273 (Figs 5C and EV5D). Under these conditions, the IGV snapshot of the NMD-target SRSF2 274 showed NMD-inducing exon inclusion and 3' UTR splicing events, which were not visible in 275 combined UPF3B KO/UPF3A KD cells (Fig 5D). Collectively, the RNA-seq data support the 276 previously observed strong NMD inhibition in response to the complete absence of both UPF3 277 paralogs and hence their proposed redundancy. 278 UPF3A supports NMD independent of a bridge function 279 Next, we aimed to analyze whether the severe effects in the dKOs are at least partly due to 280 the loss of a protein-protein interaction bridge between UPF2 and the EJC, while the presence 281 of UPF3A in the UPF3B KOs preserved this function ensuring NMD functionality. Therefore, 282 we expressed FLAG-tagged UPF2 in WT, UPF3B KO and UPF3A-UPF3B dKO cells and 283 analyzed the UPF2 interactome using mass spectrometry (Dataset EV4). Consistent with the 284 previously described interaction partners, we found many NMD factors as well as EJC proteins 285 in the UPF2 interactome in WT cells (Fig 6A). Contrary to our expectation, the three EJC core 286 components (EIF4A3, RBM8A, MAGOHB) barely co-precipitated with UPF2 in the absence of 287 UPF3B (compared to control: log2 FC = 0.58, 0.63 and 0.74, respectively; Fig 6B) and were 288 therefore strongly decreased in comparison to the WT cells ( Fig EV6A). Hence, the UPF2-289 bound UPF3A was unable to establish a stable interaction with the EJC. Surprisingly, in the 290 UPF3B KO cells the EJC-associated CASC3 still showed relatively high levels of co-291 precipitation (log2 FC = 4.42), which therefore appears to be independent of the interaction 292 with the other EJC components. In the dKOs all interactions with EJC proteins including 293 CASC3 were completely lost (Figs 6C and EV6B). The latter was also observed in a 294 comparable approach employing stable isotope labeling with amino acids in cell culture 295 (SILAC) to analyze the UPF2 interactome in the WT and dKO cells (Dataset EV5). All EJC 296 core components that were highly co-precipitated in WT cells were lost in the absence of both 297 UPF3 paralogs (Figs EV6C-E). 298 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint

18
. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. UPF2 and the EJC, which we validated in the mass spec analysis, it was surprising to see that 312 disruption of either of these interactions did not affect UPF3B's rescue capacity (Fig 6F,  interaction on NMD, we created UPF3B variants lacking that specific middle domain or 323 combined the deletion with the previously used interaction mutations (Fig 6D and E). We 324 observed a similar pattern as the UPF3B mutants examined in the previous experiment: when 325 only the middle domain was deleted, UPF3B was able to rescue NMD comparable to the WT 326 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint protein (Fig 6F lane 7). In combination with a mutation in the UPF2-or the EJC binding site its 327 function in NMD was severely impaired (lanes 8 and 9). This suggests that if the classic bridge 328 formation is inhibited by removing either of the interaction sites, UPF3B relied on the function 329 carried out by the uncharacterized middle domain. 330 As further support for the NMD promoting effect of UPF3A on NMD, we performed a rescue 331 experiment in the above used UPF3B KO UPF3A KD cells (Figs 7A and B). Expression of the 332 siRNA insensitive UPF3A construct did not only restore NMD functionality, it was even more 333 efficient than the UPF3B construct (Fig 7C lane 3 vs. lane 5), underlining our previous 334 statement: UPF3A supports and elicits NMD comparably to its paralog UPF3B. 335 Also comparable to UPF3B, UPF3A has a second naturally occurring isoform but instead of 336 skipping exon 8 (like UPF3B) it excludes its fourth exon. This isoform is transcribed in 337 approximately one third of the cases (Fig EV6G) in WT HEK 293 cells but cannot be detected 338 on protein levels. We were interested, whether this isoform was as potent to elicit NMD as the 339 full-length construct. Expression of the exon 4 UPF3A deletion construct in the UPF3 depleted 340 cells showed no rescue (Fig 7C lane 4). Hence, exon 4 must encode for an essential region 341 required for the bridge-independent function of UPF3A. 342 In view of these observations and the close proximity of exon 4 (124-157) to the middle domain 343 (147-256), we decided to investigate which effect the deletion of the homologous exon 4 in the 344 paralog UPF3B has. The UPF3B Δe4 construct behaved like the corresponding UPF3A 345 construct and showed no NMD rescue activity (lane 6). However, the expression of the N-346 terminus of UPF3B was able to restore NMD comparably to the WT protein (lane 7). This is 347 consistent with all our previous findings, since the first 279 amino acids contain the UPF2 348 binding site as well as the middle domain, which was shown to be sufficient to elicit NMD (Fig  349   6F lane 5). Due to the fact that the C-terminus contains only one interaction site and lacks exon 350 4, its incapability to rescue NMD is in line with our previous experiments. Overall, our results 351 identify exon 4 of UPF3 as a previously unnoticed region that is essential for its function in 352 NMD. 353 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint Although our results cannot fully answer this last question, we have more or less definite 379 answers for the functions of the two proteins UPF3A and UPF3B in NMD. All our data support 380 the notion that the presence of UPF3B or UPF3A is sufficient to maintain NMD activity in 381 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. Although these results were obtained in a heterologous context, mouse UPF3A does not 397 appear to be a general NMD inhibitor. 398 The different KO cells that we have generated in the course of this project enabled us to 399 conduct experiments that went beyond investigating UPF3-dependent NMD substrates in 400 human cells. Specifically, we were able to study the composition of NMD complexes without 401 UPF3B or both UPF3 proteins and to carry out rescue experiments with different UPF3A and 402 UPF3B protein variants. Not entirely unexpected, we observed that in the absence of UPF3A 403 and UPF3B, the interaction between UPF2 and the EJC is lost. This bridging by UPF3 between 404 UPF2-containing NMD complexes and the EJC was previously considered to be essential for 405 NMD. However, two observations argue against UPF3 being mainly a bridging protein. First,406 UPF3B mutants that cannot interact with either the EJC or UPF2 fully rescue NMD. Only when 407 both interaction sites were mutated, UPF3 lost its NMD function. This indicates that the 408 interaction with one of the two interaction partners is sufficient to maintain NMD. Second, we 409 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint observed that not only in UPF3 double-KO cells, but also in UPF3B KO cells, the bridge 410 between UPF2 and the EJC was lost. Although quite surprising at first glance, this is in good 411 agreement with previous results showing that the interaction between the EJC and UPF3A is 412 substantially weaker than that between the EJC and UPF3B (Kunz et al., 2006). Indeed, earlier 413 structural data also argue against a bridging function of UPF3. In the cryo-EM structure of an 414 EJC-UPF3-UPF2-UPF1 complex, UPF1 did not face towards a possible terminating ribosome 415 in the 5' direction, but instead in 3' direction (Melero et al., 2012). Therefore, one could 416 conclude that the interactions between all these proteins do not take place at a single time our view, only be explained with more complex models, which must also consider non-linear 427 relationships and potential auxiliary functions of certain regions of UPF3 (Fig 8). 428 Since the interactions of UPF3 are essential only in combination with each other, we propose 429 that UPF3A and UPF3B exert multiple functions at different time points of NMD and in 430 association with different complexes. We only consider here the previously described 431 interactions of UPF3 with UPF2, the EJC and the release factor 3 (Fig 8B). While it was 432 previously described that UPF3B interacts better than UPF3A with UPF2 (Chan et al., 2009), 433 we find both proteins in the FLAG-UPF2 IP. The amount of UPF3A does not seem to increase 434 when UPF3B is depleted, which could be due to the overexpression of UPF2 in our 435 experimental system. The interaction of UPF3 and UPF2 is also conserved in yeast and is thus 436 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021.  25 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint likely to have functional significance. However, a UPF3B mutant in which binding to UPF2 was 438 inactivated rescues NMD better than the UPF3B WT. 439

NMD complex assembly
The interaction of the middle domain of UPF3B with RF3 has only recently been described 440 (Neu-Yilik et al., 2017). Again, we find that removing only the middle domain does not 441 substantially inhibit NMD. In combination with an inactivation of the UPF2 binding, the deletion 442 of the middle domain leads to a complete inhibition of the NMD in the rescue assay. This could 443 also happen when exon 4 of UPF3B or UPF3A is removed, which is located at the junction 444 between the UPF2 binding domain and the middle domain. Therefore, UPF3A Δexon4 would 445 be a naturally occurring, NMD-inactive variant of the UPF3 proteins. 446 How might the different regions and domains communicate with each other and regulate the 447 function of UPF3 in NMD? The function of the middle domain in relation to NMD has not yet 448 been investigated. It is conceivable that UPF3B plays a minor role in translation termination 449 and that the events that trigger NMD can also occur without the middle domain -potentially 450 with a delay (Figure 8c). Therefore, the deletion of the middle domain could be tolerated in 451 isolation but would become fatal in combination with other mutations that impair additional 452 functions. With regard to the EBM, we propose that its binding stabilizes the EJC, for example 453 by preventing the interaction of the EJC with PYM1 (Fig 8C). PYM1 is a known EJC 454 disassembly factor and binds to the EJC at a surface area that overlaps with the EBM binding Likewise, the interaction of UPF3 with UPF2 might be important for the assembly of an NMD-460 inducing complex (Fig 8C) that needs to be timed with translation termination and that leads 461 to NMD activation only in the presence of the EJC. Although these suggestions may not 462 accurately reflect the molecular events during NMD, they illustrate possible functions of the 463 domains of UPF3, particularly in relation to their NMD-inactive combinations (Fig 8). Overall, 464 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint our observations fit well with a "synthetic lethal" model in which inactivation of any two domains 465 together disrupts UPF3 activity. 466 One factor whose function needs to be examined in more detail in the context of UPF3 is 467 CASC3. Our mass spectrometry analysis shows that CASC3 immunoprecipitates very well 468 with UPF2 in wildtype cells. CASC3 still partially precipitates with UPF2 in UPF3B KO cells, 469 although the interaction of the other EJC factors is reduced to background levels. In previous 470 work, we observed that UPF3B interacts less well with the EJC when CASC3 is knocked out 471 (Gerbracht et al., 2020). This indicates that an interaction between CASC3 and the UPF3 472 proteins exists that is not well understood so far. What kind of interaction this is and what 473 function it has will be interesting to address in future experiments. 474 Our own work and the work of Yi et al. have re-examined the functions of the human UPF3 475 paralogues UPF3A and UPF3B (Yi et al., 2021). Together, the studies confirmed some 476 previous findings and disconfirmed others, thereby successfully (re-)defining the role of UPF3 477 proteins in human cells. A few questions remain unanswered and need to be addressed in the 478 future, for example how exactly the middle domain supports NMD. As described above, our 479 results have implications for the understanding of the NMD mechanism, as they are 480 incompatible with, and thus exclude, certain models of NMD. In addition, they may also help 481 to better understand the link between UPF3B and intellectual disability and which domains of 482 UPF3A modulate the severity of the disease and may therefore be potential targets for therapy. 483 484 485 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint labeled amino acids Lysin and Arginine at final concentrations of 0.798 mmol/L and 0.393 620 mmol/L, respectively. The three conditions were "light" (unlabeled Lys/ Arg), "medium" (Lys4/ 621 Arg6) and "heavy" (Lys8/ Arg10). Unlabeled proline was added in all conditions to prevent 622 enzymatic Arginine-to-Proline conversion. 623

Experimental setup for SILAC with FLAG-tagged UPF2
624 Expression of FLAG-GST and FLAG-UPF2 was induced for 72 h with 1x cumate. The cells 625 lysed in 250 -400 μl Buffer E with 1 μg/ml RNase and sonicated using the Bandelin Sonopuls 626 mini20 with 15x 1s pulses at 50% amplitude with a 2.5 mm tip. Protein concentrations were 627 measured using the Bradford assay and protein samples containing 1.6-1.7 mg/ml total protein 628 were diluted. 600 μl of these samples were incubated with 30 μl Anti-FLAG M2 magnetic beads 629 (Sigma) for 2 h on an overhead shaker at 4 °C. The beads were then washed three times for 630 5 min with mild EJC-Buffer before eluting twice with 22 μl of a 200 μg/ml dilution of FLAG-631 peptides (Sigma) in 1x TBS for 10 min at RT and 200 rpm each elution step. Another elution 632 with 1x SDS loading buffer was performed to analyze pull down efficiency via Western blot. 633 The FLAG-peptide eluates were then mixed as followed: 7 μl of both light conditions, 14 μl 634 medium and 14 μl heavy. 1 volume of SP3 (10% SDS in PBS) was added and the samples 635 were reduced with 5 mM DTT and alkylated with 40 mM CAA. 636 Tryptic protein digestion was achieved by following a modified version of the single pot solid 637 phase-enhanced sample preparation (SP3) (Hughes et al., 2014). In brief, paramagnetic Sera-638 Mag speed beads (Thermo Fisher Scientific) were added to the reduced and alkylated protein 639 samples and then mixed 1:1 with 100% acetonitrile (ACN). Protein-beads-complexes form 640 during the 8 min incubation step, followed by capture using an in-house build magnetic rack. 641 After two washing steps with 70% EtOH, the samples were washed once with 100% ACN. 642 Then they were air-dried, resuspended in 5 μl 50 mM Triethylamonium bicarbonate 643 supplemented with trypsin and LysC in an enzyme:substrate ratio of 1:50 and incubated for 16 644 h at 37°C. The next day the beads were again resuspended in 200 μl ACN and after 8 min 645 incubation placed on the magnetic rack. Tryptic peptides were washed with 100% ACN and 646 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint air-dried before dissolved in 4% DMSO and transfer into 96-well PCR tubes. The last step was 647 the acidification with 1 μl of 10% formic acid, then the samples were ready for mass spec 648 analysis. 649 Proteomics analysis was performed by the proteomics core facility at CECAD via data-650 dependent acquisition using an Easy nLC1200 ultra high-performance liquid chromatography 651 (UHPLC) system connected via nano electrospray ionization to a Q Exactive Plus instrument 652 (all Thermo Scientific) running in DDA Top10 mode. Based on their hydrophobicity the tryptic 653 peptides were separated using a chromatographic gradient of 60 min with a binary system of 654 buffer A (0.1% formic acid) and buffer B (80% ACN, 0.1% formic acid) with a total flow of 250 655 nl/min. For the separation in-house made analytical columns (length: 50 cm, inner diameter: 656 75 μm) containing 2.7 μm C18 Poroshell EC120 beads (Agilent) that were heated to 50 °C in 657 a column oven (Sonation) were used. Over a time period of 41 min Buffer B was linearly 658 increased from 3% to 27% and then more rapidly up to 50% in 8 min. Finally, buffer B was 659 increased to 95% within 1 min followed by 10 min at 95% to wash the analytical column. The MS RAW files were then analyzed with MaxQuant suite (version 1.5.3.8) on standard 666 settings with the before mentioned SILAC labels (Cox & Mann, 2008). By matching against the 667 human UniProt database the peptides were then identified using the Andromeda scoring 668 algorithm (Cox et al., 2011). Carbamidomethylation of cysteine was defined as a fixed 669 modification, while methionine oxidation and N-terminal acetylation were variable 670 modifications. The digestion protein was Trypsin/P. A false discovery rate (FDR) < 0.01 was 671 used to identify peptide-spectrum matches and to quantify the proteins. Data processing and 672 statistical analysis was performed in the Perseus software (version 1.6.1.1) (Tyanova et al.,673 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint 2016). Using the One-sample t-test the significantly changed proteins were identified (H0 = 0, 674 fudge factor S0 = 0.2). Visualization was performed with RStudio (version 1.2.5033). 675 Label-free quantitative mass spectrometry 676 Twenty-four hours before expression of the FLAG-tagged constructs, the HEK 293 WT cells 677 were treated with Luciferase siRNA and the UPF3B KO clone 90 and UPF3 dKO clone 1 cells 678 were treated with siRNAs targeting residual UPF3B. The expression of either FLAG-GST or 679 FLAG-UPF2 in WT cells and FLAG-UPF2 in the clones 90 and 1 was induced for 48 h with 1x. 680 Lysis and sample preparation were performed as described above. MS analysis was 681 performed as described above with a slightly adjusted gradient as followed: 3 -30% B in 41 682 min, 30 -50% B in 8 min, 50-95% B in 1 min, followed by 10 min washing at 95%. LFQ values 683 were calculated using the MaxLFQ algorithm (Cox et al., 2014) in MaxQuant. Significantly 684 changed proteins were identified by two-sample t-testing (fudge factor S0 = 0.2). 685

686
The cells were harvested in RNAsolv reagent and total RNA extraction was performed as 687 described above. 3.0 µg total RNA were resolved on a 1% agarose/0.4 M formaldehyde gel 688 using the tricine/triethanolamine buffer system (Mansour & Pestov, 2013). Next a transfer on 689 a nylon membrane (Roth) in 10x SSC followed. The blot was incubated overnight at 65°C in 690 Church buffer containing [α-32P]-GTP body-labeled RNA-probes for mRNA reporter detection 691 (Voigt et al., 2019). Ethidium bromide stained 28S and 18S rRNA served as loading controls. 692 RNA signal detected with the Typhoon FLA 7000 (GE Healthcare) was quantified in a semi-693 automated manner using the ImageQuant TL 1D software with a rolling-ball background 694 correction. EtBr-stained rRNA bands were quantified with the Image Lab 6.0.1 software (Bio-695 Rad). Signal intensities were normalized to the internal control (rRNA) before calculation of 696 mean values. The control condition was set to unity (TPI WT for reporter assays), quantification 697 results are shown as data points and mean. 698 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint

Conflict of interest 748
None. 749 . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint . CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. CC-BY-NC 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted July 13, 2021. ; https://doi.org/10.1101/2021.07.07.451444 doi: bioRxiv preprint