A toolbox for systematic discovery of stable and transient protein interactors in baker's yeast

Developing ABOLISH—a strategy to enhance the signal-to-noise ratio for exogenous biotin ligases

To expand the arsenal of biotin-based tools available for protein interaction profiling in yeast it is essential to take into consideration endogenously biotinylated proteins (Sumrada & Cooper, 1982; Stucka et al, 1991; Hasslacher et al, 1993; Brewster et al, 1994; Hoja et al, 2004; Kim et al, 2004). This is because most biotinylated yeast proteins are highly abundant (Fig 1A) and therefore can mask many of the expected PPI assay signals on a streptavidin blot (Fig 1B) or take up a large percent of the reads from MS analyses. Since PL relies on biotinylated protein enrichment and detection, it would clearly be advantageous to reduce the background signal from the endogenously biotinylated proteins. To do this, we created a new method of endogenous biotinylation reduction that we call ABOLISH, for Auxin-induced BiOtin LIgase diminiSHing. In this method, Bpl1, the only endogenous yeast biotin ligase, is C-terminally tagged with an auxin-inducible degron (AID*, Nishimura et al, 2009; Morawska & Ulrich, 2013). Therefore, in the presence of auxin and the Oryza sativa transport inhibitor response 1 (OsTIR1, Nishimura et al, 2009) adaptor protein, the controlled and transient degradation of this essential enzyme ensues, leading to a reduction in biotinylation of its substrates.

Previously, many of the PL experiments using biotin ligases in yeast aimed to reduce the endogenous biotinylation by preculturing cells in reduced biotin (RB) media (Jan et al, 2014; Costa et al, 2018; Dahan et al, 2022). We therefore tested this condition followed by auxin addition to induce Bpl1 degradation and finally treatments with a biotin pulse (illustrated in Fig EV1A). This demonstrated that RB media alone was not an optimal condition since the initial dramatic reduction in endogenous biotinylation levels was rapidly reversed upon addition of free biotin (Fig EV1B), highlighting the limitations of this method for reducing background biotinylation. Importantly, this reversal was not observed if auxin was used to deplete Bpl1-AID*-9myc, proving that ABOLISH can achieve a more complete reduction in background biotinylation noise.

Click here to expand this figure.

Next, we wanted to track whether ABOLISH could indeed increase the sensitivity of detection when using exogenous, promiscuous, biotin ligase fusions. To do that, we chose a complex for which we could follow both stable and transient PPIs: the most recently characterised ER-resident insertase; the ER membrane protein complex, EMC (Guna et al, 2018). This highly-conserved machinery is composed of eight subunits (Emc1-7 & Emc10) in yeast (Jonikas et al, 2009) and 10 (EMC1-10) in humans (Christianson et al, 2012). Since its discovery as an insertase for moderately hydrophobic tail-anchor (TA) proteins (Guna et al, 2018; Volkmar et al, 2019), it has also been found to insert multi-pass transmembrane domain (TMD)-containing proteins into the ER (Chitwood et al, 2018; Shurtleff et al, 2018; Tian et al, 2019; Bai et al, 2020; Miller-Vedam et al, 2020; O'Donnell et al, 2020; Pleiner et al, 2020). Furthermore, it is required for the biogenesis of single-pass TMD proteins, which do not contain a signal peptide (also known as type III membrane proteins, O'Keefe et al, 2021) and a subset of TMD proteins, which traffic from the ER to lipid droplets (LD) (Leznicki et al, 2021). To this end, it has a wide, and not yet fully characterised, substrate range and a clear set of stable interactions that we can track.

To compare the various biotin ligases in their capacity to label both stable and transient interactions, and to evaluate whether ABOLISH could enhance the detection of these labelled proteins, we tagged Emc6 at its N-terminus with either BioID2-HA or TurboID-HA (Fig 1C). A third strain expressing both TurboID-HA-Emc6 and the ABOLISH system was also generated, along with three control strains in which Sbh1, rather than Emc6, was tagged. Sbh1 was selected as a control to compare against Emc6 primarily because it is an ER membrane protein of similar abundance (Weill et al, 2018) whose N-terminus (used for tagging) also faces the cytosolic side of this organelle. Furthermore, it is part of the SEC machinery required for translocation across the membrane and is therefore functionally comparable, yet distinct from, the EMC. Hence, the comparison between these samples should lead to the discovery of specific EMC interactors. All promiscuous biotin ligase tags were preceded by the constitutive, moderate, CYC1 promoter to ensure that we do not dramatically overexpress our proteins leading to false positives, and all tagged proteins ran at their expected molecular weights as determined by SDS–PAGE (Fig EV1C).

Surprisingly, we found that overnight growth in RB media resulted in a decrease in the amount of TurboID-HA-tagged proteins (Fig EV1D); thus negating the signal-to-noise advantage conferred by ABOLISH. To uncover a condition where the TurboID-tagged protein levels are not reduced but endogenous biotinylation levels are, we tested several different parameters including: overnight growth in either regular YPD or SD media containing auxin; and growth in either RB or regular SD media with auxin and biotin addition at different time-points (Fig EV1E; see legend for time-point details). We found that overnight treatment with auxin in SD media (Fig EV1E, 2^nd lane) resulted in the strongest background biotinylation reduction without a loss in TurboID-HA-Emc6. The requirement for this relatively lengthy auxin treatment is rationalised by the long half-lives (> 10 h, Christiano et al, 2014) of endogenously biotinylated proteins. Interestingly, the abundance of the ER translocon, Sec61, remained constant independent of the conditions tested. This suggests that the loss of TurboID-tagged proteins triggered by biotin depletion (in RB media) may be a regulatory adaptation to biotin starvation, rather than general protein degradation from the ER. We then calibrated the length of time for biotin treatment that would ensure sufficient labelling material and time for true PPI events to be captured by TurboID without increasing background biotinylation (Fig 1D). We found that even 4 h of exogenous biotin addition did not negate the effect of the auxin-induced depletion of endogenous biotinylated proteins and therefore this growth pipeline (using regular media and not RB media; illustrated in Fig 1E) was adopted for future ABOLISH experiments. These data collectively demonstrate that the ABOLISH method can be harnessed to reduce background biotinylation ‘noise’, paving the way for enhanced signal detection from exogenous proximity-labelling enzymes.

Comparing three biotin ligase systems identifies their ability to uncover both stable and transient protein–protein interactions by LC–MS/MS

While IPs of epitope-tagged proteins enrich for stable interactors (in this case EMC complex components), streptavidin APs should capture transient interactions labelled by exogenous biotin ligases (in this case clients inserted into the ER by the EMC or regulators of the EMC) and a subset of stable ones; depending on the protein topology and accessibility of K residues on the same side of the membrane. To directly compare the type of interactions that we can identify we analysed, either by HA-IP or streptavidin-AP, strains expressing either BioID2-HA-Emc6, TurboID-HA-Emc6, or TurboID-HA-Emc6 on the background of the ABOLISH system (Fig 2A).

Although we had previously observed the reduced background conferred by ABOLISH using Western blot (Fig 1), we first used our MS data to understand the impact of this system on proteomic experiments. We compared the intensities of the six known endogenously biotinylated proteins (Fig 1B) measured by LC–MS/MS from the strains subject to streptavidin-AP expressing TurboID-HA-Emc6 either with or without ABOLISH (Fig EV2). Consistent with our earlier observations (Fig EV1B), Arc1 was the most sensitive to ABOLISH; however, all other endogenously biotinylated proteins were also strongly and significantly reduced, with the exception of Hfa1, the levels of which did not change. Since Hfa1 abundance is very low compared with the other proteins, its insensitivity to ABOLISH has less impact on the background noise. Altogether, endogenously biotinylated proteins made up nearly half (~ 47%) of the total intensity from all proteins identified by LC–MS/MS (as measured by intensity-based absolute quantification (iBAQ)) from streptavidin-AP of the TurboID-HA-Emc6 strain without ABOLISH. Implementing ABOLISH decreased this number to ~ 27%, demonstrating how much proteomic ‘bandwidth’ is freed up by this approach thus increasing the probability of bona fide interactor identification.

Click here to expand this figure.

Next, we wanted to identify the high-confidence protein interactors from the different LC–MS/MS runs (Fig 2A). To do this, Emc6 replicate samples were compared with their Sbh1 control counterparts to find high-confidence protein identifications (Dataset EV1 and via PRIDE, PXD033348). These were defined by the following criteria: a P-value of ≤ 0.05 (streptavidin samples) or ≤ 0.1 (HA samples); a fold-change of ≥ 2; and identification by two or more unique peptides (Fig 2B). From the HA-IP, as expected, several EMC complex components satisfied these requirements (Fig 2B and C; dotted outline). From the streptavidin-AP samples, only three high-confidence hits were found by BioID2, two of which were Emc1 and Emc4 (Fig 2B and C; blue fill). This confirms that although BioID2 is able to label bona fide interactors, its capacity is limited likely due to its relatively low catalytic activity at 30°C (Kim et al, 2016). On the contrary, 13 high-confidence identifications were made by TurboID (Fig 2B and C; black, solid outline). Looking at stable interactors, Emc2 was found in addition to Emc1 and Emc4, already hinting at increased labelling functionality relative to BioID2. Importantly, the luminal EMC component, Emc7, was not found, indicating that there was no postlysis biotinylation by the cytosolic-facing TurboID. Most encouragingly, of the remaining 10 putative interactors, nine had membrane protein features classically associated with EMC clients, suggesting an increased capacity to uncover transient interactions (Fig 2C; asterisks). Eight of these are multi-pass TMD secretory pathway proteins, and the remaining protein (Alg1) is a lipid-droplet (LD) protein with a single N-terminal TMD—a characteristic recently demonstrated to define EMC-dependence (Leznicki et al, 2021).

Furthermore, incorporating the ABOLISH system enabled the detection of even more putative interactors labelled by TurboID (Fig 2B and C; yellow fill). The overlap between both TurboID strategies was very large with the same stable interactions and all nine candidate substrates being found. Another 16 high-confidence identifications were made, five of which were secretory pathway multi-pass TMD proteins (Fig 2C; asterisks). Comparing our list of putative substrates to published yeast EMC client data found by ribosome profiling (Shurtleff et al, 2018) and proteomic analysis of WT vs EMC3 KO cells (Bai et al, 2020), revealed that eight out of the 14 identified had previously been found in either study (Fig 2D) supporting the validity of our transient PPI discovery. To our knowledge, this is the first time TurboID has been successfully used in baker's yeast, and our data demonstrate that it labels both stable and transient PPIs. In addition, the ABOLISH system enhances the capacity to detect TurboID-labelled interactors.

Validating new EMC substrates using genetic tools and a natively-expressed pairwise biotinylation method

The similarity between our list of candidate EMC clients and published datasets (Fig 2D) strongly suggested that TurboID-mediated proteomics, both with and without ABOLISH, identified bona fide EMC substrates. Such substrates should be affected by loss of the complex, and indeed, it was previously shown that the abundance and/or localisation of true EMC substrates changes upon Emc3 loss (Bai et al, 2020). We therefore deleted EMC3 on a selection of our candidates and observed a strong reduction in the abundance of GFP-Alg1 (Fig 3A, top panel) and, to a lesser, but still significant, extent, Gnp1-GFP (Fig 3A, 2^nd panel). Deletion of EMC3 also changed the localisation of GFP-Pdr12 relative to the control strain (Fig 3A, 3^rd panel). Pdr12 is a plasma membrane (PM) ATP-binding cassette (ABC) transporter, which first requires insertion into the ER before trafficking to its final destination. Therefore, the accumulation of Pdr12 on the ER (shown by colocalisation with ER-resident Sec63; Fig EV3A) in the Δemc3 strain likely signifies a preinserted population at the ER surface. Interestingly, Pho84-GFP was also affected by the loss of EMC3 (Fig 3A, bottom panel). Pho84 is an inorganic phosphate transporter, which traffics along the secretory pathway to the PM. In phosphate-rich media (used in these experiments) it is known to be internalised and degraded by the vacuole and hence very weakly detectable in the WT images (Petersson et al, 1999; Hürlimann et al, 2007). EMC3 deletion, however, led to a strong accumulation of Pho84-GFP signal at internal membranes. These changes were specific to the proposed clients since localisation and abundance of the ER membrane protein, Sec63, were not affected by perturbation of the EMC (Fig EV3A). Collectively, these functional assays support these proteins as newly-validated clients of the EMC complex.

Click here to expand this figure.

More broadly, however, verifying transient interactions is, in itself, a challenging task as methods to validate PPIs (such as co-IP) are again optimised for very stable interactions. We therefore used a parallel biotinylation approach involving the BirA biotin ligase, which specifically biotinylates the AviTag sequence (Cronan, 1990; Beckett et al, 1999). In this setup, even transient protein–client interactions can be assayed in vivo and at physiological expression levels. To do this, a haploid strain expressing a BirA-tagged protein (e.g. Emc6) under its native promoter is mated with a haploid strain of the opposite mating type, which expresses a potential interactor N-terminally tagged with AviTag (also under native promoter control). The diploid strains can then be analysed for the appearance of a streptavidin-positive band that proves that BirA came sufficiently close to the AviTag (illustrated in Fig 3B). Initially, well-characterised and previously validated interactors were selected to test the utility and feasibility of this validation method: hence strains expressing either AviTag-Emc2, -Emc4, or -Spf1 (Jonikas et al, 2009; Shurtleff et al, 2018) were crossed with the BirA-Emc6 strain. To ensure the best signal-to-noise ratio in this setup, we integrated the ABOLISH system into these strains, and this clearly reduced the background signal and made the assay even cleaner (Fig EV3B and C). The ABOLISH system is therefore critically important for: lower abundance proteins; more transient interactions; proteins that are less efficiently biotinylated; or proteins whose molecular weight is similar to that of endogenously biotinylated proteins. Hence ABOLISH can also extend the dynamic range of BirA-AviTag for pairwise, gel-based assays.

This assay is highly specific since it did not falsely report an interaction for the highly abundant ER-resident protein, Pdi1 (Fig EV3D), in contrast to the clearly visible bands for Emc2, Emc4 and Spf1 (Fig EV3B and C). This made us confident that our assay reveals bona fide interactions in vivo. Indeed, Fks1, an interactor found by TurboID (Fig 2B and C) and a known EMC substrate (Shurtleff et al, 2018), was readily detected by streptavidin blot using the BirA-AviTag/ABOLISH system (Fig 3C). The AviTagged amino acid permease, Gnp1, was similarly easy to detect. Some clients required a higher contrast setting to be visualised. However, both AviTag-Pdr5 and AviTag-Pdr12 produced clear streptavidin-reactive bands compared with AviTag-Stv1, which did not produce a detectable Emc6 interaction (Fig 3C). Collectively, these data highlight the power of the BirA-AviTag/ABOLISH system for providing a rare, in vivo ‘snapshot’ of the transient interactions between the EMC insertase and both previously-confirmed (Shurtleff et al, 2018; Bai et al, 2020) and newly-validated (Fig 3A and C) substrates. More broadly it serves as a rapid, systematic pipeline for the validation of TurboID interactomes.

Generating a biotinylation toolkit: a collection of five full-genome libraries to facilitate high-throughput protein–protein interaction discovery

Through studying and comparing the promiscuous biotin ligases that can be used in yeast, we have demonstrated that TurboID, especially when combined with the ABOLISH system, serves as an unbiased tool to efficiently label stable and transient functional interactors in S. cerevisiae. This in turn can lead to the discovery of novel protein-machinery substrates, as highlighted for the EMC (Figs 2 and 3), and regulators. We have also demonstrated the suitability of the BirA-AviTag technology for assaying and validating native pairwise interactions and the capacity of this sensitive methodology to highlight even transient interactions.

To truly harness the power of these biotinylation tools and make them widely applicable, we created whole-proteome collections of yeast strains (also called libraries) using our recently developed approach for yeast library generation called SWAp Tag (SWAT) (Yofe et al, 2016; Meurer et al, 2018; Weill et al, 2018). This approach allows us to take an initial library and swap its tag to any one of our choice. Therefore, using the N′ GFP SWAT library and accompanying SWAT protocol (Yofe et al, 2016; Weill et al, 2018) we generated five whole-genome libraries (Fig 4). In the first two, each strain encodes one yeast protein fused at its N′ to a TurboID-HA tag expressed under the control of a medium-strength constitutive CYC1 promotor (Yofe et al, 2016; Weill et al, 2018) with, or without, the ABOLISH system. The third library is an N′ tag CYC1pr-BioID2-HA collection. All three libraries contain generic N′ localisation signals (signal peptides (SPs) and mitochondrial targeting signals (MTS) (Yofe et al, 2016; Weill et al, 2018)) where required. The last two full-genome libraries express N-terminally tagged proteins with either BirA or AviTag under their endogenous promoters and native N′ localisation signals, and the ABOLISH system is also integrated. In addition to PPI validation (as shown above) these libraries can, of course, be used for hypothesis-driven interrogation of interactions between any two proteins of interest.

All newly-generated libraries were subject to strict quality control checks (see Methods). Furthermore, a number of new library strains were selected and subject to SDS–PAGE analysis to confirm both protein expression and that the new tag had recombined in-frame during the SWAT process. This was demonstrated to be the case for the BioID2-HA (Fig EV4A), TurboID-HA (Fig EV4B) and TurboID-HA/ABOLISH (Fig EV4C) libraries. Hence these represent five high-coverage yeast libraries that will be freely distributed to enable high-throughput exploration, discovery and validation of stable and transient interactions throughout the yeast proteome.

Click here to expand this figure.

Reagents and Tools table

Methods and Protocols