Wtf^poison protein self-assembly is broadly conserved

To test the models of Wtf protein functions and to understand how these functions could be supported by extremely diverse protein sequences (e.g., <20% amino acid identity), we generated a large panel of wtf variants that represent over 100 million years of divergence [19]. We reasoned that functionally important features of the proteins would be conserved in functional variants (i.e., toxic Wtf^poison proteins), despite minimal amino acid conservation (Fig 1C). Twelve of the variants we assayed are wild-type or mutant alleles of wtf genes found in S. octosporus, S. osmophilus or S. cryophilus (S1 Fig). One of these genes, S. octosporus wtf25, has been previously characterized and found to encode a functional meiotic driver in its endogenous species ([19]; S1 Table). Four additional genes (S. cryophilus wtf1, S. osmophilus wtf41, and S. octosporus wtf61) have been shown to encode functional Wtf^poison and Wtf^antidote proteins in an ectopic Saccharomyces cerevisiae assay system ([19]; S1 Table). The S. cerevisiae system used to study Wtf proteins is well established and reflects phenotypes observed in the endogenous fission yeast species [15,19,25]. Finally, we assayed a panel of new mutant alleles of the previously characterized wtf4 gene from the S. kambucha isolate of S. pombe [15,22,24] (Fig 1D).

Our initial model, based on Wtf4^poison, was that Wtf^poison toxicity is linked to the ability of a Wtf^poison to self-assemble [15]. To test this idea, we assayed if other functional Wtf^poison proteins besides wild-type Wtf4^poison also self-assemble. To do this, we used the Distributed Amphifluoric FRET (DAmFRET) approach, in which proteins are tagged with the photoconvertible mEos3.1 fluorophore (referred to as mEos hereafter) and expressed in S. cerevisiae. A uniform fraction of the green mEos population is photoconverted to its red form, and protein-protein interactions are then detected as FRET (Fluorescence Resonance Energy Transfer) between the green and red fluorophores within single cells. Monomeric mEos exhibits negligible FRET and is used as a negative control (Fig 2A and 2B) [15,31].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Extremely diverged toxic Wtf^poison proteins exhibit similar self-assembly and intracellular localization.

A. Cartoon illustrating the Distributed Amphifluoric FRET (DAmFRET) assay (modified from [31]). AmFRET values for individual cells were calculated by dividing the acceptor fluorescence by the FRET fluorescence, both measured via cytometry. B. Combined AmFRET values for three technical replicates of the specified Wtf^poison-mEos proteins and monomer-mEos (negative control). X^P represents the WtfX^poison proteins tested here. The median is indicated with a solid line and the bars represent the interquartile range. For easier comparison, the values were normalized so that Wtf4^poison had a median of 1 in each experiment. The data shown here do not include outliers. See S2 Data for the complete dataset and p-values. Statistical significance: *p<0.01; t-tests with Bonferroni correction. C. A spot assay of cells serially diluted and plated on SC-TRP-URA and SC Gal-TRP-URA plates. Each strain carries an empty [TRP1] plasmid, and either an empty [URA3] plasmid (EV) or the indicated wtf^poison-mEos allele under the control of a galactose-inducible promoter. The plates were grown at 30 ℃ for 3 days. D. Representative images of the same strains depicted in ℃ were induced in galactose media for 4 hours at 30 ℃ to express the indicated mEos-tagged proteins. The images are not at the same brightness and contrast settings to clearly show localization of tagged proteins. Yellow arrows indicate endoplasmic reticulum-like localization. TL is transmitted light, and the scale bar is 4 µm.

https://doi.org/10.1371/journal.pgen.1011534.g002

We first tagged four toxic (S. cryophilus Wtf1^poison, S. osmophilus Wtf41^poison, and S. octosporus Wtf25^poison and Wtf61^poison) and one non-toxic (S. osmophilus Wtf19^poison) Wtf^poison protein at the C-terminus with mEos. The tagged proteins retained their toxic/non-toxic phenotypes when expressed in S. cerevisiae using a galactose-inducible expression system (Fig 2C) [19]. The assayed proteins all share less than 20% amino acid identity with Wtf4^poison, which we use as a positive control (S1B Fig). We found that all five newly tested Wtf^poison-mEos proteins, including the nontoxic Wtf19^poison-mEos, exhibited self-assembly (Fig 2B). It is important to note, however, that the data reported for Wtf25^poison-mEos are limited and we could not reliably assay that protein via DAmFRET. This is because only living cells are considered in DAmFRET and Wtf25^poison-mEos kills cells rapidly at low expression (e.g., they die faster than cells expressing Wtf4^poison).

We also assayed DAmFRET in partial deletion alleles of wtf4^poison-mEos (S2 and S3 Figs). Some of these alleles are described in more detail below, but for this section, we focused on exploring the potential connection between assembly and toxicity. The two toxic mutant Wtf4^poison proteins tested, Wtf4-ex5Δ^poison-mEos (S2A and S2D-F Fig) and Wtf4Δ^20-poison-mEos (S3A-E Fig) both exhibited self-assembly. Of these two proteins, Wtf4Δ^20-poison-mEos is more like wild-type Wtf4^poison-mEos, in both toxicity and degree of assembly (S2D-F and S3B-E Figs). Most non-toxic Wtf4 mutant proteins (Wtf4-TMD1Δ^poison-mEos, Wtf4-ex3Δ^poison-mEos, Wtf4-ex4Δ^poison-mEos, Wtf4-ex6Δ^poison-mEos and Wtf4-consΔ^poison-mEos) exhibited reduced assembly compared to the wild-type (S2A-E and S3A-D Figs). We did, however, find one exceptional mutant protein, Wtf4Δ^10-poison-mEos. This protein was not toxic but showed similar assembly to the wild-type protein (S3A-D Fig). We could not reliably assay the assembly of some alleles (Wtf4-TMD2Δ^poison-mEos, Wtf4-TMD6Δ^poison-mEos and Wtf4-ex2Δ^poison-mEos) due to low fluorescence signal (S2A, S2B and S4 Figs).

These results show that Wtf^poison self-assembly is broadly conserved (over 100 million years) and can be supported by a wide range of amino acid sequences (e.g., proteins sharing <20% amino acid identity, S1B Fig). Given the broad conservation of self-assembly, that all the toxic alleles exhibited assembly, and that mutant alleles with reduced toxicity also exhibited reduced assembly, our results strongly support an association between self-assembly and toxicity. Importantly, our results also demonstrate that self-assembly alone is insufficient for Wtf^poison toxicity, as we found several Wtf^poison proteins (Wtf19-mEos and Wtf4Δ^10-poison-mEos) that assemble but are not toxic.

Deletion alleles demonstrate that both conserved and non-conserved regions contribute to Wtf4^poison self-assembly and toxicity

Our deletion mutants in wtf4^poison encompass both non-conserved and conserved sequences allowing us to test the functional importance of both sequence types (Fig 1; [19–21]). Within S. pombe Wtf proteins, there is one well conserved 29 base pair segment found at the beginning of exon 3 (Fig 1B; [20]). The Wtf4-consΔ^poison-mEos protein lacks this domain and has disrupted self-assembly and is non-toxic, indicating the conserved sequence has a functional role in both Wtf4^poison assembly and toxicity (S2C-E Fig).

We similarly found that several regions that are poorly conserved in Wtf proteins are functionally important in Wtf4^poison. For example, exon 4 is not conserved within functional wtf genes, even within S. pombe [19,20,26]. Still, our results demonstrate exon 4 is functionally important in wtf4 as the Wtf4-ex4Δ^poison-mEos protein has disrupted self-assembly and is not toxic (S2A, S2D and S2E Fig). An additional variable feature of wtf genes is the number of predicted transmembrane domains, which varies between four and eight within functional Wtf proteins (S2 Table). The wtf4 gene has 6 predicted transmembrane domains and we assayed individual deletions of three (Wtf4-TMD1Δ^poison-mEos, Wtf4-TMD2Δ^poison-mEos and Wtf4-TMD6Δ^poison-mEos). All three transmembrane deletions disrupted toxicity, indicating they are also functionally important (S2B, S2D and S2E Fig).

An additional set of deletion mutants we queried in a poorly conserved region followed up on an observation by Hu et al [17] in one of the inaugural papers describing wtf drivers. In that work, Hu et al [17] described 10 amino acid C-terminal truncations of two wtf genes known as cw9 and cw27. These mutant alleles both exhibited disrupted Wtf^poison, but not Wtf^antidote activity, implicating the C-terminus in Wtf^poison function. These proteins share 62% amino acid identity in the C-terminal exon 6 with each other and, at most, 70% with Wtf4. We tested if a similar 10 amino acid truncation of wtf4 would specifically disrupt Wtf4^poison activity. This mutation gave us the exceptional wtf4-Δ¹⁰ allele described above that encodes a non-toxic Wtf4Δ^10-poison, that surprisingly exhibits wild-type self-assembly (S3A-D Fig). To expand our analyses beyond wtf4, we also made 10 amino acid C-terminal truncations of S. cryophilus Wtf1, S. osmophilus Wtf41, S. octosporus Wtf61 and Wtf25 poison. These proteins are all significantly smaller than Wtf4 (S1C Fig). Surprisingly, we found that these additional truncated proteins all retained toxicity (S3F and S3G Fig) and could be rescued by the corresponding antidotes (S5 Fig) indicating the C-termini are not universally important for Wtf^poison function.

We extended our deletion analyses of the C-terminus of Wtf4 poison and were surprised to find that deleting ten more amino acids than the ten missing in the non-toxic Wtf4Δ^10-poison protein restored toxicity. Specifically, the Wtf4Δ^20-poison protein exhibited robust self-assembly and near wild-type levels of toxicity (S3A-D Fig). A larger deletion of 29 amino acids, Wtf4ex6Δ^poison, showed reduced self-assembly and was not toxic (S3A-D Fig).

Together, our results show that even non-conserved sequences can have context-dependent importance within a Wtf^poison protein. Specifically, a feature (e.g., the residues encoded in exon 4, or the last 10 amino acids of Wtf4^poison) can be functionally important in one Wtf^poison protein but be missing or dispensable in another. This, combined with the lack of conservation within Wtf proteins, suggests that the contextually important amino acids (like the last 10 of Wtf4^poison) do not have a specific function, but can rather contribute to the overall properties of the protein. Changes in these contextually important regions can be complemented by changes elsewhere in the protein.

Intracellular localization of mutant Wtf^poison proteins is correlated with toxicity

We identified self-assembly as one feature shared by functional Wtf^poison proteins and suspected that intracellular localization could be another, since we previously found that the toxic S. octosporus Wtf25^poison-mCherry exhibited similar localization to that of toxic Wtf4^poison [19]. Specifically, both poisons show small puncta broadly distributed in the cytoplasm of S. cerevisiae cells, with minor localization to what appears to be the endoplasmic reticulum (ER; [15,19]). All functionally confirmed Wtf driver proteins have multiple predicted transmembrane domains (S2 and S3 Tables) and hence, this localization pattern may reflect the poison being trafficked from the ER through the secretory pathway, through the Golgi and trans-Golgi network [25].

To test the hypothesis that a link exists between the intracellular distribution and toxicity of Wtf^poison proteins, we imaged the wild-type Wtf^poison-mEos proteins described above in S. cerevisiae cells. The localization of the S. octosporus Wtf25^poison-mEos was similar to the previously described Wtf25^poison-mCherry described above (Fig 2D). The additional toxic Wtf^poison proteins from S. octosporus, S. osmophilus and S. cryophilus, also exhibited dispersed puncta, like the toxic Wtf4^poison-mEos control cells (Fig 2D). The nontoxic S. osmophilus Wtf19^poison-mEos, however, showed a distinct localization pattern with strong signal enrichment in what appears to be the ER (Fig 2D). These data are consistent with our hypothesis that there is a link between intracellular distribution and toxicity with toxic Wtf^poison proteins exhibiting distributed puncta.

To further test our hypothesis, we also assayed the localization of the mutant Wtf4^poison proteins mentioned above (S2F and S3E Figs). The most toxic mutant protein, Wtf4Δ^20poison-mEos, localized in distributed puncta like wild-type (S3A-C and S3E Fig). The less toxic Wtf4-ex5Δ^poison-mEos protein also formed some distributed puncta, but also exhibited more ER-like localization than wild-type (S2A, S2D and S2F Fig). Most of the non-toxic mutant Wtf4^poison proteins (Wtf4-TMD1Δ^poison-mEos, Wtf4-TMD2Δ^poison-mEos, Wtf4-ex2Δ^poison-mEos, Wtf4-ex3Δ^poison-mEos, Wtf4-ex4Δ^poison-mEos, and Wtf4-consΔ^poison-mEos) were less dispersed and exhibited a largely ER-like localization, similar to the non-toxic S. osmophilus Wtf19^poison-mEos (Fig 2C-D and S2A-D and S2F Fig). One additional non-toxic mutant, Wtf4-TMD6Δ^poison-mEos, localized to the vacuole, reminiscent of Wtf^antidote protein localization (S2B, S2D and S2F Fig). The non-toxic Wtf4-ex6Δ^poison-mEos had a unique localization pattern that combined a diffuse distributed signal with some larger protein assemblies (S3A-C and S3E Fig). Finally, the non-toxic Wtf4Δ^10-poison-mEos localization was indistinguishable from wild-type Wtf4^poison-mEos (S3A-C and S3E Fig).

In summary, all the toxic Wtf^poison proteins show a distributed puncta localization pattern, like Wtf4^poison. All except one non-toxic Wtf^poison protein showed a distinct localization pattern from Wtf4^poison, most often with enhanced ER-like localization. These results parallel our DAmFRET analyses where all toxic Wtf^poison proteins assemble and all the nontoxic proteins, except one, show reduced or lack of assembly. In both cases, the exceptional protein is Wtf4Δ^10-poison, which is non-toxic, but shows wild-type assembly and localization in cells. Despite the exception, our results show that Wtf^poison self-assembly (assayed by DAmFRET) combined with cellular localization patterns are good predictors of Wtf^poison protein toxicity (Fig 7A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Increasing poison assembly with tags suppresses Wtf^poison toxicity.

A. A cartoon of the constructs used in this experiment (C-D). The mEosNB-FTH1 construct was integrated into the genome and is under the control of a doxycycline-repressible promoter. mEosNB is a nanobody that binds mEos. The wtf^poison alleles are carried on a [URA3] plasmid and are under the control of a galactose-inducible promoter. B. Cartoon of the constructs expressed on each medium used: SC-Ura, SC Galactose-Ura + 40 mg/L doxycycline, and SC Galactose-Ura. FTH1 is a 24-mer but is depicted as an 8-mer core. C. Representative images of cells induced with galactose media (with or without 40mg/L Doxycycline) for 4 hours at 30 ℃ to express the indicated Wtf^poisonmEos proteins. The cells induced without Doxycycline (-Dox) also express mEosNB-FTH1. This strain background exhibits high autofluorescence, so we spectrally unmixed the signal to remove autofluorescence (See Methods). The images are not at the same brightness and contrast settings to clearly show localization of tagged proteins. TL is transmitted light, and the scale bar is 4 µm. D. A spot assay of cells carrying the constructs illustrated in A-C serially diluted on SC-Ura, SC Galactose-Ura + 40mg/L Doxycycline and SC Galactose-Ura plates and grown at 30 ℃ for 3 days. Each strain carries either an empty [URA3] plasmid (EV) or the indicated wtf^poison-mEos allele. These media induce the expression of mEosNB-FTH1 (i), the indicated Wtf^poison-mEos protein (ii), or both (iii), respectively. The horizontal break in the image of each plate is due to rearrangements of the images to facilitate easy comparison. All strains within a panel were grown on the same plates (i.e., one SC-TRP-URA or SC Gal-TRP-URA or SC Gal-TRP-URA + 40mg/L Doxycycline plate).

https://doi.org/10.1371/journal.pgen.1011534.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Reducing Wtf4^poison assembly with NT* tag increases toxicity and affects its rescue by Wtf4^antidote.

A. Cartoon of alleles used in this experiment (B-F). The NT* tag has a general anti-aggregation property [32]. B. A spot assay of cells serially diluted and plated on SC-TRP-URA and SC Gal-TRP-URA plates and grown at 30 ℃ for 4 days. Each strain carries both a [URA3] and a [TRP1] plasmid. The plasmids are either empty (EV) or carry the indicated wtf4 alleles under the control of galactose-inducible promoters. The horizontal break in the image for each plate is due to a rearrangement of the image to facilitate easy comparison. All strains were grown on the same plates (i.e., one SC-TRP-URA or SC Gal-TRP-URA plate). C. AmFRET values for three technical replicates of the specified Wtf4^poison-mEos alleles and monomer-mEos (negative control). The median is indicated with a solid line and the bars represent the interquartile range. For easier comparison, the values were normalized so that Wtf4^poison had a median of 1 in each experiment. Since Wtf4-20^poison-mEos cells were very low in number, we pooled two biological replicates (n=6 technical replicates). The data shown here do not include outliers. See S2 Data for the complete dataset and p-values. Statistical significance: *p<0.025, t-tests with Bonferroni correction. D. A spot assay of cells serially diluted and plated on SC-TRP-URA and SC Gal-TRP-URA plates and grown at 30 ℃ for 3 days. Each strain carries both a [URA3] and a [TRP1] plasmid. The plasmids are either empty (EV) or carry the indicated wtf4 alleles under the control of galactose-inducible promoters. All strains were grown on the same plates (i.e., one SC-TRP-URA or SC Gal-TRP-URA plate). E. Representative images of the same strains depicted in D were induced with galactose media for 4 hours at 30 ℃ to express the indicated Wtf4^poison-mEos proteins and/or Wtf4^antidote-mCherry. The images are not at the same brightness and contrast settings to clearly show localization of tagged proteins. 4^P indicates Wtf4^poison, 4^A indicates Wtf4^antidote, TL is transmitted light, and the scale bar is 4 µm. F. Pearson’s Correlation between mCherry and mEos signal in cells from E expressing the specified proteins. N>100, *** p<0.001, t-test.

https://doi.org/10.1371/journal.pgen.1011534.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Effective neutralization of Wtf4^poison requires more than a physical connection to a Wtf^antidote.

A. Cartoon of constructs used in this experiment. S. kambucha wtf4^poison was tagged at the C-terminus with either mCherry or GBP-mCherry (GBP: GFP-binding protein). S. kambucha wtf4^antidote was tagged with mEos or GFP. S. octosporus wtf61^antidote was tagged with GFP. S. octosporus wtf61^poison was tagged with mCherry. B. Experimental set up and summary of the results shown in C and D. In a matching Wtf protein pair (top), poison-antidote interaction and rescue of poison toxicity is observed. In the mismatched pair (bottom), interaction between GFP and GBP results in a forced interaction between the poison and the antidote (shown in D). This interaction is insufficient to rescue the mismatched poison (shown in C). C. Spot assay of cells serially diluted and plated on SC-TRP-URA and SC Gal-TRP-URA plates and grown at 30 ℃ for 3 days. Each strain carries both a [URA3] and a [TRP1] plasmid. The plasmids are either empty (EV) or carry the indicated wtf alleles under the control of galactose-inducible promoters. The horizontal break in the image of each plate is due to rearrangements of the images to facilitate easy comparison. All strains within a panel were grown on the same plates (i.e., one SC-TRP-URA or SC Gal-TRP-URA plate). D. Representative images of the same strains shown in C were induced in galactose for 4 hours at 30 ℃ to express the indicated Wtf proteins. E. Representative images of cells induced with galactose and 500nM β-estradiol for 4 hours at 30 ℃ to produce the indicated proteins. Rnq1-mCardinal marks the insoluble protein deposit (IPOD) that is associated with the vacuole [33–35]. In D-E, the images are not at the same brightness and contrast settings to clearly show localization of tagged proteins. 4^P indicates Wtf4^poison, 4^A indicates Wtf4^antidote, 61^P indicates Wtf61^poison, 61^A indicates Wtf61^antidote, TL is transmitted light, and the scale bar is 4 µm.

https://doi.org/10.1371/journal.pgen.1011534.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Modification of wtf4 exon 6 CDS repeats can disrupt antidote rescue.

A. Cartoon of two coding sequence repeat mutants of S. kambucha wtf4. B. Logo representing the amino acids encoded by the repeats found in exon 6 of S. pombe wtf genes from [20]. C. The amino acids encoded by the exon 6 repeats in S. kambucha wtf4 and the wtf4-rep2^sc and wtf4-rep2^A mutant alleles. D. Spot assay of cells serially diluted and plated on SC-TRP-URA and SC Gal-TRP-URA plates and grown at 30 ℃ for 3 days. Each strain carries both a [URA3] and a [TRP1] plasmid. The plasmids are either empty (EV) or carry the indicated wtf4 alleles under the control of galactose-inducible promoters. The horizontal breaks in the image of each plate are due to rearrangements of the images to facilitate easy comparison. All strains within a panel were grown on the same plates (i.e., one SC-TRP-URA or SC Gal-TRP-URA plate). E. Representative images the same strains depicted in D were induced in galactose for 4 hours at 30 ℃ to express the indicated Wtf proteins. The images are not at the same brightness and contrast settings to clearly show localization of tagged proteins. The arrows in the TL panels highlight vacuoles. 4^P indicates Wtf4^poison, 4^A indicates Wtf4^antidote, TL indicates transmitted light, and the scale bar is 4 µm. F. Pearson’s Correlation between mEos and mCherry signal in cells expressing the specified constructs from E. N>100, ***p<0.001, t-test.

https://doi.org/10.1371/journal.pgen.1011534.g006

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Analyses of Wtf proteins reveal additional functional constraints.

A. A dispersed Wtf^poison localization and poison-poison assembly can be observed across greatly diverged proteins and are associated with Wtf^poison toxicity. However, assembly and/or localization alone are insufficient to cause toxicity. In multiple examples of non-toxic proteins (an example is included in each panel), we observe either an endoplasmic reticulum-like localization or vacuolar localization. B. Modulating Wtf^poison assembly with exogenous tags affects its toxicity. While increasing Wtf^poison assembly via tethering to mEosNB-FTH1 suppresses toxicity, increasing solubility with NT* (or MBP in the case of S. kambucha Wtf4^poison) increases the toxicity. C. Multiple modalities of ineffective antidote rescue. Effective rescue of a Wtf^poison by the corresponding Wtf^antidote (e.g., Wtf4 proteins) is supported by specific co-assembly and vacuolar localization. Artificial tethering of a mismatched Wtf^poison and Wtf^antidote does not result in effective rescue. Disruption of poison-antidote co-localization (e.g., Wtf4-rep2^sc proteins) or vacuolar targeting of co-localized poison-antidote proteins (e.g., Wtf4-rep2^A proteins) also results in ineffective rescue.

https://doi.org/10.1371/journal.pgen.1011534.g007

Altering assembly properties of Wtf^poison proteins affects toxicity

To further test the model that the toxicity of Wtf^poison proteins is tied to their self-assembly, we sought to alter the assembly properties of Wtf4^poison using tags. We first aimed to increase the overall size of the Wtf4^poison assemblies. To do this, we employed a recently described tool to oligomerize mEos3-fused proteins in trans. Specifically, we expressed a fusion protein consisting of the human ferritin heavy chain protein (FTH1) and a nanobody that recognizes mEos (mEosNB) [36]. The FTH1 domain self assembles to form a 24-mer core and can form supramolecular clusters when fused to self-assembling proteins [37,38]. In our assay, we expressed Wtf4^poison-mEos from a galactose inducible promoter and the mEosNB-FTH1 from a doxycycline repressible promoter (Fig 3A and 3B). As expected, we found that Wtf4^poison-mEos formed fewer and larger puncta inside cells in the presence of mEosNB-FTH1, consistent with the mEosNB-FTH1 complexes bringing Wtf^poison-mEos assemblies into supramolecular clusters (i.e., -Dox panels, Fig 3C). Expression of the mEosNB-FTH1 alone had no effect on yeast growth (Fig 3D, panel (i)). While expression of the Wtf4^poison-mEos alone was toxic, co-expression of the mEosNB-FTH1 and Wtf4^poison-mEos markedly suppressed the toxicity of the Wtf4^poison-mEos (Fig 3D, compare panel (ii) to panel (iii)). We also observed the same suppression of toxicity with four additional toxic Wtf^poison-mEos proteins in the presence of the mEosNB-FTH1, indicating the effect was not specific to Wtf4^poison (Fig 3C and 3D). These results indicate that these supramolecular Wtf^poison assemblies are non-toxic. This effect could be due to altered intracellular localization and/or the reduced exposed surface area of the Wtf^poison assemblies.

We next attempted to increase the solubility Wtf4^poison assemblies using a modified N-terminal domain (NT*) tag derived from spidroins, a principal component of spider silk [32]. The NT* tag can decrease protein aggregation, which is the role of the native domain in spiders [39–43]. We added NT* to the N-terminus of the Wtf4^poison-mEos to generate NT*-Wtf4^poison-mEos (Fig 4A). We found that this protein had a novel combination of assembly and localization not represented in any of our other alleles. Specifically, the NT*-Wtf4^poison-mEos protein showed less self-assembly than Wtf4^poison-mEos (Fig 4C), but the NT* tag did not visibly alter the localization of the protein within cells (Fig 4E). Interestingly, the NT*-Wtf4^poison-mEos protein showed increased toxicity in cells (Fig 4B). To test if the effects of the NT* tag were specific to Wtf4^poison or more general, we also tagged S. octosporus Wtf25^poison with NT* to generate NT*-Wtf25^poison-mEos (S6A Fig). As with NT*-Wtf4^poison-mEos, we observed unaltered localization and increased toxicity of the NT*-Wtf25^poison-mEos protein, relative to wild-type Wtf25^poison-mEos (S6B and S6D Fig). We could not, however, quantify NT*-Wtf25^poison-mEos assembly by DAmFRET as the high toxicity prevented us from obtaining sufficient viable cells with mEos fluorescence (S4E Fig).

We also assayed how another solubility tag, the E. coli Maltose Binding Protein (MBP), affected Wtf^poison toxicity [44–46]. The MBP-tagged Wtf4^poison (MBP-Wtf4^poison-mEos) had the same phenotype as the NT*-tagged Wtf4^poison protein: decreased assembly, increased toxicity, unaltered localization, relative to the wild-type protein (S7 Fig). The phenotype of the MBP-tagged Wtf25^poison (MBP-Wtf25^poison-mEos) however, did not mirror that of the NT*-tagged protein. Instead, we found that the MBP-Wtf25^poison-mEos protein showed reduced, but still high toxicity. The MBP-Wtf25^poison-mEos protein also showed an altered, stronger endoplasmic reticulum-like localization, as compared to the wild-type protein (S7A-C Fig). This protein thus adds to those described above in which we see more ER-like localization associated with reduced Wtf^poison toxicity. The MBP-Wtf25^poison-mEos protein did exhibit self-assembly, but we could not compare it the wild-type protein as high toxicity limited the viable cells we could assay via DAmFRET (S4F and S7D Figs).

The bulk of our experiments are consistent with a model where the assembly properties and the distribution of Wtf^poison proteins within cells affects toxicity, with distributed, punctate assemblies exhibiting more toxicity. Increasing Wtf^poison solubility (while maintaining assembly), increases toxicity, unless localization is disrupted. Forcing Wtf^poison proteins into large, localized assemblies, suppresses toxicity. The non-toxic Wtf4Δ^10-poison protein that forms distributed assemblies, however, undermines this simple model or at least indicates that there are additional unidentified features that are essential for toxicity.

An additional possibility that we considered was that the expression levels of the Wtf^poison proteins likely contributes to the phenotypes we observed, despite expressing all alleles from a common vector backbone using a common promoter. To assay expression levels, we quantified fluorescence of the mEos tags shared by all alleles. We did not use western blots as that approach is challenging to apply to Wtf^poison proteins as they are toxic and highly hydrophobic [15,24]. Instead, we looked at the acceptor (red form of mEos) fluorescence in live cells analyzed in the DAmFRET experiments (S4 Fig; Data from Figs 2B, 4C, S2E and S3D). As expected, based on other studies [47–50], we observed significant fluorescent protein signal heterogeneity within cells expressing the monomer mEos control and each Wtf^poison protein tested (S4 Fig). We did not, however, find increased fluorescent signal in cells expressing the most toxic proteins (as determined by growth assays). On the contrary, we tended to observe less signal in the cells expressing the most toxic proteins (e.g., Wtf4^poison and NT*-Wtf4^poison). These data support a model in which cells expressing the higher levels of toxic proteins are more likely to be removed by death and are not quantified. As mentioned above, however, we also observed low signal in the cells expressing the non-toxic Wtf4-ex2Δ^poison, Wtf4-TMD2Δ^poison, and Wtf4-TMD6Δ^poison alleles (S4 Fig). The low levels of Wtf4-TMD6Δ^poison could be due to its degradation in the vacuole (S2F Fig). For the other two alleles, it is formally possible the low levels of these proteins in cells could contribute to their lack of toxicity.

Limited modularity of the Wtf^antidote-specific domain

We next wanted to test which protein features affect Wtf4^antidote function (Fig 1A and 1B). As described above, the Wtf4^antidote shares the resides encoded by exons 2–4 with the Wtf4^poison but has an additional N-terminal domain encoded by exon 1. We hypothesized that the amino acids encoded by exon 1 would be insufficient for function and that exons 2–6 would be required for protein self-assembly as they comprise the Wtf4^poison protein, which self-assembles [15]. As expected, we found that a protein consisting of only the exon 1-encoded residues linked to a mEos tag (Wtf4 Exon1-mEos) could not self-assemble (S8A and S8B Fig). The Wtf4 Exon1-mEos protein was also not trafficked to the vacuole, despite this protein harboring two PY motifs that promote ubiquitin-mediated trafficking of wild-type Wtf^antidote proteins [25] (S8E Fig). Finally, the Wtf4 Exon1-mEos protein could not rescue Wtf4^poison toxicity, which parallels recent results using distinct Wtf proteins (S8C and S8F Fig) [25]. These results demonstrate that the antidote-specific domain encoded by wtf4 exon 1 is insufficient for antidote function.

To test if the exon 1 encoded domain was modular, we generated two mutants wtf4^poison-ex1 and wtf4^poison-ex1^int. The wtf4^poison-ex1 allele moves the antidote-specific domain moved to the C-terminus of the protein (S8A Fig). The wtf4^poison-ex1^int allele moved exon 1 a more central region of the protein (beginning of exon 4). This location is between the last two predicted transmembrane domains and is not predicted to disrupt them (Fig 1B). We found that mEos tagged versions of both Wtf4^poison-ex1 and Wtf4^poison-ex1^int proteins were both non-toxic and trafficked to the vacuole, suggesting they retained at least some antidote functionality (S8D and S8F Fig). Only the Wtf4^poison-ex1, however, could neutralize the toxicity of Wtf4^poison-mCherry (S8D and S8F Fig). These results demonstrate that the domain encoded by exon 1 is at least partially modular, but that a central location within the polypeptide can disrupt the Wtf^antidote protein’s ability to neutralize a Wtf^poison. We speculate this could be because extensive, continuous, amino acid identity facilitates Wtf^poison and Wtf^antidote co-assembly.

Wtf^antidote requires more than physical linkage to effectively traffic Wtf^poison to the vacuole

We next wanted to determine if co-assembly serves merely to link Wtf^poison and Wtf^antidote proteins to allow for co-trafficking, or if co-assembly serves a more nuanced role in neutralizing Wtf^poison toxicity. To test this, we generated Wtf^poison and Wtf^antidote proteins that were physically linked, but not co-assembled, by artificially tethering Wtf^poison proteins to diverged Wtf^antidote proteins, with which they cannot co-assemble, using a combination of GFP and GFP-binding protein (GBP) tags (Figs 5A, 5B, S9A and S9B).

We first tested if tethering Wtf4^poison to S. octosporus Wtf61^antidote could neutralize the toxic Wtf4^poison. We generated Wtf4^poison-GBP-mCherry and Wtf61^antidote-GFP proteins and found the tags did not disrupt protein function. Specifically, the Wtf4^poison-GBP-mCherry protein was toxic, and could be rescued by Wtf4^antidote proteins, including Wtf4^antidote-GFP. Similarly, the Wtf61^antidote-GFP protein was able to rescue Wtf61^poison-mCherry (Fig 5C). However, Wtf61^antidote-GFP was not able to efficiently traffic Wtf4^poison-GBP-mCherry to the vacuole or neutralize its toxicity (Fig 5C and 5D). This failure to rescue is not a failure of the GFP-GBP interaction to link the proteins as we found that the Wtf61^antidote-GFP and Wtf4^poison-GBP-mCherry proteins largely co-localized, which they did not do in the absence of the GBP tag (Fig 5D). Moreover, the localization of the tethered Wtf61^antidote-GFP-Wtf4^poison-GBP-mCherry changed relative to the individual proteins: the Wtf61^antidote-GFP no longer trafficked to the vacuole and Wtf4^poison-GBP-mCherry was less distributed in cells. Effective GFP-GBP linkage is the most parsimonious explanation of these observations. To test if our results were generalizable, we also analyzed a widely diverged, independent pair of Wtf proteins (S. octosporus Wtf25^poison and S. cryophilus Wtf1^antidote). Our results with this protein pair mirrored those described above suggesting that specific antidote-poison co-assembly generally promotes proper antidote function (S9 Fig).

Given that the tethered Wtf61^antidote-GFP and Wtf4^poison-GBP-mCherry were often adjacent to vacuoles and not totally distributed in cells, as occurs when trafficking of Wtf proteins is disrupted [15], we suspected that the vacuolar targeting was still occurring, but blocked at a late step in the process (e.g., vacuole entry). Consistent with this notion, the assemblies often co-localized with the Rnq-1mCardinal protein, which is a prion protein that marks the insoluble protein deposit (IPOD) [33–35] (Fig 5E). This result was surprising to us because with a different expression system (β-estradiol induction) a considerable amount of the Wtf4^poison/Wtf4^antidote co-assemblies accumulate at the IPOD are not toxic to cells [15]. It is unclear why GFP/GBP mediated assemblies, but not Wtf/Wtf mediated co-assemblies, would be toxic at the IPOD, but we speculate about this in the discussion.

Beyond the GFP/GBP tethering experiments, we made additional observations suggesting that a specific nature of co-assembly with Wtf^poison is required for Wtf^antidote function. When characterizing the toxic NT*-Wtf4^poison protein (with the NT* domain that increases solubility), we found that this protein was not effectively rescued by wild-type Wtf4^antidote-mCherry (Fig 4D). This drastic reduction in rescue (relative to that observed with the wild-type Wtf4^poison) is striking given that the colocalization of NT*-Wtf4^poison-mEos with Wtf4^antidote-mCherry appears only mildly reduced relative to that observed between wild-type Wtf4 proteins (Fig 4E and 4F). In addition, the localization of the NT*-Wtf4^poison-mEos is altered in the presence of Wtf4^antidote-mCherry (it becomes less distributed compared to wild type Wtf4 proteins; Fig 4E). These observations suggest the NT*-Wtf4^poison-mEos and Wtf4^antidote-mCherry proteins interact, but are not efficiently trafficked into the vacuole (Fig 4E). This is analogous to the GFP/GBP linked, but unassembled, Wtf^poison and Wtf^antidote pairs described above. The effect of the NT* tag on Wtf^poison/Wtf^antidote compatibility was not, however, universal. We did not observe strong disruption of poison/antidote compatibility caused by the NT* tag on S. octosporus Wtf25^poison (NT*-Wtf25^poison allele, S6 Fig), indicating that factors affecting poison/antidote compatibility can be context dependent.

Together, our results demonstrate that a physical linkage is insufficient to ensure efficient neutralization of a Wtf^poison protein’s toxicity by a Wtf^antidote. Instead, specific co-assembly of the proteins likely both links compatible Wtf^poison and Wtf^antidote proteins and facilitates their effective co-trafficking into the vacuole.

C-terminal region supports Wtf4^antdiote function

As introduced above, Hu et al [17] previously described mutant alleles of two S. pombe wtf genes lacking the codons for the last 10 amino acids. Those mutants maintained Wtf^antidote activity [17]. To test if the C-terminal amino acids were generally dispensable for antidote function, we made wtf4^antidote truncation alleles. We found that a 10 amino acid truncation of Wtf4, (Wtf4Δ^10-antidote-mCherry) was slightly toxic to cells, as compared to the Wtf4^antidote-mCherry and empty vector controls (S10 Fig). A 20 amino acid truncation of the, Wtf4Δ^20-antidote-mCherry, showed even more toxicity, although it was still considerably less than the toxicity observed with the Wtf4^poison-mCherry. A 29 amino acid truncation, Wtf4-ex6Δ^antidote-mCherry, exhibited no toxicity (S10 Fig). These results suggest that the C-terminal 20 amino acids of Wtf4^antidote play a role in limiting the toxicity of the Wtf4^antidote protein.

We next tested if the truncation Wtf4^antidote proteins could rescue the toxicity of a wild-type Wtf4^poison-mEos protein. Neither the Wtf4Δ^10-antidote-mCherry or Wtf4-ex6Δ^antidote-mCherry proteins appreciably neutralized Wtf4^poison-mEos toxicity (S10A and S10B Fig). The Wtf4Δ^20-antidote-mCherry protein, however, rescued growth of cells expressing Wtf4^poison-mEos to a level comparable to that observed in the cells expressing only Wtf4Δ^20-antidote-mCherry. So, despite the slight toxicity of the Wtf4Δ^20-antidote-mCherry protein, it still retained some antidote function.

Interestingly, we were surprised to see that the other truncated antidotes Wtf4Δ^10-antidote-mCherry and Wtf4-ex6Δ^antidote-mCherry could also partially rescue the toxicity of Wtf4Δ^20-poison-mEos poison, although it was much less rescue compared to that of the wild-type Wtf4^antidote-mCherry protein (S10A and S10B Fig). This demonstrates that these truncated proteins retain some functionality.

Our results indicate that the C-terminus supports Wtf4^antidote function, but that some functionality remains even in the absence of the last 29 amino acids. When considered in combination with the results of Hu et al [17], our combined results indicate that the importance of the C-terminus is context dependent. Like our Wtf^poison results discussed earlier, these observations, combined with the lack of conservation of Wtf proteins, suggests that Wtf^antidote function (outside of the conserved PY motifs) relies on overall properties of the protein (e.g., lack of toxicity and ability to co-assemble with a matching Wtf^poison), rather than any functional domain.

Self-killing alleles that encode a functional Wtf^poison with an incompatible Wtf^antidote

Because Wtf proteins are encoded on largely overlapping coding sequences, a change in the shared coding sequence can simultaneously create a novel Wtf^poison protein and a matching novel Wtf^antidote protein. Our engineered mutant alleles offered the opportunity to explore if matching proteins are always compatible (i.e., if a novel toxic Wtf^poison protein is always neutralized by its corresponding Wtf^antidote protein).

We therefore explored poison and antidote compatibility more broadly within the alleles we generated. In many cases, the Wtf^antidote proteins were able to rescue their matching Wtf^poison alleles (S1 Table). One class of mutants, however, created toxic Wtf^poison proteins that were not rescued by the matching Wtf^antidote. In wtf4, these mutants changed a region of exon 6 that encodes a 7 amino acid repeat found in variable numbers (0–84 base pairs of repeat sequence) within S. pombe wtf genes (Fig 6A and 6B) [20]. The repeat region in wtf4 encodes one complete repeat plus three additional amino acids of the repeat (Fig 6C). We found that mutating this region in Wtf4 by either scrambling the amino acid order (Wtf4-rep2^sc) or by replacing the ten amino acids with alanine (Wtf4-rep2^A allele) generated toxic Wtf^poison proteins that were not neutralized by their matching Wtf^antidote proteins (Fig 6D). The Wtf4-rep2^sc-antidote-mCherry and Wtf4-rep2^A-antidote-mCherry proteins were both trafficked to the vacuole when expressed alone but did not effectively co-traffic their corresponding Wtf^poison proteins (Fig 6E). The Wtf4-rep2^{sc antidote}-mCherry and Wtf4-rep2^{sc poison}-mEos proteins showed decreased co-localization, relative to wild-type proteins, suggesting the proteins had disrupted co-assembly (Fig 6E and 6F). Alternatively, the Wtf4-rep2^{A antidote}-mCherry and Wtf4-rep2^{A poison}-mEos proteins colocalized (Fig 6F) but remained more distributed within cells relative to wild-type (Fig 6E). This distributed localization is a change from the vacuolar localization of Wtf4-rep2^{A antidote}-mCherry alone, further supporting that the Wtf4-rep2^{A antidote}-mCherry and Wtf4-rep2^{A poison}-mEos are interacting. Therefore, the incompatibility of the Wtf4-rep2^A proteins adds additional support to our earlier conclusion that a particular form of association is required between Wtf^antidote and Wtf^poison proteins to ensure the poison is effectively trafficked to the vacuole and neutralized.

We also assayed the effects of deleting the repeat region of wtf4 exon 6 alone (wtf4-rep2Δ allele), or in combination with deleting another repetitive region found in exon 3 (wtf4-rep1-2Δ allele; S11A Fig) [20]. We found that the Wtf4-rep2Δ^poison-mEos protein is toxic but is only partially rescued by Wtf4-rep2Δ^antidote-mCherry, relative to the rescue observed between wild-type Wtf4 poison and antidote proteins (S11D Fig). We found that the two Wtf4-rep2Δ proteins exhibit decreased colocalization, relative to wild-type proteins, suggesting the limited rescue is due to disrupted poison-antidote interaction (S11E and S11F Fig). Interestingly, we found that the defect in poison-antidote compatibility conferred by the deletion of the exon 6 repeats in the wtf4-rep2Δ allele is partially suppressed by also deleting the repetitive region found in exon 3. Specifically, the proteins encoded by the wtf4-rep1-2Δ allele, with both regions deleted, have near wild-type phenotypes (S11 Fig). We observed no defects in poison and antidote compatibility in the proteins encoded by an allele (wtf4-rep1Δ) lacking only the repetitive region in exon 3 (S11A, S11B, S11D-F Fig).

Finally, we also tested if the novel mutations we made in this study in the exon 3 or C-terminal repeats affected compatibility with wild-type Wtf proteins. In all cases tested, the repeat mutant Wtf^poison proteins were not neutralized by their wild-type Wtf^antidote counterparts and vice versa (S12 Fig). For example, Wtf4-rep1Δ^poison-mEos is not neutralized by Wtf4^antidote-mCherry and Wtf4^poison-mEos is not neutralized by Wtf4-rep1Δ^antidote-mCherry (S12B Fig).

The repeats in exon 6, but not in exon 3, are broadly conserved in the wtf gene family [19]. We also mutated the homologous region in S. octosporus wtf25 and found analogous phenotypes as those described above in wtf4 (S13 Fig). Like wtf4, S. octosporus wtf25 also encodes 10 amino acids in this C-terminal region (S13A-C Fig). We generated the S. octosporus wtf25-rep2^Sk allele by swapping the endogenous codons for those of wtf4. We found that this allele encoded an incompatible Wtf25-rep2^Sk-poison-mCherry and Wtf25-rep2^Sk-antidote-mEos pair (S13D Fig). Surprisingly, the individual Wtf25-rep2^Sk proteins were compatible with Wtf25 proteins (i.e., Wtf25-rep2^Sk-poison-mCherry was rescued by Wtf25^antidote-mEos and vice versa; S13D Fig). We also deleted the repeat region of S. octosporus wtf25 (to generate the wtf25-rep2Δ allele) and again found phenotypes similar to those observed in the analogous wtf4 mutant (wtf4-rep2Δ) (S14 Fig). Specifically, the Wtf25-1^antidote-mCherry showed reduced rescue of the Wtf25-rep2Δ^poison-mEos toxicity, relative to the wild-type Wtf25 protein pair (S14C Fig).

Altogether, our results further support a critical role for the repeats in Wtf^poison and Wtf^antidote compatibility and reveal that mutants in this domain can encode Wtf^poison proteins not neutralized by the matching Wtf^antidote proteins. Such alleles are important constraints on wtf gene evolution as they would contribute to infertility via self-killing.

Protein assembly plays conserved roles in Wtf protein function

The mutant analyses provided in this work expands and supports a working model in which 1) Wtf^poison toxicity is tied to the homotypic assembly of the proteins [15], 2) Wtf^antidote proteins are ubiquitinated and trafficked to the vacuole [15,25] and, 3) Wtf^antidote proteins co-assemble with their matching Wtf^poison proteins and co-traffic them to the vacuole [15,27]. Recent work has established that ubiquitin-mediated Wtf^antidote trafficking and co-trafficking with their corresponding Wtf^poison proteins are conserved within the extremely diverse gene family [25]. It was unclear, however, if homotypic protein assembly is a conserved feature of Wtf^poison or Wtf^antidote protein function.

The results of this study support the model that homotypic protein assembly plays critical roles in the function of Wtf^poison proteins. First, we observed that the ability to self-assemble was conserved amongst functional (i.e., toxic) wild-type Wtf^poison proteins from four Schizosaccharomyces species (Fig 2). Given that the functional proteins share as little as 20% amino acid identity, these observations likely reflect conserved functional importance of self-assembly. Our mutant analyses of Wtf4^poison proteins provided additional support for this model in that all toxic mutant proteins self-assembled (S2 and S3 Figs). Furthermore, we found that the Wtf^poison toxicity could be modulated by altering the assembly properties of the protein with tags (Figs 3, 4, S6 and S7). Together, our data suggest that Wtf^poison toxicity is tied to protein assembly with distributed, small assemblies showing greater toxicity than localized, larger assemblies (Figs 2–4, 7, S2, S3, S6 and S7). The recent work of Zheng et al. [25] suggests the distributed assemblies may represent localization to the trans-Golgi network.

Still, it is important to note that the Wtf4Δ^{10 poison} allele was non-toxic, despite assembling into small, distributed assemblies indistinguishable from those generated by the wild-type protein (S3 Fig). This exceptional protein highlights that even if self-assembly is critical for Wtf^poison toxicity, it is not the only factor required. In addition, this allele offers an opportunity for future work to explore features that distinguish toxic from nontoxic protein assemblies.

Our results also support an expanded, more nuanced role for homotypic protein assembly and trafficking in Wtf^antidote function. First, we found that the antidote-specific domain that contains the PY motifs is insufficient to promote vacuole trafficking, suggesting some other features of Wtf proteins are also required (S8 Fig). We posit protein self-assembly could contribute to this function. For Wtf^antidote neutralization of a Wtf^poison, we initially assumed that co-assembly of Wtf^antidote with Wtf^poison proteins served only to physically link the proteins, thus enabling the Wtf^antdiote to traffic the Wtf^poison to the vacuole [15]. We found, however, that physical linkage between a Wtf^poison and Wtf^antidote can be insufficient to ensure their co-trafficking to the vacuole and neutralization of the Wtf^poison’s toxicity. We observed this insufficiency in experiments linking two distinct pairs of non-matching Wtf^poison and Wtf^antidote proteins with GFP-GBP tags (Figs 5 and S9). In addition, we found that the wtf4-rep2^A and NT*-wtf4 alleles encode protein pairs in which poison-antidote interaction appeared largely intact, but trafficking into the vacuole and poison neutralization was disrupted (Figs 4 and 6). These experiments suggest that efficient co-trafficking of Wtf^poison/ Wtf^antidote protein assemblies requires a particular conformation or strong affinity between the interacting proteins not replicated in the ineffective Wtf^poison/Wtf^antidote combinations mentioned above (Fig 7).

Moreover, our experiments surprisingly revealed that trafficking toxic Wtf assemblies to the proper destination may not always be sufficient to ensure their neutralization. In previous work using a different induction system than that employed here, we found that much of the trafficked Wtf4^poison/Wtf4^antidote assemblies accumulated at the insoluble protein deposit (IPOD), in addition to the vacuole [15] and the cells were viable. Similar, but generally smaller, assemblies can be seen accumulating outside of the vacuole (likely the IPOD) in many of the viable cells we image expressing Wtf^antidote proteins or compatible Wtf^poison/Wtf^antidote assemblies using the GAL induction system employed in this study (e.g., Fig 6E panels expressing Wtf4^antidote and Wtf4^antidote/Wtf4^poison). Similarly, Zheng et al [25] found that vacuole localization was not essential for Wtf^poison neutralization as they observed that trafficking to the endosome can be sufficient. Here, however, we observed that Wtf4^poison-GBP/Wtf61^antidote-GFP assemblies (and likely the Wtf25^poison-GBP/Wtf1^antidote-GFP assemblies) were trafficked to the IPOD, but were still toxic (Figs 5, 7 and S9). These results suggest that factors beyond localization, perhaps assembly conformation, can also affect the toxicity of Wtf proteins.

Commonalities between Wtf proteins and other self-assembling proteins

The Wtf proteins share broad parallels with other nonhomologous proteins that form assemblies. For example, a sequence-independent common oligomeric property may underly the toxicity of unrelated amyloid proteins [51]. While Wtf^poison proteins are related to each other, their sequences are extremely diverged. We propose a common feature of their assembled forms is likely responsible for their shared toxicity. Another feature Wtf proteins share with several unrelated proteins is the capacity to form functional assemblies. Multiple amyloidogenic proteins form functional amyloids that perform diverse biological functions, including long term memory in flies [52], epigenetic inheritance in yeast [53] and biofilm formation in bacteria [54–56]. Some functionally aggregating amyloids have also shown to be toxic at intermediate stages of assembly, suggesting that protein toxicity could pose a risk in certain instances [57]. The co-expression of Wtf^antidote with the toxic Wtf^poison results in a change in its localization and the assembly properties of the Wtf^poison-Wtf^antidote complex. This suggests that protein-protein interaction within a pair of corresponding Wtf proteins is similar to functional aggregation. While the outcome of functional Wtf protein assembly is different (i.e., successful drive), the delicate interplay between toxic and non-toxic protein assemblies is similar to functional aggregating amyloids.

The use of E3 ubiquitin ligases to direct protein trafficking is an additional theme Wtf proteins share with other self-assembling proteins acting in diverse cell signaling processes, including immune response [58,59], prion disease [60] and other neurodegenerative disorders [61–67]. In multiple cases, ubiquitination of a key protein results in its aggregation, differential trafficking to specific intracellular locations or degradation, suggesting that this is a common mechanism for enabling downstream signaling processes. Additionally, a lack of ubiquitination by the ligase often results in toxic aggregates [60,63,66] or reduced functionality of the key protein [58,59], which are very reminiscent of the Wtf^poison/Wtf^antidote assemblies discussed above.

Rapidly evolving coding sequence repeats can affect Wtf^poison-Wtf^antidote compatibility

Most S. pombe wtf genes contain varying copy numbers of a sequence repeat in exon 3 [20]. The potential role of the exon 3 repeats was unclear, but all functionally validated S. pombe wtf drivers contain the exon 3 repeats (S2 Table). One wtf gene (wtf23 from the CBS5557 strain) that lacks the exon 3 repeats has been tested in S. pombe and failed to cause drive in two strain backgrounds, although it was not determined if the encoded proteins were non-functional or if the driver was effectively suppressed [20,26]. Previous work also found that a mismatch of repeat numbers in exon 3 between a Wtf13^poison (five repeats) and Wtf18-2^antidote (four repeats) could still produce a compatible poison and antidote pair [27]. In this work, we found that deleting the exon 3 repeats in wtf4 (wtf4-rep1Δ allele) produced a functional, compatible Wtf^poison and Wtf^antidote pair. The Wtf4-rep1Δ^poison, however, was not neutralized by the wild-type Wtf4^antidote, which has two repeats (S12B Fig). This indicates that the exon 3 repeats can affect Wtf^poison and Wtf^antidote compatibility in a context dependent fashion.

Many S. pombe wtf genes also contain a sequence repeat in exon 6 [20]. Repeats homologous to those found in S. pombe exon 6 can also be found in the C-termini of many genes in S. octosporus and S. osmophilus [19]. All functionally validated drivers contain repeats in S. octosporus, but one functional S. pombe driver (wtf35 from FY29033) lacks the repeats (S5 Table) [19,26]. Previous work demonstrated that a mismatch in the number of exon 6 repeats could affect Wtf^poison and Wtf^antidote compatibility in S. pombe Wtf proteins [15,27]. This current work extends previous work by showing with additional alleles of wtf4 and novel alleles of S. octosporus wtf25 that copy number mismatches in this region disrupt Wtf^poison and Wtf^antidote compatibility.

Overall, our results support the model that the rapid copy number evolution of the repeats found in wtf genes contribute to rapid innovation of novel Wtf^poison and Wtf^antidote pairs [19,20]. These highly evolvable sequences have likely contributed to the long-term evolutionary success of wtf drivers, as generating novel alleles allows frequent generation of novel drivers likely to be heterozygous (and thus drive) in crosses. Frequent driver turnover may also complicate the evolution of drive suppressors that are not other wtf genes. Still, this work reveals that these hypermutable regions come with a burden as they contain the potential to generate self-killing alleles, which are discussed below.

Wtf fitness landscape likely includes self-killing alleles

Our analyses of mutations in the repeats found in exon 6 of S. pombe genes revealed a novel self-killing phenotype in which a wtf allele can encode a toxic Wtf^poison that is not effectively neutralized by its corresponding Wtf^antidote. Such an allele is expected to lead to a dominant loss of fertility, analogous to mutations where the antidote protein expression or function is disrupted [17,22,25]. This phenotype was strongest in mutants that changed the sequence of the repeats (i.e., wtf4-rep2^sc, wtf4-rep2^A and wtf25-rep^Sk). These mutations are rather dramatic and have a low probability of arising spontaneously in nature.

A weaker version of the self-killing phenotype was, however, also observed in mutations that deleted the repeats. Interestingly, the deleterious effects of removing the exon 6 repeats in wtf4 (wtf4-rep2Δ allele) could be suppressed by also deleting the exon 3 repeats (wtf4-rep1-2Δ allele) (S12C and S12D Fig). This suppression, in addition to the one functional driver known to lack the exon 6 repeats, shows that changes in other regions of the protein can compensate for the missing repeats. Still, the existence of wtf genes without the repeats, and extensive gene conversion within the family, suggests that novel deleterious repeat deletion mutations are likely to arise recurrently in natural populations [19–21].

We may have also fortuitously sampled one largely self-killing allele from a natural population that did not have disrupted repeats, suggesting there are multiple paths to generating such alleles. The wtf41 gene from S. osmophilus encodes a toxic Wtf41^poison that is not efficiently neutralized by the corresponding Wtf41^antidote protein (S5C Fig, [19]). All together, we propose self-killing wtf alleles could contribute to recurrent, spontaneous sub-fertility or infertility. We propose this spontaneous sub-fertility could be a persistent burden on the population fitness of all Schizosaccharomyces species carrying wtf drivers.

Rapid evolution and the risk of self-killing alleles

Beyond wtf genes, there are many known killer meiotic drivers [3,5,6,8,9,11]. There are also likely many more yet to be discovered, given their accelerated pace of discovery in recent years. Although the genes causing drive in different systems are not homologous, they often share mechanistic and evolutionary themes [13,14,68]. Those themes include production of poisons/killer elements and rapid evolution. The wtf genes have illustrated that high evolvability, via nonallelic gene conversion and mutable coding sequence repeats, can facilitate the evolutionary success of meiotic drivers [19,21,69]. This work reveals that rapid evolution of killer elements also presents the risk of generating self-killing alleles. There is no reason to suppose such risks would be specific to the wtf killers. Instead, we posit that such self-killing alleles may be a widespread source of recurrent, spontaneous infertility in eukaryotes.

Cloning

We confirmed all the vectors described in this study by Sanger sequencing or by Nanopore sequencing via Plasmidsaurus. The specifics for the yeast strains used in this study are listed in S4 Table, plasmids are in S5 Table and the oligos are in S6 Table.

Strain construction

Spot assays

For spot assays, we first grew the strains for each experiment in the appropriate SC dropout media overnight at 30 °C. The next day, we measured the OD600 of each strain and normalized the OD of all the strains to an OD of ~1. We then serially diluted these normalized cultures 10-fold (up to the dilution of 10⁻⁵) in water and plated 5 μL of each dilution on both SC dropout media and SC Galactose dropout media. We imaged the plates post 72 h of growth at 30 °C, except in the case of Figs 4B, S6B and S6C, where plates were grown for an additional day at 30 °C to clearly visualize differences in poison toxicity.

For the spot assays presented in Fig 3D, we first grew the strains in 5 mL SC-URA (uracil) media overnight at 30 °C. The next day, we pelleted overnight cultures and resuspended in 1 mL of sterile water. We then used 100 μL of this as the first in the dilution series. This was diluted 5-fold, up to eleven dilutions in order to observe the toxicity/suppression clearly. We plated 5 μL of each dilution on SC-URA media, SC Galactose-URA media and SC Galactose-URA media with 40 mg/L Doxycycline (Calbiochem #324385). We imaged the plates post 72 h of growth at 30 °C.

For the spot assays presented in S7B Fig, we first grew the strains in 5 mL SC-TRP-URA media overnight at 30 °C. The next day, we pelleted overnight cultures and resuspended in 1 mL of sterile water. We then used 100 μL of this as the first in the dilution series. This was diluted 5-fold, up to 11 dilutions in order to observe the toxicity clearly. We plated 5 μL of each dilution on SC-TRP-URA media and SC Galactose-TRP-URA media. We imaged the plates post 72 h of growth at 30 °C.

Microscopy

The appropriate strains for each experiment were first grown in 5 mL of the appropriate SC dropout media overnight at 30 °C. The next day, we inoculated 3 mL of SC Raffinose dropout media with 1 mL of the saturated overnight cultures and let them grow overnight at 30 °C. On the following day, we pelleted the cultures and resuspended the cells in 3 mL SC Galactose dropout media and incubated at 30 °C. Cells were induced for 4–6 hours on SC Galactose dropout media and then imaged on a Zeiss LSM 980 confocal microscope which consisted of an Axio Observer Z1 base, through a 40 x C-Apochromat (NA = 1.2) water immersion objective. The GFP and mCherry tagged proteins were excited at 488 and 561 nm, respectively. Emission was collected in channel mode onto a GaAsP detector in photon counting mode through a 491–544 nm bandpass for GFP and a 570–632 nm bandpass for mCherry. Transmitted light was also collected. Image fields of view were zoomed optically to a ~26 µm square with 512 pixels in each dimension.

For imaging cells expressing S. octosporus Wtf61^antidote-GFP, S. kambucha Wtf4^poison-GBP-mCh and Rnq1-mCardinal, we first grew cells in SC-Trp-Ura-Leu media. The next day, we inoculated 3 mL of SC Raffinose-Trp-Ura-Leu media with 1 mL of the saturated overnight cultures and let them grow overnight at 30 °C. The following day, we pelleted the cultures and resuspended the cells in 3 mL SC Galactose-Trp-Ura-Leu + 500 nM β-estradiol (VWR, # #AAAL03801-03) media and incubated at 30 °C. Cells were induced for 4 hours and then imaged as described above, except that we also excited mCardinal tagged proteins at 640 nm and collected the emission on channel mode onto a 650–700 nm bandpass detector of the Zeiss LSM 980 confocal microscope.

Spectral unmixing for mEosNB-FTH1 strains

For imaging the mEosNB-FTH1 strains, we first grew the strains in 5 mL SC-URA+40 mg/L Doxycycline. The next day, we inoculated 3 mL of both SC Raffinose-URA and SC Raffinose-URA+40 mg/L Doxycycline media with 1 mL of the saturated overnight cultures and let them grow overnight at 30 °C. The following day, we pelleted the cultures and resuspended the cells in 3 mL of both SC Galactose-URA and SC Galactose-URA+40mg/L Doxycycline and incubated at 30 °C. The rest of the methodology was as described in the section above.

The background fluorescence for strains constructed with SZY6594 had high autofluorescence in the GFP channel. To distinguish autofluorescence from true mEos signal, we spectrally unmixed the images. First, we captured spectral images with the lambda mode on the detector of the Zeiss LSM 980 confocal microscope (same settings as above), exciting the cells at 488 nm and collecting emission over the entire visual spectrum. We then made reference spectra for mEos and autofluorescence from control cells, and then spectrally unmixed images using an in-house written plugin on ImageJ (https://research.stowers.org/imagejplugins/). The data presented in Fig 3C has transmitted light images collected on the channel mode with GFP excitation, and the spectrally unmixed images collected on lambda mode.

Image analysis to quantify colocalization of mEos and mCherry signals in tagged constructs

Cell regions of interest (ROIs) were found based on the transmitted light only using Cellpose (https://www.cellpose.org/) in Python using the “cyto” model and a diameter of 200. Pixels from these ROIs were then background subtracted using an average background signal from an ROI placed away from any cell. The Scipy Pearson Correlation for each cell ROI was then calculated for all pixels in each cell. Dead cells and aberrant ROIs were hand filtered out using Fiji (https://fiji.sc/). Find the raw data for these analyses in S1 Data.

DAmFRET

We performed the analysis similar to the methods in [15], with a few exceptions. We grew the strains in the same method described for microscopy, however, we induced the cells for 5–6 hours. We then photoconverted the cells post induction for 5 minutes, using the OmniCure S2000 Elite UV lamp for photoconversion. The total power over 5 minutes of exposure amounts to 12.048 J/cm². The sample collection and data analysis were similar to methods described in [15] with a few exceptions. We included an empty vector strain in each experiment to effectively exclude auto fluorescent cell populations. All the cells for each experiment were induced at the same time, for the same amount of time, and were grown in the same 96-well plate. Three technical replicates were included for each sample.

There was only one fluorescent population that was independent of expression level for each construct analyzed. In order to clearly visualize AmFRET values across the cells expressing different constructs, we analyzed cells above a certain Acceptor fluorescence intensity threshold set for each experiment. For each set of experimental data, the median of the Wtf4^poison-mEos AmFRET values was calculated. The difference of this median value from 1 was then added to all the datasets within an experiment. For visualizing the results of the DAmFRET experiments, plots of AmFRET values were generated using GraphPad Prism 10 (Version 10.2.3 (347)).

Statistical analysis for the DAmFRET experiments involved pooling the technical replicates and performing pairwise t-tests comparing these values to either the wild-type Wtf4^poison-mEos dataset, or the monomer dataset. For each DAmFRET experiment, we then did a multiple sampling correction (Bonferroni correction) by dividing the p value cutoff (0.05) by the number of tests performed. Find the raw data and statistical analysis for each DAmFRET experiment in S4 Data.

We used GraphPad Prism 10 (Version 10.2.3 (347)) to visualize acceptor fluorescence intensity values of live cells expressing the relevant Wtf^poison constructs presented in S6 Fig. Three technical replicates of each isolate were plotted, with the median included. Find the raw data for this analysis in S2 Data.