Introns provide a platform for intergenic regulatory feedback of RPL22 paralogs in yeast

Ribosomal protein genes (RPGs) in Saccharomyces cerevisiae are a remarkable regulatory group that may serve as a model for understanding genetic redundancy in evolutionary adaptations. Most RPGs exist as pairs of highly conserved functional paralogs with divergent untranslated regions and introns. We examined the roles of introns in strains with various combinations of intron and gene deletions in RPL22, RPL2, RPL16, RPL37, RPL17, RPS0, and RPS18 paralog pairs. We found that introns inhibited the expression of their genes in the RPL22 pair, with the RPL22B intron conferring a much stronger effect. While the WT RPL22A/RPL22B mRNA ratio was 93/7, the rpl22aΔi/RPL22B and RPL22A/rpl22bΔi ratios were >99/<1 and 60/40, respectively. The intron in RPL2A stimulated the expression of its own gene, but the removal of the other introns had little effect on expression of the corresponding gene pair. Rpl22 protein abundances corresponded to changes in mRNAs. Using splicing reporters containing endogenous intron sequences, we demonstrated that these effects were due to the inhibition of splicing by Rpl22 proteins but not by their RNA-binding mutant versions. Indeed, only WT Rpl22A/Rpl22B proteins (but not the mutants) interacted in a yeast three-hybrid system with an RPL22B intronic region between bp 165 and 236. Transcriptome analysis showed that both the total level of Rpl22 and the A/B ratio were important for maintaining the WT phenotype. The data presented here support the contention that the Rpl22B protein has a paralog-specific role. The RPL22 singleton of Kluyveromyces lactis, which did not undergo whole genome duplication, also responded to Rpl22-mediated inhibition in K. lactis cells. Vice versa, the overproduction of the K. lactis protein reduced the expression of RPL22A/B in S. cerevisiae. The extraribosomal function of of the K. lactis Rpl22 suggests that the loop regulating RPL22 paralogs of S. cerevisiae evolved from autoregulation.


Introduction
Ribosome biogenesis absorbs a substantial portion of yeast cell resources [1]. It requires 78 ribosomal proteins (RPs), approx. 200 ribosomal protein assembly factors, and~75 small nucleolar RNA genes [2]. All of this production needs to be precisely balanced and, perhaps more importantly for regulatory requirements, must be capable of coping with the rapid changes in ribosome numbers. Cells must be able to react to the availability of nutrients and to various stressors by speeding up or stopping growth [3]. In a fast growing cell, ribosomes must be assembled at a rate of 2000 per minute to a degree of high precision and efficiency [1,4], while leaving no unused components or stalled assembly intermediates that would otherwise hamper fitness or even endanger the cell [5,6].
Ribosomal proteins in S. cerevisiae are encoded by 137 genes, among which are 59 pairs of functional paralogs [7]. These paralogs originated during the events leading to whole genome duplication (WGD) in the ancestor of Saccharomyces before 100 My and have been retained to a considerable degree, representing~11% of the surviving paralog pairs [8][9][10]. It is believed that pairs of ohnologs (paralogs that originated during WGD) brought about a higher level of regulatory complexity, increasing the spectrum of responses to varying nutrient conditions [11,12]. A control system originally regulating the expression of one ancestral gene can become intergenic [13]. S. cerevisiae is an intron-poor species that has lost most of its introns (only~280 remain) [14,15]. Of these, 104 are RPG introns; therefore, given the high expression levels of RPGs [16,17], 90% of mRNA splicing is devoted to producing RPG mRNAs [1].
Although RPG paralogs mostly code for proteins of an identical or highly similar amino acid composition (35 pairs differ by less than 2 amino acids [18]), they differ markedly in their 5'-and 3'-UTR sequences and introns [19]. These differences are species/strain-specific and, at least in yeast, have been shown as subject to selection pressure [20]. Apparently, they were fixed in populations as a result of adapting to specific growth conditions and genetic backgrounds [20,21]. Some RPG pairs differ in their ORFs and their products have distinct localization patterns. These factors can affect specific aspects of cell physiology. For example, it has been reported that translation of ASH1 mRNA requires a specific subset of ribosomal protein paralogs, including Rpl22A [22]. Shifts in the relative concentrations of paralogs can lead to subpopulations of functionally different ribosomes [23]. "Specialized" ribosomes can operate in distinct subpools of mRNA substrates or in specific compartments [22,[24][25][26]. Recently, quantitative mass spectrometry of translationally active ribosomes from mouse embryonic stem cells confirmed ribosomal heterogeneity with respect to several RPs as well as their association with specific subsets of mRNAs [27].
RPG expression in S. cerevisiae is tightly coordinated with rRNA synthesis [28,29] and predominantly regulated at the levels of transcription [3,30] and splicing [31]. Systematic deletions of introns from RP genes indicate that introns affect cell fitness, rRNA processing, and the production of other paralogs in the case of duplicated RPGs [23]. Intron-containing RPGs have been shown to respond to various types of environmental stresses [21,23,31]. The mechanisms implicated in splicing-mediated regulation include the interaction of the ribosomal protein with its own (or paralogous) transcript through motifs related to rRNA structures bound by RP [32]. The ability of intramolecular pre-mRNA structures to influence splicing reactions was first established more than 30 years ago [33,34], but detailed understanding of the structural properties of introns has only recently begun to emerge [35]. Complementary sequences can base-pair over shorter or longer distances, looping out exons [36], occluding or weakening splice sites [37], or bringing splice sites into proximity [35,38,39]. In more complex scenarios, splicing signals are only modulated by the structural context, which can impact/block spliceosome assembly at later stages. One of the most extensively studied examples of intron-mediated regulation is RPL30. The Rpl30 protein inhibits its own synthesis by blocking both the splicing and translation of the RPL30 transcript [40][41][42]. Rpl30 binds an internal loop in the vicinity of 5'ss and affects U2-mediated recognition of BP, perhaps through its effects on the U1 snRNP conformation or the structure of the transcript [43]. Other examples of splicing regulation include Rps14, which binds to RPS14B (but not RPS14A) pre-mRNA and represses its splicing [44], as well as Rps9 and Rpl22, which inhibit the splicing of RPS9A and RPL22B pre-mRNA, respectively [13,45]. These inter-paralog relationships seem to have evolved independently and in some cases several times in different taxa [13].
A high proportion of RPG pairs is unique to post-WGD yeast. However, in the cases of some RPGs, paralogs have occurred throughout evolution. The Rpl22 pair is one of the bettercharacterized examples of such a tendency. The available evidence documents its existence in yeast, zebrafish, Drosophila, mice, and humans [23,[46][47][48]. Rpl22 and Rpl22-Like1 play extraribosomal roles in modulating the splicing of Smad2 pre-mRNA during zebrafish morphogenesis. They both bind Smad2 pre-mRNA but act antagonistically; their balance is important for correct splicing of exon 9 [49]. Rpl22 is required for T-cell differentiation in mice and for the emergence of hematopoietic stem cells in zebrafish [46]. Mouse Rpl22 binds to Rpl22l1 mRNA and regulates its expression [48]. During Epstein-Barr virus infection, human Rpl22 is sequestered by EBER-1 viral RNA, a condition which enhances the replicative potential of Burkitt lymphoma cells [50]. Human RPL22 has also been suggested as playing a role in binding to telomerase RNA [51]. Several RPs, including Rpl22, have been implicated in p53 activation upon stress disturbing ribosome biogenesis [52,53]. Rpl22 has been recently shown to directly bind and block MDM2 [53], which is intriguing, given the occurrence of inactivating mutations of RPL22 in various cancers [53,54]. In yeast, RPL22A deletion increases replicative lifespan [18] and affects sensitivity to oxidative stress [55,56] and acetic acid [57]. Furthermore, Rpl22 has been implicated in translation regulation of meiotic inducer IME1. The atypical 5'-UTR of the IME1 transcript stalls its translation in rpl22Δ cells [58]. Recently, Gabunilas and Chanfreau reported that RPL22 paralogs are regulated in an intergenic fashion through their introns. This regulation was mediated by Rpl22 proteins and required a structured element that did not show direct sequence homology with the rRNA binding motif to which Rpl22 binds within the ribosome [59].
In our study, we set out to analyze the roles of introns in intergenic relationships observed previously in S. cerevisiae [23]. Of the 7 paralog pairs examined, we found 1 intragenic and 1 intergenic intron-dependent effect in RPL2A and RPL22A/B, respectively. In this paper, we summarize our findings on the RPL22 pair and independently demonstrate-using proteomic analysis of Rpl22 levels, three hybrid system testing, and assays with RNA-binding mutantsthat intergenic regulation is mediated by Rpl22 protein-binding and is asymmetric with respect to introns. A and B introns are capable of inhibiting mRNA production of their own genes in an Rpl22-dependent way to 50% and 8% of WT levels, respectively. Importantly, we show that Kluyveromyces lactis RPL22, which is not a duplicated gene, is also subject to intronand Rpl22 protein-mediated inhibition in K. lactis cells. Because Rpl22 can potentially bind to other non-ribosomal targets as well as its own pre-mRNA [49], we mapped the effects of RPL22A/RPL22B gene and intron deletions on the transcriptome level. Our data suggest that Rpl22B assumes a specific role which cannot be complemented by Rpl22A.

Strains, growth conditions, transformation
The strains used in this study were derived from BY4741 and BY4742 standard laboratory strains (the list is provided in S1 Table). For gene expression analysis, yeast were grown for at least eight generations in a rich medium (YPD supplemented with adenine, YPAD) after transferring from a pre-culture until they reached the mid-exponential phase. For overexpression and primer extension analysis, cells were grown in a synthetic medium without uracil and/or histidine for two generations. Yeast transformation was performed as described [60].

Intron deletions
The delitto perfetto method was used to delete introns from RPL22 genes in the BY4741 and BY4742 strains [61]. In the first step, we amplified the URA3 gene from the pRS316 plasmid with primer pairs KA70 + KA71 and KA78 + KA79 and inserted it into the RPL22A and RPL22B intron, respectively. In the second step, cells were transformed with integrative recombinant oligonucleotides (IROs), KA72 + KA73 for RPL22A and KA80 + KA81 for RPL22B, whose integration led to the loss of the URA3 gene together with the intron sequence. These cells were selected using 5-fluoroorotic acid (5-FOA). To delete the hairpin from the RPL22A intron, we used IROs KA74 + KA75. KlRPL22 intron was removed from JLQ36 strain using the procedure described below (simplified). URA3 cassette was amplified with primers JL479 and JL480 from pJet1-RPL22A-RPL22Bi-URA3. URA3 + transformants were left overnight on YPAD plates before selection on 5-FOA. Successful intron/hairpin deletions were confirmed by PCR and sequencing. The list of all oligonucleotides and plasmids used in this study is provided in S2 and S3 Tables, respectively.

Intron and gene replacement
Intron and gene swaps were performed using an approach similar to the method used by Längle-Rouault [62]. Briefly, we prepared DNA cassettes containing the sequence to be inserted, flanked by 45-50 bp arms complementary to the immediate surroundings of the destination site. The 3' flanking sequence was included twice as a direct repeat separated by the URA3 selection marker. While URA3 was used to select transformed cells, it was also prone to pop out via homologous recombination between the two copies of the 3' flanking sequence. These cells were then selected on a medium containing 5-FOA. The genetic manipulation was confirmed by sequencing. S4 Table summarizes the oligonucleotides used and the mode of creation for particular constructs. Two of the replacement cassettes were constructed in the pJet1 plasmid. PCR fragments produced with primers F1 + R1 (5' part containing the sequence to be inserted + flanking nucleotides) and F2 + R2 (3' part-the URA3 marker and the second copy of the 3' flanking sequence) were first inserted individually into the pJet1 plasmid using the CloneJet kit (Thermo Fisher Scientific). The respective F1-R1 fragments were then excised by digestion with BamHI and BglII and ligated into the corresponding F2-R2-bearing constructs digested with BglII. The other cassettes were assembled by PCR with primers F1 and R2 using an overlap of 24 bp between the parts.

Plasmid construction
To generate the RPL22 overexpression constructs, the respective RPL22 ORFs were PCRamplified from genomic DNA from intronless strains (KAY61 strain and primers FN12 + FN13 for RPL22A; KAY67 strain and primers FN14 + FN15 for RPL22B), digested with NcoI/BamHI, and cloned into plasmid pVTU260 digested with the same enzymes. KlRPL22 was amplified from Kluyveromyces lactis IFO1267 cDNA using primers FN34 and FN35 and cloned in the same manner. For experiments using K. lactis, expression cassettes containing RPL22A, RPL22B, their mutated versions, and KlRPL22 were released from pVTU260 by PaeI digestion and inserted into the pCXs22 plasmid [63].
Mutations in the Rpl22 RNA binding site [64] were designed to change lysines in the protein sequence 73-GKYLKYLTKKYLKKNQL-89 to glutamates (73-GEYLEYLTEEYLEENQL-89) by changing the lysine codons to GAA. DNA sequences coding for the Rpl22 RNA-binding mutants flanked with NcoI and BamHI restriction sites were synthesized by GeneArt (Thermo Fisher Scientific) and used to replace WT RPL22 ORFs in the pVTU260 plasmid with the corresponding restriction enzymes.
To prepare the splicing reporters, we replaced the BamHI/EcoRI COF1 insert in the pOG71 plasmid [39] with BamHI and EcoRI-digested products of PCR amplification with primers MO09 + MO10 for the RPL22A intron and primers MO11 + MO12 for the RPL22B intron using BY4741 genomic DNA as a template. These manipulations resulted in plasmids p423GPD-RPL22A-CUP1 and p423GPD-RPL22B-CUP1, which were verified by restriction analysis and sequencing.
Vectors p3HR2 and pACT2 used in the yeast three-hybrid assays were donated by the Wickens Laboratory. To create hybrid proteins, intronless versions of RPL22A, RPL22B, and their RNA-binding mutants were re-cloned from the pVTU260-based expression vectors into the pACT2 vector using the NcoI and BamHI restriction sites. The intronic sequences were PCR-amplified from genomic DNA (strain BY4741) and cloned into the SphI site of vector p3HR2.

RNA isolation, reverse transcription and real-time PCR
RNA from cells grown to the mid-exponential phase in the YPAD medium was isolated using the MasterPure Yeast RNA Purification Kit (Epicentre Biotechnologies), including the DNase I treatment step, essentially according to the manufacturer's instructions. Two micrograms of RNA were converted to cDNA using RevertAid reverse transcriptase (Thermo Fisher Scientific) with random hexamer primers unless indicated otherwise. Real time-qPCR reactions were performed using the MESA GREEN qPCR MasterMix Plus for SYBR Assay No ROX (Eurogentec). All reactions were run in triplicate using the LightCycler 480 II (Roche). Relative pre-mRNA and mRNA quantities were calculated using the ΔΔCt method [65]. For intron and gene replacement experiments, primers were designed in order to distinguish between the various forms of RPL22 transcripts present in the cells. Specificities of the primer pairs used are listed in S5 Table. All primer pairs were tested for lack of (i) unspecific amplification under standard qPCR conditions and (ii) amplification of other related templates mentioned in S5 Table. The contribution of the second copy of RPL22B in the strain where the intron and exon 2 of RPL22A were replaced with the corresponding part of RPL22B was calculated as the difference between the relative abundances of total RPL22B mRNA (primers KA50 + JL385) and RPL22B mRNA transcribed from the RPL22B locus (JL360 + JL385).
The statistical analysis was performed on ΔC t data using the t-test with correction for multiple testing. Statistical tests were conducted with the "R" statistical package version 3.2.3 (www. r-project.org/) using the t.test() function with parameters "paired = FALSE, alternative = 'two. sided' ". The function p.adjust() with Holm correction was used; P<0.05 was considered significant.

Primer extension analysis of relative splicing efficiency
The analysis was performed according to the protocol described in [39]. Briefly, RNA was isolated using the MasterPure Yeast RNA Purification Kit (Epicentre Biotechnologies). Three to five micrograms of RNA were transcribed into cDNA using RevertAid reverse transcriptase (Thermo Fisher Scientific) with the YAC06 and YU14 [66] primers labeled using γ 32 P-ATP and the T4 polynucleotide kinase (Promega). cDNA molecules were then separated on 8% polyacrylamide gel with 7 M urea together with a FX174 HinfI DNA labeled marker. Radioactive signals were captured on Imaging Screen-K (BioRad) and detected using the Typhoon FLA imager (GE Healthcare Life Sciences).

Proteomic analysis
For mass spectrometry (MS) analysis, cells were grown in 50 ml of YPAD to the mid-exponential phase, collected by centrifugation (1000 g, 3 min), washed with 1 volume of water, frozen in liquid nitrogen, and stored at -80˚C. Cells were disintegrated by 5 rounds of beating (20 s, speed 4, 5 min cooling between rounds) with glass beads in the FastPrep-24 instrument (MP Biomedicals) in 450 μl of 100 mM triethylammonium bicarbonate buffer with 1% sodium deoxycholate. Cell debris was removed from the lysate by centrifugation at 1000 g for 1 min at 4˚C, followed by centrifugation at 2000 g for 5 min at 4˚C; the lysate was frozen and stored at -80˚C. Peptides were generated by trypsin digestion, analyzed on Thermo Orbitrap Fusion (Q-OT-qIT, Thermo), and quantified using MaxQuant software (version 1.5.3.8) [67] in our core facility.

Yeast three-hybrid protein-RNA interaction testing
Interactions of Rpl22 proteins with their RNA ligands were assayed in the YBZ1 yeast strain as described in [68]. Transformants containing plasmids expressing the respective hybrid protein and target RNA were assayed for expression of the HIS3 reporter gene by testing the ability of 10-fold serial dilutions of cell cultures to grow on selective media. To distinguish between interactions of different strength, activity of the HIS3 reporter was assessed in the presence of a competitive inhibitor, 3-aminotriazole.

RNA structure modeling
For RNA structure modeling, RNAFold [69] was used with default settings, except for the temperature being set to 30˚C.

Transcriptome profiling
Cells grown in the exponential phase for at least eight generations in YPAD medium at 30˚C were harvested by centrifugation (1000 g, 3 min, room temperature); cell pellets were stored at -80˚C. Cells were disrupted using glass beads in the FastPrep-24 instrument (MP Biomedicals) and total RNA was obtained using the acidic phenol:chloroform extraction method as described previously [70].
RNA was further purified with the MasterPure Yeast RNA Purification Kit (Epicentre Biotechnologies), starting with the DNase I treatment step. Strand-specific paired-end RNA-Seq libraries were constructed from poly(A)-enriched RNA samples by BGI Genomics using their standard procedure. RNA-Seq libraries were sequenced (100 nt, paired-end) by BGI Genomics using the Illumina HiSeq platform. Raw reads were filtered after sequencing, including the removal of adapter sequences, contamination, and low-quality reads. Three biological replicates were analyzed for each strain. The sequencing data are available in the ArrayExpress database (www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-5275.

Introns in some, but not all, RPGs affect their gene expression under standard cultivation conditions
The roles of RPG paralogs in ribosome biogenesis and translation, both in unicellular eukaryotes and in Metazoa, remain an important issue [20]. We tested 7 pairs of paralogous gene pairs-RPL22, RPL16, RPL37, RPL17, RPL2, RPS0, and RPS18 -previously reported to be regulated in an intergenic and intron-dependent way [23]. From diploid strains containing heterozygous intron deletions in each paralog, we prepared haploids lacking the intron in either the A paralog, B paralog, or in both and then assayed RPG mRNA abundance ( Fig 1A). Intron deletions in all but two of the tested candidate pairs had little impact on expression levels of the manipulated gene or its paralog. Intron deletion in RPL2A led to a very strong inhibition of its own expression, which points to the existence of an intragenic regulatory loop, while intron deletions in RPL22A/B showed intergenic effects. Phenotypes of the intron-manipulated RPL22 strains were reproducible across cultivation conditions (YPAD versus synthetic media), irrespective of the strain background (BY4741/BY4742 versus JPY10I) or the number of generations between the inoculation and the collection of cells for analysis. Further analysis of additional environmental stresses and metabolic states may clarify the differences between our study and the previous results of Parenteau et al. [23]. Notably, there are also discrepancies between studies with regard to changes of intron deletions in the RPS9 and RPS14 paralogs [13,44,23]. In summary, while some paralog pairs display intron-dependent intergenic relationships, the roles of introns in other pairs may be less obvious and should not be simply inferred.
Introns mediate asymmetric intergenic regulation of the RPL22A and RPL22B paralogs by Rpl22 proteins Because we were able to observe an intergenic relationship only with the RPL22A/B gene pair, we next concentrated on it. We prepared intronless versions of RPL22A and RPL22B in the BY4741 and BY4742 genetic backgrounds, compatible with yeast deletion collection [77]. As shown in Fig 1 and S1 Fig, the effects of RPL22 intron deletions in the strains derived from BY4741 and BY4742 are in agreement with the results obtained using strains constructed by J. Parenteau [23]. The ratio of RPL22A to RPL22B mRNA abundance was 93 to 7 in WT ( Fig  1B). Intron deletion from RPL22A (designated rpl22aΔi) led to a~1.6-fold increase in RPL22A mRNA, accompanied by a dramatic (~19-fold) decrease of RPL22B mRNA. Removal of the intron from RPL22B (rpl22bΔi) enhanced the abundance of its own mRNA 5.4-fold, while RPL22A expression decreased 1.3-fold.
The expression of RPL22A did not differ considerably from the WT level when the RPL22B gene was deleted, but it did differ considerably when the RPL22A intron was also absent. On the other hand, removing RPL22A increased RPL22B mRNA to a similar abundance as in rpl22bΔi (40% of total RPL22 mRNA in WT). Still, the level of total RPL22 mRNA was lower in rpl22aΔ than in WT, which suggests that these cells are unable to attain the normal level of RPL22 expression by transcription from the RPL22B locus alone. We also tested the levels of pre-mRNA in all the above strains (S1 Fig). mRNA downregulation of RPL22A and RPL22B in rpl22bΔi and rpl22aΔi, respectively, was not accompanied by dramatic increases in corresponding pre-mRNAs. In contrast, upregulation of RPL22B mRNA in rpl22aΔ, which apparently represents the maximum expression attainable from that locus, was accompanied by a drop in pre-mRNA levels.
To test whether mRNA levels would also be reflected in protein abundance, we analyzed the relative amounts of Rpl22A and Rpl22B proteins in strains with different RPL22 alleles using mass spectrometry. The proteomic data corresponded to those for mRNA levels (compare Fig 1B and 1C). Rpl22A was the major Rpl22 protein in cells. Its abundance increased in cells where its gene did not contain an intron as well as in the rpl22bΔ strain where Rpl22B was missing. On the other hand, Rpl22B accounted only for 4% of the total Rpl22 level in the WT strain. However, when its intron-mediated regulation was abrogated, the Rpl22B level rose tõ 17.5%. Both RPL22 genes contain long introns (389 nt in RPL22A, 321 nt in RPL22B). The distance between the branch-point and 3' splice-site is longer in RPL22A (56 nt) than in RPL22B (17 nt). The two introns also differ in their donor site sequences, which both deviate from the GTATGT consensus: GTATGA in RPL22A and GTACGT in RPL22B. Interestingly, there is potential for a stable secondary structure (hairpin) to be formed between the branch-point and 3' splice site of the RPL22A intron (S2 Fig). However, when we prepared a strain from which the hairpin was removed (rpl22aΔH), the levels of RPL22 mRNAs and pre-mRNAs in that strain did not differ from WT (S1 Fig).
The asymmetric behavior of these introns led us to analyze the splicing efficiency of RPL22derived splicing reporter constructs in strains with manipulated RPL22 genes (Fig 1D and 1E). In these reporters, the strong constitutive GPD promoter drove transcription of exon 1 (including a part of 5'-UTR) followed by the intron and a part of exon 2 of either RPL22A or RPL22B fused to the CUP1 gene. Using a primer extension, we observed high-efficiency RPL22B reporter splicing in strains from which the RPL22A paralog was absent (Fig 1E), which supports the hypothesis that Rpl22A stalls the expression of RPL22B by blocking the splicing step. This inhibition probably occurs during splice-site recognition or during the early steps of spliceosome assembly, as we did not observe any lariat-intermediate accumulation.
and intragenic intron dependency, respectively. Plots show mean fold changes of "A" (a) and "B" (b) paralog mRNA in mutants with intron deletion (Δi) in "A", "B" or both paralogs as determined by RT-qPCR relative to WT. Data were normalized to SPT15 expression and to the RNA level in WT cells. Error bars represent s.d. from two (RPL17, RPS0, RPS18) or three (RPL2, RPL16, RPL37) biological replicates. The statistical significance of the difference between WT strain and a strain bearing intron deletion is indicated as (2) for P 0.01 and (3) for P 0.001 based on the t-test with Holm correction for multiple testing (see Methods). Pre-cultures of colonies from freshly dissected spores were diluted to such a low OD so that they could undergo 10 generations before they were harvested at mid-exponential phase. In independent experiments, intron deletion mutants of RPL22, RPL16 and RPL37 were grown in YPAD or synthetic medium for 2 or 10 generations, respectively. The two cultivations gave essentially the same results as above (data not shown). (B) RPL22 mRNAs are negatively regulated by their introns and respond in an intergenic way to the manipulations of their paralogous counterparts. The plot shows mRNA levels of RPL22 paralogs normalized to WT. The RPL22A to RPL22B ratio was calculated from the same data as shown in S1 Fig Indeed, splicing of the RPL22B reporter was inhibited in strains with intron deletion in RPL22A (i.e., with increased Rpl22A protein levels), whereas it was intermediate in WT. In contrast to RPL22B, the relative changes of RPL22A reporter mRNA accumulation were much less pronounced (Fig 1D). To summarize, we documented that RPL22 introns inhibit expression of their genes, albeit to a different extent. Rpl22 protein levels corresponded with mRNA levels.
RPL22 pre-mRNAs are controlled by the nonsense-mediated decay pathway Regardless of changes in mRNA quantity, RPL22 pre-mRNA levels were mostly unaffected (S1 Fig). To determine which RNA degradation pathways would be involved in the metabolism of RPL22 RNA species, we analyzed changes in mRNA and pre-mRNA levels of RPL22A and RPL22B in strains with deletions of genes involved in several RNA degradation pathways ( Fig  2). Members of the TRAMP complex (air1Δ, air2Δ, trf4Δ, trf5Δ), exosome complex (rrp6Δ, ski7Δ, hbs1Δ), and THO/TREX complex (swt1Δ, thp1Δ, thp2Δ) did not affect the levels of RPL22 RNA species. On the other hand, deletions of cytoplasmic nonsense-mediated decay (NMD) pathway members (upf1Δ, upf2Δ, and upf3Δ) and cytoplasmic 5'-3' exonuclease xrn1Δ increased pre-mRNA levels of RPL22A~4 fold and RPL22B~8 fold.
To obtain more detailed information about RPL22 RNA metabolism, we analyzed doublemutant strains with a deletion of selected RNA surveillance genes (air1Δ, air2Δ, dbr1Δ, and upf1Δ) together with RPL22 intron manipulations (S3 Fig). Impairment of the pathways in air1Δ, air2Δ, and dbr1Δ did not change the expression profiles of RPL22 mRNAs and pre-mRNAs beyond the effect of intron manipulations. Importantly, RPL22B pre-mRNA was 8 Fig 2. RPL22 pre-mRNAs are targeted by the nonsense-mediated decay pathway. RPL22A and RPL22B expression was assessed by RT-qPCR (reverse transcription followed by quantitative PCR) in strains with deletion in genes representing various pathways of RNA metabolism. cDNA was prepared using random hexamers and oligo(dT) 18-mers. Values for each amplicon were normalized to WT levels and are represented by a horizontal line. The means of biological triplicates, quadruplicates (air1Δ), and duplicates (ski7Δ, swt1Δ, thp1Δ, and thp2Δ) are shown. The statistical significance of the difference between WT strain and a mutant strain is indicated as (1) for P 0.05 and (2) for P 0.01 based on the t-test with Holm correction for multiple testing (see Methods).
https://doi.org/10.1371/journal.pone.0190685.g002 times more abundant in rpl22aΔi upf1Δ than in WT cells, even though RPL22B mRNA levels remained very low. This further supports the hypothesis that RPL22B pre-mRNA splicing is blocked by elevated levels of Rpl22.
We also took advantage of a mutant in splicing factor PRP45 that causes pre-mRNA accumulation due to impaired spliceosome assembly [78]. We compared the effects of this splicing factor mutation (prp45 (1-169)) with that of upf1Δ, upf3Δ, and xrn1Δ on pre-mRNA accumulation of RPL22A/RPL22B, ECM33, and COF1 transcripts (S4 Fig). Unlike ECM33, where prp45 (1-169) elicited~6-fold pre-mRNA accumulation (NMD alleles had little effect), RPL22B showed the opposite behavior. The RPL22A response was intermediate. We suggest that these data reflect the low proportion of splicing events per pre-mRNA molecule generated from RPL22B as well as the dependence of RPL22 metabolism on NMD. In summary, we suggest that both the NMD pathway and 5'-3' cytoplasmic degradation are responsible for pre-mRNA degradation of both RPL22A and RPL22B in WT cells.
Overexpression of Rpl22A/B, but not their mutant versions, downregulates the production of RPL22 mRNAs in an intron-specific way We attempted to manipulate RPL22 mRNA levels by expressing the cDNA of each paralog from a multicopy plasmid under the control of the strong ADH1 promoter. Increased levels of Rpl22A repressed RPL22B and RPL22A to approx. 10% and 50% of empty vector control, respectively. Rpl22B showed similar efficacies (Fig 3A, S6 Table). Intronless variants of RPL22A/RPL22B did not respond to the inhibitory effects of Rpl22A/Rpl22B overproduction, which confirms the intron requirement. To determine whether Rpl22 protein interaction with its primary transcript would be necessary for regulation, we used the same expression system to produce Rpl22 proteins mutated in their putative RNA-binding domain according to a previous study of human RPL22 [64]. Production of these modified proteins (see S5 Fig) proved inconsequential to endogenous RPL22 mRNA levels (Fig 3).
We confirmed the capacity of Rpl22 proteins to inhibit splicing using primer extension analysis in rpl22aΔ-rpl22bΔ double-mutant and WT strains. The strains carried the RPL22B splicing reporter and the expression plasmid with or without RPL22A/RPL22B cDNA. We observed high-efficiency splicing of the RPL22B reporter transcript in the rpl22aΔ-rpl22bΔ strain, which was strongly blocked by the ectopic production of either Rpl22A or Rpl22B, but not of their RNA-binding mutants (Fig 3B). The same effect was also observed in WT cells, albeit at a smaller amplitude because of the partial splicing block mediated by endogenous Rpl22. Stability of the expressed proteins was confirmed using Western blots (S5 Fig). These results show that introns exert inhibitory effects on RPL22A and RPL22B mRNA accumulation and that this effect depends on the level of Rpl22 protein(s).

Introns swapped between RPL22 genes maintain their regulatory potential
To separate the effects of introns from the corresponding coding regions as well as sequences surrounding the locus, we transplanted introns between the RPL22A and RPL22B genes. Replacing the RPL22A intron with its RPL22B counterpart led to the decrease of RPL22A mRNA and the concomitant derepression of RPL22B. Reciprocal transplantation increased RPL22B mRNA (Fig 4, S7 Table), which indicates that the RPL22A intron confers less stringent inhibition. We then ectopically produced an excess of Rpl22 proteins to test the sensitivity of the chimeric genes to regulation. Overproduction of either Rpl22A or Rpl22B reduced the mRNA abundance of both endogenous RPL22 genes. There was, however, a significant Regardless of the locus in which the gene resided and the identity of the coding region surrounding the intron, the presence of the RPL22A intron led to mRNA reduction of approx. 50%, while the RPL22B intron resulted in a drop of approximately 90% (Fig 4).
We also replaced the RPL22A gene (intron + exon 2) with a second copy of RPL22B. We took advantage of the fact that RPL22 exon 1 codes only for 4 amino acids, identical between the paralogs; there is only 1 mismatch among the 12 nucleotides. Therefore, replacing intron Rpl22A or Rpl22B. The "AΔi", "BΔi", and "AΔi BΔi" symbols denote strains lacking introns in one or both RPL22 paralogs. Gray fields indicate that data were not determined. For details and statistical analysis see S6 Table. (B) Primer extension analysis of the relative splicing efficiency of a reporter derived from RPL22B (left panel). The effect of overproducing Rpl22A, Rpl22B, and their RNA-binding mutants was compared between a strain lacking endogenous RPL22 genes and WT. The right panel shows the relative abundance of a spliced reporter mRNA signal normalized to the U14 snoRNA loading control and to the level of mRNA in WT cells with an empty vector. M is the DNA size marker. Error bars represent s.d. from 3 biological replicates. The statistical significance of the difference between empty vector strain and an overexpression strain is indicated as (1) for P 0.05 and (2) for P 0.01 based on the t-test with Holm correction for multiple testing (see Methods).
https://doi.org/10.1371/journal.pone.0190685.g003 Configurations of chimeras at the RPL22A and RPL22B loci are described for each row in the center. All values were normalized to the WT strain bearing an empty vector (top left cell in each heatmap). Leftmost columns ("vector") reflect only the effect of genetic background, while A/B columns show the effect of overexpression. cDNA was prepared using random hexamers and oligo(dT) 18-mers. The primers used to detect RPL22B mRNA were specific for the transcript derived from the endogenous RPL22B locus. The detection of the RPL22B mRNA transcribed from the A locus (underlined) was accomplished using a specific set of primers (see last row of the right panel). For details and statistical analysis see S7 Table. https://doi.org/10.1371/journal.pone.0190685.g004 + exon 2 (up to the stop codon) effectively swaps the whole coding region. The expression of recombinant RPL22 was higher than that of the endogenous copy, apparently due to a stronger promoter of RPL22A. Irrespective of the expression level, both RPL22B genes responded to Rpl22A/Rpl22B overexpression in unison, producing mRNA in similar ratios between empty vector control and overexpression constructs (Fig 4). To reiterate, we confirm that introns determine the sensitivity of RPL22 expression to Rpl22-mediated inhibition.

Rpl22 proteins interact with the RPL22B intron in a structured region downstream of 5'ss
Because the interrelationship between RPL22 paralogs was intron-and protein-dependent, we examined some of the potential Rpl22-intron interactions using the yeast three-hybrid system (3H) [68]. We prepared constructs covering three separate regions of the RPL22B intron (see Fig 5) and tested their interaction with Rpl22 proteins as well as their RNA-binding mutants. As shown in Fig 5, we were able to detect the interaction of the "I2" region (nucleotides 165 to 236 of the RPL22B intron) with both Rpl22A and Rpl22B. The interaction was comparable in strength to iron regulatory protein 1 (IRP1), with the iron-responsive element (IRE) as the positive control [68]. Rpl22A/Rpl22B RNA-binding mutants, which were successfully produced in the reporter strain (see S6 Fig), did not show any interaction. We also attempted to express the whole RPL22B intron in the reporter strain, but no interaction was detected. We assume that because of the stretches of oligo-T, RNA was most likely not produced due to terminating RNA Pol III used in the three-hybrid system [79]. The inhibitory effect of Rpl22 on RPL22B pre-mRNA is thus most likely mediated by its binding to a structure within the 72nt region between 5'ss and BP.

Intron-mediated regulation of RPL22 is conserved in Kluyveromyces lactis
Kluyveromyces lactis is a member of the "Saccharomyces complex" [8,80]. In contrast to the Saccharomyces sensu stricto complex taxa, its ancestor did not undergo WGD (see Introduction); K. lactis has only singleton RPGs in its genome [81]. We therefore wanted to investigate whether Rpl22 would be capable of regulating its splicing in K. lactis. We ectopically overproduced K. lactis Rpl22 (further referred to as KlRpl22) in S. cerevisiae. As shown in Fig 6A, the K. lactis protein repressed mRNA abundance of endogenous RPL22A and RPL22B to the same extent as if it were one of the S. cerevisiae paralogs. To achieve conditions more closely reflecting the physiological situation, we replaced the RPL22A gene (intron + exon 2) with the corresponding part of KlRPL22. If the K. lactis protein were unable to complement Rpl22A with regard to its inhibitory capacity, the outcome would be the upregulation of RPL22B mRNA. However, in this case RPL22B mRNA levels remained at the WT level (Fig 6B). After producing evidence that KlRpl22 is able to regulate S. cerevisiae RPL22 genes, we addressed the question whether the exogenous gene could itself be regulated. The strain harboring KlRPL22 in the RPL22A locus was transformed with pVTU260 plasmids overproducing either Rpl22A or Rpl22B. Each of the constructs led to the downregulation of KlRPL22 mRNA to about 40%. In a strain lacking an intron (i.e., where the RPL22A intron and exon 2 were replaced by KlRPL22 exon 2 only), overexpression of the Rpl22A or Rpl22B protein did not reduce KlRPL22 mRNA abundance.
To prove that the mechanism of RPL22 regulation is conserved in K. lactis, we investigated whether KlRPL22 would respond to Rpl22 overproduction in K. lactis cells. We observed that plasmid-driven expression of RPL22A, RPL22B, and KlRPL22 caused a reduction in KlRPL22 mRNA levels (Fig 6D). Mutated versions of RPL22A or RPL22B had no effect, proving that the inhibition is dependent on the intact RNA-binding capacity of Rpl22. These results show that the KlRPL22 singleton of K. lactis is capable of intragenic regulation.

Transcriptome analysis of rpl22 mutants
We measured the impact of RPL22 gene manipulations at the transcriptome level in order to determine other potential (non-RPL22) regulatory intron targets and paralog-specific changes in mRNAs. We performed RNA-Seq analysis of transcriptomes from WT, rpl22aΔ and rpl22bΔ knock-out, and rpl22aΔi and rpl22bΔi intron deletion strains. The results are summarized in Fig 7. We observed a much more pronounced change in the transcriptome of rpl22aΔ compared to rpl22bΔ (1490 vs 186 significant differentially expressed genes (DEGs); Fig 7A  and S8 Table). This is in accordance with the consensus that Rpl22A is the major paralog, reflecting the severity of growth defects in the individual rpl22 knock-out strains (the doubling times in YPD of rpl22aΔ, rpl22bΔ, and WT are 150.2 min, 90.2 min, and 90.7 min, respectively) [18]. When introns were deleted from the RPL22 genes, 50 DEGs were identified in rpl22aΔi, but only 2 DEGs in rpl22bΔi (upregulated RPL22B and downregulated RPL22A; Fig 7A). Interestingly, manipulating the expression of the two RPL22 paralogs had antagonistic effects on the transcriptome. These were partial, but significant, overlaps between DEGs upregulated in rpl22aΔ and DEGs downregulated in rpl22bΔ, and vice versa (Fig 7A). Furthermore, when we reviewed all 186 DEGs up-or downregulated in rpl22bΔ and examined their expression in rpl22aΔ, these genes typically showed antagonistic regulation between the two rpl22 deletion strains (Fig 7B and 7C). The RNA-Seq data confirmed the dramatic increase of RPL22B mRNA in rpl22aΔ cells, but did not reveal additional genes as candidates for direct Rpl22Adependent and intron-mediated inhibition.
We estimated the splicing efficiencies of endogenous RPL22 genes from our RNA-Seq data using a method described previously [75]. The splicing efficiency changes in the manipulated strains were in agreement with qPCR measurements and splicing reporter results. Low levels of Rpl22 protein in rpl22aΔ facilitated unrestrained splicing of RPL22B, as also evidenced by lower pre-mRNA levels (see S1 Fig). Interestingly, this was accompanied by changes in the proportions of two alternatively spliced minor isoforms, which we identified in the RNA-Seq data (Fig 8B). The 5'ss-alternative isoform was produced with higher frequency in WT (where the splicing of RPL22B was partly inhibited) than in rpl22aΔ (where the inhibition was lifted). The 3'ss-alternative isoform, which shares 5'ss with the major product, behaved similarly to the major product. While the existence of these alternative isoforms (both of which contain STOP codons) has been reported previously [82], their differential behavior with respect to Rpl22 inhibition is here documented for the first time.
To better understand the specific processes affected by the Rpl22 paralogs, we performed GO enrichment analysis of the respective sets of up-and downregulated DEGs (S9 Table). We found that ribosome biogenesis genes were enriched both among DEGs upregulated in rpl22aΔ and among DEGs downregulated in rpl22bΔ. Also, DEGs upregulated in rpl22bΔ or rpl22aΔi were enriched for genes involved in sugar metabolism (mostly trehalose synthesis) . Regions tested using the three-hybrid system are shown in color. The I1 region (green) represents nucleotides 11 to 123 of the RPL22B intron; I2 (red)-nt 165 to 236; I3 (blue)-nt 256 to 321. (B) Intron region 2 (I2) interacts with WT Rpl22A and Rpl22B but not with their RNA-binding mutants using the yeast three-hybrid system. RPL22A ("A") and RPL22B ("B") or their mutated versions ("Amut", "Bmut") in combination with different parts of the RPL22B intron were assayed for expression of the HIS3 reporter gene, which is activated in the presence of protein-RNA interaction. 10-fold serial dilutions of cells were spotted on plates with increasing concentrations of 3-aminotriazole (3-AT). "-U", "-L", and "-H" denote the lack of uracil, leucine, and histidine in the medium. IRE and IRP served as positive controls. p3HR2 is the empty plasmid for bait RNA expression and pACT2 is the plasmid for expression of the Gal4 activation domain.
https://doi.org/10.1371/journal.pone.0190685.g005 and in phosphate metabolism. Furthermore, rpl22aΔ DEGs were specifically enriched for genes involved in mitochondrion organization (downregulated) and amino acid metabolism (upregulated). On the other hand, the deletion of RPL22B led to the upregulation of stressresponse genes and the downregulation of factors important for ribosome biogenesis (different from those upregulated in rpl22aΔ; Fig 8A). The data suggest that the minor RPL22B acquired a paralog-specific role, as its deletion produces a distinct transcription phenotype at~WT total RPL22 levels.

Discussion
Deciphering the physiological significance of the RPG regulon of S. cerevisiae may contribute toward our understanding of how the underlying framework of the genome is used in evolutionary adaptations [83]. The selection of strains that best respond to the contradictory challenges of stressors and nutrients is perhaps the reason for the high retention rate of paralogs among yeast RPGs [3,84]. The panel of nuclear-encoded mitoribosomal protein genes does not contain ohnologs [85,86]. The number of proteins translated in mitochondria is limited but, perhaps more importantly, it was the adaptation to anaerobic environments, such as insect guts, which drove this ohnolog-based strategy [81]. RPG transcripts differ markedly in their untranslated regions and the nucleotide diversity in the 5'-UTRs of RPGs is higher in comparison to housekeeping genes [20]. Duplicated genes contribute more resistance to deleterious mutations not only because the two alleles can substitute for each other, but also because the regulatory network is more able to respond by rewiring [87,88]. More variability in regulatory regions means a more diversified protein interactome [89] and better adaptivity of the network as a whole [90]. Such adaptations may require paralog pairs that differ only in regulatory parts of genes or by one or two amino acids, as is the case for 42 of the 59 RPG duplexes. One exciting possibility is that paralogous genes generate proteins that are paralog-specifically modified. Their posttranslational modifications may be executed differentially based on differences in untranslated regions and/or translation efficiencies between paralogous mRNAs [91]. Remarkably, human RPGs cannot substitute for their yeast counterparts [92] despite the high degree of amino acid sequence conservation, which suggests the existence of yeast-specific networkbased adaptations. While the high retention rate of introns attests to their importance for organisms, the underlying reason for this retention warrants further research.

Intergenic regulation of RPL22A/B is splicing-mediated and asymmetric
We monitored mRNA and pre-mRNA levels in strains with various combinations of intron and whole gene deletions in RPL22A and RPL22B paralogs (Fig 1). While RPL22A mRNA ranged from 160% to 60% of WT, the amplitude for RPL22B was much larger (500% to 5% of RPL22B WT), which suggests that RPL22B concentration provides the "readout" for regulatory purposes. Our results are in agreement with the research carried out by Gabunilas and Chanfreau [59] and were obtained independently of their results using partially different methods.
Proteomic analysis of the manipulated strains shows here for the first time that Rpl22A/ Rpl22B protein levels match the ratios of corresponding mRNAs (Fig 1C). Using splicing reporters (Fig 1) as well as knock-ins of chimeras in the RPL22 loci (Fig 4), we demonstrate that splicing of the RPL22B intron is stringently regulated by Rpl22 levels, whereas the RPL22A reporter is much less sensitive (Fig 1). The reporters contained endogenous intron sequences expression of KlRPL22 as well as RPL22A and RPL22B in K. lactis. cDNA was prepared using random hexamers and oligo(dT) 18-mers. The statistical significance of the difference between empty vector strain and an overexpression strain is indicated as (1) for P 0.05, (2) for P 0.01, and (3) for P 0.001 based on the t-test with Holm correction for multiple testing (see Methods).
https://doi.org/10.1371/journal.pone.0190685.g006 Significance was determined by one-sided Wilcoxon signed rank test (P value is given below each rpl22aΔ panel). Strain genotypes are indicated below the boxplots. The RPL22B gene itself was excluded from this analysis (right panel). (C) The DEGs downregulated in rpl22bΔ as compared to WT (blue data points) are less affected in rpl22aΔi than in rpl22bΔ, however their expression changes in the same direction in both mutants. The fold change versus WT of this group of as well as parts of 5'-UTR, exon 1, and part of exon 2. These results complement the data of Gabunilas and Chanfreau [59], where constructs used to measure splicing lacked alternative 5'ss and contained only intronic sequences. Our results are also in agreement with the conclusion that RPL22A/B regulation occurs at the pre-mRNA splicing step. This does not exclude the possibility that, under stress or under conditions affecting growth, more complex regulation (involving translation efficiency [55], or protein stability) takes place.
In exponentially growing WT cells, splicing of both paralogs is apparently subject to negative regulation by Rpl22 (Figs 1 and 3). Therefore, free Rpl22 concentration must be well above dissociation constant for RPL22B intron binding. Under conditions of balanced RP production and ribosome biogenesis, Rpl22 concentration in the nucleoplasm is determined by the pool of free Rpl22 [93] and by the affinities of Rpl22 for its binding sites on large subunits or their maturation stages. A sudden excess of free Rpl22 would mildly inhibit RPL22A splicing and further lower the expression of Rpl22B (i.e., the B/A ratio would become very low). According to our findings, this is the situation which affects the transcription of genes involved in ribosome biogenesis (Fig 8). It is possible that the fluctuations in free Rpl22B concentrations help to fine-tune ribosome assembly during disturbances caused by nutrition or environmental changes.
Intron mediated-regulation is insensitive to Rpl22A/B amino acid differences and both Rpl22 proteins may bind to a structured region between 5'ss and BP in the RPL22B intron when assayed in the yeast three-hybrid system (I2; Fig 5). In contrast, the 3'ss-proximal hairpin structure in the RPL22A intron (S2 Fig) had no role in splicing regulation (S1 Fig). In all instances documented thus far, the extraribosomal functions of Rpl22 (see Introduction) are mediated through RNA binding, most likely involving secondary transcript structures and the rRNA-binding domain of the protein. Human Rpl22 has been shown to bind to a broadly defined short hairpin loop (5'-NNNNG (N) 7 CUNNN-3') [94], which it can target within its own mRNA, the intron of Smad2, or the human Epstein Barr virus, EBER-1 RNA [48][49][50]. The secondary structures mediating Rpl22 inhibition in the RPL22B and RPL22A introns warrant further study. There is no obvious similarity to the rRNA structure or the SELEX motif identified for the human protein [94,95]. However, given the three-hybrid results (Fig 5) and the fact that RNA-binding mutants lack inhibitory potential (see Figs 3 and 4), there is a strong indication that intron RNA forms a structure that can be recognized by Rpl22. The situation may be more complex, as the structure can be complemented by other proteins or RNA, such as parts of the intron. Alternatively, it may comply with Rpl22-binding requirements, despite the lack of any apparent sequence similarity. Intriguingly, the intron-Rpl22 interaction specifically inhibited usage of the major, but not the minor, 5'ss ( Fig 8B).
We observed~4-8-fold increases in RPL22A/RPL22B pre-mRNA levels in the mutants defective for the cytoplasmic NMD pathway and 5'-3' degradation (Fig 2, S3 Fig). This indicates that RPL22 pre-mRNAs are unable to compete for the splicing machinery in the presence of Rpl22. The data also imply that the inhibitory complexes containing the pre-mRNAs are stable until the time the pre-mRNAs are subjected to NMD in the cytoplasm. Our observations extend the data of Gabunilas and Chanfreau [59]. Unlike these authors, we observed no tendency for an increase of RPL22 mRNA in the XRN1 mutant, which may perhaps reflect the cultivation conditions-dependent availability of free Rpl22.
We summarize that splicing inhibition of RPL22A/RPL22B is operational at both introns, albeit to a different degree, and that it is dependent on Rpl22 RNA-binding propensity but DEGs thus correlates with the negative change of Rpl22B expression (0.1% in rpl22aΔi to zero in rpl22bΔ) but not with the change of total Rpl22 (see the right panel). The RPL22A and RPL22B genes themselves were excluded from this analysis.
https://doi.org/10.1371/journal.pone.0190685.g007 The upper plot shows read counts for each spliced RPL22B product, normalized to sequencing library sizes (RPM, reads per million). The lower plot shows independent of the differences in 19 amino acids between the Rpl22A and Rpl22B proteins. These results both support the conclusions drawn by Gabunilas and Chanfreau [59] and further extend their findings. We were able to demonstrate the effects of ectopically produced Rpl22A/B proteins in the rpl22aΔbΔ background. We also showed that RNA-binding mutant versions of Rpl22 are devoid of regulatory activity. Using the yeast three-hybrid system, we located Rpl22 binding to a region which overlaps with the fragment described by Gabunilas and Chanfreau [59].

Intron-dependent regulation of S. cerevisiae RPL22 most likely evolved from autoregulation
Kluyveromyces lactis is part of the "Saccharomyces complex" (which did not undergo WGD) and has a single-copy RPL22 homolog [8]. The alignment of RPL22A, RPL22B, and KlRPL22 shows that all three introns share a common motif (S2 File, underlined), which is also present in the RPL22 introns of other Saccharomycotina species (not shown; [59]). This motif (i) lies partly within the I2 construct that interacts with Rpl22 in the three-hybrid system ( Fig 5) and (ii) borders the element important for splicing inhibition identified by Gabunilas and Chanfreau [59]. We assayed the regulatory properties of the K. lactis intron and protein in both S. cerevisiae and K. lactis. When the K. lactis intron was introduced together with exon 2 into the S. cerevisiae RPL22A locus, its splicing was repressed at high Rpl22 protein levels similar to the endogenous RPL22A intron. We also showed that while the KlRpl22 protein was capable of regulating RPL22A and RPL22B in S. cerevisiae (Fig 6A and 6B), Rpl22A/B as well as KlRpl22 inhibited splicing of KlRPL22 in K. lactis (Fig 6D). These results prove that K. lactis RPL22 may be regulated by an intragenic loop, whereby the KlRpl22 protein level modulates splicing of its own transcript. If we assume that a similar situation existed in the ancestor of S. cerevisiae before WGD, then, hypothetically, such autoregulation would have become intergenic and asymmetric after WGD (see Fig 8C). Therefore, the basis of the RPL22 (auto)-regulatory mechanism is older than the difference between the S. cerevisiae Rpl22 proteins. We contend that Rpl22 acquired its intron-dependent function in the lineage ancestral to the "Saccharomyces complex".
Rpl22A and B differ from KlRpl22 by 14 and 23 amino acids, respectively, and from each other by 19 amino acids, with most of the differences in the C-terminal part. The differences between A and B seem not to affect the capacity of the protein to inhibit splicing of either intron (see Figs 1,3 and 4). KlRpl22, which is more similar to Rpl22A, is also capable of recognizing S. cerevisiae pre-mRNAs. The degree of inhibition of KlRPL22 splicing is more similar to the responsiveness of RPL22A (to~40%). We speculate that after WGD the expression of the genes eventually became asymmetric, turning RPL22B into a minor paralog. We also hypothesize that splicing-mediated regulation might have contributed to the establishment of the asymmetry. Apparently, only after the duplication was a stringent inhibition (of one of the introns) enabled. It is notable that both RPL22A and RPL22B contain the non-canonical 5'ss sequences, GTATGA and GTACGT, respectively, which are conserved across the S. cerevisiae sensu stricto group. The singleton KlRPL22 contains a canonical site. A similar situation exists in the meiotic MER1 regulon, where Mer1-regulated genes retain non-consensus splice sites the percentage of each spliced isoform. Data from pooled samples, adjusted for the number of biological replicates, are shown (n WT = 4, n rpl22a = 3). (C) Model summarizing the regulatory interrelationship between RPL22 paralogs. We assume that both Rpl22 proteins can inhibit splicing of either RPL22A or RPL22B pre-mRNA. Splicing of RPL22B pre-mRNA is much more sensitive to the inhibitory effect of Rpl22. This model represents the situation in WT cells, where Rpl22A is the major Rpl22 protein. Thus, its level contributes considerably to the splicing regulation of the minor RPL22B paralog but also partially to controlling its own expression. The Rpl22B level is low but can still provide minor regulation of its own as well as Rpl22A production.
https://doi.org/10.1371/journal.pone.0190685.g008 across the species of the group [96]. Another feature of asymmetry between RPL22 paralogs is the presence of two Rap1 binding sites in their 5'-regulatory region. RPL22B apparently can only use one, as its 3'-proximal site has a C/A mutation of one of the two key cytosines within the binding consensus (CACCCATACAT [97]).
The regulation of KlRPL22 poses a question whether the K. lactis singletons corresponding to the other regulated RPG pairs are always regulated in autogenic fashion. We intend to test this possibility in future experiments. While RPL22A/B pair and its kin might represent autogenic regulation turned intergenic, other RPs might have acquired their extraribosomal roles (i.e., RNA structures as binding partners) only after WGD. RPL2A (see Fig 1A), which seems to be regulated in a paralog-specific autogenic mode, is a candidate for such a scenario.

Genome-wide effects of RPL22 manipulations
We were intrigued by the possibility of other nonribosomal-RNA regulatory targets for RPs [49]. Rpl22 may regulate the splicing of other introns, such as those of other RPGs, in addition to acting on its own gene and paralog. We therefore decided to compare transcriptome changes in situations with differing paralog ratios. While we did not find changes that would indicate additional intron-containing regulatory targets of Rpl22, our data illustrate the requirement of Rpl22 for ribosome biogenesis and also include DEGs involved in metabolism. They are also in partial agreement with previously reported Affymetrix expression microarray profiles of rpl22aΔ that include amino acid and carboxylic acid metabolism among the upregulated DEG categories [22]. Previously, Rpl22 has been implicated in cell-fate decisions in response to environmental signals [55]. Cells with reduced Rpl22 levels have been shown to exhibit a reduced 60S:40S particle ratio and accumulated "halfmers" on polysomes, which indicates that mRNA-bound 40S particles lack their 60S counterparts [58]. RPL22A is also among the RPs that affect the replicative lifespan and confer resistance to tunicamycin-induced ER stress [18].
Deletion of RPL22B, which leaves cells with A-only Rpl22 at the~WT mRNA level, up-/ downregulated 122/63 DEGs. We found this group of transcripts very interesting because they represent a response to the removal of the Rpl22B paralog without deviating from the~WT Rpl22 level (Fig 1). The downregulated DEGs exhibited factors involved in the biogenesis of both ribosomal subunits, rRNA metabolism, subunits of RNA Pol I and III, nucleotide salvage and synthesis pathway genes, tRNA modification factors, and factors involved in proteosynthesis. The upregulated DEGs contained genes that are involved in stress response and metabolism, including nucleotide and carbohydrate metabolism. RPL22A intron deletion, which downregulates RPL22B and upregulates total RPL22 concentration, yielded a substantial overlap with the rpl22bΔ phenotype ( Fig 7A). Our data thus suggest that Rpl22B assumes a paralog-specific role that affects ribosome biogenesis.
RPG pairs coding for different protein isoforms can be used to modify ribosome composition and function (see Introduction). RPL22 appears as a pair of paralogs in unrelated taxa analogically to, for example, RPS9 [13]. In such cases, subfunctionalization involves proteins as well as UTRs and introns. The differences can affect Rpl22 functions during ribosome biogenesis and/or during translation within the large ribosomal subunit [22,25,58] or they may be related to the extraribosomal tasks of Rpl22 proteins.
Gabunilas and Chanfreau report that MMS-induced DNA stress or disruption of rRNA transcription causes splicing-mediated downregulation of RPL22B mRNA (and Rpl22B production) [59]. Such effects downregulate ribosome biogenesis, which should increase free Rpl22 levels and thus RPL22B inhibition (see the model in Fig 8C). The early response to oxidative stress may also involve changes in translation efficiency of specific transcripts that are enriched for TTG (Leu) codons [55], including RPL22A. The translation effect is brought about by Tmr4 methyltransferase, which produces tRNA LEU(CAA) with methylated C at the wobble position. This modification specifically potentiates anticodon-codon binding, thus enhancing the translation of some, albeit not all, TTG-enriched mRNAs [55,98]. While RPL22A is TTG enriched, RPL22B is not. In fact, RPL22B is the most distant codon-usage outlier of the cytoplasmic RPGs [99]. Hypothetically, the Rpl22A/B ratio may thus predetermine the amplitude and character of the reaction to oxidative stress. The immediate-early translation response requires RPL22A (and further enhances the excess of A over B). This should synergize with the inhibition of biogenesis (see Fig 8A). Restarted ribosome biogenesis should lead to decreased levels of RPs (including Rpl22) in the nucleoplasm, eventually unblocking RPL22B splicing. Increased Rpl22B should then further enhance ribosome biogenesis.
The extraribosomal roles of RPs [18,100] are far from widespread and not all RPs are equally suited for their tasks. Why is it that Rpl22 (or Rpl30, Rps9, among others [13,41]) binds its own transcript and regulates its splicing whereas other (RNA binding) RPs seem not to do so? We argue that due to the heavy constraint on the variability of the protein with respect to its ribosomal function, it must have been the variability on the RNA level which gave rise to Rpl22 binding structures in the transcriptome. This variability should be similar for all RPGs, however. Perhaps the discerning factor was the position of the protein in a key node that regulates ribosome biogenesis [4,101,102]. Such an intron-RP interaction would have been more likely to be fixed in cell population once its impact on splicing improved fitness [19,20]. Alternatively, it may have been the availability of the protein in the nucleoplasm, because of the association of the protein with the ribosome on its periphery or at a lower affinity, which made it available to sample other interactions (or "evolutionary opportunities" [103]). Evidence for the extraribosomal roles of Rpl22s in divergent taxa throughout evolution perhaps illustrates the exploitation of such opportunities [49,50,104]. It would be of interest to uncover more examples and to learn what predisposes their (re-)occurrence in organisms. Moreover, the regulated splicing of RPL22A/RPL22B (see Fig 8C) may serve as a model for establishing a regulatory relationship between two paralogous genes in evolution.  [2] or deletions of components of the cytoplasmic 5'-3' RNA degradation pathways, and combinations of these mutations affect mRNA levels of RPL22A, RPL22B, ECM33 and COF1 genes only to a limited extent (with the exception of xrn1Δ, which lowers the mRNA level by up to 50% in some cases). (B) The extent of RPL22A and RPL22B pre-mRNA accumulation in strains mentioned above differs from that of ECM33 and COF1 genes. mRNA and pre-mRNA levels were measured by RT-qPCR with primers spanning exon-exon junctions or exon-intron junctions, respectively. Data were normalized to TOM22 gene and expressed as mean fold change relative to WT level (indicated by a horizontal line). Error bars represent s.d. of two biological replicates for all RPL22-related analyses and for ECM33-and COF1-related analyses in xrn1Δ and xrn1Δ prp45  strains. In the remaining ECM33-and COF1-related measurements, error bars represent s.d. of 3 biological replicates. The statistical significance of the difference between WT strain and either prp45(1-169) strain or pooled data (indicated by horizontal bolts) of NMD mutants or NMD prp45