In this study, we present experiments that lead to four main conclusions: First, we show that VIP3 is the bona fide SKI8 homolog of Arabidopsis; second, we demonstrate that next generation sequencing of random-primed RNA samples with short reads can be used to estimate the turnover of mRNA transcripts; third, we show that the phenotypic aspects of VIP3 function in Paf1c and the SKI complex can be separated; and fourth, we provide evidence that the dual role of SKI8 homologs in Paf1c and the SKI complex appears to depend on the species-specific cellular context.
Our interest in VIP3
originates from the discovery of the zwg
mutant that segregated in the Arabidopsis Sav-0 accession. It seems unlikely that the 7 bp deletion in vip3zwg
represents indeed an allelic variant recovered from a natural environment because of its detrimental phenotypic consequences. A haplo-insufficient beneficial effect of vip3zwg
could explain maintenance of the allele by balancing selection, however, we did not observe any obvious phenotypes in the heterozygous plants that would support this idea. Rather, it appears likely that vip3zwg
is a spontaneous allele that has arisen during the propagation of the Sav-0 accession in stock centers starting in the 1960s over several decades 
At the outset of our study, it was still unclear whether VIP3
is indeed the Arabidopsis SKI8
homolog. While its role in epigenetic regulation of FLC
transcription through association with the Paf1c complex had been well documented 
, its potential role in the SKI complex had not been characterized. Because of the evolutionary distance between higher plants, yeast and mammals this could not be considered a given, in particular as VIP3 and SKI8 fall into an abundant class of structurally similar WD40 repeat proteins. This was underlined by the finding that ScSki8 is by far not the closest VIP3 homolog in yeast. For instance, position-specific iterated BLAST identifies more than two dozen yeast proteins that are more homologous to VIP3 than ScSki8 (e.g., a 93.2 score, 69% coverage and 5×10−24
e-value for PRP4 as compared to 48.9 score, 40% coverage and 5×10−8
e-value for ScSki8). It is only our functional analyses that suggest that VIP3 is indeed the bona fide
ScSki8 homolog. Consistent with a potential role in both Paf1c and the SKI complex, we found that VIP3 is present in both the nucleus and cytoplasm, and in at least two protein complexes of distinct size. The larger peak fractions around 690 kDA could represent Paf1c, whereas the peak around 300 kDa could represent the SKI complex 
. A third peak around even smaller (100–200 kDa) size fractions could represent partial components of these complexes or VIP3 dimers, which might accumulate in excess as the GFP-VIP3
transgenes were typically expressed at higher levels than endogenous VIP3
. Interestingly, immunoprecipitation of GFP-VIP3 after gel filtration identified association with the Arabidopsis SKI3 homolog, but not the SKI2 homolog. This might mean that the SKI complex dissociates into sub-components during gel filtration and/or that SKI2 is lost during immunoprecipitation washes. Alternatively, it could reflect the fact that SKI8 interaction with SKI3 is direct, while interaction with SKI2 is indirect 
. However, when directly immunoprecipitated from total protein extract, AtSKI3 as well as AtSKI2 was pulled down in our stringent conditions, underlining that VIP3 is indeed part of the SKI complex.
The notion that VIP3 is a functional subunit of the SKI complex is supported by our genome-wide analysis of mRNA stability in vip3zwg
mutants. To estimate mRNA turnover was foremost a technical challenge, because it meant that standard cDNA synthesis using poly-T oligonucleotides directed against the 3′ poly-A tail of mRNAs could not be applied. This also abolished the inherent selection of the mRNA fraction for sequencing from the much larger amount of ribosomal or transfer RNAs. Instead, to also capture mRNAs undergoing 3′ to 5′ degradation, cDNA was synthesized with random-priming, and the mRNA fraction was enriched by removing ribosomal RNAs through capture columns. High throughput sequencing of the cDNA samples and subsequent read mapping onto the reference transcriptome revealed that our method efficiently enriched the mRNA fraction, which generally represents 1–2% in total RNA samples, about 5 to 10-fold. The relative read abundance along transcripts is to some degree determined by technical biases, such as the directionality of cDNA synthesis 
. However, it should also reflect the steady state equilibrium between mRNA synthesis and breakdown considering that primers were not limiting in cDNA synthesis and that poly-A tails provide priming sites but are not included in the sequence analysis. Generally, the coverage profiles displayed a decrease from 5′ to 3′, suggesting that exosome-mediated 3′ to 5′ degradation is the main driver of mRNA breakdown 
. To quantify the stability of individual transcripts, we defined a 5′ to 3′ coverage index, which was generally >1, consistent with the overall profile. The comparison of the 5′-most 20% of a transcript versus its 3′-most 20% was designed to avoid skewed values in the case of poorly covered transcripts, and indeed comparatively few outliers were observed. In some cases, these reflected obvious mismappings because of repetitive or redundant sequences (e.g. retrotransposon borders), while in others mismapping might have occurred because of the relaxed stringency that was required to map mRNA sequences from a divergent accession onto the reference transcriptome 
. Overall, the patterns as well as the quantitative difference between the wild type and vip3zwg
samples were robust, even if more selective criteria were applied or if other indexes were considered, such as linear fitting of read coverage. Thus, the index values suggest that in the vip3zwg
sample the relative abundance of intact 3′ ends as compared to 5′ ends is higher, pointing to a shifted steady state equilibrium between mRNA transcription and degradation. This finding is consistent with the generic role of the SKI complex in exosome activation 
and was particularly evident in the group of the most prominent exosome targets, termed the “hidden transcriptome” 
. In summary, our data support the idea that VIP3 is a SKI complex component that affects mRNA stability and that random-primed RNA-Seq is a valid approach to estimate mRNA turnover.
The implication of VIP3 in the SKI complex suggests that the vip3 phenotype should reflect the combination of VIP3 function in both Paf1c and the SKI complex. The availability of a mutant in the AtSKI2 gene, which can be unequivocally identified by homology searches, enabled us to disentangle the two activities. Interestingly, Atski2 plants displayed dwarfism, but neither early flowering nor aberrant flower development. Thus, the latter aspects of the vip3 phenotype should primarily result from impaired Paf1c function. It is noteworthy however that the Atski2 dwarf phenotype is not as severe as in vip3, and that growth defects have also been observed in mutants of other Paf1c components. It thus appears likely that the SKI complex-related growth defects in vip3 are aggravated by the additionally impaired Paf1c activity.
To clarify more directly which portions of the vip3
phenotype are attributable to impaired Paf1c or SKI complex function, we sought to exploit the fact that ScSki8 does not associate with Paf1c in yeast 
and presumably also not in Arabidopsis. However, to our surprise ScSki8 was able to fully rescue all aspects of the vip3
phenotype. Thus, it appears that in the cellular context of Arabidopsis, ScSKI8 can fulfill VIP3's role in Paf1c. This could mean that other factors determine whether SKI8 is recruited to Paf1c or not, and that in this sense Arabidopsis is closer to mammals than yeast. Indeed we also tried to complement vip3zwg
by constitutive expression of the mouse SKI8 homolog, WDR61. However, for unknown reasons, we never managed to recover transgenic plants in repeated transformation attempts, which could mean that WDR61 expression is poisonous for Arabidopsis. Thus, while cellular context must play an important role, SKI8 function might to some degree also depend on inherent features. Future experiments to determine the interaction patterns of different SKI8 homologs and derivative point mutants of interest are a promising avenue to clarify this issue in detail.