Mitochondrial-Nuclear DNA Interactions Contribute to the Regulation of Nuclear Transcript Levels as Part of the Inter-Organelle Communication System

Nuclear and mitochondrial organelles must maintain a communication system. Loci on the mitochondrial genome were recently reported to interact with nuclear loci. To determine whether this is part of a DNA based communication system we used genome conformation capture to map the global network of DNA-DNA interactions between the mitochondrial and nuclear genomes (Mito-nDNA) in Saccharomyces cerevisiae cells grown under three different metabolic conditions. The interactions that form between mitochondrial and nuclear loci are dependent on the metabolic state of the yeast. Moreover, the frequency of specific mitochondrial - nuclear interactions (i.e. COX1-MSY1 and Q0182-RSM7) showed significant reductions in the absence of mitochondrial encoded reverse transcriptase machinery. Furthermore, these reductions correlated with increases in the transcript levels of the nuclear loci (MSY1 and RSM7). We propose that these interactions represent an inter-organelle DNA mediated communication system and that reverse transcription of mitochondrial RNA plays a role in this process.


Introduction
Mitochondria have a central role within the metabolic systems of cells. In yeast (Saccharomyces cerevisiae), as in other organisms, the mitochondrial organelle contains a genome that encodes an essential subset of the electron transport chain components [1] that are necessary for respiratory growth [2].
The mitochondrial genome has drastically reduced in size during the course of evolution to the point that ,98% of the genes required for mitochondrial function are encoded within the nuclear chromosomes [3]. Consequently, mechanisms must exist to coordinate and control the expression of the nuclear-and mitochondrial genome-encoded genes in order to maintain and control mitochondrial function [4,5]. Intriguingly, despite the fact that the majority of mitochondrial genes have transferred to the nuclear genome, transfer of mitochondrial DNA (mtDNA) to the yeast nucleus remains an on-going process with mtDNA being used to repair double stranded breaks in yeast nuclear chromosomes under certain conditions [6]. Unstable mitochondrial plasmids have also been observed to transfer into the yeast nucleus [7,8] in a nuclear gene (e.g. YME1, YME2) dependent manner [8][9][10]. The nuclear functions of these transferred mtDNAs are unknown, however elevated mitochondria to nucleus DNA migration rates correlate with accelerated chronological aging in yeast [11].
Distal regulatory regions (e.g. enhancers) are known to loop within chromosomes in order to interact with the promoter region of the genes that they control [12]. Furthermore, enhancers can also interact with promoters on different chromosomes to control gene expression [13,14]. These types of inter-and intrachromosomal interactions can be captured using proximity-based ligation methodologies (e.g. Chromosome Conformation Capture (3C) [15]) that incorporate high resolution (i.e. ,2 Å [16]) crosslinking of interacting DNA strands, restriction digestion, dilution, and ligation to identify DNA sequences that interact within a cell. Using a proximity-based ligation method that we developed to observe the global set of genome wide interactions (Genome Conformation Capture (GCC)), we previously observed that nucleic acids of mitochondrial origin interact with nuclear loci (hereinafter referred to as Mito-nDNA interactions) in S. cerevisiae [17]. Surprisingly these inter-organelle, Mito-nDNA interactions are frequent and statistically significant suggesting that they perform a hitherto unrecognized role within yeast cells [17]. Furthermore, analysis of one of these interactions demonstrated carbon source dependence [17]. Intriguingly, the quality and quantity of mitochondrial DNA has been shown to affect patterns of nuclear transcription [18,19] and replication [20] in yeast.
In this study we explore the hypothesis that inter-organelle interactions respond to the metabolic status of the cell to regulate nuclear transcript levels. Using GCC we identify dramatic differences in both the frequency and identities of inter-organelle interactions occurring in S. cerevisiae during growth on glucose, galactose (i.e. respiro-fermentation [21,22]), and glycerol lactate (i.e. solely respiration). We also demonstrate that Interactions between mitochondrial genes (i.e. COX1 and Q0182, a dubious mitochondrial ORF) and nuclear encoded loci (i.e. MSY1 and, RSM7, respectively), are dependent upon a functional electron transport chain and mitochondrial encoded reverse transcriptase machinery. Finally, the levels of the nuclear encoded MSY1 and RSM7 gene transcripts are increased when the interaction frequency is reduced by the knockout of mitochondrial reverse transcriptase activity. On the basis of these results we propose that reverse-transcription mediated inter-organelle DNA interactions are a novel form of communication between mitochondria and the nucleus.

Results
We previously captured Mito-nDNA interactions in S. cerevisiae cells grown in glucose by GCC [17]. A detailed investigation of one of these Mito-nDNA interactions (between the COX1 gene (Mt: 24872-26193 bp) and the nuclear encoded MSY1 gene (Chr XVI; 365496-365760 bp), herein after denoted COX1-MSY) [17] demonstrated that it was carbon source dependent. Therefore, we hypothesized that Mito-nDNA interactions would alter, on a global scale, according to the cell's metabolic state, and in particular, the carbon source used for growth. Thus, we used GCC to generate comprehensive maps of the Mito-nDNA interactions in S. cerevisiae during exponential growth in media containing glucose, galactose, or glycerol lactate. Two biological replicates were prepared and analyzed for each condition. Interaction networks were constructed from 36  Statistical and experimental methods were used to determine if the Mito-nDNA interaction patterns could have been generated by experimental noise alone, which would be expected to produce random pairings of fragments from the two genomes. In silico simulations (100,000) were performed [17] to determine the maximum count of a particular interaction that would be observed under this random noise model, given the same number of sequences, interactions and fragments as in the experimental data. These results showed that the real dataset deviates from a random distribution and, therefore, we conclude that the interaction patterns cannot be attributed to noise alone under any of the conditions, in each case with a p-value less than 10 25 . Subsequently, we performed analyses to determine what frequency individual interactions have to achieve before they are deemed to be present at a level above experimental noise (Methods S1). As a result, we identified 8678 statistically significant interactions occurring between the nuclear and mitochondrial genomes during glucose growth, 1780 during galactose growth, and 8153 during growth in glycerol lactate ( Table 1). The numbers of interactions in each condition did not correlate with the measured mitochondrial copy number (Table S1). Biological replicates for each condition were highly correlated for statistically significant interactions (R 2 = 0.78, 0.93, 0.93, respectively; Figure S1 and Methods S1). Accordingly, sequences from biological replicates were combined and reanalyzed.
To experimentally control for spurious inter-molecular ligation events during the GCC process, samples were spiked with two ligation controls during library preparation. The first ligation control consisted of PCR products amplified from Escherichia coli or Lambda bacteriophage (Table S2) that were added (1:1 ratio with the nuclear genome copy number) before the GCC ligation step. These controls were designed to estimate the frequency of random inter-molecular ligation events during GCC library preparation. A maximum of 47 separate ligation events were observed, none of which occurred at levels above the statistically defined experimental noise threshold. The second ligation control consisted of the addition of pUC19 plasmid to the sample following the GCC ligation in order to control for random ligation events during highthroughput sequencing library preparation. We observed a maximum of six interactions between pUC19 and the rest of the genome; again none of these interactions were above the statistically defined experimental noise threshold. These controls show that random inter-molecular ligations occur at very low frequencies that are below our noise threshold for significant interactions. This is true even for the high copy number rDNA and mitochondrial DNA elements and thus provides empirical evidence that random ligation events during sample preparation do not account for the interactions we observe.
Significant interactions were separated into two pools, those which occur between the mtDNA and the nuclear ribosomal DNA repeats (Mito-rDNA), and those between mtDNA and unique nuclear loci (Mito-nDNA; Table 1). The rDNA repeats form part of the nucleolus and encode the rRNA component of the cytosolic ribosomes. The rDNA repeats constitute ,9.8% of the yeast genome; yet, the Mito-rDNA interactions constitute 95.8%, 52.4%, and 84.5% of the total interactions between the nuclear and mitochondrial genomes in glucose, glycerol lactate, and galactose, grown cells, respectively. There does not appear to be an interaction ''hotspot'', with Mito-rDNA interactions evenly spread across the 9.1 kb rDNA repeat (data not shown). Hence, Mito-rDNA interactions are over-represented within the data-set and are carbon source dependent (Table 1). We also observed considerable alterations to the numbers of Mito-nDNA interactions. Moreover, the mitochondrial regions that are involved in interactions with the nDNA are not uniformly distributed across the mitochondrial genome ( Figure S2).
In order to determine whether the observed changes in Mito-nDNA interactions are chromosome specific, we asked whether nuclear chromosome length correlates with the number of interactions identified for each individual chromosome. The number of Mito-nDNA interactions per nuclear chromosome is highly correlated with chromosome length in the glycerol lactate condition, but not in glucose or galactose ( Figure 1). This discrepancy is mainly due to the deviation of chromosome X from the trend during growth in glucose and galactose. Intriguingly, the increase in mtDNA interactions with chromosome X is accounted for by a single nuclear MspI fragment that There was a .10 fold increase in the number of Mito-nDNA interactions during growth in glycerol lactate (respiration) as compared to growth in glucose and galactose (respiro-fermentation). Growth on galactose resulted in less Mito-nDNA and Mito-rDNA interactions combined, compared to the other two conditions. Statistically significant DNA-DNA interactions were divided according to whether the mtDNA was interacting with nuclear rDNA, or with unique nuclear loci. Corrections for the copy numbers of the rDNA repeats and the mitochondrial genome were incorporated into the significance calculations (Methods S1). doi:10.1371/journal.pone.0030943.t001 encompasses the promoter region and part of the coding sequences of two divergent ORFs: one uncharacterized ORF (YJR115W), and RSM7 which encodes a mitochondrial small subunit ribosomal protein. Numerous mtDNA MspI fragments, including fragments surrounding or overlapping the COX1, COX3, VAR1 and SCE1 genes, interact with this one region on chromosome X. Yeast mitochondrial escape mutants (YME) [8] have been previously implicated in an elevated rate of transfer of unstable mitochondrial plasmids to the yeast nucleus [7][8][9][10]. Therefore, we predicted that the YME pathway was the source of mtDNA fragments interacting with the nuclear genome, and that mutations within this pathway would result in an increase in the frequency of inter-organelle DNA interactions. To test this prediction, we used quantitative 3 C to compare the frequency of the strong COX1-MSY1 interaction (identified in [17]) in S. cerevisiae YME knockout mutants (i.e. Dyme1, Dyme2). Contrary to expectations, we observed a significant decrease in the frequency of the COX1-MSY1 interaction in the Dyme1 strain as compared to the wild-type (Ttest [Paired P(T, = t) one-tail, n = 4] p = 0.010; Figure 2). Deletion of YME2 or a functionally unconnected nuclear gene (ADE2) did not significantly affect the COX1-MSY1 interaction frequency (t-test [paired P(T, = t) one-tail, n = 4] p = 0.377 and 0.103 respectively; Figure 2). These results suggest that the source of the mtDNA that participates in the Mito-nDNA interactions is not the unstable mitochondrial plasmids that were previously identified as escaping the mitochondria for the nuclear compartment.
Deletion of YME1 results in an elevated rate of mitochondrial turn-over as well as an abnormal globular mitochondrial morphology [9,23]. Therefore, it is possible that this fragmented mitochondrial phenotype contributes to the reduction in the COX1-MSY1 interaction frequency we observed in the yme1 deletion strain. To test this we arrested yeast cells with a-factor, which results in a fragmented mitochondrial network [24] that is phenotypically similar to the one observed in yme1 deletion strains [23]. We also deleted the mitochondrial fission gene (MDV1) to create strains that are unable to correctly fragment mitochondria [25,26]. Interestingly, we observed a similar reduction in the COX1-MSY1 interaction frequency upon a-factor induced synchronization ( Figure S3). However, the interaction frequency measured in the Dmdv1 strain was intermediate between that observed for the wild-type and Dyme1 strains, and not significantly different from either (T-test [Paired P(T, = t) one-tail, n = 4] wt-Dmdv1 p = 0.143, Dmdv1-Dyme1 p = 0.210; Figure 2). Therefore, it is unlikely that mitochondrial fragmentation is directly responsible for the observed changes in COX1-MSY1 interaction frequency.
We next postulated that an abnormal mitochondrial morphology, coupled with elevated mitochondrial turnover would result in a disturbance of the mitochondrial ATP synthesis pathway, and this may explain the observed reduction in the frequency of the COX1-MSY1 interaction. Therefore, we tested the inter-organelle COX1-MSY1 interaction for ATP dependence by treating yeast cells with an electron transport chain uncoupling agent, 2,4-Dinitrophenol (DNP), at a concentration (5 mM) that inhibits respiration but allows fermentation ( Figure S4). We observed a significant time-dependent decrease in the frequency of the COX1-MSY1 interaction in the presence of DNP (t-test p,0.05; Figure 3A), as measured by quantitative 3 C. However, an interaction between two nuclear loci (nDNA-nDNA; Chr VII: 868673-873686 bp -Chr IX: 172565-173311 bp) was also shown to be affected by treatment with DNP ( Figure 3B). The observed dependence of the nDNA-nDNA interaction on a proton gradient across the mitochondrial membrane, and thus mitochondrial ATP synthesis, suggests that formation of these DNA-DNA interactions is ATP dependent.
The region of the COX1 gene involved in the COX1-MSY1 interaction overlaps the non-essential group II mitochondrial aI5c intron. There are four Group II introns present within yeast mitochondria (aI1, aI2, bI1, and aI5c) and these introns encode functional splicing, reverse transcriptase, or endonuclease machinery [27][28][29][30][31]. Only the aI1 and aI2 introns encode reverse transcriptase [32] activity while the aI5c intron encodes endounuclease activity [33,34] but not reverse transcriptase activity [35]. Therefore, we postulated that reverse transcription of the mitochondrial group II introns might be involved in the COX1-MSY1 interaction. To test this we measured the COX1-MSY1 interaction using quantitative 3 C on a strain which only contained the COX1 aI5c intron (i.e. 161-U7 GII-0 aI5c; Figure 4A). We observed a 40-60% decrease in the inter-organelle COX1-MSY1 interaction in the GII-0 aI5c strain relative to the wild type ( Figure 4B). We concluded that the partial dependence of the COX1-MSY1 interaction upon the presence of the group II introns reflects a role for reverse transcription in the interorganelle interactions. However, part of the COX1-MSY1 interaction remains independent of mitochondrial encoded reverse transcriptase.
It remained to be seen if interactions involving other mitochondrial loci required the presence of the type II introns and hence reverse transcriptase. We examined the interaction frequency between a dubious mitochondrial ORF (Q0182; mtDNA [65783-65903 bp]), that does not contain any group II introns and therefore is not altered in the 161-U7 GII-0 strain, and the nuclear encoded RSM7 (Chr X [638756-640423 bp]) gene. The Q0182-RMS7 interaction frequency decreased in the absence of the mitochondrial group II introns ( Figure 4C). These results Interaction frequency between the mitochondrial COX1 and nuclear MspI fragments was assayed by quantitative 3 C (Methods S1) in wild-type (S. cerevisiae BY4741), Dyme1 (BY4741 Dyme1), Dyme2 (BY4741 Dyme2), Dade2 (BY4741 Dade2) and Dmdv1 (BY4741, Dmdv1) strains. Interaction values were corrected for mitochondrial genome copy number (see Methods) and are expressed as percentages of wild-type (set at 100%) +/2 standard error of the mean (n = 4). Deletion of an unconnected gene (ade2) did not significantly affect interaction frequency. T-tests (paired P(T, = t) one-tail, n = 4) were performed to determine the significance of observed variations: wild-type: Dyme1 p = 0.01; wild-type: Dyme2 p = 0.377; wild-type: Dade2 p = 0.103; wild-type: Dmdv1 p = 0.143; Dyme1: Dmdv1 p = 0.210. Only Dyme1 demonstrated a significant difference. doi:10.1371/journal.pone.0030943.g002 Figure 3. A functional electron transport chain is required to maintain the interaction between the mitochondrial COX1 and nuclear MSY1 loci. Uncoupling of the electron transport chain was achieved by 2,4-dinitrophenol (5 mM) treatment of exponentially growing S. cerevisiae in synthetic complete media, containing glucose or galactose, for the indicated time ( Figure S4). COX1-MSY1 A) and nDNA-nDNA B) interaction frequencies were determined by quantitative 3 C analyses using fluorescent probes (see Methods S1). Interaction values in A) were corrected for mitochondrial genome copy number while those in B) were corrected for nuclear genome copy number (see Methods). Interaction values were expressed as percentages of the untreated sample (set at 100%) +/2 standard error of the mean (n = 3). doi:10.1371/journal.pone.0030943.g003 confirm that this phenomenon is not restricted to interactions involving COX1.
Our results suggested that the nucleic acids of mitochondrial origin which participate in the Mito-nDNA interaction are reverse transcribed from RNA intermediates prior to transfer to the nucleus as cDNAs. However, it remained possible that the Mito-nDNA interactions we observed did not involve inter-organelle transfer. Rather, these interactions might have been completely or partially due to interactions between nuclear loci and mitochondrial sequences that had been integrated into the nuclear genome (i.e. nuclear-mitochondrial sequences (NUMTs)). To rule out the possibility that NUMTs were involved, we performed quantitative 3 C analyses in strains without (i.e. 161-U7 GII-0) the COX1 aI5c intron ( Figure 4A and B). Removal of aI5c, and hence the probe site ( Figure 4A), resulted in complete loss of detectable COX1-MSY1 interactions ( Figure 4B). This confirmed that the COX1-MSY1 interaction involves DNA that is directly derived from the mitochondrial genome and not a NUMT.
The number of significant (Methods S1) Mito-nDNA interactions increased .10-fold in respiring (i.e. glycerol lactate grown) cells, relative to glucose or galactose grown cells (Table 1). This increase was not due to a higher number of sequence reads for the respiring sample. Thus, a greater number of unique nuclear loci connect to mtDNA during respiratory growth when the mitochondria are most active. This result, coupled with the need for a functional electron transport chain and reverse transcriptase machinery, led us to hypothesize that the Mito-nDNA interactions are functional in nature, and specifically that they are capable of controlling the transcript levels of the nuclear loci with which they interact. To test this we performed quantitative reverse transcriptase PCR (qRT-PCR) to determine the transcript levels of the nuclear encoded MSY1 and RSM7 genes in WT cells, the mitochondrial group-II intron knockout mutant (161-U7 GII-0), and strain 161-U7 GII-0 aI5c ( Figure 4A). We found that the population transcript level of the MSY1 gene is significantly higher (t-test, two-sample unequal variance, one-tail, n = 2, p = 0.0007) in strain 161-U7 GII-0 ( Figure 5A), which does not contain the probe site and, therefore, has no detectable COX1-MSY1 interaction ( Figure 4A and B), thus identifying the maximum transcript level in the absence of detectable inter-organelle interactions. Critically, we observed a similar population level increase in MSY1 transcript levels following the removal of the type II introns, except aI5c (i.e. strain 161-U7 GII-0 aI5c; Figure 5A). A similar increase was observed for RSM7 transcripts in both the 161-U7 GII-0 and 161-U7 GII-0 aI5c strains relative to the WT ( Figure 5B), consistent with the effects of intron deletion on the Q0182-RMS7 interaction level ( Figure 4C). By contrast deletion of MRS1, which is involved in mitochondrial group I intron splicing [36,37], had no effect on either MSY1 or RSM7 transcript levels ( Figure 5C), or the COX1-MSY1 interaction frequency ( Figure S5). Thus, strains lacking The finding that inter-organelle interactions affect nuclear transcript levels necessarily predicts that the deletion of yme1, which reduces the frequency of the COX1-MSY1 interaction (Figure 2), should correlate with an increase in nuclear MSY1 transcript levels. Indeed increases in MSY1 transcript levels, and other genes involved in mitochondrial gene expression and the biogenesis of the respiratory chain, have been identified within yeast cells containing the yme1 deletion growing with a mixed respiro-fermentative metabolism [38].

Discussion
In this study we have shown that a large number of nuclear loci interact strongly and reproducibly with Mitochondrial DNA in S. cerevisiae and that the spectrum of these interactions is dependent upon the carbon source on which the yeast are grown. Interestingly, we find that Mito-nDNA interactions are significantly reduced when group II mitochondrial introns that contain reverse transcriptase machinery are deleted. This suggests that the mitochondrial DNA that is involved in these inter-organelle interactions is cDNA that has been reverse transcribed from mitochondrial RNAs. Critically, we demonstrate that suppression of inter-organelle DNA-DNA interactions correlates with elevated transcript levels for the interacting nuclear gene and a reproducible albeit small increase in growth rate ( Figure S6), suggesting that these interactions are biologically relevant and play a role in regulating nuclear gene expression. This is further supported by previous observations that yeast nuclear transcription responds to the presence or absence of mitochondrial genome sequences [18,19]. From these results we propose that the Mito-nDNA interactions act as part of an inter-organelle communication system to signal mitochondrial metabolic state and regulate gene expression. While this DNA based inter-organelle communication may seem surprising, there is a large body of evidence demonstrating the presence of mitochondrial DNA in the nucleus and supporting the on-going nature of this transfer [7][8][9][39][40][41][42][43][44][45]. Thus, it is plausible that the process of mitochondrial DNA transfer has evolved into a functional signaling mechanism. In the case of the glucose and galactose dependent COX1-MSY1 and Q0182-RSM7 interactions we have shown a repressive role for Mito-nDNA interactions in the control of nuclear transcript levels. However, there is no reason to assume that all interactions are repressive.
The observation that mito-nDNA interactions correlated with chromosome length would be consistent with non-specific interactions. However, this relationship is also consistent with the hypothesis that the interactions are enriched with elements which are themselves evenly distributed across the genome. Yeast genes, ARS's, and nuclear encoded mitochondrial genes fulfill this criterion ( Figure S7). Importantly, we observed that the nuclear fragments involved in the interactions are enriched for regions that overlap genes with mitochondrial functions in glucose (p = 0.08) and glycerol-lactate grown cells (p, 10 28 , Table S3), in agreement with earlier observations [17]. The condition specific significance of these enrichments can be interpreted as reflecting the regulatory roles of these interactions as part of a functional signaling mechanism.
Our global network analyses were obtained using proximitybased ligation methodologies (i.e. GCC and q3C) with a single restriction enzyme (i.e. MspI). Duan et al. used different restriction enzymes (i.e. EcoRI and HindIII) to interrogate yeast genome structure [46]. A comparison of the interaction frequencies  (Table S4) despite the fact that the interaction frequencies do not correlate quantitiatively ( Figure S8). Duan et al.'s investigation of yeast genome organization did not report any Mito-nDNA interactions [46]. However, the absence of these interactions was due to the author's omission of the mitochondrial genome from their analyses because there is more than one copy of the mitochondrial genome per cell. This form of repetition is surmountable because interactions can be mapped onto a single scaffold, unlike elements which are repeated at different locations within the genome sequence. Re-examination of the datasets published by Duan et al. [46] identified a number of conserved Mito-nDNA between our datasets (Table S4). Interactions between COX1-MSY1 and Q0182-RSM7 (Data S1 and S2, respectively) were observed in both datasets, at levels above or just below the cut-off. The observation that Mito-nDNA interactions are present in these datasets is important for several reasons: 1) the methodologies while both based on proximity-ligation were considerably different, particularly with respect to enrichment and cell conditions; and 2) EcoRI cuts the yeast genome less frequently and at different positions to MspI. While this goes some way towards relieving the likelihood that the interaction network is dependent on the restriction profile it does not completely alleviate this possibility, particularly because only one Mito-nDNA interaction was observed in the HindIII datasets (Data S3). The failure to identify more of these interactions in the HindIII datasets supports our conclusion that only parts of the mitochondrial genome are transferred into the nucleus and interact with the nuclear DNA. Because HindIII only cuts the yeast mitochondrial genome two times it is not unexpected that it does not cut within the transferred parts. Furthermore, the HindIII results support the finding that the interactions are not due to random ligation because random interactions would occur irrespective of fragment length.
It has been shown that DNA sequences which share stretches of sequence homology can associate in vitro [47,48]. Theoretically it is possible that such a mechanism could allow the sequence dependent association of fragments prior to ligation in our experiments. However, only three pairs of interaction fragments that we detected showed any homology upon blast analysis (Data S6) and these were not the interactions that we investigated. Therefore, this mechanism is unlikely to contribute significantly to our results.
Deletion of the group II introns suggests a role for mitochondrial encoded RNA intermediates in the inter-organelle interactions. This implies that the regions of the mitochondria that are involved in these interactions should predominantly be open reading frames (ORFs). However, there was no bias for mitochondrial ORFs being involved in the Mito-nDNA or Mito-rDNA interactions we observed. Superficially, this argues against the transfer occurring through reverse-transcribed cDNAs. However, yeast mitochondrial genes are transcribed as polycistronic transcripts from 14 ATATAAGTA consensus promoters and possibly another 5 non-consensus promoter sites around the mitochondrial genome [49]. Hence, a large percentage of the mitochondrial genome is physically transcribed [50][51][52] and therefore available to act as a template for reverse transcription.
The incomplete ablation of the interactions following the deletion of the mitochondrial group II introns raises the possibility that not all Mito-nDNA interactions involve reverse-transcribed mitochondrial sequences. If mitochondrial mini-circles contribute to inter-organelle interactions [8], we should have seen an increase in interaction frequency in the Dyme1 mutant. The fact that we did not indicates that this form of mitochondrial DNA transfer has a negligible role in the signaling pathway we are proposing. Ongoing transfer despite the loss of the mitochondrial reverse transcriptase can be explained by: 1) cytoplasmic or nuclear reverse transcription of mRNA released from damaged mitochondria, or 2) the presence of other retrotransposon or retroviral encoded reverse-transcriptase of either mitochondrial or nuclear origin within the mitochondrial matrix. Such a mechanism is supported by the identification of remnants of nuclear derived copia-, gypsyand LINE-like retrotransposon elements within Arabidopsis mitochondria [53,54].
It is unclear whether the mtDNA that participates in the Mito-nDNA interactions is transferred by a direct connection between the mitochondrial and nuclear organelles or by uptake from the cytoplasm. Uptake from the cytoplasm is feasible given that unstable mitochondrial plasmids are first released into the cytoplasm [7,9], and high success rates are generally attained for yeast transformation which involves passage through the cytoplasm [55,56]. However, Ricchetti et al. demonstrated that the mtDNA mediated repair of nuclear double strand breaks is independent of Dyme1 mutations [6] and therefore occurs through another, possibly direct, transport mechanism. Direct transport from the mitochondria to nuclear compartments could occur as a result of a tethering/transport complex that physically links mitochondria to the endoplasmic reticulum [57].
In conclusion, several important questions are raised by this work. Firstly, what is the mechanism by which the Mito-nDNA interactions affect changes in transcript levels? Is the effect mediated by physical interaction between the mitochondrial derived cDNA and the nuclear locus or by more indirect means? It is possible that transcription factors may be sequestered or recruited to locations of activity through interactions with the mitochondrial DNA. Secondly, does this inter-organelle signaling pathway operate on a general level or function just to regulate conditionally essential genes (e.g. RMS7 and MSY1)? Thirdly, while we demonstrated that specific interactions are dependent upon reverse transcription, whether this is true for all the Mito-nDNA and Mito-rDNA interactions remains to be determined. Finally, the universal significance of these interactions remains to be determined, particularly given the non-ubiquitous distribution of group II introns within higher eukaryotic mitochondria. The pervasive presence of reverse transcription within higher eukaryotic cells leads us to propose that this phenomenon will be widespread and that it deserves further investigation.

Genome Conformation Capture (GCC)
GCC was performed according to [17]. Refer to the supplementary methods for a detailed description. Briefly, chromatin was prepared from 15 sets of 10 8 (i.e. a total of 1.36610 9 ) cross-linked cells. Chromatin was digested with MspI (Fermentas) and ligated (T4 ligase; Invitrogen). Crosslinks were reversed in the presence of proteinase K (final concentration 7-11 mg, Roche). Samples were treated with RNase A (final concentration 10 mgml 21 ) prior to purification by phenol:chloroform (1:1 v/v, three times) and column extraction (Zymo Clean and Concentrator, Zymo Research). Paired-end sequencing (36 bp) was performed on 5 mg DNA using the Illumina Genome Analyzer platform (Allan Wilson Centre, Massey University, New Zealand & Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland).
External controls were added at two steps in the GCC protocol to control for random ligation events. The first ligation control, a linear DNA fragment with a free MspI site at one end (Methods S1), was added in a 1:1 ratio with the nuclear genome prior to the addition of ligase. The second ligation control (1610 6 molecules of pUC19) was included prior to RNase A treatment as a control for the sequencing step ligation.

GCC Network Assembly
Network assembly was performed using Topography v1.19 (available on request [17]). The SOAP [59] algorithm was used to position paired end sequences and single ends, which contain an MspI restriction site, onto the S. cerevisiae reference genome (Methods S1). No mismatches were allowed.

Bioinformatic analyses
Bioinformatic and statistical analyses (see Methods S1) were performed on chromosomal interactions involving the nuclear and mitochondrial genomes for which the sequences mapped uniquely onto the reference genome. Connections with the ribosomal DNA (rDNA), 2 micron plasmid and mitochondrial genomes were considered as unique because they could be positioned to a ,1 MB region of Chromosome XII, the 6318 bp 2 micron plasmid or the ,85 kbp mitochondrial genome, respectively [17]. All statistical analyses involving 2 micron plasmid, mitochondrial, or rDNA sequences included copy number corrections (Methods S1). Other repetitive elements, such as LTRs and tRNAs, were omitted from the analysis.
Dedicated interaction standards (concentration from: 2 ng ml 21 -2610 215 g ml 21 ) were prepared by PCR amplification (from S. cerevisiae BY4741) of the interacting regions, followed by MspI digestion and ligation of the two interacting partner fragments.
For comparison, all samples were presented as a percentage of wild-type, following standardization for: 1) the amount of a15c intron-containing DNA (i.e. mitochondrial copy number); or 2) the number of nuclear genomes (determined using the single copy GAL1 locus; [61]; Primer sequences are listed in Table S2. This standardization was performed to correct for alterations to mitochondrial genome stability and the rates of appearance of rho 2 or rho 0 strains. This is critical as inter-organelle interactions are dependent upon the presence of the mitochondrial genome (see 161-U7 GII-0 results). The method of standardization depends upon the interaction being investigated (i.e. COX1-MSY1 interactions were standardized by mitochondrial genome copy number while nuclear-nuclear locus interactions were standardized by GAL1 copy number). No significant differences were observed when inter-organelle interactions were standardized by mitochondrial or nuclear copy number (data not shown).

Quantitative Reverse Transcription-PCR
qRT-PCR standards were amplified from S. cerevisiae BY4741 genomic DNA (Table S2). PCR products were purified (Zymo DNA clean and concentrator TM -5 kit according to manufacturer's instructions). The concentration of each qRT-PCR standard was determined by Nano-drop and used to make dilutions ranging from 4.0-4.0610 25 ng/ml. qRT-PCR reactions were performed using One Step SYBRH Ex Taq TM qRT-PCR Kit according to the manufacturer's instructions (TaKaRa). The qRT-PCR was run with the following protocol: 42uC, 5 min; 95uC, 10 sec; 406 [95uC, 5 sec; 60uC, 31 sec] 95uC, 15 sec; 60uC, 1 min; 95uC, 15 sec. All transcript levels were standardized to nuclear ACT1 and expressed as percentage of wild-type (set at 100%) +/2 standard error of the mean.

Supplementary files
The following additional data are available with the online version of this paper . Supplementary material file contains: Figures  S1, S2, S3, S4, S5, S6, S7, S8, Tables S1, S2, S3, S4, S5, Methods S1 and Supplementary references. Additional data files: contain the analysis of the Duan data (Data S1, S2, S3, S4, S5); and the mitochondrial blast analyses (Data S6); a list of all uniquely mapping mito-nDNA and mito-mito interactions and their interaction strength (Data S7); a list of all mapped mito-nDNA and mito-mito interactions and their interaction strength (Data S8); and a list of nuclear genes with a mitochondrion annotation in gene ontology (Data S9). Sequences are available from GEO (accession number GSE34132). Figure S1 Biological Repeats correlate well at the MspI restriction fragment level. Two biological repeats were performed for each condition; A) glucose, B) glycerol lactate, and C) galactose. R 2 values are as follows; Glucose 0.78, glycerol lactate 0.93, and galactose 0.93. Scatter plots were constructed from statistically significant (p = ,0.0004) interactions involving only MspI fragments which could be uniquely positioned on the reference genome. Adjacent interactions have been omitted as we are unable to distinguish between true adjacent interactions and those which are the result of simply sequencing across an uncut MspI site. Circularized fragments (i.e. self interactions) have also been omitted. (DOC) Figure S2 Inter-organelle interactions vary with metabolic state and do not occur evenly across the mitochondrial genome. Interaction frequency was graphed as a percentage of the total number of interactions in the sample, according to segment length. To test whether Mito-nDNA interactions have a uniform distribution (i.e, the total number of interactions in a segment is proportional to its length) we aggregated consecutive restriction fragments to create 58 sections that were expected to have at least 5 interactions under the null hypothesis of uniformity. A Chi-squared goodness of fit test was performed, and the distribution of the interactions was shown to deviate significantly from uniformity (p,0.0001, 57 df) for all conditions, Thus, Mito-nDNA interactions are not uniformly distributed across the mitochondrial genome. The linearized mitochondrial genome is shown for comparison of the interaction frequency with mitochondrial ORF and inter-genic sequence positions. Metabolic conditions were as follows: A) respirofermentation (glucose), B) respiro-fermentation (galactose), and C) respiration (glycerol lactate). Only statistically significant unique interactions between the mitochondrial genome and nuclear chromosomes were included in this analysis (p#10 25 ; n = 2). Interactions with the rDNA and 2-micron plasmid were removed. D) Nuclear genome interactions are not enriched over mitochondrial open reading frames. We compared the numbers of nuclear genome interactions with mitochondrial inter-and intra-genic regions to determine if the interactions across the mitochondrial genome were enriched over the open reading frames (ORFs). Galactose displays a larger number of interactions with mitochondrial ORFs but the difference is not statistically significant. Interactions were assigned proportionally to inter-and intra-genic regions to obtain a ratio of inter-genic to intra-genic interactions and expressed as percentages. tRNAs were not deemed intra-genic. Interestingly, the galactose sample exhibited 7% and 13% more inter-organelle interactions involving the COX1 ORF than glycerol lactate and glucose, respectively. Thus, while there is no obvious preference for interactions with mitochondrial ORFs, interactions involving COX1 show differences between the datasets. (DOC) Figure S3 The COX1-MSY1 Mito-nDNA interaction is cell cycle dependent. Cells were synchronized at three different cell cycle phases by treatment with a-factor (3.4 mm), Hydroxyurea (100 mM), or Nocodazole (15 mgml 21 ); G1, S, G2/M, respectively). Mito-nDNA interaction frequency, between the representative mitochondrial and nuclear MspI fragments, was assayed by quantitative 3 C (see Methods S1). Interaction values were corrected for mitochondrial genome copy number (see Methods). Interaction values are expressed as percentages of the untreated sample (set at 100%) +/2 standard error of the mean (n = 3). (DOC) Figure S4 5 mM 2,4-Dinitrophenol (DNP) inhibits respiratory growth but does not prevent growth of fermenting S. cerevisiae BY4741 cells. S. cerevisiae BY4741 cultures were grown (50 ml, 30uC, 160 rpm) on glucose (fermentation) or glycerol/lactate (respiration) to an Optical density (600 nm; OD 600 ) of 0.600. Cultures were diluted to an OD 600 of 0.150 (50 ml final volume) in their respective media. 5 mM DNP (final concentration) was added to two of the cultures, while two remained untreated. The cell growth was monitored (OD 600 ) for a further 11.5 hours, with the exception of the untreated glucose culture which was only grown for 4 hours. (DOC) Figure S5 Deletion of MRS1 (BY4741 Dmrs1), a nuclear gene involved in splicing mitochondrial type-I introns, has no significant effect on the frequency of the COX1-MSY1 interaction in glucose grown yeast cells. Interaction frequency was expressed as percentages of the wild type S. cerevisiae strain BY4741 (WT, set at 100%) +/2 standard error of the mean (n = 3). (DOC) Figure S6 Deletion of group II introns results in an increase in growth rate. Growth rates were determined for Saccharomyces cerevisiae strains (161-U7, 161-U7 GII0, and 161-U7 GII0 +aI5c; Figure 4a) grown in SC+2% glucose (30uC and 160 rpm). Cultures were inoculated to an initial optical density (OD 600 ) of 0.05 from overnight cultures. The OD 600 was measured every two hours for 10 hour. Data represent the mean 6 SD (n = 3). (DOC)  Methods S1 This file contains supplementary information for methods used in this manuscript.

(DOC)
Data S1 This text file contains the significant interactions that were identified as occurring between the mitochondrial genome and the region surrounding the nuclear MSY1 locus in the dataset prepared by Duan et al. [46]. The number of instances for any interaction that had to be seen was set at . = 3 for the Mse1 and . = 4 for the Msp1 datasets.

(TXT)
Data S2 This text file contains the significant interactions that were identified as occurring between the mitochondrial genome and the region surrounding the nuclear RSM7 locus in the dataset prepared by Duan et al. [46]. The number of instances for any interaction that had to be seen was set at . = 3 for the Mse1 and . = 4 for the Msp1 datasets.

(TXT)
Data S3 This text file contains the one interaction that was identified as occurring between the mitochondrial and nuclear genomes in chromatin cut with HindIII and subsequently Mse1.

(DAT)
Data S4 This file contains all the interactions that were identified by re-analysis of the Duan et al. [46] datasets for chromatin digested with EcoRI and subsequently Mse1. All listed interactions were all above the cut-off which was set at . = 3.

(DAT)
Data S5 This file contains all the interactions that were identified by re-analysis of the Duan et al. [46] datasets for chromatin digested with EcoRI and subsequently Msp1. All listed interactions were all above the cut-off which was set at . = 4.

(DAT)
Data S6 This file contains the results of a blastN analysis of mito-nDNA interacting fragments. Only three pairs of interaction fragments that we detected showed any homology upon blast analysis. The smaller restriction fragment from each interaction pair was compared to the longer fragment by blastn using default parameters. The length, score, and evalue for each comparison was recorded. Comparisons that showed no similarity were given an evalue score of 10.