Widespread temperature sensitivity and tRNA decay due to mutations in a yeast tRNA.

Microorganisms have universally adapted their RNAs and proteins to survive at a broad range of temperatures and growth conditions. However, for RNAs, there is little quantitative understanding of the effects of mutations on function at high temperatures. To understand how variant tRNA function is affected by temperature change, we used the tRNA nonsense suppressor SUP4oc of the yeast Saccharomyces cerevisiae to perform a high-throughput quantitative screen of tRNA function at two different growth temperatures. This screen yielded comparative values for 9243 single and double variants. Surprisingly, despite the ability of S. cerevisiae to grow at temperatures as low as 15°C and as high as 39°C, the vast majority of variants that could be scored lost half or more of their function when evaluated at 37°C relative to 28°C. Moreover, temperature sensitivity of a tRNA variant was highly associated with its susceptibility to the rapid tRNA decay (RTD) pathway, implying that RTD is responsible for most of the loss of function of variants at higher temperature. Furthermore, RTD may also operate in a met22Δ strain, which was previously thought to fully inhibit RTD. Consistent with RTD acting to degrade destabilized tRNAs, the stability of a tRNA molecule can be used to predict temperature sensitivity with high confidence. These findings offer a new perspective on the stability of tRNA molecules and their quality control at high temperature.

INTRODUCTION

tRNAs are highly evolved to efficiently and accurately incorporate amino acids during translation. Mutations in tRNA often result in reduced function, with serious consequences to cell viability and to human health; for example, over 230 mitochondrial tRNA mutations are associated with human disease, including mutations in every region of the molecule (Ruiz-Pesini et al. 2007). Two major properties of tRNA are conserved among all domains of life. First, virtually all tRNAs have a similar overall structure (Kim et al. 1974a; Westhof et al. 1985; Giegé et al. 2012) for efficient recognition by different parts of the translation machinery, for flexibility in passage through the ribosome (Valle et al. 2003; Schmeing et al. 2009; Zhou et al. 2013), and for stability through multiple rounds of translation (Alexandrov et al. 2006; Whipple et al. 2011; Dewe et al. 2012). Second, all tRNAs have unique sequence elements for efficient and highly selective decoding of mRNA and for charging by their cognate tRNA synthetase (Giege et al. 1998; Cochella and Green 2005; Ledoux et al. 2009; Ling et al. 2009; Shepotinovskaya and Uhlenbeck 2013). tRNAs are also subject to ubiquitous post-transcriptional modifications, and these aid in all aspects of tRNA function (Pütz et al. 1994; Helm et al. 1999; Johansson et al. 2008; Demeshkina et al. 2010; Maehigashi et al. 2014; Grosjean and Westhof 2016; Rozov et al. 2016).

To ensure that the structural requirements for tRNA are met, tRNAs are subject to several quality control pathways. In the yeast Saccharomyces cerevisiae, these pathways recycle certain hypomodified tRNAs back to the nucleus (Shaheen and Hopper 2005; Takano et al. 2005) for modifications during biogenesis (Ohira and Suzuki 2011) and for possible repair (Kramer and Hopper 2013); degrade wild-type (WT) and certain hypomodified pre-tRNAs by the nuclear surveillance pathway (Kadaba et al. 2004, 2006; Gudipati et al. 2012); and degrade mature tRNAs through the rapid tRNA decay (RTD) pathway if they lack specific modifications or have destabilizing mutations (Alexandrov et al. 2006; Chernyakov et al. 2008; Whipple et al. 2011; Guy et al. 2014). In humans there is also evidence that RTD occurs at high temperature (Watanabe et al. 2013).

Although it is well known that RNA stability has a direct effect on RNA function at different temperatures, there is little quantitative information on the effects of mutations on RNA function at different temperatures. This problem has been addressed to some extent in quantifying regulation of gene expression by RNA thermometers in response to mutations that affect stability of regulatory helices (Chowdhury et al. 2003; Kortmann and Narberhaus 2012; Cimdins et al. 2014). More typically, RNA function at different temperatures is only analyzed qualitatively, as in the analysis of destabilizing mutations that counteract the cold sensitivity of a stabilizing stem mutation of U6 snRNA (Fortner et al. 1994), and the analysis of reduced growth at high temperatures due to several destabilizing tRNA variants (Whipple et al. 2011). Although high-throughput studies have been used to quantify the effects of sequence variation on the function of a catalytic RNA (Pitt and Ferre-D'Amare 2010) and of a tRNA (Guy et al. 2014; Li et al. 2016) under a specific set of conditions, there have been no such studies on how sequence variation affects RNA function at different temperatures.

Analysis of the tolerance of RNAs to temperature is inherently important for two reasons. First, all organisms grow well over a broad range of temperatures and therefore require cellular components such as tRNAs to function relatively uniformly throughout these conditions. Yeast, for example, grow relatively well from 15°C to 39°C. Second, there is likely strong evolutionary pressure on the thermal stability of tRNAs and other functional RNAs, given the positive correlation between G–C content (Galtier and Lobry 1997) or predicted RNA structural stability (Lu et al. 2006) and the optimal growth temperature of an organism.

We report here the first comprehensive analysis of the effects of sequence variation on the function of an RNA, the tRNA suppressor SUP4. We previously developed a high-throughput approach to comprehensively quantify the effects of mutations on the function of SUP4 and to define the spectrum of mutations that trigger tRNA decay by the RTD pathway at a specific temperature (Guy et al. 2014). In this work, we extend this approach to define the effects of sequence variation on SUP4 function at a higher temperature, and to evaluate the connections between temperature sensitivity and susceptibility to the RTD pathway. We find evidence that temperature sensitivity of tRNA variants occurs frequently and is associated with thermodynamic parameters and RTD.

RESULTS

High-throughput quantification of tRNA temperature sensitivity

We examined the effects of high temperature on the function of the yeast nonsense suppressor tRNA SUP4, using our previously described approach, which measures the relative in vivo function of each tRNA in a library of variants (Guy et al. 2014). SUP4 tRNA function was evaluated using a version of the RNA-ID reporter (Dean and Grayhack 2012) expressing GFPoc (suppressible by SUP4) and RFP, each transcribed in opposite directions under the control of the bidirectional P promoter (Supplemental Fig. S1A). Thus, cells bearing a more functional SUP4 tRNA variant will have a higher ratio of GFPoc to RFP (Guy et al. 2014).

We grew the yeast library of SUP4 variants to log phase at 37°C and sorted cells by their GFP/RFP score into four bins using fluorescence activated cell sorting (FACS) (Supplemental Fig. S1B). We extracted DNA from each bin, deep sequenced the SUP4 gene to determine the distribution of each variant among the bins, and converted the distribution to a GFPSEQ score that reflects the variant's function relative to WT SUP4 under the same conditions (Supplemental Table S1; see Materials and Methods). Variants were categorized as nonfunctional (GFPSEQ < 0.026, the smallest value that could be distinguished from zero), marginally functional (0.026 ≤ GFPSEQ < 0.18), substantially functional (0.18 ≤ GFPSEQ < 0.9), and highly functional (GFPSEQ ≥ 0.9) (Guy et al. 2014).

GFPSEQ37 values for individual variants were strongly correlated among four biological replicates, with R2 values ≥0.9 for every pairwise comparison of data sets (Supplemental Table S1; Supplemental Fig. S2A), which is comparable to the reproducibility between replicates in our previous data sets at 28°C (Guy et al. 2014). For further analysis, we chose the biological replicate that contained the largest number of unique variants, replicate 1 (labeled 37deg in Supplemental Table S1). As an additional quality-control measure, we reconstructed 12 variants from the 37°C data set and measured their GFP/RFP ratio, normalized to WT SUP4, by flow cytometry (GFPFLOW) at 37°C. As with the previous 28°C data set, GFPFLOW37 and GFPSEQ37 were strongly correlated, with an R2 value of 0.86 (Supplemental Fig. S2B).

After filtering for data quality (see Materials and Methods), the 37°C data set contained 25,191 variants, including 212/213 possible single variants, 9321 double variants, and 8441 triple variants (Supplemental Table S2). Of these, 24,282 variants were also present in our 28°C data set (Guy et al. 2014). The high degree of overlap between the 28°C and 37°C data sets reflects saturation in sampling of the library.

The majority of SUP4 variants are temperature sensitive

Almost all of the variants decreased in function in MET22+ cells, defined here as WT, at 37°C relative to 28°C (Supplemental Table S2; Supplemental Fig. S3; Fig. 1A). Indeed, only seven of the 47 highly functional single variants at 28°C remained highly functional at 37°C, and a similar loss of function was observed among the substantially functional single variants and among all double variants. To quantify temperature sensitivity, we defined the temperature sensitive (TS) ratio as GFPSEQ28/GFPSEQ37 and considered a variant TS if this ratio was ≥2. A variant could be assigned a TS ratio only if its GFPSEQ28 score was ≥0.052, twice the smallest GFPSEQ28 value (0.026) distinguishable from background (Guy et al. 2014). Using this scoring metric, we assigned TS ratios to 76 single variants (Fig. 1B) and 821 double variants (Supplemental Table S2). Temperature sensitivity was the rule rather than the exception, with 57% (43/76) of the single variants and 75% (613/821) of the double variants classified as TS. Temperature sensitivity occurred in tRNA variants with a wide range of GFPSEQ28 scores including 59.1% (169/286) of the highly functional single and double variants, 74.2% (216/291) of the substantially functional variants, and 84.6% (271/320) of the marginally functional variants. The lower prevalence of temperature sensitivity among the highly functional variants could be due in part to the saturation of GFPSEQ28 at GFPFLOW scores >0.45, which was previously observed (Guy et al. 2014), and would reduce our ability to resolve modest temperature sensitivity among these variants.

FIGURE 1.

The majority of SUP4 variants are temperature sensitive in WT cells at 37°C. (A) Most SUP4 single variants have reduced GFPSEQ values at 37°C relative to 28°C. GFPSEQ37 scores in WT cells were compared to GFPSEQ28 scores previously obtained in WT cells (Guy et al. 2014). Variants below the dashed line decreased in function at 37°C, while variants above increased in function. Variants in loops, blue; variants in stems, solid red (base pair preserved) or open red (base pair broken). (B) Temperature-sensitive SUP4 single variants are widespread. Cloverleaf heatmap depicting TS scores for SUP4 single variants in WT cells, as measured by GFPSEQ28/GFPSEQ37. TS variants, shades from white to orange; non-TS variants, teal; variants not scored, stippled. Nucleotides 34–37 and the post-transcriptionally added CCA were not subject to variation. TS difference (GFPSEQ28–GFPSEQ37) is indicated by the thickness of each variant's wedge. (C) The calculated ΔΔG° thresholds for variant function are higher at 28°C than at 37°C. For each single and double variant, the GFPSEQ is plotted against its calculated ΔΔG° at 37°C (top panel) and 28°C (lower panel). Note that 95% of functional variants have ΔΔG° less than that of the gold line; red line, 99%.

TS variants were most often located in stems (Fig. 1B). The vast majority (79%, 27/34) of single mutant stem variants that could be scored were TS, and the seven stem variants that were not (C27U, A29G, G51A, C52U, C64U, C65U, and G69A) all preserved canonical base pairing (including G–U pairs), with the exception of G51A, which forms a C–A mismatch in the T-stem. In contrast, variants in loop regions were less frequently temperature sensitive; only 38% (16/42) of the scoreable single variants with loop mutations were TS, and eight of these 16 were located adjacent to stems where they could potentially affect the stability of the helix by forming stacking interactions (Turner and Mathews 2010). This difference in temperature sensitivity for stem vs. loop variants represents a significant enrichment in stem mutations associated with a TS phenotype (χ2 test, P < 0.05), supporting a role for structure in the TS phenotype.

Double variants that were TS likewise frequently bore mutations in or adjacent to stems. Among the scoreable double variants, 76.7% (551/718) of the variants with at least one mutation in or adjacent to a stem were TS, while only 60.2% (62/103) of the variants with mutations only in loops were TS (P < 0.005).

Interestingly, variants at locations of dihydrouridine modification were particularly likely to be non-TS. This category includes all variants at residues U16, U17, U20a, U20b, two variants at residue U47, and one variant at residue U20. Together, these variants accounted for the majority (15/26) of the non-TS single variants with loop mutations.

Given the strong connection between temperature sensitivity and stem mutations, which are known to affect structural stability, we tested if temperature sensitivity was a result of a more stringent threshold for correct folding imposed by high temperature. For this analysis, we used two approaches. First, we calculated the ΔΔG° for the secondary structure of each variant, defined as its predicted change in stability from the native SUP4 structure for each variant, as estimated by the folding free energy change nearest neighbor model of RNA stability (Turner and Mathews 2010). We observed that tRNA function was highly dependent on ΔΔG°, particularly at 37°C; at 37°C, 95% of single and double variants that were at least marginally functional had ΔΔG° <6 kcal/mol, compared to a threshold of <8 kcal/mol at 28°C (Fig. 1C, gold lines). There was a similar trend based on a 99% cutoff (Fig. 1C, red lines). Second, we calculated a normalized ensemble defect for each variant, which uses a scale of zero to one to reflect the propensity for an RNA secondary structure to misfold, where one represents completely misfolded RNA (Zadeh et al. 2011), and observed the same trend as ΔΔG°. At 37°C, 95% of single and double variants that were at least marginally functional had ensemble defects less than 0.16 (Supplemental Fig. S4, gold lines), compared to an ensemble defect cutoff of 0.21 for SUP4 variants at 28°C (Guy et al. 2014), and a similar trend held for a 99% cutoff (Supplemental Fig. S4, red lines).

A46G variants are often cold sensitive

Although the overwhelming majority of SUP4 single and double variants were temperature sensitive, 11 variants had GFPSEQ scores that increased more than twofold at 37°C, indicative of cold-sensitivity (CS ratio >2) (Supplemental Table S2). Many of these CS variants included the A46G mutation. The A46G single variant was much more active at 37°C, with a GFPSEQ37 score of 0.279 compared to 0.026 at 28°C (CS ratio of 10.71), and seven other double mutant variants with an A46G mutation also had high-CS ratios, ranging from 13.27 to 2.13 (Supplemental Table S2). Subsequent examination by flow cytometry showed that the A46G single variant was active at 37°C (GFP/RFP, 0.324), but had near background levels of activity at 28°C (0.020) and 23°C (0.013) (Fig. 2A). The CS phenotype of A46G also extends to one of three tested double variants with this mutation, using GFPFLOW; A46G and A13C A46G had CS ratios of 5.72 and 2.68, respectively, while A22U A46G and C59G A46G had CS ratios of 1.91 and 0.83, respectively (Fig. 2B).

FIGURE 2.

The A46G mutation confers cold sensitivity. (A) The SUP4 A46G single variant is cold sensitive. The A46G variant was subjected to flow cytometry at 23°C, blue; 28°C, cyan; and 37°C, red, and scatterplots were generated from GFP and RFP values for 10,000 cells. (B) Double variants containing the A46G mutation are cold sensitive. Variants as indicated were subjected to flow cytometry in biological triplicate, and the median GFPFLOW was determined at 28°C, blue; and 37°C, red.

Temperature sensitivity is linked to the rapid tRNA decay pathway

We speculated that the RTD pathway might be responsible for the high prevalence of temperature sensitivity that we observed in WT (MET22+) cells, since RTD degrades tRNAs that are destabilized due to lack of certain modifications or to the presence of mutations that adversely affect structure, and is generally more efficient at higher temperatures (Chernyakov et al. 2008; Whipple et al. 2011; Guy et al. 2014).

To define RTD substrates at high temperature, we tested the library of SUP4 variants at 37°C in a met22Δ mutant, which lacks a functional RTD pathway (Chernyakov et al. 2008; Whipple et al. 2011; Dewe et al. 2012), as we previously did at 28°C (Guy et al. 2014). This data set allowed us not only to evaluate the relationship between temperature sensitivity and RTD in WT cells, but also to evaluate temperature sensitivity in the absence of RTD in the met22Δ mutant (Supplemental Fig. S5). Replicates of the GFPSEQ data set derived from the library of SUP4 variants in a met22Δ strain had a strong correlation (R2 > 0.9) at 37°C (Supplemental Fig. S2A), and we used replicate #2 for the comparisons below. We scored variants for RTD at 37°C based on the RTD37 ratio (GFPSEQ37met22Δ/GFPSEQ37MET22+). We defined variants with an RTD37 ratio ≥2 as RTD substrates, and therefore only scored variants with a GFPSEQ37met22Δ ≥0.052, twice the background (Guy et al. 2014). This filter resulted in an RTD37 score for 72 single variants and 480 double variants.

We found that 24/72 single variants were RTD substrates at 37°C (Fig. 3A), 13 of which were previously identified as RTD substrates at 28°C. Of the 11 single variant RTD substrates unique to 37°C, six contained mutations in the acceptor stem, four of which change G–C pairs to less stable G–U wobble pairs (C3U, C66U, C67U, and C69U), and two of which form mismatches in the U4–G69 wobble pair (U4A and U4G). Of the remaining five variants that were uniquely RTD37 substrates, two had mismatch mutations in other stems (G62U and A28U), two affected the nucleotides immediately adjacent to the D-stem (A13G, A22G), and one was in the variable arm (A46U).

FIGURE 3.

RTD is widespread at 37°C and associated with temperature sensitivity. (A) RTD substrates at 37°C include variants throughout the tRNA. Cloverleaf heatmap depicting RTD scores for SUP4 single variants, as measured by GFPSEQ37met22Δ/GFPSEQ37 WT. RTD Substrates (RTD score ≥2), shades of white to green; non-RTD substrates, purple. RTD difference (GFPSEQ37met22Δ–GFPSEQ37 WT) is indicated by the thickness of the wedge for each variant. (B) Strongly TS variants are overwhelmingly substrates for RTD at 28°C or 37°C. Venn diagram showing the relationship between variants that are scored as strongly TS in WT (TS ratio ≥4) (blue) and variants that are scored as RTD substrates at 37°C or 28°C (red). Overlap region (purple). (C) Strongly non-TS variants are often not substrates for RTD at 28°C or 37°C. Venn diagram showing the relationship between variants that are strongly not TS in WT (TS ratio ≤1.25) (blue) and variants that are RTD at 37°C (red). Overlap region (purple). (D) Schematic of poison primer extension in the presence of ddCTP to separate endogenous tRNATyr from integrated SUP4 variants. (E) SUP4 variants identified as RTD substrates at 37°C have increased tRNA levels in a met22Δ strain. Poison primer extension was used to quantify tRNA levels from the bulk RNA of cells grown in steady state at 28°C or 37°C. SUP4 tRNA levels were quantified relative to WT tRNATyr by determining percent read-through (% RT).

To understand the relationship between temperature sensitivity in WT cells and RTD, we examined the overlap between temperature sensitivity and RTD at either 28°C or 37°C, for single and double variants that were present in each data set. Strikingly, 91.0% (161/177) of the variants that were strongly TS in WT cells (TS ratio ≥4) were RTD substrates at either 28°C or 37°C (Fig. 3B), whereas only 40.3% (29/72) of the variants that were strongly not TS in WT cells (TS ratio <1.25) were RTD substrates at either temperature (Fig. 3C). This high degree of association between temperature sensitivity and RTD was even more pronounced for higher thresholds of TS ratios. For example, 96.5% (111/115) of the variants that were very strongly TS in WT cells (TS ratio >6) were RTD substrates at either 28°C or 37°C. Conversely, decreasing the threshold reduced the percentage of TS WT variants that were RTD substrates, to 84% for TS ratio ≥3 and to 55.5% for TS ratio ≥2. The strong association between temperature sensitivity in WT cells and RTD susceptibility suggests that the majority of temperature sensitivity in WT cells is due to increased susceptibility to the RTD pathway.

We examined the association between RTD and temperature sensitivity for two variants (A28U and U4G) that were TS in WT cells and subject to RTD at 37°C, by measuring their tRNA levels using a poison primer extension assay (Fig. 3D). This analysis showed that each variant had substantially reduced tRNA levels in a WT strain at 37°C compared to 28°C (Fig. 3E, cf. lanes 7 and 9, and lanes 11 and 13), but had increased tRNA levels in a met22Δ strain at 37°C compared to a WT strain at 37°C (cf. lanes 9 and 10, and lanes 13 and 14); in contrast SUP4 itself was largely unchanged in all of these conditions (lanes 3–6).

SUP4 variants are frequently temperature sensitive in a met22Δ mutant

Although much of the temperature sensitivity of WT cells was connected to RTD at 28°C or 37°C, we also observed pervasive temperature sensitivity of SUP4 variants in met22Δ cells (Supplemental Table S2), in which RTD is inhibited. Of the 96 single variants that could be scored for temperature sensitivity in met22Δ cells, 53 were TS (Fig. 4A; Supplemental Fig. S6A), while 72.7% (885/1217) of the double variants were TS in a met22Δ strain (Supplemental Fig. S6B). As we observed for TS variants in WT cells, TS variants in met22Δ cells overwhelmingly had at least one mutation in or adjacent to stems, accounting for 98% (52/53) of the TS single variants and 93% (823/885) of the TS double variants (Supplemental Table S2).

FIGURE 4.

Temperature sensitivity is common in met22Δ cells, and is associated with RTD. (A) The majority of SUP4 variants are temperature sensitive in a met22Δ strain. Cloverleaf heatmap depicting TS scores for SUP4 single variants in met22Δ as measured by GFPSEQ28/GFPSEQ37. Shading and symbols as in Figure 1B. (B,C) SUP4 variants that are TS in met22Δ are almost all substrates for RTD at 28°C or 37°C. Venn diagram showing the relationship between variants that are RTD at 28°C/37°C (blue) and variants that are TS in met22Δ (red), with overlap, purple, for a TS ratio >2 (B) and a TS ratio >4 (C). (D) SUP4 variants identified as TS in met22Δ have decreased tRNA levels at 37°C compared to 28°C. Poison primer extension was used to quantify tRNA levels from the bulk RNA of cells grown in steady state at 28°C or 37°C.

There was also an unexpectedly strong association between variants that were TS in met22Δ strains and those that were TS in WT strains. Of the 52 TS single variants in met22Δ cells that had GFPSEQ scores in all data sets, 26 were also TS in WT cells, and all but three of the remaining variants were not scoreable for temperature sensitivity in WT cells because their GFPSEQ28 scores were less than the required cutoff of 0.052. This trend extended both to double variants and to strongly TS single and double variants. For example, of the 525 double variants that were TS in met22Δ cells and had GFPSEQ scores in all data sets, 252 were TS in WT cells and 229 of the remaining 273 were not scoreable for temperature sensitivity (Supplemental Table S2).

Temperature sensitivity is associated with tRNA decay in met22Δ cells

The strong associations between the set of WT TS variants and the sets of met22Δ TS variants and of RTD variants presented the possibility that the set of met22Δ TS variants was also associated with the set of RTD variants. Indeed, among single and double variants that had GFPSEQ scores in all data sets, 81.1% (468/577) of met22Δ TS variants were also subject to RTD at 28°C or 37°C (Fig. 4B), and this ratio rose to 94.2% (277/294) for met22Δ variants that were strongly TS (Fig. 4C).

Examination of several of the variants that were TS in a met22Δ strain showed that the TS was frequently, but not always, associated with loss of tRNA. For several variants we observed substantially reduced tRNA levels in a met22Δ strain grown at 37°C, relative to 28°C, including G68C, A71U, C66A, U2C, and C5U (Fig. 4D; Supplemental Fig. S7A,B). For these tRNA variants, as well as for the U41C C59A double variant, decay occurred within 3 h (Supplemental Fig. S7C); in contrast, for some variants that were TS in a met22Δ strain, such as the A38U U39C and the A38U G45A variants, there was no observed decay in this timeframe.

Three lines of evidence suggest that the decay of met22Δ TS variants does not involve the nuclear surveillance pathway, which targets pre-tRNA and is the only other known pathway responsible for tRNA decay (Kadaba et al. 2004, 2006; Gudipati et al. 2012). First, flow cytometry analysis of U42G, G68C, and G7U variants demonstrated that the met22Δ TS GFPFLOW phenotype was not suppressed by deletion of TRF4 (Fig. 5A), a known component of the nuclear surveillance pathway (Kadaba et al. 2004). Second, decay of the G68C variant occurred at the same rate with or without the transcription inhibitor thiolutin (Fig. 5B), suggesting that decay occurs at the level of mature tRNA. Third, decay of the G68C variant, as measured by tRNA levels, was not suppressed by deletion of either TRF4 or RRP6 (Fig. 5C), another component of the nuclear surveillance pathway (Kadaba et al. 2004).

FIGURE 5.

The nuclear surveillance pathway is likely not responsible for temperature dependent tRNA decay of tested variants in met22Δ mutants. (A) Deletion of TRF4 in a met22Δ strain does not decrease the TS ratio measured by GFPFLOW for three variants analyzed. GFPFLOW28/GFPFLOW37 for met22Δ, blue, and trf4Δ met22Δ, red. (B) The SUP4G68C variant is similarly degraded in a met22Δ strain after shift to 37°C in the presence or absence of thiolutin. met22Δ mutants bearing SUP4 G68C and WT variants as indicated were treated with thiolutin or buffer, shifted to 37°C for indicated times, and levels of SUP4 and WT tRNATyr were analyzed from bulk RNA of harvested cells by poison primer extension. (C) The SUP4 G68C variant is degraded at 37°C in a met22Δ strain and in trf4Δ and rrp6Δ derivatives. SUP4 G68C tRNA levels were measured in a met22Δ, met22Δ trf4Δ, and a met22Δ rrp6Δ strain 4 h after shift to 37°C, as described in B.

Since the decay of met22Δ TS variants did not appear to be due to the nuclear surveillance pathway, we considered the possibility that the RTD pathway could still function in a met22Δ strain, although previous results had suggested otherwise (Chernyakov et al. 2008; Dewe et al. 2012). For example, a trm8Δ trm4Δ mutant is temperature sensitive because the RTD pathway degrades tRNAVal(AAC) due to the lack of m7G46 and m5C49, and this phenotype is fully suppressed by a met22Δ mutation (Chernyakov et al. 2008). To examine whether RTD could act in a met22Δ strain, we transformed a trm8Δ trm4Δ met22Δ strain with a plasmid overexpressing the 5′–3′ exonuclease Rat1 under galactose control. Consistent with the ability of RTD to act in a met22Δ mutant, overexpression of Rat1 and its partner Rai1 in media containing galactose caused temperature sensitivity in a trm8Δ trm4Δ met22Δ strain, whereas no difference was observed in media containing glucose, or with a vector control (Fig. 6). Similarly, the temperature sensitivity of a tan1Δ trm44Δ strain caused by decay of tRNASer(CGA) lacking ac4C12 and Um44 by the RTD pathway is prevented by a met22Δ mutation, and overexpression of Rat1 and Rai1 caused mild temperature sensitivity in a tan1Δ trm44Δ met22Δ strain (Fig. 6). The same increased temperature sensitivity was observed when Xrn1, the other 5′–3′ exonuclease of the RTD pathway, was overexpressed in these same strains (Fig. 6). These results argue that met22Δ suppression of RTD is not absolute, and can be overcome under certain conditions.

FIGURE 6.

5′–3′ Exonucleases of the RTD pathway can function in a met22Δ strain. trm8Δ trm4Δ, and tan1Δ trm44Δ mutants and their met22Δ derivative strains (and a WT control) were transformed with plasmids expressing RAT1/RAI1 or XRN1 under control of the galactose promoter (or a vector control), and growth was analyzed in media containing galactose or glucose at the indicated temperatures.

Temperature sensitivity results from a disruption of tRNA structure

Given the high frequency of temperature sensitivity of stem variants in both WT and met22Δ strains, we modeled the importance of thermodynamic features of secondary structure in the prediction of temperature sensitivity using a random forest model (Breiman 2001). The input features included ΔΔG° and the ensemble defect (Zadeh et al. 2011) for each of the single and double variants at 22°C, 28°C, 37°C, and 45°C, as well as the location of the mutation within the molecule. The model was tested using a variation of leave-one-out cross validation that excluded from the training set all variants that shared any mutation with the variant to be tested (see Materials and Methods). Additionally, we tested a simple model that used only ΔΔG° as input, analogous to the model previously used to predict RTD (Guy et al. 2014).

Both the random forest and simple ΔΔG° models predicted temperature sensitivity with good accuracy. The receiver-operator curve of the temperature sensitivity prediction in WT cells using the random forest model had an area under the curve (AUC) of 0.73 (Fig. 7A, blue line), and comparable performance on met22Δ cells, with AUC of 0.71 (Fig. 7B, blue line). The simple ΔΔG° model performed slightly poorer than the random forest model, with an AUC of 0.70 (Fig. 7A, green line) for temperature sensitivity in WT cells and an AUC 0.68 (Fig. 7B, green line) in met22Δ cells. The performance of these predictions suggests that thermodynamic features are strongly predictive of the temperature sensitivity of SUP4 variants, and that the random forest model can better discriminate between TS and non-TS.

FIGURE 7.

Temperature sensitivity can be predicted from thermodynamic properties of the tRNA sequence. (A) Receiver-operator curve for prediction of temperature sensitivity in WT cells, using multiple features and a random forest model (blue), or a simple model using only ΔΔG° (green). (B) Receiver-operator curve for prediction of temperature sensitivity in met22Δ cells, as in A.

Epistasis analysis suggests an important role for the N59 residue

We examined the effects of temperature on genetic interactions in double variants in the WT and met22Δ strains. As we did previously (Guy et al. 2014), we analyzed genetic interactions by assigning each double variant an epistasis score [Epistastis = GFPSEQDouble – (GFPSEQSingle 1 × GFPSEQSingle 2)], which quantified the function of the double variant, relative to that expected from the product of the GFPSEQ scores of the individual single variants. As before, we defined positive epistasis as an epistasis score ≥0.18 and negative epistasis as a score less than or equal to −0.18.

To compare epistasis in WT cells at 28°C and 37°C, we examined the 8799 double variants with GFPSEQ scores at both temperatures (Supplemental Table S3). We observed 49 positive epistatic interactions at 37°C (compared to 39 at 28°C), 21 of which were found at both temperatures, and 28 of which were exclusively at 37°C (Supplemental Fig. S8A). Eight of the 67 double variants with positive epistasis at either 28°C or 37°C had an A44U mutation. We previously noted that the A44U variant improved the function of variants with an anticodon stem mutation, presumably by forming a stronger pairing at the 26–44 tertiary pair, but weakened function of variants with mutations in other parts of the molecule (Guy et al. 2014). Remarkably, another 28/67 double variants with positive epistasis at either 28°C or 37°C had a C59 mutation, most frequently C59A (17/67).

We found similar trends from analysis of positive epistasis of the 5069 double variants in a met22Δ background with scores at both 28°C and 37°C. Consistent with the strong suppression of RTD by a met22Δ mutation (Chernyakov et al. 2008; Whipple et al. 2011), we identified more positive epistatic interactions in this background; 133 at 37°C and 85 at 28°C (Supplemental Fig. S8B), compared to 49 and 39, respectively, in WT. As observed in WT cells, positive epistasis in met22Δ cells most frequently included the C59A and A44U mutations, as well as base pair restoration mutations. In addition the positive epistasis data sets in met22Δ cells included the U4C mutation, which is a known node of positive epistasis in WT cells because this mutation stabilizes the U4–G69 pair (Guy et al. 2014), but was not included here in WT cells because of our more stringent sequencing cutoffs. At 37°C in a met22Δ strain there were 89 double variants with positive epistasis unique to 37°C, including 34/89 with a U4C mutation, 10 with a C59A mutation, and 19 with an A44U mutation.

Further analysis of the variants with positive epistasis suggested that mutation of C59 improves some aspect of tertiary structure, since an overwhelming fraction of the other double variants with positive epistasis affected known or likely structural elements. For example, of the 21 double variants with positive epistasis in WT cells at both temperatures, the 13 variants lacking a C59 mutation included seven that restore base pairs, three with a mutation that restores or modifies a tertiary pair (U17G G19U, A29U A44U, and A28U A44U) (Guy et al. 2014), two that affect the 13–22 pair found in many other tRNAs (A13U A22G and A13U A22U), and one that affects two residues involved in tertiary interactions (A9G G45C) (Supplemental Table S3). Indeed, all but five of the 67 double variants with positive epistasis at either temperature had mutations in C59 or in residues involved in structure. Moreover, this trend continued in the positive epistasis data set in the met22Δ strain. Of the 173 variants with positive epistasis at 28°C or 37°C, 24 (13.9%) had the C59A mutation (C59G and C59U mutations did not pass quality control for this library). The other 149 variants included 23 that restored a base pair, 43 with a stabilizing U4C mutation, 24 with an A44U mutation likely stabilizing the 26–44 tertiary pair, 40 affecting a tertiary interaction found in tRNAs and 19 variants with unexplained effects. Based on the very strong association in the positive epistasis data set of C59-containing double variants with variants containing known or likely structural mutations, we speculate that C59 mutations have a stabilizing effect on structure.

We ruled out the possibility that mutations of C59 increase transcription of SUP4 tRNA. The internal Pol III promoter of tRNA genes includes Box A nucleotides 8–21 and Box B nucleotides 53–61, with no defined role for N59 (Allison et al. 1983; Geiduschek and Tocchini-Valentini 1988; Marck et al. 2006). Consistent with the lack of a transcription role, we found that a C59A mutation did not significantly increase SUP4 tRNA levels at either temperature, but dramatically increased tRNA levels of a de-stabilized G50A variant at both temperatures, which is in accord with the positive epistasis score of G50A C59A at both 28°C and 37°C in WT (Supplemental Fig. S9). Thus, the most parsimonious explanation is that the C59A mutation confers stability to SUP4 tRNA rather than causing an increase in transcription.

DISCUSSION

One conclusion that arises from this work is the prevalence of temperature sensitivity among tRNATyr variants in the context of SUP4 suppression in WT cells. tRNAs are known to be stable, with half-lives in yeast on the order of 9 h (Gudipati et al. 2012), and half-lives in metazoans and protozoans on the order of days (Nwagwu and Nana 1980; Kanerva and Mäenpää 1981; Karnahl and Wasternack 1992). Moreover, tRNAs have a high tolerance for mutations, maintaining some function in a number of different single and double variants (Guy et al. 2014). However, our finding that 73.1% (656/897) of scoreable single and double variants were temperature sensitive emphasizes the extreme degree to which tRNA function is reduced by an increase of only 9°C, from 28°C to 37°C.

There are two notable exceptions to the prevalence of temperature sensitivity among variants. First, mutation of uridine residues in the D-loop that are normally modified to dihydrouridine were relatively benign. Since dihydrouridine is prevalent in psychrophiles and leads to increased conformational flexibility (Dalluge et al. 1996, 1997), it is possible that at high temperature the dihydrouridine contribution to tRNA flexibility is less important and that mutation of these residues is, therefore, benign. Second, the A46G mutation is associated with cold sensitivity as a single variant or in the context of several double variants. One interpretation of this result is that A46G leads to an inhibitory structure at lower temperature that is disrupted at higher temperature. Since G46 of tRNAPhe is involved in a stabilizing base triple with G22 of the C13–G22 pair (Kim et al. 1974b), it is possible that the G46 variant of SUP4 forms an inhibitory triple with G23 of the C12–G23 pair, which is otherwise expected to pair with A9.

Another conclusion is that a large fraction of the temperature sensitivity of SUP4 variants in WT (MET22+) cells can be attributed to RTD. Indeed 55.5% of variants with a TS ratio ≥2 were RTD substrates at 28°C or 37°C, and this value rose to 84%, 91.0%, and 96.5% for TS ratios ≥3, ≥4, and ≥6, respectively. This high degree of association of temperature sensitivity with RTD emphasizes that the RTD pathway is responsible for the majority of the temperature-dependent quality control for this tRNA species. In support of this claim, two TS variants of SUP4 that were RTD substrates had reduced tRNA at 37°C compared to 28°C, which was partially rescued by a met22Δ mutation, directly implicating the RTD pathway in the temperature sensitivity. Since the 5′-3′ exonucleases of the RTD pathway are conserved in eukaryotes (Nagarajan et al. 2013), and there is evidence for RTD in humans at high temperature (Watanabe et al. 2013), we speculate that the RTD pathway might also have a role in temperature dependent quality control throughout eukaryotes.

The association of RTD with temperature sensitivity might also be extended to met22Δ mutants, since an overwhelming majority of met22Δ TS variants (81.1%) were also substrates for RTD at 28°C or 37°C. This temperature sensitivity in met22Δ mutants could be due to RTD, because decay is occurring in most of the tested variants, and because the nuclear surveillance pathway does not contribute to this decay in several tested variants. Furthermore, both Rat1 and Xrn1 can act in met22Δ mutants when overexpressed, implying that the exonucleases are not fully inhibited by this mutation; this conclusion is consistent with the observations that RAT1 is an essential gene (Amberg et al. 1992) and xrn1Δ mutants grow poorly (Kim et al. 1990), whereas a met22Δ mutant is healthy (Chernyakov et al. 2008). However, it is also possible that decay occurs in these SUP4 variants in a met22Δ strain due to another as yet unidentified pathway.

Our ability to predict temperature sensitivity of SUP4 variants in a WT strain by calculation of ΔG° may be a consequence of the tight connection between temperature sensitivity and RTD, and between RTD and instability leading to exposure of the 5′ end (Whipple et al. 2011; Guy et al. 2014). However, we note that it is difficult to determine if the important free energy parameter for prediction of temperature sensitivity is ΔΔG° (the difference in ΔG° between the variant and the WT variant) or ΔG° (the calculated free energy change of folding for each variant). Since the two parameters differ only by the constant represented by ΔG° of the WT variant, their relative importance cannot be distinguished computationally. ΔΔG° may be the more important factor since predicted ΔG° values of tRNAs naturally exhibit a range of more than 10 kcal/mol within the same organism (Fig. 8; Chan and Lowe 2009). However, it is possible that variabilities in the contributions of modifications and specific tertiary interactions in different tRNA species would buffer calculated ΔG° values.

FIGURE 8.

Distribution of predicted folding free energies for tRNA sequences from diverse species. Structure alignments were downloaded from the genomic tRNA database (Chan and Lowe 2009) and their energies were calculated using RNAstructure 5.8. Shown are all cytoplasmic tRNA sequences from chimpanzee (Pan troglodytes), mouse (Mus musculus), budding yeast (S. cerevisiae), and maize (Zea mays).

One such tertiary interaction that could vary in different tRNAs could involve N59. N59 of SUP4 may play a previously unappreciated role in tRNATyr structure, since 28 of 67 variants with positive epistasis in WT cells at either temperature had C59 mutations (17 C59A, 9 C59G, and 2 C59U), and all but five of the remainder were mutated in known secondary interactions or tertiary interactions documented in different tRNAs. The positive epistasis of variants with C59 mutations is general since it affected variants with mutations in every region of the tRNA. This positive epistasis of variants with C59 mutations was also prevalent in variants scored in a met22Δ strain. Furthermore, although the B Box of the internal pol III promoter covers the region of nucleotides 53–61, nucleotide 59 has not been implicated as a promoter recognition element either by phylogenetic analysis (Marck et al. 2006) or by experiment (Geiduschek and Tocchini-Valentini 1988), consistent with our primer extension analysis. One interpretation of our results is that C59 destabilizes SUP4 structure. N59 and N60 are known to stack on the Levitt base pair, the conserved N15–N48 tertiary base pair found in all tRNAs (Ladner et al. 1975; Giegé et al. 2012), but there are differences in the angle of the stacking in different tRNAs, which could change stability with different N59 residues.

The preponderance of temperature sensitivity among SUP4 variants revealed in response to a 9°C temperature increase was unexpected. Previous physical analysis of tRNAs showed first melting transitions for unfolding of tertiary and secondary structures at or above 50°C, albeit in buffers that differ from the cellular environment (Coutts et al. 1974; Crothers et al. 1974; Hilbers et al. 1976). Indeed, variants in our library with a ΔΔG° as modest as ∼2.0 kcal/mol could result in greater than or equal to twofold loss of function at temperatures as low as 37°C. We interpret these data to mean that the RTD pathway examines tRNAs to detect small changes in structure rather than folding transitions. In support of this interpretation, we previously showed that small differences in predicted stability of tRNASer(CGA) variants could also provoke decay in vivo at elevated temperatures, increasing the exposure of the tRNA 5′ end to Rat1 and Xrn1 exonucleases (Whipple et al. 2011), which presumably allows the exonucleases to gain a foothold (Jinek et al. 2011). Subsequent analysis will be needed to address the mechanism by which small predicted stability changes can provoke drastic changes in stability in vivo.

MATERIALS AND METHODS

Yeast strains

The yeast strains used for SUP4 library construction, FACS, and analytical flow cytometry (YK 380-1: BY4741 can1::P and its met22Δ::KanR derivative YK 391-1) were as previously described (Guy et al. 2014). Individual SUP4 variants in plasmid AB230-1 were introduced into strains YK 380-1 and YK 391-1 by digestion with the restriction enzyme Stu1 (NEB), linear transformation at the ADE2 locus, and selection on S-His dropout medium. Three biological isolates were picked and saved at −80°C in YPD +8% DMSO.

Plasmids

Individual SUP4 variants were constructed by ligation of overlapping oligomers (IDT) into the tRNA gene cassette of plasmid AB230-1 after digestion with Bgl II and XhoI to insert the tRNAs into the corresponding flanking regions of the tH(GUG)G2 gene, as previously described (Guy et al. 2014).

The final inserted sequence for SUP4 was as follows: 5′-agatctaacaaagttcataaagaaattaCTCTCGGTAGCCAAGTTGGTTTAAGGCGCAAGACTTTTATCACTACGAAATCTTGAGATCGGGCGTTCGACTCGCCCCCGGGAGAttttttcctcgag-3′ where the SUP4 exon sequence is capped and underlined, the anticodon is in bold, the intron is in italics, and the 5′ leader and the 3′ trailer derived from the tH(GUG)G2 flanking region is in lower case, with the flanking Bgl II and Xho I sequences underlined (Whipple et al. 2011). Variants constructed from SUP4 all contained one or more mutations in the exon region.

Analytical flow cytometry

Strains were grown overnight at 28°C in YP medium containing 2% raffinose and 2% galactose supplemented with 80 mg/L adenine (YP Raff/Gal + Ade) for ∼12 h and then used to inoculate fresh prewarmed YP Raff/Gal + Ade at 28°C and 37°C for another ∼12 h of growth. Cells were diluted during growth as needed so that they were maintained in log phase, and were analyzed at an OD600 between 0.8 and 1.2. Samples were then analyzed by flow cytometry as previously described (Guy et al. 2014). The GFPFLOW score for analyzed variants was derived from the median GFP divided by the median RFP for 10,000 recorded events with an RFP fluorescence above the cutoff of 5 × 103; standard deviations were derived from the analysis of biological triplicates in this manner and variants were normalized to the GFPFLOW score of SUP4 measured in their respective conditions of analysis.

SUP4 library construction and analysis

The libraries of yeast SUP4 variants in WT (YK380-1) and met22Δ (YK391-1) strains were constructed previously, and contained 3% random mutations in SUP4 residues 1–33 and 38–73 (Guy et al. 2014). Each library was comprised of ∼200,000 SUP4 variants that were transformed, derived from a master library of 325,000 E. coli plasmids. To analyze the yeast libraries, ∼4.5 million cells of the yeast SUP4 libraries were grown in YP medium containing 2% raffinose supplemented with 80 mg/L adenine for 24 h at 28°C and then diluted to an OD600 of 0.05 into YP Raff/Gal + Ade at 37°C. Cells were sorted by FACS on an Aria-11 cell sorter (BD Biosciences) at the University of Rochester Medical Center Flow Cytometry Core facility. The cells were sorted as previously described (Guy et al. 2014) into four bins based on GFP fluorescence whose borders were based on three control strains grown at 28°C with reproducible fluorescence intensities termed “GO” (RNA-ID Reporter without a 5′ stop codon in GFP), “STOP” (RNA-ID Reporter with a 5′ stop codon in GFP and no suppressing tRNA), and “CGA3” (RNA-ID Reporter with the inhibitory codons CGA3 at the 5′ end of GFP); we also only collected cells with a minimum RFP fluorescence >5 × 103 (Dean and Grayhack 2012; Guy et al. 2014). In total, at least 2 million cells were collected for each library and then plated onto YPD medium for 3 d at 25°C before being scraped, pooled, and stored at −80°C. Genomic DNA was isolated from frozen aliquots of the stored cells for each bin and used for deep sequencing.

Isolation of bulk RNA

WT and met22Δ strains with integrated SUP4 variants were grown in YPD media at 28°C or 37°C and harvested at OD ∼1, or at defined time points after shift from 28°C to 37°C, or after addition of thiolutin to 5 µg/mL and shift to 37°C. Bulk low-molecular-weight RNA was extracted from ∼2 OD-mL pellets by hot phenol extraction followed by ethanol precipitation and resuspension in ddH2O as previously described (Jackman et al. 2003).

Primer extension analysis of SUP4 variants

Poison primer extension assays were carried out as previously described (Guy et al. 2014), using one of two 5′ end labeled primers (complementary to nucleotides 57–37 [P5] or 62–43 [P7] of SUP4) in the presence of ddCTP. Reaction products were resolved on a 15% polyacrylamide 7 M urea gel, and then dried and exposed to a phosphorimager plate for analysis.

Sequencing

The SUP4 gene with 27 5′ and 16 3′ nucleotides was amplified for 20 cycles (10 sec at 98°C, 30 sec at 52°C, and 30 sec at 72°C) from ∼1 to 3 µg of genomic DNA as previously described (Guy et al. 2014).

Sequence assembly and quality filtering

Sequences were trimmed and demultiplexed using a custom script. Forward and reverse reads were combined using Enrich version 0.2. Reads with a phred score lower than 30 were removed from subsequent analysis.

Calculation of GFPSEQ

GFPSEQ was calculated as previously described (Guy et al. 2014). The number of reads in each bin was tabulated using Enrich version 0.2. The read counts for each variant in each bin were converted to an estimated number of cell counts by multiplying the frequency of each read in the bin by the number of cells collected for that bin. Cell counts for each variant were converted to a GFPSEQ score by taking the average GFP/RFP for each bin, weighted by the estimated number of cells for that bin, and normalized to a GFPFLOW value measured for the WT sequence. That is,

Data quality control

Data for individual variants were filtered for quality as previously described (Guy et al. 2014). A variant was discarded if it had fewer than 100 reads or fewer than 30 estimated cell counts

As we previously observed, PCR chimerism can lead to spurious WT reads for some variants. This is visible as a bimodal distribution of cell counts in the four bins. Variants were discarded if the number of cell counts in bins 1 and 4 were within 10-fold of one another and bins 1 and 4 were the bins with the highest number of cell counts. A manual inspection of the double variants indicated this filtering scheme was sufficient to remove the variants with invalid scores.

An interactive graphic allowing visualization of function under each condition, RTD, TS, and epistasis is available at https://rna.urmc.rochester.edu/tRNA/2017-10-02_interactive_trna.html.

Statistical significance of enrichment of TS variants in stems

The statistical significance of the enrichment of TS variants in stem regions was assessed using a χ2 test, with a Type I error rate, α, set to 0.05.

Thermodynamic calculations

The RNAstructure software package version 5.8 (Reuter and Mathews 2010) was used for all thermodynamic calculations, using folding free energy changes adjusted to the appropriate temperature (Lu et al. 2006).

The ΔΔG° value for each variant was calculated with the nearest neighbor rules, using the simplified non-logarithmic formation of the multibranch loop energy used by the structure prediction algorithms (efn2 flag --simple) (Mathews et al. 2004; Reuter and Mathews 2010). Because SUP4 contains numerous G–U wobble pairs, we included a recently updated set of parameter values for G–U base pair stacks (Chen et al. 2012). Nucleotides in the reference structure were unpaired in the calculation if they were mutated in a way that prevented canonical A–U, G–C, or G–U base pairing.

The normalized ensemble defect for each variant is given by where n is the length of the nucleotide sequence (in this case 78 nt), is the estimated probability that the nucleotide at position i is unpaired, is the estimated probability the nucleotide at position i is paired to its correct pairing partner in the reference structure, and πi is an indicator variable taking a value of one if the nucleotide at position i is paired in the reference structure and zero if it is unpaired in the reference structure (Zadeh et al. 2011). Thus, the normalized ensemble defect is the average probability that a nucleotide is not forming the desired secondary structure (either a specific pair or being unpaired).

Prediction of temperature sensitivity

Temperature sensitivity was predicted for each variant for which temperature sensitivity could be assessed using the RandomForestClassifier model from scikit-learn (Pedregosa et al. 2011) with input features of ΔΔG° at 28°C, ΔΔG° at 37°C, ED at 28°C, and ED at 37°C. Additional features tested included ΔΔG° and ED at 22°C and 45°C, and indicator variables indicating presence of a mutation in the acceptor stem, D-stem, anticodon stem, T-stem, a mismatch position adjacent to any stem, or in a loop not adjacent to a stem.

Cross validation was performed using the leave-one-out method. For each variant in the data set, the model was retrained on the set of variants that contained no mutations in common with the variant to be tested, and temperature sensitivity was predicted for the remaining variant.

The random forest classifier produces a value between zero and one that represents the model's confidence that this example belongs to the positive class (temperature sensitive variants, in this case). For the reported accuracy scores, a variant was considered “predicted TS” if the model predicted 50% confidence that the variant was TS. Receiver-operator characteristic curves were generated by assessing sensitivity and false positive rate at every possible threshold of confidence values.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.