Impaired phloem loading in genome-edited triple knock-out mutants of SWEET13 sucrose transporters

Crop yield depends on efficient allocation of sucrose from leaves to seeds. In Arabidopsis, phloem loading is mediated by a combination of SWEET sucrose effluxers and subsequent uptake by SUT1/SUC2 sucrose/H+ symporters. ZmSUT1 is essential for carbon allocation in maize, but the relative contribution to apoplasmic phloem loading and retrieval of sucrose leaking from the translocation path is not known. We therefor tested whether SWEETs are important for phloem loading in maize. Here we identified three leaf-expressed SWEET sucrose transporters as key components of apoplasmic phloem loading in Zea mays L. Notably, ZmSWEET13 paralogs (a, b, c) are among the highest expressed genes in the leaf vasculature. Genome-edited triple knock-out mutants are severely stunted. Photosynthesis of mutants was impaired and leaves accumulated starch and soluble sugars. RNA-seq revealed profound transcriptional deregulation of genes associated with the photosynthetic apparatus and carbohydrate metabolism. GWAS analyses may indicate that variability in ZmSWEET13s is correlated with agronomical traits, specifically flowering time and leaf angle. This work provides support for cooperation of three ZmSWEET13s with ZmSUT1 in phloem loading in Zea mays L. Our study highlights these three ZmSWEET13 sucrose transporters as possible candidates for the engineering of crop yield. One Sentence Summary Three SWEET sucrose transporter paralogs are necessary for phloem loading in maize.


Introduction 1
Crop yield is critical for human nutrition, yet the underlying machinery that ultimately 2 determines yield potential is still not understood. Crop productivity under ideal 3 conditions is determined by the efficiency with which plants intercept light, convert it into 4 chemical energy, translocate photosynthates, and convert these to storage products in 5 harvestable organs (Zhu et al., 2010). In many crops, sucrose is the primary form for 6 translocation inside the conduit (the phloem). A combination of SWEET-mediated efflux 7 from phloem parenchyma and subsequent secondary active sucrose import by SUT 8 sucrose/H + symporters is thought to create the driving force for pressure gradient-driven 9 phloem transport and retrieval of sucrose leaking along the translocation path (Chen et 10 al., 2015a). 11 There is a debate regarding the mechanisms of phloem loading in crops. Sucrose is 12 thought to follow one of three routes: (i) apoplasmic loading via plasma membrane 13 transporters, (ii) symplasmic loading via diffusion through plasmodesmata, or (iii) 14 polymer trapping via enzymatic addition of galactinol, which is thought to impair back-15 diffusion through plasmodesmata (Turgeon & Wolf, 2009;Chen et al., 2015a). Some 16 mechanisms may coexist, as suggested by anatomical studies which have found thin 17 and thick walled sieve tubes in monocots, cell types that may differ regarding the 18 primary loading mechanism (Botha, 2013). 19 In Arabidopsis, a SWEET/SUT-mediated apoplasmic mechanism appears important for 20 phloem loading (Chen et al., 2012(Chen et al., , 2015a. SWEETs are a class of seven 21 transmembrane helix transporters that function as hexose or sucrose uniporters (Xuan 22 et al., 2013). Multiple SemiSWEETs and SWEETs have been crystallized, and 23 AtSWEET13 has been proposed to function in complexes via a 'revolving door' 24 mechanism to accelerate transport efficacy (Feng & Frommer, 2015;Latorraca et al., 25 2017;Han et al., 2017). In Arabidopsis, SWEET roles include phloem loading, nectar 26 secretion, pollen nutrition, and seed filling (Chen et al., 2012;Sun et al., 2013;Lin et al., 27 2014;Sosso et al., 2015). In rice, cassava, and cotton, SWEETs act as susceptibility 28 factors for pathogen infections (Chen et al., 2010;Cohn et al., 2014;Cox et al., 2017). 29 AtSWEET11 and 12 are likely responsible for effluxing sucrose from the phloem 30 parenchyma into the apoplasm (Chen et al., 2012). Sucrose is subsequently loaded 31 against a concentration gradient into the SECC via the SUT1 sucrose/H + symporter 32 (a.k.a. AtSUC2), powered by the proton gradient created by co-localized H + /ATPases 33 (Riesmeier et al., 1994;Gottwald et al., 2000;Slewinski et al., 2009;Srivastava et al., 34 2009). Though the fundamental involvement of SUT transporters in phloem loading has 35 been demonstrated using RNAi and knock-out mutants in Arabidopsis (also in potato, 36 tobacco, tomato, and maize (Riesmeier et al., 1994;Bürkle et al., 1998;Srivastava et 37 al., 2009;Chen et al., 2015a), atsweet11,12 and atsuc2 (sut1) mutants only showed 38 partially impaired phloem loading. 39 In monocots, including all cereal crops, the situation is less clear. In maize, the phloem-40 expressed ZmSUT1 (Baker et al., 2016) (phylogenetically in the SUT2 clade) appears to 41 be critically important for phloem translocation (Slewinski et al., 2009), whereas rice 42 ossut1 mutants and RNAi lines had no apparent growth or yield defects (Ishimaru et al., 43 2001;Scofield et al., 2002;Eom et al., 2012). As a result, there is an ongoing debate 44 regarding the mechanisms behind phloem loading in cereals (Braun et al., 2014;Regmi 45 et al., 2016). 46 Here we identified a set of three close paralogs of SWEET13 from Z. mays L. as 47 essential sucrose transporters for phloem loading. 48

Material and Methods 49
Plant material and growth conditions 50 zmsweet13a, zmsweet13b and zmsweet13c alleles were obtained with a CRISPR-Cas9 51 construct targeting a sequence (5'-GCATCTACAAGAGCAAGTCGACGG-3', the 52 underlined CGG for PAM) conserved in all three paralogs in the 3rd exon as described 53 (Char et al., 2017). T0 plants were selfed or outcrossed to B73 and plants which did not 54 contain the CRISPR construct were selected for further analysis. T1, T2, and T3 plants 55 homozygous for all three mutated genes (zmsweet13abc) were selected along with 56 wild-type siblings. Height was assessed by weekly measurement from the soil surface 57 to the top of the highest fully-developed leaf. Wild-type "siblings" were descendants of 58 the Hi-II plants transformed and outcrossed once to B73, which in the T1 generation did 59 not carry the CRISPR-Cas9 construct or any detectable mutations. Triple mutant plants 60 either descended from selfed T0 Hi-II plants or outcrossed once to B73. The mutant 61 phenotype was unaffected by the genetic difference. Mutants and wild-type plants were 62 grown side by side, in greenhouses under long-day conditions (16h day/8h night, 28-63 30 C), and in 2016 in a summer field at Carnegie Science (Stanford, California, USA). 64

Genotyping of rice and maize plants 65
Genomic DNA was extracted from leaves using a Qiagen Biosprint 96. PCR was 66 performed with the Terra PCR Direct Red Dye Premix Protocol (Clontech Laboratories) 67 with melting temperatures of 60 °C, 64 °C, and 62.5 °C for ZmSWEET13a, b, and c, 68 respectively (for primers see Table S2). Amplicons of relevant regions of the CRISPR-69 Cas9 targeted ZmSWEET13 alleles were sequenced by Sequetech (Mountain View, 70 CA). Chromatograms were analyzed using 4Peaks (www.nucleobytes.com/4peaks/). 71

Phylogenetic analyses 81
The evolutionary history was inferred by using Maximum Likelihood with a JTT matrix-82 based model. The tree with the highest log likelihood (-3000.1) is shown. The 83 percentage of trees in which associated taxa clustered together is shown next to the 84 branches. Initial tree(s) for the heuristic search were obtained by Neighbor-Joining to a 85 matrix of pairwise distances with the JTT model used for estimation. The analysis 86 involved 16 polypeptides sequences. A minimum of 95% site coverage was required so 87 that no more than 5% alignment gaps, missing data, and ambiguous bases were 88 allowed at any position. There were a total of 252 positions in the final dataset. 89 Evolutionary analyses were conducted in MEGA6. 90

Soluble sugar analyses 91
Flag leaves were harvested from mature plants at 7:00 am. 70 mg of liquid nitrogen-92 ground tissue was incubated for 1 hour with 1 ml of 80% ethanol on ice with frequent 93 mixing. Samples were spun for 5 min at 4 °C at 13,000 g, and supernatant was 94 removed. This step was repeated once. The liquid supernatant was subsequently dried 95 in a vacuum concentrator and re-suspended in water. Sucrose, glucose and fructose 96 were measured using NAD(P)H-coupled enzymatic methods using a plate reader 97 M1000 (Tecan), with measured values normalized to fresh weight. Starch quantification 98 was performed as previously described (Sosso et al., 2015). 99

Starch staining 100
Flag leaves collected at 7:00 am were were boiled in 95% ethanol for approximately 30 101 min (until chlorophyll pigments disappeared). Cleared leaves were submerged in 102 saturated Logol's iodine (IKI) solution for 15 min, rinsed twice with H 2 O, and imaged 103 with a Lumix GF1 camera (Panasonci, Kadoma, Osaka, Japan). The IKI solution used 104 for starch staining was made by adding 1 g of iodine and 1 g of Potassium Iodide to 100 105 mL H 2 O. 106

RNA isolation and transcript analyses 107
RNA was extracted using the Trizol method (Invitrogen). First strand cDNA was 108 synthesized using Quantitect reverse transcription Kit (Qiagen). qRT-PCR to determine 109 expression level was performed using the LightCycler 480 (Roche), and the 2 −ΔCt 110 method for relative quantification. Wild-type maize and zmsweet13abc flag leaves were 111 sampled at 5:00 pm. Primers in the last exon and the 3′ UTR of ZmSWEET13a, b, and c 112 (Supplementary Table 2) were used for qRT-PCR to determine gene expression levels. 113 Internal references were Zm18s and ZmLUG. 114

FRET sucrose sensor analysis in HEK293T cells 115
ZmSWEET13a, b, and c coding sequences were cloned into the Gateway entry vector 116 pDONR221f1, followed by LR recombination into pcDNA3.2V5 for expression in 117 HEK293T cells. HEK293T cells were co-transfected with ZmSWEET13a, b, or c in 118 pcDNA3.2V5 and the sucrose sensor FLIPsuc90µ∆1V (Chen et al., 2012) using 119 Lipofectamine 2000 (Invitrogen). For FRET imaging, HBSS medium was used to 120 perfuse HEK293T/FLIPsuc90µ∆1V cells with defined pulses containing 20 mM sucrose 121 in buffer. Image acquisition and analysis were performed as previously described (Chen 122 et al., 2012). AtSWEET12 was used as a positive control. Negative control were empty 123 vector transfectants. 124

Analyses of photosynthetic rates 136
Licor LI-6800 measurements were taken at mid-day under normal greenhouse 137 conditions (28 °C, PAR 1000, 60% relative humidity). 2 cm diameter disc of leaf was 138 clamped inside the Licor measurement chamber and relative levels of CO 2 inside and 139 outside of the chamber were measured, with µmol m −2 s −1 CO 2 absorbed by leaf 140 segment inside chamber used as a proxy for photosynthesis rate. Measurements were 141 made at the tips of Leaf 7 -Leaf 10 at mid-day. 142

Candidate gene association study 143
To test whether sequences at SWEET loci are associated with phenotypic variations in 144 the maize population, we analyzed the maize diversity panel composed of 282 inbred 145 lines (HapMap3 SNP data (Bukowski et al., 2015) for the panel from the Panzea 146 database (www.panzea.org)). We filtered SNP data (MAF > 0.1; missing rate < 0.5) 147 using PLINK (Purcell et al., 2007) and calculated a kinship matrix using GEMMA (Zhou 148 & Stephens, 2012) using the filtered SNP set. GWAS was performed by fitting a mixed 149 linear model using GEMMA, where the kinship matrix was fitted as random effects in the 150 model. An FDR approach (Benjamini & Hochberg, 1995) was employed to control the 151 multiple test problem with a cutoff of 0.05. Linkage disequilibrium of SNPs in our 152 candidate genes with significant association SNPs was calculated using PLINK (Purcell 153 et al., 2007). 154

RNA-seq and data analysis 155
zmsweet13abc triple mutants and WT siblings were grown in soil under greenhouse 156 conditions. Total RNA was isolated from flag leaf tissues using acidic phenol extraction 157 as described previously (Eggermont et al., 1996). Purification of poly-adenylated mRNA 158 using oligo(dT) beads, construction of barcoded libraries, and sequencing using Illumina 159 HiSeq technology (150 bp paired-end reads) were performed by Novogene 160 GeneCounts). Unique reads were filtered by mapping quality (q20) and PCR duplicates 166 removed using Samtools (v. 1.3.1). Gene expression was analyzed in R (v. 3.4.1) using 167 DEseq2 software (v. 1.16.1) (Love et al., 2014). Genes were defined as differentially 168 expressed by a two-fold expression difference with a p-value, adjusted for multiple 169 testing, of < 0.05. Accession numbers for the RNA-Seq data in the Gene Expression 170 Omnibus database will be made available. 171

Results 172
To directly test whether SWEETs are involved in phloem loading in maize, we evaluated 173 the role of leaf-expressed maize SWEETs in carbon allocation. We identified three 174 SWEET13 paralogs (GRMZM2G173669: ZmSWEET13a, GRMZM2G021706: 175 ZmSWEET13b, GRMZM2G179349: ZmSWEET13c) as the most highly expressed 176 SWEETs in maize leaves (Fig. S1). ZmSWEET13a and b are located in tandem on 177 chromosome 10 in a region syntenic with the OsSWEET13 locus in rice, while 178 ZmSWEET13c is on chromosome 3 (Fig. S2). Interestingly, maize appeared to be one 179 of few cereals carrying three SWEET13 paralogs in its genome, along with S. bicolor 180 and T. urartu (Figure 1a; Fig. S3). ZmSWEET13a, b and c are preferentially expressed 181 in bundle sheath/vein preparations rather than mesophyll (Fig. S4), as is ZmSUT1. 182 ZmSUT1 and the three SWEETs showed higher expression in leaf tips (Fig. S5). We 183 tested the transport activity of the three SWEETs by coexpressing each with sucrose 184 FRET (Förster resonance energy transfer) sensors in human HEK293T cells (Chen et 185 al., 2010(Chen et 185 al., , 2012. All three SWEETs mediated sucrose transport in mammalian cells 186 (Fig. 1b). To test whether these SWEETs were part of (i) intercellular translocation or 187 (ii) intracellular sugar sequestration similar to the Arabidopsis SWEET2, 16 or 17 188 Klemens et al., 2013;Guo et al., 2014;Chen et al., 2015b), we 189 tested their subcellular localization in transiently transformed tobacco cells, and found 190 that they localized preferentially to the plasma membrane (Fig. 1c). 191 Recently, ZmSWEET13 had been implicated as a possible key player in C4-192 photosynthesis in grasses (Emms et al., 2016). To test their role in maize, we designed 193 guide RNAs that target a conserved region within a transmembrane domain, assuming 194 that such mutations would lead to complete loss of function. We generated single 195 knock-out mutants, as well as combinations of mutant alleles, using CRISPR-Cas9 (Fig.  196   S6). Genome editing allowed us to recover two mutant alleles of ZmSWEET13a, four of 197 ZmSWEET13b and three for ZmSWEET13c. The majority of mutations were caused by 198 single nucleotide insertions in the target sequence. All mutations create premature 199 STOP codons leading to truncated polypeptides (at amino acid 129 in the fourth of 200 seven transmembrane domains) (Fig. S6). T2 lines carrying homozygous mutations in 201 all three genes were characterized by severe growth defects (Fig. 2a). The growth 202 phenotype was verified in subsequent generations in the greenhouse and a single field 203 season. Single and double mutants showed slight growth defects, while triple mutants 204 showed substantial defects: plants were severely stunted with shorter, narrower leaves 205 ( Fig. 2a-c). Leaves were chlorotic, accumulated 5x more starch and 4x more soluble 206 sugars compared to wild-type (Fig. 3a-c), likely a consequence of impaired phloem 207 loading. Accumulation of sugars and starch occurred primarily in mesophyll and bundle 208 sheath cells and strongly impacted photosynthesis, even in greenhouse-grown mutant 209 plants (Fig. 3d, e; Fig. S7). In the field, triple mutants from five independent allelic 210 combinations presented even more severe phenotypes, with extreme chlorosis, 211 massive anthocyanin accumulation and extremely stunted growth; in several cases 212 resulting in lethality (Fig. S8). SWEET13 mRNA levels were drastically reduced in all 213 three ZmSWEET13s, as quantified by RNA-seq and qRT-PCR ( Fig. S9 and Fig. 2d). In 214 summary, the strong phenotype of the triple mutant is consistent with maize using 215 predominantly an apoplasmic phloem loading mechanism. 216 Despite the severe defects, triple mutant plants grown in the greenhouse (as well as a 217 subset in the field) exported sufficient sugars from leaves to produce viable seeds. A 218 possible explanation for the viability of the triple mutants could be compensation by 219 other sucrose-transporting clade III SWEETs. To test this hypothesis and to obtain 220 insights about possible physiological changes in the mutants, we performed an RNA-221 seq on flag leaves of wild-type (Hi-II outcrossed once to B73) and triple mutant plants 222 (Hi-II background). Notably, we did not observe significant enrichment of mRNA of clade 223 III SWEETs, arguing against transcriptional compensation by other clade III SWEETs 224 (Fig. S9). Consistent with impaired photosynthesis and chlorosis, mRNA levels of 225 multiple genes encoding functions in the light harvesting complex and 226 chlorophyll/tetrapyrrole biosynthesis were substantially reduced in triple mutants (Fig.  227   S10 and Fig. S11). Furthermore, in line with the accumulation of starch and soluble 228 sugars in leaves, we found that transcripts related to carbohydrate synthesis and 229 degradation were affected in the triple mutants (Fig. S12). 230 A recent study has found that the Arabidopsis homolog AtSWEET13 (although 231 phylogenetically not the closest homolog of ZmSWEET13) can also transport the plant 232 hormone gibberellin (Kanno et al., 2016). The observed phenotypes for the triple 233 zmsweet13 knock out mutants in maize are consistent with a primary role in sucrose 234 transport and distinct from the ones observed in the Arabidopsis sweet13;14 double 235 mutant; i.e., male sterility, increased seedling and seed size (Kanno et al., 2016). 236 To determine if variation in the ZmSWEET13 genes may account for differences in 237 agronomically important traits in existing maize lines, we conducted a genome-wide 238 association study (GWAS) using phenotypic traits obtained from maize diversity panel 239 (Flint-Garcia et al., 2005). We obtained genotypic data from maize HapMap3 SNPs 240 (Bukowski et al., 2015) and filtered out SNPs having minor allele frequency < 0.1 and 241 missing rate > 0.5, leaving ~13 million SNPs for analyses. We performed GWAS using a 242 mixed linear model approach (Zhou & Stephens, 2012), where kinship calculated from 243 the genome-wide SNPs was fitted as the random effects. The SNPs that passed the 244 FDR threshold of 0.05 and showed linkage disequilibrium (R 2 > 0.8) with ZmSWEET13s 245 genes were considered significant associations. SNPs in ZmSWEET13s were 246 significantly associated with ear-related traits (i.e. ear rank number and ear height) and 247 developmental traits (i.e. days to silk, days to tassel, middle leaf angle, and germination 248 count) (Fig. S13 and Fig. S14). While these results are compatible with a key role of 249 ZmSWEET13s in carbon allocation, it will be necessary to determine whether 250 polymorphisms in these genes or flanking regions are causative for these traits. 251

Discussion 252
Phloem sap of many monocots and dicots contains very high sucrose concentrations, 253 and it is thought that this gradient creates the drive for phloem translocation of sucrose 254 and all other molecules in the phloem sap. Inhibition of the production of the SUT1 255 protein, by RNAi or mutations, typically leads to stunted growth and accumulation of 256 carbohydrates in leaves (Riesmeier et al., 1994;Bürkle et al., 1998). Chlorosis and 257 inhibition of photosynthesis, which often accompany the general growth defects, may 258 either be due to feedback inhibition of photosynthesis or may be a consequence of 259 nutrient deficiencies caused by the reduced supply of carbohydrates to the root system 260 (Ainsworth & Bush, 2011). SUTs function as sucrose/H + symporters and appear to fulfill 261 two roles: (i) loading of the SECC with sucrose in source leaves, and (ii) retrieval of 262 sucrose that diffuses out of the SECC as a consequence of the high sucrose 263 concentration in the SECC relative to surrounding tissues. SUTs import sucrose from 264 the cell wall space, implying the existence of transporters that efflux sucrose into the cell 265 wall space preceding uptake by SUTs. AtSWEET11 and 12 are candidates for this role 266 in Arabidopsis: they appear to function as uniporters and can thus serve as cellular 267 efflux systems. Both SWEETs were highly expressed in leaves, localize to the phloem 268 parenchyma, and atsweet11;12 mutants were smaller and accumulated starch in 269 leaves. However, the phenotype was relatively weak, implying leaky mutations, 270 compensation by other transporters, or the coexistence of other phloem loading 271 mechanisms. Other mechanisms could include symplasmic transport, or yet unknown 272 processes. Lastly, is possible that SUTs play a major role in sucrose retrieval along the 273 path in addition to their role in phloem loading, while ZmSWEET13's are thought to be 274 only involved in sucrose efflux prior to SUT1 uptake into phloem. Would a mutant of 275 ZmSWEET13 present weak defects similar to those in atsweet11;12 plants, or a more 276 severe phenotype equivalent that in zmsut1 mutants? 277 Here, we show that maize has three paralogs in clade III of the SWEET family that are 278 among the most highly expressed genes in leaf tissue. They derive from relatively 279 recent gene duplication events: sorghum and wheat have three copies per genome, 280 while Brachypodium and rice each have only one. The comparatively high number of 281 SWEET13s had been attributed to specific roles in C4 photosynthesis, however the 282 presence of three SWEET13s in wheat puts this interpretation into question. Evidence 283 that the maize SWEET13s cooperate in phloem loading is based on two key 284 observations: a severe growth defect similar to that of zmsut1 mutants, and massive 285 accumulation of free sugars and starch in leaves. These phenotypic effects are also 286 similar to the RNAi phenotypes in potato and tobacco (Riesmeier et al., 1994;Bürkle et 287 al., 1998). The observed growth defect in maize is much more severe than that of the 288 atsweet11;12 mutant in Arabidopsis, and comparable to that of the zmsut1 mutant 289 (Slewinski et al., 2009;Chen et al., 2012). We thus propose that the three 290 ZmSWEET13s and ZmSUT1 play dominant roles in phloem loading. 291 It is noteworthy, that the combined zmsweet13abc mutations are not lethal, since the 292 plants still produce fertile viable offspring, implying additional mechanisms for phloem 293 loading in maize. While it is possible that other transporters might compensate, it is 294 unlikely that other SWEETs take over such roles, as judged by the lack of induction of 295 other clade III SWEET genes in the mutants. Therefore, maize likely uses either 296 symplasmic or other loading mechanisms in parallel. 297 It is still not clear whether SWEET13 triplication mainly serves to increase the amount of 298 SWEET proteins in the same cells (e.g. phloem parenchyma), or if each SWEET13 299 transporter mediates efflux from a specific cell type and loading is achieved in a multi-300 tier fashion. This question is of particular interest because in situ hybridization 301 experiments identified SUT1 in companion cells, xylem and phloem parenchyma, as 302 well as the bundle sheath. To address this question, we generated translational reporter 303 gene fusions and included the first three introns. However GUS activity or GFP 304 fluorescence were not detectable in any of the transformants carrying fusions for any of 305 the three SWEET13s (data not shown). We therefore hypothesize that additional 306 elements are required for proper expression of these constructs. 307 Another interesting question is whether maize can serve as a model for phloem loading 308 in rice, barley and wheat. Surprisingly, RNAi of the rice homolog of ZmSUT1 did not 309 lead to a detectable effect on the phenotype of the sporophyte (Ishimaru et al., 2001). 310 Thus, it remains a matter of debate whether rice uses predominantly apoplasmic and 311 symplasmic or other mechanisms simultaneously. It will therefore be important to study 312 the role of SWEET homologs in rice and other crops. It is noteworthy in this context that 313 OsSWEET11 and 15 are expressed preferentially in the caryopsis and act as key 314 players in apoplasmic unloading processes in developing rice grains (Yang, parallel 315 submission) (Ma et al., 2017). 316 Finally and most importantly, it will be interesting to see whether overexpression of 317 clade III SWEETs in leaves of cereals may help to increase yield potential. 318 Chronogram branch divergence time-points are in million years (Emms et al., 2016).
(B) Sucrose transport activity by ZmSWEET13a, b and c in HEK293T cells coexpressing the FLIPsuc90µΔ1V (sucrose). Cell were transfected to express sensors only as negative control, or to co-express AtSWEET12 as positive controls for sucrose.
HEK cells were subjected to a 20mM sucrose pulse for 3 minutes, (mean ± s.e.m., repeated independently four times with comparable results).