Oral samples as non-invasive proxies for assessing the composition of the rumen microbial community

RESEARCH ARTICLE Oral Samples as Non-Invasive Proxies for Assessing the Composition of the Rumen Microbial Community Ilma Tapio 1 , Kevin J. Shingfield 1¤ , Nest McKain 2 , Aurélie Bonin 3 , Daniel Fischer 1 , Ali R. Bayat 1 , Johanna Vilkki 1 , Pierre Taberlet 3 , Timothy J. Snelling 2 , R. John Wallace 2 * 1Green Technology, Natural Resources Institute Finland, Jokioinen, Finland, 2Rowett Institute of Nutrition and Health, University of Aberdeen, Aberdeen, United Kingdom, 3Laboratoire d'Ecologie Alpine, CNRS, Grenoble, France ¤Current address: Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Aberystwyth, United Kingdom *[email protected].uk Abstract Microbial community analysis was carried out on ruminal digesta obtained directly via rumen fistula and buccal fluid, regurgitated digesta (bolus) and faeces of dairy cattle to assess if non-invasive samples could be used as proxies for ruminal digesta. Samples were collected from five cows receiving grass silage based diets containing no additional lipid or four different lipid supplements in a 5 x 5 Latin square design. Extracted DNA was analysed by qPCR and by sequencing 16S and 18S rRNA genes or the fungal ITS1 amplicons. Faeces contained few protozoa, and bacterial, fungal and archaeal communities were substantially different to ruminal digesta. Buccal and bolus samples gave much more similar profiles to ruminal digesta, although fewer archaea were detected in buccal and bolus samples. Bolus samples overall were most similar to ruminal samples. The differences between both buccal and bolus samples and ruminal digesta were consistent across all treatments. It can be concluded that either proxy sample type could be used as a predictor of the rumen microbial community, thereby enabling more convenient large-scale animal sampling for phenotyping and possible use in future animal breeding programs aimed at selecting cattle with a lower environmental footprint. Introduction Ruminant livestock production has a large environmental footprint because of emissions of the greenhouse gas, methane, and to high nitrogenous emissions in urine and faeces [1]. The primary source of these emissions is rumen microbial metabolism [2]. Thus, understanding the ruminal microbiota is a vital prerequisite for improving the environmental credentials of meat and milk production. The introduction of high-throughput sequencing techniques has opened new ways to explore complex microbial ecosystems, including the rumen [3–6]. Sampling ruminal digesta is essential to enable the technology benefits to be realized. Rumen sampling PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 1/15 OPEN ACCESS Citation: Tapio I, Shingfield KJ, McKain N, Bonin A, Fischer D, Bayat AR, et al. (2016) Oral Samples as Non-Invasive Proxies for Assessing the Composition of the Rumen Microbial Community. PLoS ONE 11 (3): e0151220. doi:10.1371/journal.pone.0151220 Editor: Robert J Forster, Agriculture and Agri-Food Canada, CANADA Received: September 21, 2015 Accepted: February 23, 2016 Published: March 17, 2016 Copyright: © 2016 Tapio et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data are available from the Dryad database (accession number http://dx.doi. org/10.5061/dryad.1b07d). Funding: This work was supported by the European Commission 289319. Competing Interests: The authors have declared that no competing interests exist. can be carried out by oral intubation, but this is an unpleasant procedure for the animal and also results in a sample that is often heavily contaminated with saliva [7]. Rumenocentesis provides valid samples but involves puncturing the abdominal wall with a needle and removing digesta by syringe, also undesirable in terms of animal welfare [8], and restricts the amount of sample that can be collected. The most reliable samples are obtained from animals that have been surgically modified by fitting a ruminal cannula [7,9], but this requires skilled surgery, dedicated animal facilities and in most countries requires formal governmental permission. In any case, rumen fistulation is impractical for sampling large numbers of animals. Ruminants regurgitate ruminal contents regularly in order to chew the partially digested plant material [10,11]. The chewed bolus is then swallowed for further microbial degradation. It might be expected, therefore, that the microbiome of the mouth could represent a reflection of the ruminal microbiome. If so, collection of small samples of oral fluid could be used as a proxy for assessing the microbial ecology of the rumen, avoiding the need for more invasive sampling procedures. Our hypothesis was that the oral sample will contain microbes from the regurgitated bolus that results from rumination and that the microbial composition of the bolus might be representative of the ruminal community. Another proxy that has been investigated before is faeces. The faecal community is significantly different to that of the rumen [12– 15], but nonetheless there may be indicators that could prove useful, analogous to the presence of faecal archaeol, a membrane lipid of ruminal archaea, being used as a marker for ruminal methanogenesis [16–18]. The aim of the present experiment was to compare the communities in these potential alternative samples in order to evaluate their usefulness as proxies for direct sampling of ruminal digesta. A paper has been published recently [19] in which the same hypothesis was explored using buccal samples in sheep. The present paper confirms the conclusions of that paper concerning the validity of buccal sampling, in this case using dairy cows, and further amplifies the investigation by comparing bolus and faecal samples. Materials and Methods Animal experimentation All experimental procedures were approved by the National Ethics Committee (Hämeenlinna, Finland) in accordance with the guidelines established by the European Community Council Directives 86/609/EEC [20]. The experiment was conducted between 15 February 2012 and 3 July 2012 at Natural Resources Institute Finland (formerly MTT Agrifood Research Finland), Jokionen Finland (60.8° N, 23.5° E; altitude 103 m). Five Finnish Ayrshire cows fitted with rumen cannula (#1C, i.d. 100 mm, Bar Diamond, Inc., Parma, ID) of (mean ± SE) 4 ± 0.6 parity, 63 ± 11.7 d in milk, and 705 ± 25.5 kg live weight were used in a 5 × 5 Latin square with 28-d experimental periods. Treatments comprised total mixed rations based on grass silage (forage: concentrate ratio 60:40 on a dry matter (DM) basis) containing no additional fat (CO) or 50 g/kg diet DM of methyl esters of myristic acid (MA), rapeseed oil (RO), safflower oil (SO) or linseed oil (LO). Lipid supplements replaced concentrate ingredients. Each period comprised 2 d adaptation, 21 d supplementation, and 5 d washout to minimize treatment carry-over effects. Samples for the analysis of rumen microbial composition were collected at 15.00 h on d 20 and 09.00 h on d 22 of each period. Collection and processing of samples Ruminal digesta samples were collected from four regions (anterior dorsal, anterior ventral, posterior dorsal, and posterior ventral) within the rumen-reticulum. Immediately after collection, ruminal digesta samples were mixed thoroughly and squeezed through 2 layers of Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 2/15 cheesecloth. Five hundred μl of rumen liquid were mixed with 1 ml of phosphate buffered saline-glycerol (30% v/v) buffer (PBS-gly) and immediately frozen at -80°C. Regurgitated ingesta (bolus) samples were collected as close in time as possible to that of rumen samples. Depending on the rumination behaviour of each cow, the time after rumen sampling varied among the animals but did not exceed 20 min. Bolus samples were processed in the same way as rumen samples. Buccal samples (effectively saliva mixed with bolus particles) were collected at the same time as bolus using sponge swabs. Three collection methods were investigated: (i) samples were collected using the BuccalAmp DNA extraction kit (Epicentre) and (ii) using the Performagene Livestock (PG-100) kit (DNA genotek) and processed following the manufacturer’s protocols, or (iii) samples were collected using Performagene Livestock sponge swabs, immediately submerged in 1 ml of PBS-gly buffer and frozen at -80°C. Processing of samples based on method (iii) was the best in terms of DNA quality and quantity, and therefore used for sampling. For technical reasons, buccal swabs could not be taken during the first experimental period. Fresh faeces were collected by stimulating rectal activity at the time around rumen sampling. Twenty five g of faeces were preserved in 50 ml of PBS-gly buffer and stored at -80°C. Total genomic DNA was extracted from 1 ml of mixed d 20 and d 22 sample (in case of rumen, bolus and buccal swab) or 30 mg of faeces following the protocol of Yu and Morrison [21]. Quantitative PCR of 16S and 18S rRNA genes DNA concentrations were determined with a NanoDrop ND 1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE). DNA was diluted to 0.1 ng/μlin5μg/ml herring sperm DNA for amplification with universal bacterial primers UniF and UniR [22] and 1 ng/μlin 5μg/ml herring sperm DNA for amplification of other groups [23]. Quantitative PCR was carried out using a BioRad CFX96 as described by Ramirez-Farias et al. [24]. Primer sets and target species are described in Table A in S1 Text. Amplification of archaeal 16S RNA genes was carried out using the primers described by Hook et al. [25] and calibrated using DNA extracted from Methanobrevibacter smithii PS, a gift from M. P. Bryant, University of Illinois. Amplification efficiency was evaluated using template DNA from Roseburia hominis A2-183 (DSM 16839 T ) for the universal bacteria and Clostridium Cluster XIVa calibrations, Faecalibacterium prauznitzii A2-165 (DSM 17677 T ) for Clostridium Cluster IV, and Bacteroides thetaiotaomicron VPI-5482 (DSM 2079 T ) for Bacteroidetes. Protozoal 18S rRNA gene amplification was calibrated using DNA amplified from bovine rumen digesta with primers 54f and 1747r [26]. Coverage of qPCR primers was checked from original references and by use of the Probe Match tool of the Ribosome Database Project [27]. Bacterial abundance was calculated from quadruplicate Ct values using the universal bacterial calibration equation. Pairwise differences between treatments and between samples within treatments were evaluated by a simple t-test. Pvalues have been adjusted for multiple testing using the Benjamini-Hochberg method [28]. Amplicon preparation and sequencing Primers used for PCR amplification of bacteria and archaea 16S rRNA genes, ciliate protozoa 18S rRNA genes and anaerobic fungi ITS1 genes were designed in silico using ecoPrimers [29], the OBITools software suite [30](http://www.grenoble.prabi.fr/trac/OBITools) and a database created from sequences stored in GenBank. For each sample, PCR amplifications were performed in duplicate. An eight nucleotide tag unique to each PCR duplicate was attached to the primer sequence, in order to enable the pooling of all PCR products for sequencing and the Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 3/15 subsequent assignation of sequence reads to their respective samples. PCR amplicons were combined in equal volumes and purified (QIAquick PCR purification kit, Qiagen, Germany). Amplicon libraries were prepared in Fasteris SA (Geneva, Switzerland https://www.fasteris. com/dna/) using the TruSeq Nano DNA HT Sample Prep Kit from Illumina using a protocol with only five PCR cycles (https://www.fasteris.com/dna/?q=content/metafast-protocolamplicon-metagenomic-analysis). All markers were sequenced using the MiSeq technology from Illumina, which produced 250-base paired-end reads, except for the archaea marker, which was sequenced on the Illumina HiSeq platform, generating 100-base paired-end reads. In silico analysis showed that shorter archaea amplicons did not compromise ability to identify archaea at species level. Sequence analysis and taxonomic assignment Alignment of paired-end reads, sample assignment and removal of sequences with ambiguous nucleotides and sequences of lengths outside the empirical sequence length distribution were performed with the OBITools software suite. Sequences were deposited in Dryad database under accession number http://dx.doi.org/10.5061/dryad.1b07d. Sequences were clustered into operational taxonomic units (OTU) at 97% similarity using UCLUST [31] and filtered for chimeric reads using ChimeraSlayer (bacteria, protozoa and archaea) or UCHIME (fungi) as implemented in QIIME pipeline v1.7.0 [32]. Taxonomy was assigned using the BLAST method [33]. Bacterial OTUs taxonomy was assigned using the Greengenes 12_10, archaeal—RIM-DB database [34], that of ciliate protozoa using the SILVA 18S database [35] and anaerobic fungi were assigned using a curated fungal ITS reference database [36] kindly provided by AgResearch Ltd (Palmerston North, New Zealand). Singleton OTUs were removed and the data from each sample were rarefied to the similar sequencing depth prior to further analyses using QIIME. Pairwise taxonomy comparisons were performed by computing Pearson correlation coefficients as implemented in QIIME. In total, 20 rumen-bolus-buccal swab-and faecal samples collected from the same animals at the same time were compared. Scatter plot analysis was done using R v2.15.0 [37]. For creating microbial co-occurrence networks in rumen and alternative sampling sites, the SparCC microbial association network inference tool [38] was used to calculate correlation coefficients between all bacteria, archaea, ciliate protozoa and fungi at the genus or the deepest identifiable taxonomic classification level. Correlations were derived where X = (x_ij), i, j = 1,2,...,m is the resulting correlation matrix with x_ij being the pair wise correlation between microbes i and j and m being the total amount of compared microorganisms. An adjacency matrix A = (a_ij), i,j = 1,2,...,m was determined where a_ij = 1, if x_ij>= 0.25, a_ij = -1, if x_ij <= -0.25 and a_ij = 0 else. Once determined, the adjacency matrix was used to construct a co-occurrence network, where each node represents a taxon while the edges between the nodes represent positive/negative correlations between taxa. Communities within the networks were identified by applying the leading eigenvector [39] method using the R-Package 'igraph' [40]. The pairwise similarities between the community structures of networks have been evaluated for each treatment and sampling site separately using the adjusted rand index [41] and visualized as a heatmap, where brighter colours refer to larger rand values indicating closer similarity. Results In total, 80 samples, collected from four sampling sites, the rumen, buccal fluid, bolus and faeces were compared. Samples were collected from 5 lactating cows used in a 5 × 5 Latin Square with 28-d experimental periods to evaluate 5 experimental diets. Treatments comprised total Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 4/15 mixed rations based on grass silage containing no additional fat (CO) or supplemented with methyl myristate (MA), rapeseed oil (RO), safflower oil (SO) or linseed oil (LO). Diets were formulated to induce changes in rumen microbial populations to provide a robust test of the suitability of sampling proxies. qPCR analysis of 16S and 18S rRNA genes from ruminal, buccal, bolus and faecal samples The abundance of different microbial groups was compared by qPCR across treatments and sample types (Fig 1; Tables B and C in S1 Text). Faecal samples differed markedly from corresponding ruminal samples, in that although archaea were present at a similar abundance, protozoa were virtually absent, and total bacteria were higher. Clostridium Cluster IV was on average 5× more abundant in faeces than in ruminal digesta. Bacteroidetes showed a correspondingly lower abundance. Buccal swab samples contained similar abundances of total bacteria to ruminal samples, but the proportion of Bacteroidetes tended to be higher and Clostridium Clusters IV and XIVa lower. Archaea were 0.48× as abundant in swab samples compared to ruminal samples. Protozoal 18S rRNA abundance appeared to be 1.7× higher in swab samples. Samples from the bolus of regurgitated digesta were generally closer in profile to ruminal digesta samples removed via the ruminal fistulae. Archaea were 0.80×, protozoa were 0.75× and total bacteria were 1.22× the abundance in ruminal digesta, although differences were not statistically significant due to high variability between samples. The proportions of the different classes of bacteria were similar. No differences (FDR <0.05) were detected due to treatment in any of the sample types, except that ciliate protozoa tended to be decreased in rumen and oral samples by dietary MA supplements, with a compensatory increase in total bacteria. Microbial composition of different sampling sites by rRNA gene amplicon analysis Twenty rumen-bolus-buccal swab-faeces samples sets were collected for sampling site comparisons. In total, 7,305,504 high quality sequencing reads across all 4 major microbial groups (bacteria, archaea, ciliate protozoa and anaerobic fungi) were generated. The number of sequences assigned to each microbial group is reported in Table D in S1 Text. Microbiota composition is presented both as relative abundance in extracted DNA (Fig 2) and the difference in abundance relative to that determined in ruminal digesta (Fig 3; Fig A in S1 Text). One hundred and four genus-like groups of bacteria were identified in the total dataset. Fifty of these had an average abundance above 0.5% in at least one of the treatments and sampling sites but only 16 could be identified to the genus level. The remaining 54 groups with average abundance below 0.5% were pooled at the lowest common taxonomic level. In total, they accounted for less than 5% of all sequencing reads. The archaeal community was represented by 9 genera consisting of 15 groups at the species level, while ciliate protozoa and fungal populations were represented by 14 groups at the genus level, respectively. The bacterial communities in ruminal, bolus and buccal swab samples were remarkably similar to each other and affected little by treatment. The phylum Bacteroidetes was similar in rumen and bolus samples (ca. 45% of total sequences) but was 10% higher in buccal swabs and 20% lower in faeces. These differences were mainly related to the proportion of Prevotella, which was the most abundant bacterial genus in rumen-bolus-swab samples (Fig 2). The Rikenellaceae family and the phylum Actinobacteria were detected in faeces but not in other samples. Bolus samples also represented a close match to rumen samples in the abundance of Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 5/15 Firmicutes (35%) while in faecal samples Firmicutes accounted for 70% of all sequences, with the Ruminococcus genus being the most abundant (Fig 2). Out of eight distinct bacterial groups within the Lachnospiraceae family, the most obvious differences among the sampling sites were related to a group of bacteria classified only at family level (Lachnospiraceae1). This group was overrepresented in bolus, underrepresented in swabs and absent in faecal samples. Differences between rumen and swab samples were related to lower abundance of Clostridia1,Ruminococcaceae1,Coriobacteriales groups and Butyrivibrio genus in swabs, accounting for 1–2.5% lower abundances compared to ruminal digesta (Fig 3a). In the archaeal communities, Methanobrevibacter dominated in rumen-bolus-swab samples but the proportion of Mbb.gottschalkii was underrepresented and Mbb.ruminantium overestimated in bolus and buccal swab samples compared to ruminal samples (Figs 2and 3b). Methanosphaera was the second most common genus among archaea in these samples. In faeces, the Methanocorpusculum genus, not detected in rumen-bolus-swab samples, accounted for up to 53% of sequencing reads, with Methanobrevibacter and Methanosphaera genera being less abundant. Five distinct groups within the Methanomassiliicoccaceae family were detected in rumen-bolus-swab samples at similar abundance below 0.1% (Fig 3b). Diet had little influence on the archaeal community of the different sample types (Fig 2). Ciliate protozoa composition in bolus samples matched closely rumen samples but differences were observed in ciliate taxon abundance. Entodinium was more abundant whereas Isotricha and Metadinium minorum were less abundant in bolus compared to rumen samples. Buccal swab samples contained a high proportion of Diplodinium (>33%) compared to rumen Fig 1. qPCR of 16S and 18S rRNA genes in buccal swabs, regurgitated digesta (bolus), ruminal digesta and faeces. Samples collected from lactating cows fed a grass silage based diet containing no additional fat (CO) or supplemented with 50 g/kg diet dry matter of methyl myristate (MA), rapeseed oil (RO), linseed oil (LO) or safflower oil (SO). Results are expressed as copy number per ng of extracted DNA. Error bars represent SD, n = 4 per treatment, except for four buccal samples, one each from CO, RO, LO and SO, from which satisfactory amplification was not obtained. doi:10.1371/journal.pone.0151220.g001 Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 6/15 samples (<1%), whereas the proportion of uncultured Entodinium was lower by 16–23%. Furthermore, the relative abundance of Metadinium minorum and Raabena bella in buccal swabs was below 1% compared to 6–17% in rumen and bolus samples, while Polydiniella mysorea was detected in swabs (1–2%) but only in trace amounts in other sample types. No rumen specific protozoa were observed in faecal samples. The much lower abundance of faecal protozoa included the parasitic ciliate protozoa, Balantidium coli and Blastocystis sp. that were absent from rumen-bolus-swab samples. Treatment had little influence on the protozoal community of the different sample types (Fig 2). Dietary lipid supplements had a profound effect on the fungal population composition, particularly methyl myristate (Fig 2). Apart from Orpinomyces 1b found only in faeces at an Fig 2. Relative abundance of bacteria, archaea, protozoa, and fungi based on amplicon sequencing of 16S-18S rRNA genes, and ITS1 sequences in rumen, bolus, buccal swabs and faeces. Samples collected from lactating cows fed total mixed rations based on grass silage containing no additional fat (CO), or supplemented with 50 g/kg dry matter of methyl myristate (MA), rapeseed oil (RO), safflower oil (SO) or linseed oil (LO). Data reported based on the mean of 4 animals per dietary treatment. doi:10.1371/journal.pone.0151220.g002 Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 7/15 abundance below 3%, the composition of the fungal population in rumen-bolus-swab samples was similar. Differences were related to variation in abundance of Caecomyces1, KF1, Neocallimastix1 and SK3 fungal groups. The closest match was observed between rumen and bolus samples (Fig 3b). Correspondence analysis of microbial composition between sample types To evaluate how well the microbial communities in bolus, buccal swabs or faecal samples represented that in ruminal digesta, scatter plots were generated (Fig 4). In the scatter plots the relative abundances of every microbial taxon from each animal, representing all 3 non-invasive sampling sites, were compared with rumen samples. The strength of similarity between sampling sites was estimated as an overall Pearson correlation coefficient. The correlations derived indicated that bolus samples matched most closely the rumen composition for bacteria (r value = 0.99), while little overall correspondence was observed between faecal and rumen samples. Even though differences in abundance were observed among the more common bacterial groups in the buccal swab-rumen comparison, the overall correlation was high nonetheless (r value = 0.98). Patterns of similarity for archaea and anaerobic fungi among sampling sites was similar as for bacteria (Fig 4a and 4b). The largest difference between bolus and buccal samples compared with rumen digesta was for ciliate protozoa, while no relation to faecal protozoal composition was detected (Fig 4b). Fig 3. Changes in microbial abundance between rumen and the three alternative sampling sites. Each row represent genus-like microbial group for a) bacteria and b) archaea, ciliate protozoa, and fungi, while each dot represents individual cow. Differences were calculated as relative abundance in buccal swab minus abundance in rumen (red ), faeces—rumen (green ) and bolus—rumen (blue ) based on n = 20 per sampling site. doi:10.1371/journal.pone.0151220.g003 Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 8/15 Microbial co-occurrence analysis The potential of collecting bolus, faeces or buccal swabs as a reliable alternative to rumen sampling was examined through the generation of microbial association networks. Analysis was performed for each dietary treatment and on each sample type. Only correlations with SparCC |r| 0.25 were used for constructing the networks. Microbial communities identified within the networks of alternative sampling types were compared with rumen communities. Similarities of the networks are presented as a heatmap (Fig B in S1 Text). Microbial networks identified in bolus and swab samples did not offer an exact match to the network composition observed in rumen samples, while faecal microbial interactions were not comparable to the other sampling sites. Dietary treatments appeared to have an effect on the direct or indirect interactions between microbial taxa, with changes in response to MA showing the most distinct differences. Use of archaeal abundance as a biomarker of methane emissions The qPCR data for individual animals were used to calculate the ratio of abundance of archaea and bacteria, and the effects of dietary treatment and different sample types were compared. Inclusion of additional lipid in the dietary treatment decreased the relative proportions of archaea relative to bacteria in rumen samples from MA (P= 0.013) and LO (P= 0.033) treatments, with the responses to MA being the most pronounced, lowering the ratio of ruminal archaea:bacteria abundance by 36% compared to the control (Table E in S1 Text). The effects of lipid supplements were also evident in the other sample types, but the archaeal abundance was significantly lower in all other samples compared to ruminal digesta (Table E in S1 Text). When individual animals were compared across treatments, a closer correlation existed Fig 4. Scatter plot analysis of differences in relative abundance, estimated for each microbial taxon in all individual animals and across sample types, respectively. a) Each bacterial taxon in samples collected from all cows is represented as separate () in the lower triangle, while archaea are presented as (Δ) in the upper triangle; b) Anaerobic fungi are marked as (□) and ciliate protozoa as (●). The average Pearson correlation coefficient is indicated in the bottom-right corner of each diagram. doi:10.1371/journal.pone.0151220.g004 Proxies for Rumen Samples PLOS ONE | DOI:10.1371/journal.pone.0151220 March 17, 2016 9/15