Sampling.

Fourteen leaf samples were collected from four locations in Israel (Table 1): by the Mediterranean Sea (Mediterranean), an arid site north of the Dead Sea (Dead Sea arid), a site in the Negev Desert highlands (Negev), and an oasis by the Dead Sea (Dead Sea oasis). In addition, samples were collected from two locations in the United States: a tree located in a cultivated garden on the island of Martha's Vineyard, MA (Martha's Vineyard) and a site in Davis, northern California (California). Not all tree species were present in all locations (Table 1). In all cases, duplicate samples were collected from multiple trees.

Leaves were collected between 11:00 a.m. and 2:00 p.m. from different parts of each tree, at random, into sterile paper envelopes; the samples were returned to the laboratory and processed within 2 to 5 h of sampling. The leaves were placed inside 50-ml sterile plastic test tubes (Falcon) and immediately immersed in sterile phosphate-buffered saline (PBS) medium (4 g of leaf/40 ml of PBS buffer, pH 7.4). Bacteria were dislodged from the leaves using a sonication tub (Transistor/ultrasonic T7; L&R Manufacturing Co.) for 2 min at medium intensity and by vortexing six times for 10 s each time at 5-min intervals. The leaf wash (LW) was separated from the leaf debris by decanting and kept for analysis. Chlorophyll content of the leaves was measured as described by Lorenzen (23).

DNA extraction.

Leaf washes were filtered on a 0.22-μm-pore-size membrane filter (Millipore) which was subjected to total community DNA extraction, using a Soil Microbial DNA extraction kit (Zymo Research, Orange, CA).

16S/18S rRNA tag pyrosequencing.

Details of the 16S/18S rRNA tag pyrosequencing method have been described elsewhere (1,9,14,35). In short, DNA samples were PCR amplified (30 cycles) using three sets of primers (Table 2) flanking either the bacterial and archaeal V6 hypervariable region or the eukaryotic V9 hypervariable region in the small-subunit rRNA gene. Each ribosomal primer set is flanked by pyrosequencing linker sequences and by a 5-nucleotide key for sample identification among mixed amplicon libraries. An equimolar mix from each sample was used to prepare pyrosequencing beads via emulsion PCR. Beads were loaded onto a picotiter plate (estimated at >1,000,000 beads/plate), and pyrosequencing was performed using a Genome Sequencer FLX System (Roche).

To decrease error rates, the FLX system output was passed through several quality filters additional to the ones built into the system (13). The following reads were omitted from microbial community analyses: reads with ambiguous nucleotides, reads shorter than 50 nucleotides, reads lacking the sample key and/or primer sequence at either end, reads with erroneous primer sequences at either end, and reads that could not be unambiguously assigned to a sample.

The remaining sequences were assigned the taxonomic classification of the most similar reference sequences in the V6 reference database (V6 RefDB) (9), based on a global alignment of the query sequence against the reference sequences in the V6 RefDB (Global Alignment for Sequence Taxonomy [GAST]) (14). In cases where a tag was equidistant to multiple reference sequences, it was classified to the level of the most resolved taxon shared by at least two-thirds of the reference sequences nearest to that tag.

All 14 samples were sequenced using bacterial primers, and five samples were sequenced using eukaryotic primers. Only two samples (the two MediterraneanT. nilotica trees) were sequenced using an archaeal primer set (Table 2) as these were the only samples from which an archaeal 16S rRNA gene PCR product was obtained.

To estimate the species richness of samples independently of taxonomic assignment, all of the tag sequences from a single sample or from categorized groups of samples were aligned using a pairwise alignment method and clustered using a 2% single-linkage preclustering methodology followed by average-linkage clustering based on pairwise alignments (15), a clustering method designed to minimize operational taxonomic unit (OTU) inflation caused by sequencing errors. Rarefaction curves, the Chao1 richness estimator, an abundance-based coverage estimator (ACE) (7), and the Shannon diversity index (H′) were calculated using mothur (34). Comparisons between community dissimilarity and environmental conditions were carried out using environmentally fitted nonmetric multidimensional scaling (NMDS) plots (27) and partial Mantel correlations (21), both using the Vegan package in R (28). Mantel tests were used since values in dissimilarity matrices are not independent. Partial Mantel tests were used in order to isolate the effect of correlated parameters from one another.P values were calculated using 100,000 permutations on rows and columns of dissimilarity matrices.

Chemical analysis of the leaf environment.

Leaf wash for chemical analysis was prepared as described above, replacing PBS with double-distilled water (DDW). Electrical conductivity (EC) was measured using a conductivity meter (S30 Seveneasy Conductivity; Mettler, Toledo, OH). The correlation between EC values (mS/cm) and the concentration of Na⁺ ions in the leaf wash (mg/liter) was established using an inductively coupled plasma optic emission spectrometer ([ICP/OES] Optima 3000; Perkin Elmer, Waltham, MA) and was found to be linear. pH was determined using a pH meter equipped with a combination glass electrode (model 420, Orion; ThermoOrion). Total dissolved organic carbon (DOC) was determined using a Formacs total organic carbon high-temperature combustion analyzer (Skalar Analytical B. V. Breda, The Netherlands), following the removal of inorganic carbon by lowering the pH to <2.0.

Nucleotide sequence accession number.

The sequences determined in this study have been deposited in the NCBI Short Read Archive under accession numberSRA026088.2.

Chemical composition of the leaf surface deposits.

Results of chemical analysis of leaf washes (LWs) are presented inTable 3. Electrical conductivity (EC), a proxy for salinity, was significantly higher in the Dead Sea arid samples than at other sampling sites (t = 0.016,P = 0.01). This may be attributed to the high July-August temperatures in the Dead Sea area, averaging a daily maximum of 39°C (compared with 31 to 32°C in the Negev and the Mediterranean coastal plain [19]), coupled with low water availability at the site. Also, a significant difference in pH values appears to exist among tree species (one-way analysis of variance [ANOVA],P < 10⁻⁵). Leaf washes fromT. aphylla trees were alkaline (mean pH, 9.5) while leaf washes fromT. nilotica andT. tetragina were neutral (mean pH, 7.15). The concentration of dissolved organic carbon (DOC) on the leaf surface was high in all cases, varying between 1 and 4 mg of C/g of leaf. For comparison, the sugar concentration on tomato leaves was determined at 1.55 μg/g (26). Organic carbon concentrations reported in this study, as well as those previously published forT. aphylla (4,30), are thus approximately 3 orders of magnitude higher. In addition to providing rich carbon and energy sources for the phyllosphere bacteria, it is likely that these compounds also aid in desiccation and osmotic stress responses.

Sequence recovery and microbial abundance.

Our sequencing effort yielded 158,980 bacterial V6 sequences, ∼60 bp in length, from 14 samples, averaging 11,355 sequences per sample. Of these sequences, 31,547 were identified as chloroplast and mitochondrial 16S rRNA derived from the host tree (Tables 4 and5) and were removed from the data set. In the eukaryotic V9 data sets (Table 2), 14,294 out of the 48,673 sequences were identified as plant 18S rRNA in a total of five samples.

The distribution of organelle-derived sequences was not random. The mean ratio between bacteria-derived and organelle-derived amplicons of Dead Sea arid trees is 0.31 ± 0.17, compared with 292 ± 496 for Mediterranean trees. A similar trend appears in the eukaryotic data set: the mean ratio between eukaryote-derived amplicons and plant-derived amplicons of Dead Sea arid trees is 0.80 ± 0.17, compared with 184 ± 224 for Mediterranean trees. Some of this variation can be explained by differences in chloroplast content between sites (170 ± 49 mg of chlorophyll/g of leaves in Mediterranean trees and 70 ± 16 mg of chlorophyll/g of leaves in Dead Sea trees). However, this difference cannot explain the approximately 1,000-fold difference observed in the ratio between bacteria-derived amplicons and chloroplast-derived amplicons. A possible explanation is that this ratio is caused mainly by a variability of 2 to 3 orders of magnitude in microbial biomass between the Dead Sea and Mediterranean sites. This hypothesis is supported by counts of CFU on LB agar plates containing 5% NaCl. Mediterranean trees harbored 2 × 10⁵ to 6 × 10⁵ CFU per gram of leaves, while Dead Sea trees harbored only 4 × 10³ CFU/g.

It should be noted that the taxonomic resolution of the eukaryotic hypervariable region V9 was distinctly lower than that of the bacterial V6 region: only 7.3% of eukaryotic V9 could be assigned by GAST to taxonomic ranks below order. In comparison, 86.0% of bacterial V6 sequences could be assigned to family, and ∼70% could be assigned at the genus level.

Archaeal amplicons were obtained only from the two MediterraneanT. nilotica samples, yielding 13,281 and 14,107 V6 sequences. Over 99% of the sequences were identified as members of the familyHalobacteriaceae, and 92% of these were assigned to three known genera within this family (see Table S1 in the supplemental material).

Species richness and alpha diversity.

Bacterial species richness across all samples was estimated by the ACE richness estimator (7) to be 5,265 when operational taxonomic units (OTUs) were clustered with a 6% dissimilarity cutoff, and 8,754 at a 3% dissimilarity cutoff (see Fig. S1 in the supplemental material). Taxonomic assignments of the reads identified 788 genera. Interestingly, using simple average-neighbor clustering, the clustering method we have used prior to “ironing out the wrinkles” (15), resulted in an estimated 9,322 OTUs with a 3% dissimilarity cutoff, indicating that OTU inflation due to sequencing errors was less that 10%.

Categorizing samples according to tree species (Fig. 1A) and geographical location (Fig. 1B) clearly indicates that bacterial species richness and diversity correlated with geographic location rather than withTamarix species. The ACE richness estimate for the Dead Sea arid samples was 1,963 OTUs (mean, 1,239 OTUs per tree) at 94% similarity, almost twice as high as the ACE value for Mediterranean samples (1,115 OTUs; mean, 500 OTUs per tree). The ACE values forT. nilotica from all locations compared with those forT. aphylla from all locations were similar: 2,018 forT. nilotica (mean, 865 OTUs per tree) and 1,593 forT. aphylla (mean, 581 OTUs per tree). Shannon diversity indices displayed a similar pattern, with Dead Sea samples being about twice as diverse as Mediterranean samples (Fig. 1).

Previous estimates of microbial richness of the phyllosphere of a single tree species, obtained by Sanger sequencing of 16S clone libraries, were 1 to 2 orders of magnitude lower (8,20,30) but could not serve for a valid comparison of richness with the study presented here due to the difference in sampling depths between sequencing methods. Recently, pyrosequencing was applied in the study of phyllosphere microbial diversity among a variety of tree species, albeit with a lower sampling depth, reaching 600 to 1,500 reads per sample (32). While a rarefaction analysis was not presented, the authors reported that they identified 164 to 424 OTUs with a 3% dissimilarity cutoff among the first 750 reads. In comparison,Tamarix leaf samples yielded lower species richness, between 71 to 286 OTUs, for an equivalent sampling depth. This difference may be partially attributed to the targeting of a different hypervariable region (10), as well as to differences in clustering methods. While it is tempting to link the lowerTamarix species richness to the extreme abiotic conditions on its leaves, evidence points otherwise: the Dead Sea sites, which had the highest temperature and salinity among the samples in this study, also displayed the highest taxonomic richness.

Species richness of eukaryotes (see Table S2 in the supplemental material) was an order of magnitude lower than that of bacteria. In contrast to bacteria, geographic location did not seem to affect eukaryotic species richness: ACE richness estimates for Dead Sea and Mediterranean trees were similar (170 and 176 OTUs, respectively).

Species diversity.

Beta diversity, a measure of the proportion of unique taxa between different samples, was calculated according to Yue and Clayton (41), providing a weighted distance measure that accounts for the relative abundance of taxa in each sample, and the results were clustered using the Kitsch algorithm. The use of a weighted algorithm is necessary in such an analysis due to the variety of sample sizes. The same analysis, performed using other weighted distance measures such as Morisita-Horn (12) and an unweighted one (6), yielded very similar results (data not shown). Two related non-Tamarix samples were added to the analysis: a sample from the gut of aT. nilotica phyllosphere insect,Oxyrrachis versicolor, collected from a tree in an oasis by the Dead Sea (O. M. Finkel et al., unpublished data), and a sample taken from Dead Sea water during a 1992 algal bloom (5). To evaluate the taxonomic depth of differences and similarities between samples, these analyses were performed at three levels: order, family, and genus. Very similar patterns were seen for the different taxon levels. The order-level analysis is presented inFig. 2, while that of family and genus can be found in the supplemental materials (see Fig. S2 and S3). Analysis at a species resolution would yield little information as most reads could not be assigned at this level.

Two main clusters emerged for all three taxonomic levels: (i) a Mediterranean cluster that includes phyllosphere communities from the Mediterranean and Negev samples, with the Martha's Vineyard sample forming an outgroup within this cluster, and (ii) a Dead Sea cluster encompassing all samples from sites by the Dead Sea, with the sample designated California 1 and theO. versicolor sample forming outgroups. While California 2 and Dead Sea water were also included in this cluster, their shared branch lengths were so short that they should each be considered an autonomous cluster. The Mediterranean cluster appears rather uniform, characterized by small differences among its members. The Dead Sea cluster is characterized by greater branch lengths, coinciding with increasing dissimilarity with increasing geographical distance. All three Dead Sea arid samples clustered together while all other samples each formed their own clusters, with the Dead Sea oasis sample as the nearest relative to the Dead Sea arid cluster. Surprisingly, the Dead Sea water sample and, to a lesser extent, theO. versicolor sample, did not form outgroups to the entire data set although their shared branch lengths with other members of their cluster are considerably shorter than their unique branch lengths.

To test which environmental or geographical parameters correlated with community dissimilarity, an NMDS plot was drawn for the Israeli samples only (Fig. 3). As the sampling sites were in different climatic regions, the changes in salinity, mean daily temperature, and latitude appear to have the same direction. In order to isolate the effect of each of these variables, partial Mantel tests were performed. Mean daily temperature (R = 0.87,P = 0.00045) and geographic location (R = 0.404,P = 0.0125) were both found to be significantly correlated with bacterial community dissimilarity, indicating that while the majority of the difference between the Dead Sea cluster and the Mediterranean cluster appears to be due to climatic factors, geographic isolation plays a significant role as well.

The differences between sampling sites have manifested themselves in both the richness and diversity of microbial species from all three domains of life. Such variability within and between samples should be taken into account in attempting to estimate the global diversity of phyllosphere bacteria since it is not based on interspecies variability alone. In a recent communication, Knief et al. (18) also showed that geographic location may play a more important role than host species in determining microbial epibiont composition. However, other recent evidence has shown that this is not necessarily a general rule.Pinus ponderosa leaves in the United States and Australia hosted bacterial communities very similar to each other compared with sympatric trees of different species within the same genus, the exact opposite of the trend we have seen forTamarix (32).

Figure 4 displays the main taxonomic orders comprising the different samples, highlighting differences between the two geographical/climatic clusters. A more detailed list of the identified population components is presented in supplemental Table S3. The dominant bacterial order at the Mediterranean site wasOceanospirillales, in which the genusHalomonas was the most abundant, comprising over 90% of the bacterial population onT. aphylla trees and 50 to 60% of the population onT. nilotica andT. tetragina. This genus is also dominant in the two other samples that clustered together with the Mediterranean samples, Negev and Martha's Vineyard. The majority of interspecies differences can be seen within proteobacteria (Fig. 5):T. nilotica andT. tetragina support an abundance of anotherHalomonadaceae genus,Chromohalobacter, as well asSalinisphaera (Gammaproteobacteria) andRhodobacterales (Alphaproteobacteria). The latter two genera are both nearly absent onT. aphylla.

While the Dead Sea cluster is characterized by greater diversity than the Mediterranean cluster, both within and between samples, most samples appear to be dominated by members of the two Gram-positive orders,Actinomycetales andBacillales. It should be noted that the four most dominant bacterial orders found in the Dead Sea proper—Bacillales (95% of the data set),Actinomycetales,Rhizobiales, andBurkholderiales—are all relatively abundant onTamarix trees from the same region (Fig. 4).

The absence ofHalomonas, the most dominant genus on Mediterranean, Negev, and Martha's Vineyard samples, from the Dead Sea and California samples is intriguing. From a phenotypic standpoint, the abundant presence ofHalomonas in most samples is not surprising. Most known members of this genus are motile, can grow in up to 20% salinity, and are able to utilize glycerol and trehalose (25), found in abundance onTamarix leaves (4). The one trait that is exclusively shared by all Dead Sea samples is temperature, which is considerably higher than in the other sampling sites, averaging summer daily maxima of 39°C. It is possible that these temperatures or another related factor (such as water availability) inhibits growth ofHalomonas in this area. The sites with abundantHalomonas all have relatively high air humidity in common, and thus the salt crystals on their leaves are more likely to absorb moisture. Indeed, recent results indicate thatHalomonas isolates and culture-independentHalomonas 16S rRNA sequences were found in samples from the Dead Sea area taken the following winter, when temperatures were lower and humidity higher (19) (data not shown). However, it has been reported (25) that several strains ofHalomonas can grow at up to 45°C. Further investigation of localHalomonas isolates is required in order to test the environmental variables that affect their growth.

An analysis performed on unicellular eukaryotic sequences revealed a similar dependence of their abundance on the environment to that of bacteria, as samples from differentTamarix species and the same environment clustered together (Fig. 6). The dominant order in all of these samples is the fungalAscomycota. Within this order, families were distributed according to geographic location:Lecanoromycetes andTrichocomaceae were found in all three Dead Sea samples but were absent in the Mediterranean. In contrast,Onygenales were present in samples from the Mediterranean and the Dead Sea oasis but were absent from the Dead Sea arid sample (see Table S2 in the supplemental material).

An important question that remains unanswered is what fraction of the community structure can be attributed to local features, as opposed to limitations on dispersal of the microbes. In other words, what role does geographic distance play in the microbial community shifts between locations (11,24,29)? The depth of 16S tag sequencing and the rare biosphere that it reveals have partially answered that question. Sequences assigned to an unidentifiedHalomonas species dominate the Mediterranean, Negev, and Martha's Vineyard samples but are also found in small numbers in Dead Sea and California samples. This observation excludes migration barriers as the factor responsible for the differences observed. An opposite example of a site-restricted taxon is much more difficult to identify.Blattabacteriaceae, a little studied family ofFlavobacteriales, are found in relatively small numbers (16 to 47 reads, ∼0.4%) in all samples from the Dead Sea area but are completely absent from any of the other samples, despite the fact that sampling depth in the Dead Sea area was considerably lower. This might suggest that this bacterial family does not migrate as easily as others do. To more carefully determine the extent to which the differences between communities observed in this study can be attributed to local environmental features as opposed to limitations on dispersal, the role that geographic distance plays in the community shift observed between locations should be further addressed (3). This can be done by sampling trees in sites that are as similar as possible in their biotic and abiotic surroundings but are geographically distant from each other. A further insight as to how “island-like” the phyllosphere environment actually is can be gained from comparing microbial communities onTamarix to those of sympatric species of different genera, to those of the soil in which the trees grow, and to the air that connects them.

• 1.Amaral-Zettler L. A., McCliment E. A., Ducklow H. W., Huse S. M. 2009. A method for studying protistan diversity using massively parallel sequencing of V9 hypervariable regions of small-subunit ribosomal RNA genes. PLoS One 4:e6372. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 2.Andrews J. H., Harris R. F. 2000. The ecology and biogeography of microorganisms of plant surfaces. Annu. Rev. Phytopathol. 38:145–180 [DOI] [PubMed] [Google Scholar]
• 3.Andrews J. H., Kinkel L. L., Berbee F. M., Nordheim E. V. 1987. Fungi, leaves, and the theory of island biogeography. Microb. Ecol. 14:277–290 [DOI] [PubMed] [Google Scholar]
• 4.Belkin S., Qvit-Raz N. 2010. Life on a leaf: bacterial epiphytes of a salt-excreting desert tree, p. 393–406In Seckbach J., Grube M. (ed.), Symbioses and stress. Springer, Dordrecht, Netherlands [Google Scholar]
• 5.Bodaker I., et al. 2010. Comparative community genomics in the Dead Sea: an increasingly extreme environment. ISME J. 4:399–407 [DOI] [PubMed] [Google Scholar]
• 6.Bray J. R., Curtis J. T. 1957. An ordination of the upland forest communities of souther Wisconsin. Ecol. Monogr. 27:325–349 [Google Scholar]
• 7.Chao A., Lee S. M. 1992. Estimating the number of classes via sample coverage. J. Am. Stat. Assoc. 87:210–217 [Google Scholar]
• 8.Delmotte N., et al. 2009. Community proteogenomics reveals insights into the physiology of phyllosphere bacteria. Proc. Natl. Acad. Sci. U. S. A. 106:16428–16433 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 9.Dethlefsen L., Huse S., Sogin M. L., Relman D. A. 2008. The pervasive effects of an antibiotic on the human gut microbiota, as revealed by deep 16S rRNA Sequencing. PLoS Biol. 6:e280. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 10.Engelbrektson A., et al. 2010. Experimental factors affecting PCR-based estimates of microbial species richness and evenness. ISME J. 4:642–647 [DOI] [PubMed] [Google Scholar]
• 11.Green J., Bohannan B. J. M. 2006. Spatial scaling of microbial biodiversity. Trends Ecol. Evol. 21:501–507 [DOI] [PubMed] [Google Scholar]
• 12.Horn H. S. 1977. Measurement of “overlap” in comparative ecological studies. Am. Nat. 914:419–423 [Google Scholar]
• 13.Huse S. M., Huber J. A., Morrison H. G., Sogin M. L., Mark Welch D. 2007. Accuracy and quality of massively parallel DNA pyrosequencing. Genome Biol. 8:R143http://genomebiology.com/content/8/7/R143 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 14.Huse S. M., et al. 2008. Exploring microbial diversity and taxonomy using SSU rRNA hypervariable tag sequencing. PLoS Genet. 4:e1000255. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 15.Huse S. M., Welch D. M., Morrison H. G., Sogin M. L. 2010. Ironing out the wrinkles in the rare biosphere through improved OTU clustering. Environ. Microbiol. 12:1889–1898 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 16.Jackson E. F., Echlin H. L., Jackson C. R. 2006. Changes in the phyllosphere community of the resurrection fern, Polypodium polypodioides, associated with rainfall and wetting. FEMS Microbiol. Ecol. 58:236–246 [DOI] [PubMed] [Google Scholar]
• 17.Kinkel L. L., Wilson M., Lindow S. E. 1995. Effect of sampling scale on the assessment of epiphytic bacterial populations. Microb. Ecol. 29:283–297 [DOI] [PubMed] [Google Scholar]
• 18.Knief C., Ramette A., Frances L., Alonso-Blanco C., Vorholt J. A. 2010. Site and plant species are important determinants of the Methylobacterium community composition in the plant phyllosphere. ISME J. 4:719–728 [DOI] [PubMed] [Google Scholar]
• 19.Kurtzman D., Kadmon R. 1999. Mapping of temperature variables in Israel: a comparison of different interpolation methods. Climate Res. 13:33–43 [Google Scholar]
• 20.Lambais M. R., Crowley D. E., Cury J. C., Bull R. C., Rodrigues R. R. 2006. Bacterial diversity in tree canopies of the Atlantic forest. Science 312:1917. [DOI] [PubMed] [Google Scholar]
• 21.Legendre P., Legendre L. 1998. Numerical ecology, 2nd Engl. ed Elsevier, Amsterdam, the Netherlands [Google Scholar]
• 22.Lindow S. E., Brandl M. T. 2003. Microbiology of the phyllosphere. Appl. Environ. Microbiol. 69:1875–1883 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 23.Lorenzen C. J. 1967. Determination of chlorophyll and pheo-pigments: spectrophotometric equations. Limnol. Oceanogr. 12:343–346 [Google Scholar]
• 24.Martiny J. B. H., et al. 2006. Microbial biogeography: putting microorganisms on the map. Nat. Rev. Microbiol. 4:102–112 [DOI] [PubMed] [Google Scholar]
• 25.Mata J. A., Martinez-Canovas, Quesada J. E., Béjar V. 2002. A detailed phenotypic characterisation of the type strains of Halomonas species. Syst. Appl. Microbiol. 25:360–375 [DOI] [PubMed] [Google Scholar]
• 26.Mercier J., Lindow S. E. 2000. Role of leaf surface sugars in colonization of plants by bacterial epiphytes. Appl. Environ. Microbiol. 66:369–374 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 27.Minchin P. R. 1987. An evaluation of relative robustness of techniques for ecological ordinations. Vegetatio 71:145–156 [Google Scholar]
• 28.Oksanen J., et al. 2008. Vegan: community ecology package. R package version 1.15-1. R Foundation for Statistical Computing, Vienna, Austria:http://vegan.r-forge.r-project.org/ [Google Scholar]
• 29.Prosser J. I., et al. 2007. The role of ecological theory in microbial ecology. Nat. Rev. Microbiol. 5:384–392 [DOI] [PubMed] [Google Scholar]
• 30.Qvit-Raz N., Jurkevitch E., Belkin S. 2008. Drop-size soda lakes: transient microbial habitats on a salt-secreting desert tree. Genetics 178:1615–1622 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 31.Redford A. J., Fierer N. 2009. Bacterial succession on the leaf surface: a novel system for studying successional dynamics. Microb. Ecol. 58:189–198 [DOI] [PubMed] [Google Scholar]
• 32.Redford A. J., Bowers R. M., Knight R., Linhart Y., Fierer N. 2010. The ecology of the phyllosphere: geographic and phylogenetic variability in the distribution of bacteria on tree leaves. Environ. Microbiol. 12:2885–2893 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 33.Ruppel S., Krumbein A., Schreiner M. 2008. Composition of the phyllospheric microbial populations on vegetable plants with different glucosinolate and carotenoid compositions. Microb. Ecol. 56:364–372 [DOI] [PubMed] [Google Scholar]
• 34.Schloss P. D., et al. 2009. Introducing mothur: open source, platform-independent, community-supported software for describing and comparing microbial communities. Appl. Environ. Microbiol. 75:7537–7541 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 35.Sogin M. L., et al. 2006. Microbial diversity in the deep sea and the underexplored “rare biosphere.” Proc. Natl. Acad. Sci. U. S. A. 103:12115–12120 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 36.Waisel Y. 1961. Ecological studies on Tamarix aphylla. L. Karst. Plant Soil 13:356–364 [Google Scholar]
• 37.Waisel Y. 1991. The glands of Tamarix aphylla: a system for salt secretion or for carbon concentration? Physiol. Plant. 83:506–510 [Google Scholar]
• 38.Whipps J. M., Hand P., Pink D., Bending G. D. 2008. Phyllosphere microbiology with special reference to diversity and plant genotype. J. Appl. Microbiol. 105:1744–1755 [DOI] [PubMed] [Google Scholar]
• 39.Yadav R. K. P., Papatheodorou E. M., Karamanoli K., Constantinidou H. I. A., Vokou D. 2008. Abundance and diversity of the phyllosphere bacterial communities of Mediterranean perennial plants that differ in leaf chemistry. Chemoecology 18:217–226 [Google Scholar]
• 40.Yang C. H., Crowley D. E., Borneman J., Keen N. T. 2001. Microbial phyllosphere populations are more complex than previously realized. Proc. Natl. Acad. Sci. U. S. A. 98:3889–3894 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 41.Yue J. C., Clayton M. K. 2005. A similarity measure based on species proportions. Commun. Stat. Theor. Methods 34:2123–2132 [Google Scholar]

Supplementary Materials