The Hallman et al.  definition of endophytic bacteria requires “surface-disinfested plant tissue” or extraction from the plant. “Disinfestation” by killing all the epiphytic bacteria may be effective when culture-dependent protocols are used, but is not appropriate in culture-independent protocols, such as the present one, since the DNA or RNA of dead epiphytes, if not removed, would still be amplified by bacteria-specific PCR. For those organs, like tubers, whose outer layers can be easily peeled off, endophytic bacteria can be isolated from inside of the plants unambiguously. However, peeling the epidermis off leaves, while possible, is not practical for a study like the present one. Therefore, to study leaf endophytic bacterial communities, it is critical to dislodge epiphytic bacteria from the leaf surfaces as far as possible. We have dislodged epiphytes using methods similar to those reported by others [13, 26–28]. Since we did not test the rinse water for rDNA amplicons, we cannot be sure that we have removed all epiphytic bacteria. However, the observation that the complexities of the populations (Additional file 1: Table S5) were substantially lower than those reported for leaf epiphytic bacteria [29, 30] suggests that most epiphytes have been removed.
Past studies have applied multiple enzyme digestion T-RFLP to environmental bacterial community research [31–33]. Some studies have focused on the rhizosphere, rhizoplane and the epiphytic phyllosphere bacterial communities using fingerprint techniques of 16S rRNA genes, especially the rhizosphere of single cultivated plant species including potato and rice [34–36] and the phyllosphere of soybean, rice and maize [6, 37]. The present research is the first to apply single digestion T-RFLP to leaf endophytic bacteria in multiple host species. Multi-enzyme studies depend on a reliable T-RFLP database to deduce species information; however most T-RFLP databases are still developing, so that a large proportion of novel bacteria, which are highly abundant in the environment, may not be matched using current databases . Although closely related bacterial species will usually produce the same T-RF, one or more other distinct taxonomic groups may also produce the same T-RF. Therefore variation in abundance of a T-RF may be due to changes in one of the represented taxonomic groups, while a second is unchanged. Multi-enzymes are used in an effort to make taxonomic assignments; however taxonomic assignments are not necessary for identification of the factorial influences on the leaf endophytic bacterial communities, as studied in this work. Single digestion T-RFLP peaks represent OTUs (Operational T-RFLP Unit) that provide information on the diversity of leaf endophytic bacteria in different environments.
In order to assess the abilities of T-RF OTUs present in individual plants to compete with other bacteria, we focused on the relative amounts of T-RF OTUs in different plants only in those plants in which they were found. The APE of a T-RF in one host species was defined as the average proportion of a T-RF in all the samples of one plant species which have this T-RF. Calculating APE rather than regular average proportion can avoid the problem of underestimation of the abundance of a T-RF in one host species due to non-infection of the bacterial species represented in some samples. The APE of a T-RF can more accurately reflect the overall compositions of leaf endophytic bacterial communities in a plant species than can methods that include absence in the analysis.
In this research, we explored the diversity of leaf endophytic bacteria in selected plant species over time and the physical environment, in order to propose a model describing how multiple factors influence endophytic bacterial communities. Past studies have found the plant genotype and growth conditions have significant impacts on the rhizosphere bacterial communities [34–36] and on the phyllosphere bacterial communities [6, 38]. Here we considered three major influencing factors: host plant species, time and sampling sites. The distributions of leaf endophytic bacteria must be influenced by many factors; however, we hypothesized that these three major factors include most variables affecting community composition. We analyzed leaf endophytic bacterial communities from samples differing in these factors by pCCA and MANOVA of T-RFs and comparisons of the average amounts of T-RFs present in samples.
The factor of host plant species includes the effect of inner biochemical environment and physiological features of the host plant. The results show that the communities in the two grass species, P. virgatum and S. nutans, are similar to one another and distinct from those in the non-grass species. This may be due to similar environments inside grass plants, different from those inside the other plants. The coevolution and codivergence of host plants and leaf endophytic bacterial communities may also contribute to the similarities and differences in the leaf endophytic bacterial communities from different host species. The expectation of a major influence of host plant species on the communities was supported by distinct T-RF patterns from each host species (Figure 1 and Additional file 1: Table S5), by the results of pCCA which assigned half of the total variation to plant species, and the APE analysis (Table 3).
The time factor includes changes in the physical environment, such as temperature, humidity, irradiance and wind speed, and the dynamics of host plant growth. Jackson and Denney  studied the annual and seasonal variation of phyllosphere bacteria and found that compared to significant seasonal variation, the annual variation was not significant. Yadav et al.  also found that the mature leaves have higher populations of phyllosphere bacteria than young leaves. These studies motivated us to consider the seasonal variation of plant-associated bacteria. The pCCA examination of T-RFs treating sampling date as the environmental factor implicated it as a significant factor (Figure 2). The impacts of sampling date on the distribution of plant-associated bacteria were also seen in the average numbers of T-RFs at different sampling dates (Table 2). The temporal variations in relative abundance of different T-RFs suggest that during host plant growth, the structures of plant leaf-associated bacterial communities are also developing to respond to the changes of the inner biochemical environments of host plants and the variations of the weather and overall environment. The host species selected for study begin growth in late April or May. The ratios of the standard deviations of the average number of T-RFs to the average number are smaller in June and July than those in May and August, indicating that the plant-associated bacterial communities are more stable and complex when the host plants are growing in the peak of summer.
The factor of physical environment includes the soil and geobiochemical conditions, the effect of surrounding plants and animals, and the burning and grazing history of the sampling field, records of the latter of which are available. Again, pCCA attributed a significant contribution of sampling site to the total variation (Figure 2b) consistent with T-RF profile differences for the same plant species on the same date (Figure 1).
We recognize that the three targeted factors may not account for all the variation in the communities and that we did encounter a residual variation. Sources of this variation could include: occasional animal disturbance, insect-induced damages and other factors that cannot be measured accurately and parameterized in a mathematical model. Nevertheless, we suggest that the three-factor model describes an important part of the variation of plant-associated bacteria. The plant-associated bacterial communities are not static, but dynamic and evolve with host plants and environments.