Trends in bacterial and fungal communities in ant nests observed with Terminal-Restriction Fragment Length Polymorphism (T-RFLP) and Next Generation Sequencing (NGS) techniques—validity and compatibility in ecological studies

Introduction

Microbes are the most numerous organisms on earth (Bertrand et al., 2011). Their small size and the difficulty of cultivating many (if not most) species has spurred the development of molecular techniques to study the taxonomy, community structure and functions of natural microbial communities. Several techniques such as Automated Ribosomal Intergenic Spacer Analysis (ARISA) (Ramette, 2009), Temperature Gradient Gel Electrophoresis (TGGE), Denaturing Gradient Gel Electrophoresis (DGGE) (Nocker, Burr & Camper, 2007) and Terminal Restriction Fragment Length Polymorphism (T-RFLP) (Liu et al., 1997; Kent et al., 2003; Cao et al., 2013) have been developed for the assessment of complex microbial communities. The semi-quantitative properties (Blackwood et al., 2003; Aiken, 2011), high reproducibility, low cost (Thies, 2007) and the technical and analytical straightforwardness of T-RFLP has cemented it as one of the leading fingerprinting methods for several decades (Schütte et al., 2008). T-RFLP has successfully been used to evaluate the effects of spatial and temporal variation in the structure of microbial communities (Schütte et al., 2008; Robinson et al., 2009; Barnes et al., 2016). However, to obtain taxonomic identity of microbial community members, T-RFLP is usually combined with time-consuming and expensive preparation and sequencing of clone libraries (Van Elsas & Boersma, 2011).

Recent advances in high-throughput DNA sequencing techniques have made NGS (Next Generation Sequencing) methods such as Illumina MiSeq/HiSeq, IonTorrent and 454 pyrosequencing attractive alternatives for analysis of microbial communities (Teeling & Glöckner, 2012). In contrast to T-RFLP, sequence data generated by these methods reveal taxonomic identity to the extent that designated sequence data is available (Weissbrodt et al., 2012; De La Fuente et al., 2014; Prakash et al., 2014). However, NGS datasets can be extensive, require more complex bioinformatic interpretation and the protocols and methods used for microbial NGS analysis are far from standardized (Amend, Seifert & Bruns, 2010; Tedersoo et al., 2015; Balint et al., 2016). Complementing T-RFLP with NGS techniques can, however, offer a convenient way of generating and analyzing metagenomic microbial data, including the addition of taxonomic information achieved by NGS sequencing, to the T-RFLP fingerprints. This is especially useful for longitudinal studies, where substantial T-RFLP-fingerprinting data, including data generated prior to the NGS era, are available. Furthermore, different environments facilitate and promote microbiomes of different composition and diversity and the analysis should be carefully adapted to the specific study at hand (Buttigieg & Ramette, 2014; Faust et al., 2015; Weiss et al., 2016). The few studies to date on bacteria, indicate that the patterns and trends in data generated by the two techniques are largely congruent (rumen bacteria (De La Fuente et al., 2014), bacteria in Antarctic soil (Van Dorst et al., 2014), bacteria in tropical soil (Supramaniam et al., 2016)). Studies that utilize both T-RFLP and NGS data to explore fungal communities appear absent from the literature, as do studies comparing both bacterial and fungal communities using both techniques.

Ants are frequent in most terrestrial environments (Vander Meer, 2012), often being referred to as ecosystem engineers due to their skills in manipulating the abiotic and biotic properties of soil (Dauber, Schroeter & Wolters, 2001; Jurgensen et al., 2008). While building their nests, the bioturbation and the selective picking of appropriate construction materials create a soil chemistry that differs significantly from the one of the background soil (Kilpeläinen et al., 2007; Dostál et al., 2005). The distinct nest environment is clearly reflected by the nest microbiome: the nests of several mound building ant species contain a substantially higher microbial biomass (Dauber, Schroeter & Wolters, 2001), more numerous and diverse fungal colonies (Duff et al., 2016) and significantly different microbial assemblages (Boots et al., 2012), than those of the reference soils.

The aim of this study was to evaluate the congruence of Illumina MiSeq and T-RFLP data, based on the bacterial 16s and fungal ITS areas of the rRNA genes from complex environmental communities. To do this, the microbial communities inside the nests of the mound building ant Formica exsecta (Nylander, 1846) were sampled over a two-year period. The bacterial and fungal communities in the F. exsecta mounds have not been studied previously, but here molecular microbial data of active nest mounds is reported, generated with both methodologies. The relative abundance patterns, diversities and sampling efficiencies generated by the two molecular techniques were compared, and the similarity of the functional, spatial, and temporal patterns were evaluated. The extent to which T-RFLP can be complemented with the taxonomic information gained from the NGS data was also assessed.

Materials & Methods

Results

The T-RFLP-analysis generated 129 (HaeIII) and 120 (MspI) unique bacterial T-RFs (n = 37), and 102 (HaeIII) and 124 (MspI) fungal T-RFs (n = 36), respectively (Table 1). The Illumina MiSeq sequencing yielded in total (i.e., for all samples) 1,896,920 bacterial reads (16S sequences), and 3,274,825 fungal reads (ITS2 sequences), of high quality (n = 38 for both bacteria and fungi) (Table S1 and SI_Bioinformatics). These clustered into 4,699 unique bacterial and 2,315 unique fungal OTUs (Table 1). The Good’s estimate of coverage for the T-RF data suggested that 97.9% of the bacterial, and 93% of the fungal diversity had been captured by the method, whereas the coverage for the OTUs was 99.9% and 99.4%, respectively (Table 1). The species accumulation curves (Fig. 1) showed a similar pattern, with the curves for the T-RFs being generally less flattened towards the end of their curves, particularly so in the fungal data. Only the bacterial OTU communities reached asymptote, suggesting high sampling efficiency (Fig. 1). The mean Shannon-Wiener diversity index for the T-RFs was lower than for the OTUs for both the bacteria (3.35 vs. 5.69) and fungi (2.71 vs. 3.48) (Table 1). The Pareto-Lorenz curves suggested a high Fo, indicating a high functional organization, for both the bacterial and fungal communities, regardless of the method used (Fig. 2). In the bacterial data, 20% of the T-RFs and OTUs accounted for 85% and 87% of the species abundance, respectively (Fig. 2A). The corresponding values for the fungal T-RFs and OTUs projected 73% and 95%, respectively (Fig. 2B).

The PCoAs indicated clustering of bacterial communities within islands and nests, in both the T-RF and OTU data, but the clustering was more pronounced in the OTU data (Figs. 3A and 3B). The horseshoe pattern in the bacterial NGS data remained in the PCoA analysis when it was repeated with log-transformed data, and also when the analysis was performed on alternative distance matrices (Figs. S2A–S2E), suggesting a true pattern rather than an artefact. A clustering by month (Fig. S3) was indicated for the T-RFs, which seemed mainly driven by the values of September. No such indication was seen for the OTUs. The PERMANOVA confirmed the effect of month for the bacterial T-RFs. The PERMANOVA suggested compliance between the two methods, showing significant effects of island and nest for both the T-RF and OTU data (Table 2). The effect of year was not significant, neither for the T-RFs nor the OTUs (Table 2). The PCoAs from the fungal T-RF and OTU data indicated clustering across both island and nest (Figs. 3C and 3D), with the fungal OTU data showing a much more pronounced clustering than the T-RF data (Figs. 3C and 3D). The PERMANOVA showed significant effects of island and nest for both T-RFs and OTUs in the fungal data, suggesting full compliance between the two methods for all the tested effects (Table 2). The Mantel tests showed moderate to good correspondence between the T-RF and the OTU data, being significantly correlated (Pearson’s rho = 0.743, P ≤ 0.001, and 0.574, P ≤ 0.001) for the bacterial and the fungal data sets, respectively.

The relative abundances of the T-RFs, and the equivalent number of OTUs, ranked highest to lowest (encompassing 52% (HaeIII) and 50% (MspI) of the bacterial, and 78% and 81% of the fungal OTU sequences), were highly and significantly correlated (Pearson’s rho = 0.949 (HaeIII), and 0.849 (MspI), respectively) for the bacterial data, and 0.953 and 0.904, respectively for the fungal data; P ≤ 0.001 in all cases; Figs. 4A–4D. Of all the bacterial OTUs, 89% were assigned identities at the phylum, and 53% at the family level (Table 3), whereas the corresponding values for the fungal OTUs were 85%, and 39%, respectively (Table 3, Figs. S4A–S4B).

The 200 most abundant bacterial OTUs, on which the virtual restriction was performed, comprised approximately 60% of all bacterial OTU sequences. Taxonomic information (based on the taxonomic identification of the Illumina OTUs) was available for 66% of the 200 bacterial OTUs down to the level of family. The enzyme HaeIII successfully cut nearly all (99%) of the most abundant 200 OTUs, whereas MspI cut 86%. Matching virtual T-RF with the T-RFs generated from soil samples returned 93 taxonomic items, with restriction patterns from both enzymes (Table S2). Seventeen different restriction patterns were found, three of which were shared between several phyla, whereas 14 patterns were unique at the order level, and six of these at the family level (Table S2). No taxonomic information was available for three of the restriction patterns below the class level, and one pattern returned phylum only. Four of the six identified families (Burkholderiaceae, Caulobacteraceae, Nocardioidaceae and Streptomycetaceae) belonged to the bacterial families ranked as the most abundant according to the Illumina MiSeq data (Table 3).

The corresponding fungal data, based on the 200 most abundant fungal OTUs, on which the virtual restriction was performed, comprised approximately 87% of all fungal OTU sequences Taxonomic information down to the family level was available for 50% of the fungal OTUs. A virtual HaeIII restriction site was found in 97%, and an MspI restriction site in 93% of the fungal OTUs. Matching the virtual T-RFs with the T-RFs generated from nest mound soil samples returned 41 taxonomic items, with restriction patterns from both enzymes. These taxonomic items represent 14 distinct restriction patterns, all of which were unique at the level of phylum (Table S3). Nine unique patterns were identified at the order level, and 8 patterns were further identified to the family level. Four patterns were identified to phylum only. Of the identified families, the Tremellales_incertae_sedis and Venturiaceae were among the most abundant ones according to the Illumina MiSeq data (Table 3).

Discussion

In this study, T-RFLP fingerprinting and Illumina sequencing were used to analyze the bacterial and fungal communities in the nest mounds of the ant F. exsecta. The results show that the combination of the two approaches is well suited to study microbial communities of high complexity, and that they provide similar, yet complementary information. With both methods, the microbial communities showed high and similar functional organization, and similar spatiotemporal patterns across both nest mounds and sampling occasions. Comparisons between fungi and bacteria also revealed that bacterial communities were more comprehensively sampled than fungal communities.

The number of T-RFs obtained in this study were similar in magnitude to other comparable studies of soil microbial communities in polar, pasture, meadow, agricultural and forest soils, both for bacteria (Blackwood et al., 2003; Frey et al., 2009; Anderson et al., 2011; Van Dorst et al., 2014), and fungi (Schwarzenbach, Enkerli & Widmer, 2007; Boots et al., 2012; Van Dorst et al., 2014). Similarly, the number of microbial sequences, the sum of refined OTUs, and the resultant unique OTUs identified in this study, were in good correspondence with recent NGS studies of soil bacteria (Lazzaro, Hilfiker & Zeyer, 2015; Supramaniam et al., 2016) and fungi (Meiser, Bálint & Schmitt, 2014). In general, the number of defined, unique OTUs was much higher than the number of respective T-RFs (approximately 36 times the bacterial, and 19 times the fungal T-RFs). Nevertheless, when ranked per relative abundance, the high Pearson correlation suggest similar proportions of the most abundant T-RFs and OTUs, for both bacteria and fungi. This does not automatically mean that the most abundant T-RFs correspond with the most abundant OTUs or vice versa, but it indicates that the most abundant taxa were obtained by both techniques, in similar relative proportion.

Comparisons between the sampling coverage, i.e., Good’s estimates, and the species accumulation curves of fungi and bacteria, revealed that bacterial communities were more comprehensively sampled than fungal communities. Notably, the asymptote was not reached for neither fungi nor bacteria using T-RFLP, suggesting undersampling by this technique (Zhou et al., 2008). Only a small fraction of microbial diversity is sampled with this (and similar) methods (Jankowski, Schindler & Horner-Devine, 2014), resulting in low detection sensitivity and flatter species accumulation curves. Comparatively, the asymptote was reached for bacteria, and the curve better saturated for fungi using NGS, highlighting the power of this method to detect rarer taxa. Consequently, the diversity estimates were also lower for fungi than for bacteria. Technical factors likely to contribute to the observed differences in sampling success include the length of sequenced fragments, the choice of genes sequenced, and primer specificity. The technical requirement of using different primer pairs for both methods, resulted in fragments of differing length. The bacterial Illumina sequences were ∼500 bp (Timonen et al., 2017), and the T-RFLP sequences ∼1,200 bp (Muehling et al., 2016) long, respectively, whereas the fungal target sequence was 280–310 bp long (Ihrmark et al., 2012) for both methods. Thus the longer fragment can be expected to provide more phylogenetic information, leading to a finer tuned OTU clustering. For the T-RFLP differentiation, the longer fragment could provide an increased number of enzyme cutting sites, which could partly explain the higher yield of the bacterial T-RFs compared to the fungal T-RFs. In addition, the bacterial 16S gene is highly conserved, and successfully targets most bacterial phyla (Větrovský et al., 2013). Developing fungal primers with similar attributes has proven more challenging (Martin & Rygiewicz, 2005; Asemaninejad et al., 2016), as fungal markers fulfilling the criteria of being of suitable length and specificity, yet not discriminating against any fungal taxa are not available (Lindahl et al., 2013). This is further compounded by the fact that a substantial proportion of the fungal ITS sequences in the International Nucleotide Sequence Database are incomplete, with fragmentary regions in the ITS1 and ITS2 subsections (Nilsson et al., 2009). From a biological standpoint, it is nevertheless possible that the fungal communities are less diverse, owing to abiotic factors (De Vries & Shade, 2013; Wagg et al., 2014).

Illumina sequencing and T-RFLP fingerprinting captured the same patterns of functional organization in the nest mounds of F. exsecta. The Pareto-Lorenz curves suggested a high functional organization of the microbial communities inside the nest mounds. This suggests that the number of dominant species within the community is relatively low (Wittebolle et al., 2009), given that a high functional organization signals numerical dominance by a few species (Mertens, Boon & Verstraete, 2005; Marzorati et al., 2008; Wittebolle et al., 2008; Gonzalez-Gil & Holliger, 2011). This agrees with our finding that 50% of the fungal sequences, and 60% of the bacterial sequences were found among the 200 most abundant OTUs. Such communities are often specialized, but potentially susceptible to disturbance, requiring a long recovery time for regaining their functionality. Conversely, a community of high species evenness is expected to cope better with environmental disturbances (Marzorati et al., 2008). The high level of functional organization suggests that the microbial communities in the nest mounds may be specialized and strongly affected by the habitat with which they are associated (Edwards et al., 2011; Boots & Clipson, 2013; Cai et al., 2014). Further study to confirm an existent influence of the F. exsecta nest environment on the bacterial and fungal communities, compared with the effect of a reference substrate, would be required.

Both techniques also captured similar community patterns in bacteria and fungi. The results suggest that microbial communities were primarily structured at the level of nest mounds, with a more distinct clustering in the OTU data. Structure at the level of island was also found, but this effect may be compromised by the low sample size, and a more extensive study with a larger number of sampling sites would be required to confirm this effect. In this study, the higher sequencing depth (i.e., higher number of OTUs generated) likely contributed to the slightly lower correspondence between NGS and T-RFLP matrices, compared to that found in e.g., Pilloni et al. (2012) and De La Fuente et al. (2014). A higher sequencing depth is likely to reveal rarer taxa, which would not have been defined in earlier studies, and are inaccessible to methods such as T-RFLP.

Although even greater taxonomic details for the T-RFs, and a better agreement between the two methods could be achieved by the use of additional enzymes (Dunbar, Ticknor & Kuske, 2001; Edwards & Turco, 2005), the overall taxonomic information obtained was comparable to those reported for extensive T-RFLP studies involving clone libraries (Axelrood et al., 2002; Youssef & Elshahed, 2009; Lin et al., 2011). The correlation between the relative abundance of T-RFs, and the equivalent number of the most abundant OTUs was significant both for bacteria and fungi, which corroborates the congruence between the methods. The outcome of the correlation served as a workable guideline for the number of OTUs (200) that were forwarded to the in silico T-RF cuttings. For both bacteria and fungi, the virtual T-RFLP generated restriction patterns from the majority of this subset of OTU sequences obtained by the Illumina sequencing. The bacterial OTUs were assigned to 13 different phyla, with roughly half (52%) being assigned to the phyla of Proteobacteria and Actinobacteria, both of which were also captured by the matching of T-RFs. At lower taxonomic levels, T-RFs and the virtual restriction cutting patterns corresponded to the taxonomic families of Burkholderiaceae, Caulobacteraceae, Nocardioidaceae and Streptomycetaceae. In the fungal data, the Illumina sequencing captured four phyla, whereas, based on the virtual restriction, the T-RFLP only captured two phyla, the Ascomycota and Basidiomycota. At lower taxonomic levels, the virtual restriction captured the families Tremellales incertae sedis, and Venturiaceae, both of which were the most abundant fungal families captured by the Illumina sequencing.

The bacterial and fungal taxa captured by our analysis are abundant in soil environments (Acosta-Martínez et al., 2008; Buée et al., 2009; Anderson et al., 2011; Voglmayr et al., 2011; Lanzén et al., 2015; DeAngelis et al., 2015). Furthermore, the bacterial genera belonging to Burkholderiaceae have also been shown to be associated with ant genera such as, Camponotus (He et al., 2011), Cephalotes (Russell et al., 2009; Kautz et al., 2013), and Tetraponera (Van Borm et al., 2002). Several genera representing both Proteobacteria and Actinobacteria, and sequences matching Burkholderia were also present in the transcriptomic data of F. exsecta (Johansson et al., 2013). The two fungal families that were captured by both T-RFLP and Illumina (Tremellales incertae sedis and Venturiaceae) are not currently defined in the literature as having any association with ants. However, the literature on fungi associated with ants is focused solely on the fungus-farming ants and the restricted number of fungal species they cultivate, or the few insect pathogenic fungi studied in connection with ants (Hughes et al., 2004).

Conclusions

Thorough assessment of complex microbial communities is challenging, and the use of multiple methods is often required in order to achieve a comprehensive outcome. This study describes certain steps at which technical choices during the process have substantial impact on the outcome, pinpointing some plausible reasons for non-conformity between the two methods. Overall, this study shows that both T-RFLP and Illumina sequencing are suitable for analysis of the fungal and bacterial communities in nest mounds of ants, but when the two techniques are combined, it can provide an even more robust dataset. Furthermore, this study validates the use of both techniques when addressing the topic of microbial communities in complex environments such as ant nest mounds, both independently or to support findings within comparable studies. In particular, when comprehensive longitudinal studies based on T-RFLP exist, supplementing the fingerprinting data with NGS is a practical solution. Although refining the results gained by the two methods into a reliable synthesis requires testing of overall compliance, T-RFLP can be trusted to reveal the same general community patterns as NGS (Illumina MiSeq), and is a good option if resources for NGS are limited.

Supplemental Information