Summary
While recent efforts to catalogue Earth’s microbial diversity have focused upon surface and marine habitats, 12% to 20% of Earth’s bacterial and archaeal biomass is suggested to inhabit the terrestrial deep subsurface, compared to ∼1.8% in the deep subseafloor 1–3. Metagenomic studies of the terrestrial deep subsurface have yielded a trove of divergent and functionally important microbiomes from a range of localities 4–6. However, a wider perspective of microbial diversity and its relationship to environmental conditions within the terrestrial deep subsurface is still required. Here, we show the diversity of bacterial communities in deep subsurface groundwater is controlled by aquifer lithology globally, by using 16S rRNA gene datasets collected across five countries on two continents and from fifteen rock types over the past decade. Furthermore, our meta-analysis reveals that terrestrial deep subsurface microbiota are dominated by Betaproteobacteria, Gammaproteobacteria and Firmicutes, likely as a function of the diverse metabolic strategies of these taxa. Despite this similarity, evidence was found not only for aquifer-specific microbial communities, but also for a common small consortium of prevalent Betaproteobacteria and Gammaproteobacterial OTUs across the localities. This finding implies a core terrestrial deep subsurface community, irrespective of aquifer lithology, that may play an important role in colonising and sustaining microbial habitats in the deep terrestrial subsurface. An in-silico contamination-aware approach to analysing this dataset underscores the importance of downstream methods for assuring that robust conclusions can be reached from deep subsurface-derived sequencing data. Understanding the global panorama of microbial diversity and ecological dynamics in the deep terrestrial subsurface provides a first step towards understanding the role of microbes in global subsurface element and nutrient cycling.
Main text
Understanding the distribution of microbial diversity is pivotal for advancing our knowledge of deep subsurface global biogeochemical cycles 7, 8. Subsurface biomass is suggested to have exceeded that of the Earth’s surface by an order of magnitude (∼45% of Earth’s total biomass) before land plants evolved, at ca. 0.5 billion years ago 9. Integrative modelling of cell count and quantitative PCR (qPCR) data and geophysical factors indicated in late 2018 that the bacterial and archaeal biomass found in the global deep subsurface may range from 23 to 31 petagrams of carbon (PgC). These values halved previous efforts from earlier that year10 but maintained the notion that the terrestrial deep subsurface holds ca. 5-fold more bacterial and archaeal biomass than the deep marine subsurface. Further, it is expected that 20-80% of the possible 2-6 × 1029 prokaryotic cells present in the terrestrial subterranean biome exist as biofilms and play crucial roles in global biogeochemical cycles 10, 11.
Cataloguing microbial diversity and functionality in the terrestrial deep subsurface has mostly been achieved by means of marker gene and metagenome sequencing in coals, sandstones, carbonates, and clays, as well as deep igneous and metamorphic rocks 4–6, 12–20. Only recently has the first comprehensive database of 16S rRNA gene-based studies targeting terrestrial subsurface environments been compiled 10. This work focused on updating estimates for bacterial and archaeal biomass, and cell numbers across the terrestrial deep subsurface, but also linked the identified bacterial and archaeal phylum-level compositions to host-rock type, and to 16S rRNA gene region primer targets 10. While highlighting Firmicutes and Proteobacterial dominance in the bacterial component of terrestrial deep subsurface, no further taxonomic insights were gained. However, genus-level identification is critical for understanding community composition, inferred metabolism and hence microbial contributions of distinct community members to biogeochemical cycling in the deep subsurface 18, 21–23. Indeed, such genus-specific traits have been demonstrated as critical for understanding crucial biological functions in other microbiomes 24, and genus-specific functions of relevance for deep subsurface biogeochemistry are clear 25, 26.
So far, the potential biogeochemical impacts of microbial activity in the deep subsurface have been inferred through shotgun metagenomics, as well as from incubation experiments of primary geological samples amended with molecules or minerals of interest 13, 19, 20, 27–30. Recent studies of deep terrestrial subsurface microbial communities further suggest that these are metabolically active, generally associated with novel uncultured phyla, and potentially directly involved in carbon and sulphur cycling 31–36. Concomitant advancements in subsurface drilling, molecular methods and computational techniques have aided the exploration of the subsurface biosphere, but serious challenges remain mostly related to deciphering sample contamination by drilling methods and sample transportation to laboratories for processing 37, 38. The logistical challenges inherent to accessing and recovering in situ samples from hundreds to thousands of metres below surface complicate our view of terrestrial subsurface microbial ecology 39.
In this study, we capitalize on the increased availability of 16S rRNA gene amplicon data from multiple studies of the terrestrial deep subsurface conducted over the last decade. We apply bespoke bioinformatic scripts to generate insights into the microbial community structure and controls upon bacterial microbiomes of the terrestrial deep subsurface across a large distribution of habitat types on multiple continents. The deep biosphere is as-yet undefined as a biome - elevated temperature, anoxic conditions, low levels of organic carbon, and measures of isolation from the surface photosphere are some of the criteria used albeit without a consensus. For this work a more general approach has been taken to define the terrestrial deep subsurface as the zone at least 100 m from the surface 40, 41.
Meta-analysis of the terrestrial deep subsurface microbiome
Here, we were able to compare datasets encompassing different 16S rRNA gene hyper-variable regions, and derived from different DNA extraction methodologies, facilitated by closed-reference Operational Taxonomic Unit (OTU)-picking of each study individually using the same 16S rRNA gene reference database. This procedure begins to address technically confounding variables by limiting taxonomy assignments to only the archaeal and bacterial diversity listed in the chosen database and precludes the discovery of novel taxa.
The finalized meta-analysis dataset comprised of 16S rRNA data from seventeen aquifers in either sedimentary- or crystalline-host rocks, from depths spanning 94 m to 2300 m below land surface (mbls), targeting mostly groundwater across 5 countries and two continents (Supplementary Table 3). Nine DNA extraction techniques were used in these studies, ranging from standard and modified kit protocols (e.g. MOBIO® PowerSoil, see Table 1) to phenol-chloroform and CTAB/NaCl based methods 42–47. Finally, 6 different primer pair amplified regions of the 16S rRNA gene, in 454 pyrosequencing and Illumina sequencing, were used to generate the datasets.
Initial processing of 187 retrieved samples revealed 24,632,035 chimera-checked sequences 17, 28, 43, 44, 48–50. SILVA 123-aided closed-reference OTU-picking yielded 6,975 OTUs associated to 598,341 sequences following exclusion of singleton OTUs and samples containing 2 or less OTUs. The final dataset following stricter contamination-aware filtering (cf. Methodology) was comprised of 70 samples and 2,207 OTUs (513,929 sequences, 2.54% of the initial sequences), where Archaeal reads comprised 1.5% of the total number of reads.
Trends in taxonomic diversity
Among a total of 45 detected bacterial phyla, Proteobacteria were seen to dominate most community profiles in this dataset (Figure 1). The most abundant proteobacterial classes (Alpha-, Betaproteobacteria, Delta-, Gammaproteobacteria) represented 57.2% of the total number of reads, with 13.4% of these assigned to class Clostridia (Firmicutes). A general prevalence of Betaproteobacteria and Gammaproteobacteria in the deep biosphere may be explained by the diverse metabolic capabilities of taxa within these clades. Families Gallionellaceae, Pseudomonadaceae, Rhodocyclaceae and Hydrogeniphillaceae within Betaproteobacteria and Gammaproteobacteria are suggested to play crucial roles in deep subsurface iron, nitrogen, sulphur and carbon cycling across the world 43, 51, 52. The relative abundance of order Burkholderiales (Betaproteobacteria) in surficial soils has previously been correlated (R2=0.92, ANOVA p-value <0.005) with mineral dissolution rates, while genus Pseudomonas (Gammaproteobacteria) is widely known to playing a key role in hydrocarbon-degradation, denitrification and coal solubilisation in different locations53–55.
Mean grouped proportion values indicated that Betaproteobacteria were the most abundant proteobacterial class in most host rocks, representing 26.1% of all reads in the dataset. While Betaproteobacteria accounted for 53.96% of the community profile for sub-bituminous coals, Gammaproteobacteria dominated volatile bituminous coals (49.1% of the profile, Figure 1). The dominance of Betaproteobacteria and Gammaproteobacteria in coals builds on culture-based evidence of widespread degradation of coal-associated complex organic compounds by these classes56–59.
Firmicutes were represented in large part by class Clostridia and mostly associated with sedimentary aquifers (i.e. sandstone, dolomite, siltstone, shale – Figure 1). This class includes ubiquitous anaerobic hydrogen-driven sulphate reducers also known to sporulate and metabolize a wide range of organic carbon compounds that have been found to dominate extremely deep subsurface ecosystems beneath South Africa and likely globally given the pervasive high levels of H2 in similar geologic settings 13, 60–62. Clostridia from metagenomes have been detected from the terrestrial deep subsurface and inferred to have the physiological capabilities needed to thrive in these environments 20. Adaptation to extreme environments in Clostridia is posed to be driven by varied metabolic potential, sporulation ability, and capacity for CO2- or sulphur-based autotrophic H2-dependent growth 22, 63.
Lithological controls on community structure
Microbial community structure and composition in soils depend on fine-tuned geochemical, physical and hydrogeological conditions that influence microbial presence and metabolism 64. This relationship also appears to be reflected in the global subsurface, where host-rock lithology is evident as a primary control on community structure (Figures 2 and 3). Indeed, most host-rocks (10 out of 15 in this dataset) have, on average, more unique OTUs than they share with other host-rocks (Figure 2). Particularly, in sulphide-rich schists, 73% of the OTUs are, on average, unique to the host-rock. The role of host-rock lithology is further evidenced (Figure 3) as some of the host-rocks clustered at a 95% confidence interval, suggesting closely related microbial communities within similar lithologies, despite other environmental factors such as depth or location. Further, 50.6% of Jensen-Shannon distances ordinated (Figure 3) were significantly explained by aquifer lithology (ADONIS/PERMANOVA, F-statistic=4.65, p-value <0.001, adjusted Bonferroni correction p-value <0.001); thus showing that rock type was the primary variable defining microbial community structure. Other environmental features such as absolute sample depth and medium-scale location (i.e. state, region of the sampling site) explained only 3.08% and 2.78% of the significant metadata-driven variance in microbial community structure, respectively (ADONIS/PERMANOVA, F-statistic=3.95, 3.57 p-value <0.001, adjusted Bonferroni correction p-value <0.001). This suggests that depth-related changes in temperature and pressure are not significant controls on community structure. The relationship of community structure to hydrogeochemical parameters remains an area for future investigation – since hydrogeology and fluid geochemistry are also strongly controlled by lithology via water-rock reactions.
Metadata variables that were unavailable for all samples in the dataset were excluded from the statistical analyses, thus further insights into the significance of other environmental variables was not possible. Nevertheless, this is the first large-scale evidence that deep subsurface microbial community taxonomy appears host-rock-specific. Given the importance of chemolithotrophic metabolisms in dark, subsurface environments, the unique chemical and mineralogical compositions within different aquifer lithologies impart strong controls over microbiomes associated with mineral surfaces and porewaters 43, 62, 65–67. Indeed, direct utilisation of mineral surfaces or dissolved species from minerals for respiration and/or metabolism has been shown to be critical in localised subsurface environments 30, 35, 68, 69. Due to low numbers of samples for some host-rock lithologies in this dataset (e.g. one sample each for siltstone, sandstone, shale and chlorite-sericite-schist), it is not possible to ascertain that microbiome specificity is generalisable to all deep subsurface aquifer types on Earth (see Figure 3). Nevertheless, this study provides the first large-scale evidence that, at a global scale, lithology surpasses depth in shaping deep subterranean microbial communities.
A core terrestrial deep subsurface microbial community?
Analysis of prevalence across the dataset revealed that seven OTUs, all affiliated to genus Pseudomonas, were present in more than 25 and up to 41 samples (see Supplementary Figure 1, Supplementary Table 2). Network analysis (Table 2) highlighted a Pseudomonas OTU highly connected to other OTUs in the dataset. Further, BLAST70 results indicated that recovered sequences for OTUs affiliated to this genus were generally associated to marine and terrestrial soil and sediments (cf Supplementary Table 4, Supplementary Figure 4). Four OTUs affiliated to Burkholderiales (Betaproteobacteria), the second most prevalent order in the dataset, were also found to be connected to up to 34 other OTUs. Genus Thauera (Betaproteobacteria, Rhodocyclales), represented by a single OTU, was the second most central to the dataset. Finally, network and prevalence analysis highlighted the putative importance classes Betaproteobacteria and Gammaproteobacteria may have in the deep subsurface, since taxa affiliated to these were highly connected across the dataset (Table 2). These observations suggest genus-level taxonomy may be relevant to shaping subterranean microbial communities irrespective of host-rock lithology.
The metabolic plasticity of Pseudomonadales and Burkholderiales orders has been demonstrated 71–74 and may be a catalyst for their apparent centrality across the terrestrial deep subsurface microbiomes analysed in this study. These bacterial orders may represent important keystone taxa in microbial consortia responsible for providing key substrates to other colonizers in deep subsurface environments 75, 76. In particular, given the number of highly central Pseudomonas-affiliated OTUs and the prevalence of this genus in the dataset, we suggest that this genus may be key in establishing conditions for microbial colonization in many terrestrial subsurface environments. Genus Pseudomonas and possibly several members of Burkholderiales may therefore comprise an important component of the global core terrestrial deep subsurface microbial community.
Challenges from contamination
16S rRNA gene PCR-based approaches for characterizing microbial diversity in low biomass environments benefit from the sensitivity afforded by PCR, at the cost of vulnerability to contamination 77. Here, we used the prominence of sequences associated with phototrophic taxa as an indicator of either ingress of surface microbiota or contamination during sample processing. The discovery of potentially photosynthetic taxa in the initial dataset, namely 46 OTUs classified as Chloroplast (Cyanobacteria) was read as a sign that further bioinformatics-driven precautions should be taken, despite recent evidence of some cyanobacterial presence in some locations within the deep subsurface 78, 79. Specifically, the presence of other phototrophic members of phyla Chloroflexi and Chlorobi as well as classes Rhodospirillales (Alphaproteobacteria) and Chromatiales (Gammaproteobacteria) informed the decision to filter the dataset to hold only OTUs represented by more than 500 sequences and present in at least 10 samples. Recent recommendations for quality control of 16S rRNA gene datasets also support filtering-based approaches when applied to low biomass subsurface environments 38. This constraint reduced the dataset to the 70 samples and 2,207 OTUs (513,929 sequences) used for the meta-analysis (Table 1), and also reduced the number of prospective contaminants by half, although only ∼26% of the reads associated to Chloroplast-like sequences were removed (17 OTUs, 1958 reads).
Collecting contamination-free samples from the deep subsurface is difficult but important for cataloguing the authentic microbial diversity of the terrestrial subsurface. This study follows recent recommendations for downstream processing of contaminant-prone samples originated in the deep subsurface (Census of Deep Life project - http://codl.coas.oregonstate.edu/), where physical, chemical and biological, but also in-silico bioinformatics strategies to prevent erroneous conclusions have been highlighted 38, 80, 81. This study also follows frequency-based OTU filtration techniques similar to those recommended in Sheik et al. (2018) 38 and designed to remove possible contaminants introduced during sampling or during the various steps related to sample processing. The pre-emptive quality control steps hereby undertaken support a non-contaminant origin for taxa analysed in this dataset following careful in-field and laboratorial contamination-aware procedures carried out in each study. As such, the predominance of typically contaminant taxa affiliated e.g. to genus Pseudomonas was accepted as a true trend in the microbial ecology of the terrestrial deep subsurface.
No evidence was found for DNA extraction and PCR procedures significantly affecting microbial community structure in this meta-analysis (6.01% of the microbial community structure cumulatively [ADONIS/PERMANOVA, F-statistic=3.85, 3.23, p-value <0.01, adjusted Bonferroni correction p-value <0.001] vs. 50.6% from host rock lithology). In spite of this, a general convergence in DNA extraction methods would help further reduce methodology based variation and to standardize downstream analysis of deep subsurface microbial datasets 82, despite the practical challenges of each host-rock matrix and local geochemical conditions.
In the near future, the advent of recently developed techniques for primer bias-free long read 16S rRNA and 16S rRNA-ITS gene amplicon long-read-based sequencing may initiate a convergence of molecular methods from which the deep subsurface microbiology community would benefit greatly 83, 84. The future of large-scale, collaborative deep subsurface microbial diversity studies should encompass not only an effort towards standardization of several molecular biology techniques but also the long-term archival of samples 85. This will permit re-analyses using updated or unified methods after collection, where methodological variations would be controlled, and robust conclusions would more easily be achieved.
Conclusions
A global scale meta-analysis addressing the available 16S rRNA gene-based studies of the deep terrestrial subsurface revealed the dominance of Betaproteobacteria, Gammaproteobacteria and Firmicutes across this biome. Further, aquifer lithology was identified as the main driver of deep subterranean microbial communities. Depth and location were not significant controls of microbial community structure at this scale. Finally, evidence for a core terrestrial deep subsurface microbiome population was recognised through the prevalence and centrality of genus Pseudomonas (Gammaproteobacteria) and several other genera affiliated to class Betaproteobacteria. The adaptable metabolic capabilities associated to the above-mentioned taxa may be critical for colonizing the deep subsurface and sustaining communities. The terrestrial deep subsurface is a hard-to-reach complex ecosystem crucial to global biogeochemical cycles. This study attempts to consolidate a global-scale understanding of taxonomical trends underpinning terrestrial deep subsurface microbial ecology and geomicrobiology.
Methodology
Data acquisition
The Sequence Read Archive database of the National Center for Biotechnology Information (SRA-NCBI) was queried for 16S rRNA-based deep subsurface datasets (excluding marine and ice samples, as well as any human-impacted samples); available studies, were downloaded using the SRA Run Selector. Studies were selected considering the metadata and information on sequencing platform used – i.e., only samples derived from 454 pyrosequencing and Illumina sequencing were considered. Analysis of related literature resulted in the detection of other deposited studies previous search efforts in NCBI-SRA failed to detect. Further private contacts allowed access to unpublished data included in this study. The final list of NCBI accession numbers, totalling 222 samples, was downloaded using fastq-dump from the SRA toolkit (https://www.ncbi.nlm.nih.gov/sra/docs/toolkitsoft/).
As seen in Table 1, required metadata included host-rock lithology, general and specific geographical locations, depth of sampling, DNA extraction method, sequenced 16S rRNA gene region and sequencing method. Any samples for which the above-mentioned metadata could not be found were discarded and not considered for downstream analyses.
Pre-processing of 16S rRNA gene datasets
A customised pipeline was created in bash language making use of python scripts developed for QIIME v1.9.1 86, to facilitate bioinformatic analyses in this study (see https://github.com/GeoMicroSoares/mads_scripts for scripts). Briefly, demultiplexed FASTQ files were processed to create an OTU table. Quality control steps involved trimming, quality-filtering and chimera checking by means of USEARCH 6.1 87. Sequence data that passed quality control were then subjected to closed-reference (CR) OTU-picking on a per-study basis using UCLUST 87 and reverse strand matching against the SILVA v123 taxonomic references (https://www.arb-silva.de/documentation/release-123/). Closed-reference OTU picking excludes OTUs whose taxonomy has not been found in the 16S rRNA gene database used. Although this limits the recovery of prokaryotic diversity to the recorded in the database, cross-study comparisons of microbial communities generated by different 16S rRNA gene primers are made possible. This conservative approach classified OTUs in each study individually to the common 16S rRNA gene reference database the merge of all classification outputs. A single BIOM (Biological Observation Matrix) file was generated using QIIME’s merge_otu_tables.py script. The BIOM file was then filtered to exclude samples represented by less than 2 OTUs using filter_samples_from_otu_table.py, as well as OTUs represented by one sequence (singleton OTUs) by using filter_otus_from_otu_table.py. In an attempt to reduce the impacts of potential contaminant OTUs from the dataset, the post-singleton filtered dataset was further filtered to include only OTUs represented by at least 500 sequences and present in at least 10 samples overall using filter_otus_from_otu_table.py.
Data analysis
All downstream analyses were conducted using the phyloseq (https://github.com/joey711/phyloseq) package within R, which allowed for simple handling of metadata and taxonomy and abundance data 88–90. Merged and filtered BIOM files were imported into R using internal phyloseq functions, which allowed further filtering, transformation and plotting of the dataset (see https://github.com/GeoMicroSoares/mads_scripts for scripts).
Briefly, following a general assessment of the number of reads across samples and OTUs, tax_glom (phyloseq) allowed the agglomeration of the OTU table at phylum-level. For the metadata category-directed analyses, function merge_samples (phyloseq) created averaged OTU tables, which permitted testing of hypotheses for whether geology or depth had significant impacts on microbial community structure and composition. Computation of a Jensen-Shannon divergence PCoA (Principal Coordinate Analysis) was achieved with ordinate (phyloseq) which makes use of metaMDS (vegan) 91, 92. All figures were plotted making use of the ggplot2 R package (https://github.com/tidyverse/ggplot2), except for the UpsetR plot in Figure 2, which was plotted with package UpsetR (https://github.com/hms-dbmi/UpSetR).
Author Contributions
ARS developed the methodology, collated and analysed the data, and wrote the manuscript. AE and AM conceived the study, supervised AS and helped write the manuscript. Other authors provided data from field sites used in the global meta-analysis. All authors contributed, edited and approved the final manuscript.
Conflict of Interest
We declare no conflict of interest.
Supplementary Methodology
Phylogeny of Pseudomonas representative sequences
Representative 16S rRNA gene sequences for Pseudomonas OTUs in the dataset were isolated by retrieving OTU IDs affiliated to this genus in the final dataset using the subset_taxa function within phyloseq6. The SILVA 123 database was then queried against the list of OTU IDs and the results deposited in a FASTA file. Outgroup sequences of genus Sphingomonas (Proteobacteria, Alphaproteobacteria, Sphingomonadales, Sphingomonadaceae) were obtained directly from the SILVA database (https://www.arb-silva.de/) and added along with the retrieved Pseudomonas 16S rRNA gene sequences to a final FASTA file.
Using MEGA77, an alignment was performed using the MUSCLE8 algorithm and 8 iterations of the UPGMB (combines Neighbour-Joining9 and UPGMA - Unweighted Pair Group Method with Arithmetic mean) clustering method. An optimal Neighbour-Joining9 tree (cf. Supplementary Figure 4) was then created using 500 bootstrapped replicates and the Maximum Composite Likelihood (MCL)10 method to calculate evolutionary distances.
Acknowledgements
The work was funded by a National Research Network for Low Carbon Energy and Environment (NRN-LCEE) grant to ACM and AE from the Welsh Government and the Higher Education Funding Council for Wales (Geo-Carb-Cymru). Deep borehole samples from Nevada and California, USA (e.g. Nevares Deep Well 2 and BLM-1) were obtained with help in the field from Alexandra Wheatley, Jim Bruckner, Jenny fisher and Scott Hamilton-Brehm, and technical assistance and funding from the US Department of Energy’s Subsurface Biogeochemical Research Program, the Hydrodynamic Group, LLC, the Nye County Nuclear Waste Repository Program Office (NWRPO), the US National Park Service, and Inyo Country, CA. Samples from a mine in Northern Ontario Canada were obtained with funding from the Natural Sciences and Engineering Research Council of Canada and the assistance of Thomas Eckert, and Greg Slater of McMaster University. The Census of Deep Life (CoDL) and Deep Carbon Observatory (DCO) projects are acknowledged for a range of studies used in this analysis, as well as the sequencing team at the Marine Biological Laboratory (MBL). Disclaimer: Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
References
- 1.↵
- 2.
- 3.↵
- 4.↵
- 5.
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.
- 15.
- 16.
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.
- 33.
- 34.
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.
- 46.
- 47.↵
- 48.↵
- 49.
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.
- 55.↵
- 56.↵
- 57.
- 58.
- 59.↵
- 60.↵
- 61.
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.
- 73.
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.
- 90.↵
- 91.↵
- 92.↵