Based on previous sampling experiences in the Tabernas Desert and sequencing tests performed in the laboratory, a detailed roadmap for the expedition and the experimental procedures was designed (Figure 1). The total duration of the expedition was less than 60 h, including traveling (∼25% of hands-on time) and two nights. The rest of hands-on time was spent on library preparation (∼28%), sequencing and basecalling (∼26%), sampling and setup (13%), and data analysis (8%). The first set of sequencing data was generated approximately 24 h after sample collection.

Twelve biocrust and two bulk soil samples (“control” samples) were collected and analyzed through full-length 16S rRNA gene sequencing using the ONT MinION platform. A total of 1,657,804 raw reads were generated. After length and quality filtering, an average of 101,972 ± 20,949 sequences per sample were obtained (min: 50,051; max: 128,282; median Q = 10.3). Reads were subsequently analyzed by using a custom pipeline (Spaghetti), which was inspired by previous works (Cuscó et al., 2018; Santos et al., 2020; Urban et al., 2021). Spaghetti relied on minimap2 (Li, 2018) alignments against the SILVA v. 138 database (Quast et al., 2013), and taxonomic assignments were obtained in ∼2 h. Other alignment tools were tested as alternatives to minimap2, but they were discarded for different reasons: BLAST took ∼26 h to finish a ∼1M reads analysis, while LAST exceeded the available laptop’s RAM (16 Gb).

Spaghetti data analysis and visualization pipeline generated several plots designed to provide a rapid overview of the taxonomy and the diversity of the samples (Supplementary File 1). At the phylum level, biocrust samples were dominated by Cyanobacteria (∼34.5% of average relative abundance), Bacteroidota (∼22.7%), Proteobacteria (∼19.2%), Acidobacteriota (∼6.0%) and Actinobacteriota (∼4.7%), while soil samples were mainly characterized by Actinobacteriota (∼24.8%), Acidobacteriota (∼18.6%), Proteobacteria (∼14.2%). Planctomycetota (∼14.2%) and Gemmatimonadota (∼7.3%) (Supplementary Figure 1 and Supplementary Table 1).

As expected, a higher variability in the microbiome composition was detected at the genus level, with an uncultured Cyanobacteriales (∼4.7% of average relative abundance), Hymenobacter (∼3.9%), an uncultured Chroococcidiopsaceae (∼3.8%), an uncultured Spirosomaceae (∼3.7%) and Adhaeribacter (∼3.5%) being the most dominant taxa for biocrust samples. Moreover, a considerable amount of reads (∼14.0%) were assigned to chloroplasts in these samples. On the other hand, soil samples were mainly characterized by Rubrobacter (∼5.8%), Vicinamibacteraceae (∼4.8%), an uncultured Pirellulaceae (∼4.7%), Pyrinomonadaceae RB41 (∼4.4%), an uncultured Vicinamibacterales (∼4.1%), and a low presence of reads assigned to chloroplasts (∼0.15%) (Figure 2A and Supplementary Table 2).

Beta diversity analyses showed that biocrust and soil samples were clearly distinguishable at the microbiome level. Moreover, samples tend to cluster based on their sampling location (X1, X2, X3, X4 or X6), instead of other characteristics (i.e., color and shape of the biocrust) (Figure 2B). Alpha diversity indices were used to identify the most and least rich and diverse samples, which were X1.8/X1.3/c.1 and X6.2/X2.1/X4.1, respectively (Figure 2C).

Once the general taxonomic and diversity profiles were obtained, special attention was paid to 29 bacterial genera that had proven to be radiation- and/or desiccation-resistant according to the literature (Montero-Calasanz et al., 2013; Yu et al., 2015; Deng et al., 2016; Etemadifar et al., 2016; Paulino-Lima et al., 2016; Golinska et al., 2020; Tanner, 2020). The objective of this analysis was to identify those samples which maximized the richness and abundance of those radiation- and/or desiccation- resistant taxa, since they should hold a greater potential for isolating and discovering microbial strains and substances of biotechnological interest.

Overall, the number of radiation- and desiccation-resistant genera detected in the samples by Nanopore sequencing was high, ranging from 23 (X2.1 and X2.2) to 29 (X1.1 and X1.8) (Figure 3A). Although some of the taxa were present in low abundance (< 0.01%), the selected bacteria accounted for 11.5% of the relative abundance of the samples, in average (Figure 3B). Biocrust profiles were dominated by Hymenobacter (∼3.9% of the total relative abundance), Sphingomonas (∼2.7%), Rubellimicrobium (∼1.7%), Microvirga (∼1.3%) and Rubrobacter (∼1%). The two bulk soil samples were mainly characterized by the presence of Rubrobacter (∼5.8%), Arhtrobacter (∼0.9%) and Sphingomonas (∼0.7%) (Figure 3 and Supplementary Table 2).

After analyzing all the results provided by the pipeline, additional samples were collected. This time, bioprospecting activities focused on obtaining biological replicates of three selected samples: (a) biocrust X1.1, with the highest number of radiation- and desiccation-resistant genera (29); (b) bicrust X2.1, with the lowest number of radiation- and desiccation-resistant genera (23); and (c) bulk soil c.1, taken as a control for comparisons between biocrust and bulk soil samples.

GPS positions of the original samples were traced back and samples were identified based on the pictures that were taken on the first sampling day. Finally, two additional replicates were collected for each type of sample.

Back in the laboratory after the expedition, all the collected samples (n = 20) were cultured under three different conditions: (1) Tryptic Soy Agar (TSA) medium, (2) SSE/HD medium (SSE/HD), and (3) SSE/HD medium + uninterrupted artificial light (SSE/HD + light). A total of 166 strains comprising 50 different genera were isolated and identified through Sanger sequencing of the partial 16S rRNA gene. The bacterial colonies displayed differences in morphology and appearance, with white, yellow, pink, red, orange and brown being the most predominant colors (Supplementary Table 3). Initially, samples cultured on SSE/HD + light did not display any microbial growth after 4 weeks of incubation. For that reason, plates were removed from the artificial light, and a few days later, different bacterial colonies started to grow.

The genus Arthrobacter was the most represented in the microbial collection, with up to 37 isolates belonging to this taxonomic group (Figure 4). A total of 15 strains, which were mainly isolated from soil samples, were classified as Streptomyces. Other predominant genera in the collection were Pseudoarthrobacter (9 isolates), Kocuria (6), Bacillus (6), Skermanella (5), Blastococcus (5), and Belnapia (5). At the sample level, biocrusts collected from Location 1 (X1) presented the highest number of bacterial isolates. Specifically, samples X1.1B (21 isolates/15 unique genera), X1.2 (17/13), X1.3 (16/12), and X1.1 (16/10) showed the highest diversity of cultured bacteria. On the other hand, samples X3.2 (0/0), X6.2 (0/0), X4.1 (2/2), and X2.1A (2/2) presented the lowest diversity of isolates (Figure 4).

The taxonomic profiles obtained by Nanopore sequencing were compared to the results from the molecular identification of the isolated strains. Overall, Nanopore sequencing and culture-based data correlated well. In fact, only 14 out of the 166 isolated strains belonged to genera that were not detected in the original sample by in situ microbiome sequencing (Figure 4 and Supplementary Table 3). Interestingly, three of the isolated genera (Mycolicibacterium, Lentzea, and Sinorhizobium) were not detected in any sample of the dataset. After revising the database used for assigning the taxonomy of the reads (see section “Materials and Methods”), a mislabeling of those taxa at the genus level was detected. Specifically, Mycolicibacterium was labeled as Mycobacterium, Lentzea as Lechevalieria and Sinorhizobium as Ensifer. These three genera were indeed detected by Nanopore sequencing in all the samples where the strains were isolated from Supplementary Table 2. Finally, it is worth highlighting that some of the most abundant radiation-resistant bacteria detected by in situ 16S rRNA sequencing (i.e., Hymenobacter, Rubrobacter, Rubellimicrobium, Microvirga, Truepera…) were not cultured from any sample. Indeed, only 50 out of the 441 genera (11.3%) with an average relative abundance > 0.01% according to Nanopore sequencing were represented in the microbial culture collection.

Among the selected samples, X1.1 yielded the highest number of total cultured strains (48), the highest number of different cultured genera -as deduced by partial 16S rRNA gene Sanger sequencing- (26), and the highest number of cultured genera classified as radiation- or desiccation-resistant according to literature (8) (Figure 5). In contrast, and as expected considering the results from in situ sequencing (Figure 3B), X2.1 samples displayed the lowest diversity of cultured bacteria and radiation- and desiccation-resistant genera. Moreover, almost all the genera isolated from X2.1 samples were also isolated from X1.1 (Figure 5B, Supplementary Figure 2, and Supplementary Tables 4, 5), thus confirming the hypothesis that this sample was less valuable from the bioprospecting point of view. A different profile of bacteria was isolated from C1 bulk soil samples (Supplementary Figure 2), with only one radiation-resistant genus -Sphingomonas- cultured exclusively from this type of sample (Figure 5B). Interestingly, the relative abundance of Sphingomonas was higher in all the biocrust samples than in bulk soil (Figure 3A), although this genus was isolated only from samples C1B and X1.8.

Focusing on the culture conditions, 124 strains were isolated from TSA (38 different genera), 24 from SSE/HD (17 different genera), and 18 from SSE/HD + light (13 different genera) (Figure 6 and Supplementary Table 6). Nevertheless, strains isolated from SSE/HD + light presented a significantly lower similarity to their closest type strain than strains isolated from TSA (FDR adjusted p-value < 0.05; Mann–Whitney U test), based on partial 16S rRNA gene sequencing. In fact, ∼89% of the strains isolated from SSE/HD + light showed a similarity lower than 98.7% to their closest neighbor, a common threshold for defining new species (Chun et al., 2018), compared to ∼46 and 66% displayed by TSA- and SSE/HD-isolated strains, respectively (Figure 6B).

Sampling was carried out in November 2020 at the Tabernas Desert Natural Park (Almeria, Spain), under the permission of the competent authorities. Biocrust and bulk soil samples were collected in two different days. Biocrust samples were gathered using a laboratory spatula that was sterilized with ethanol 96% immediately before collecting each sample. Bulk soil (∼5 cm deep) was directly introduced into sterile falcon tubes. On the first day, fourteen different samples were taken, and then analyzed through in situ microbiome sequencing. Based on the results, six additional samples were gathered in the second sampling day. These samples were, indeed, biological replicates of the least and most promising samples based on sequencing data. Metadata (geolocation, type of sample, appearance and pictures) was collected and associated to each sample (Supplementary Figure 3).

Requirements for DNA extraction, PCR amplification, library preparation and sequencing were evaluated, and a minimum laboratory setup was designed accordingly (Supplementary Table 7). The necessary equipment fitted in the trunk of a compact car, and it was transferred to an apartment in Viator (Almería, Spain), 15 km away from the Tabernas Desert, where the mobile laboratory was established and all the experimental and data analysis procedures were carried out. The apartment was equipped with electricity, internet connection, a fridge and a freezer.

Approximately 0.25 g of the samples were used to perform DNA extraction with the DNEasy Power Soil Kit (QIAGEN, Germany, Cat. No.: 12888) according to the manufacturer’s instructions, with an additional incubation step at 65°C after the addition of the C1 solution. DNA was resuspended in 30 μL of sterile Mili-Q water. Qubit x1 dsDNA High-Sensitivity Assay kit (Qubit 2.0 Fluorometer, Thermo Fisher, Waltham, United States, Cat. No.: Q33230) was used for DNA quantification. PCR amplification of the full-length bacterial 16S rRNA gene (V1-V9; ∼1.45 kbp) was carried out by using the S-D-Bact-0008-a-S-16 (5′-AGR GTT YGA TYM TGG CTC AG-3′) and S-D-Bact-1492-a-A-16 (5′-TAC CTT GTT AYG ACT T-3′) primers (Klindworth et al., 2013), which were tailed with the ONT Universal Tags: 5′-TTT CTG TTG GTG CTG ATA TTG C-3′ for forward primer, and 5′-ACT TGC CTG TCG CTC TAT CTT C-3′ for reverse primer. The PCR reaction mix for each sample consisted of 22 μL of H₂O, 25 μL of NZYTaq II 2x Green Master Mix (NZYTech, Lisboa, Portugal, Cat. No.: MB358), 1 μL of both forward and reverse primers and 1 μL of template DNA. For the negative control, 1 μL of Mili-Q water was used instead. The following conditions were used for PCR: initial denaturation (94°C; 1 min); amplification (35 cycles) comprising denaturation (95 °C; 1 min), annealing (49 °C; 1 min) and extension (72°C; 2 min); final extension (72 °C; 10 min). The resulting amplicons were purified with the NucleoMag kit for PCR clean up with magnetic beads (Macherey-Nagel, Germany, Cat. No.: 744100.4). Magnetic beads were used at 0.5 x concentration, and manufacturer’s instructions were followed.

Barcodes were added by employing the PCR Barcoding Expansion Pack 1-96 (ONT, Oxford, United Kingdom, Cat. No.: EXP-PBC096). PCR mix consisted of 22 μL of H₂O, 25 μL of NZYTaq II 2x Green Master Mix, 1 μL of the specific barcode and 2 μL of the purified DNA. The following conditions were used for PCR: initial denaturation (95°C; 3 min); amplification (15 cycles) comprising denaturation (95°C; 15 s), annealing (62°C; 15 s) and extension (72°C; 90 s); final extension (72°C; 5 min). Amplicons were purified with the NucleoMag kit and quantified with the Qubit x1 dsDNA High-Sensitivity Assay kit. Finally, an equimolar pool of amplicons was prepared for library construction.

The Ligation Sequencing Kit (ONT, Oxford, United Kingdom, Cat. No.: SQK-LSK109) was used to prepare the sequencing library. Briefly, the NEBNext FFPE DNA Repair Mix (New England Biolabs, Ipswich, United States, Cat. No.: M6630) was used for DNA repair and end-prep. Then, a purification with the NucleoMag kit was carried out. Finally, adapter ligation and clean-up was performed by following the ONT SQK-LSK109 protocol.

A R9.4.1 MinION flow cell (ONT, Oxford, United Kingdom, Cat. No.: FLO-MIN106D) was primed and loaded as indicated by the manufacturer. Sequencing was performed during ∼6.5 h. Reads were basecalled with MinKNOW software (v. 20.06.5; core v. 4.0.5) using Guppy’s (v. 4.0.9) fast basecalling model, and sequences with Q < 7 (default threshold implemented in MinKNOW) were discarded.

Reads were analyzed with Spaghetti, a custom pipeline for automatic bioinformatic analysis of Nanopore sequencing data and semi-automatic exploratory analysis and data visualization. Briefly, Spaghetti bioinformatic pipeline consists of the following steps:

A detailed explanation of the pipeline and the specific commands that were used can be found on Spaghetti’s GitHub repository.⁴

Spaghetti data visualization and analysis module was mainly based on the phyloseq R package (v. 1.30.0) (McMurdie and Holmes, 2013). For alpha diversity tests, all the samples were rarefied to the lowest library size (50,051 reads/sample) to mitigate uneven sequencing depth. For beta diversity, Principal Coordinates Analysis (PCoA) were created using the Bray-Curtis dissimilarity metric and relative abundances. Heatmaps were produced with ampvis2 (v. 2.6.5) (Andersen et al., 2018). Custom figures were created using ggplot2 (v. 3.3.1). Plotly (v. 4.9.2.1) was used for producing interactive plots. Venn diagrams were obtained using an online tool.⁵

All the analyses were run on a MSI GF63 Thin 9SC-047XES laptop (CPU: Intel Corei7-9750H, 6 core, 12 threads; RAM: 16GB; SSD: 512 Gb; Graphics Card: GeForce GTX 1650).

Upon arrival at the laboratory, the samples were homogenized by mixing 1 g of the sample with 1 mL of sterile Phosphate Buffered Saline (PBS) and serial dilutions up to 10^–7 were performed. Then, 50 μL of the 10^–2 to 10^–7 dilutions were spread on Petri dishes containing either TSA medium (composition in g/L: 15.0 tryptone, 5.0 soya peptone, 5.0 sodium chloride, 15.0 agar) or SSE/HD 1:10 medium (composition detailed on the DSMZ media database, medium number 1,426). In the case of the SSE/HD 1:10 medium, duplicates of each dilution were cultured, with one of the replicates being incubated under uninterrupted artificial light and the other replicate being incubated, together with the TSA plates, under natural light. All plates were incubated in oxygenic conditions and at room temperature.

Individual colonies were selected based on their color and morphology from the TSA and SSE/HD 1:10 plates incubated under natural light after 6, 11, 18, 30 and 35 days of incubation (Supplementary Table 3). These colonies were re-streaked on fresh culture medium to isolate them in pure culture. Most of the isolates were obtained from TSA medium and from the more concentrated dilutions (10^–2 and 10^–4). Regarding the samples cultured on SSE/HD 1:10 under uninterrupted artificial light, these were removed from the artificial light after 4 weeks of incubation as they did not display any microbial growth. A few days after removal, different bacterial colonies started to grow. These colonies were re-streaked on fresh culture medium and isolated in pure culture. All pure strains were cryo-preserved in glycerol (20% glycerol in an over-night culture of the strain) at −80°C for further uses.

A loopful of each isolate, grown on solid medium, was resuspended in 100 μL of sterile Milli-Q water and subjected to a rapid DNA extraction that consisted of three cycles of boiling and freeze-thawing. Then, a PCR was performed to amplify the 16S rRNA gene using the following universal primers: 8F (5′-AGAGTTTGATCCTGGCTCAG-3′) (Edwards et al., 1989) and 1492R (5′-GGTTACCTTGTTACGACTT-3′) (Stackebrandt and Liesack, 1993). The following conditions were used for PCR: initial denaturation (95°C; 5 min); amplification (24 cycles) comprising denaturation (94°C; 15 s), annealing (48°C; 15 s) and extension (72°C; 90 s); final extension (72°C; 5 min).

Amplicons were visualized by electrophoresis in a 1% agarose gel stained with GoldView DNA Safe Stain (UVAT Nerium Scientific, Valencia, Spain) (100 V, 30 min). Amplicons were precipitated overnight at –20°C in a mixture of isopropanol 1:1 (vol:vol) and potassium acetate 1:10 (vol:vol) (3M, pH 5). The next day, DNA was pelleted by centrifugations for 10 min at 12,000 rpm, then washed with 70% ethanol and resuspended in 15 μL of sterile Milli-Q water. Amplicons were tagged using the BigDye^® Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Carlsbad, CA, United States) and sent for Sanger sequencing of the partial 16S rRNA gene at the SCSIE (Serveis Centrals de Suport a la Investigació Experimental) of the University of Valencia (Spain), using the same universal primers as previously mentioned (8F and 1492R).

All resulting sequences were edited with UGENE v.33 (Okonechnikov et al., 2012) to remove low quality base calls, and taxonomic identification was performed using the BLASTn tool and the 16S ribosomal RNA sequences (Bacteria and Archaea) database (NCBI). Finally, clones were dereplicated using the BLASTn tool to compare each partial 16S rRNA sequence to the rest of strains belonging to the collection of microorganisms established in this project. Any strain displaying > 99.9% similarity to another strain in the collection and isolated from the same sample was considered to be a replicate and therefore discarded from the collection. This was performed to avoid an overestimation of the culturable diversity, as bacterial clones of the same species are not relevant for the microbial collection. The comparison between results from Nanopore sequencing and microbial culture collection was based on taxonomic information. Nanopore and Sanger 16S rRNA gene sequences were taxonomically classified independently, as described above. Then, the genus-level profiles were evaluated to find those taxa that had been identified by both approaches.