​​Sediment samples were collected from the upper (oxic) layer of an artificial lake in Northwest Greenland, outside of Thule Airbase (N 76°32.659’ W 68°27.458’). This lake was chosen based on a preliminary survey of multiple locations around Thule Airbase in 2013. For this study, two sample sites within the lake were chosen, designated as Rich Lake 1 (RL1) and Rich Lake 2 (RL2). Sediment from each site was combined for cultivation. Markers (plastic pipes dug into the ground) were placed at both sites to ensure continuous sampling from the same location throughout the season. The sites were 30 feet from each other, about 15 cm from the water edge, with 1–3 mm of water above the sediment. Sediment samples were collected at various time points during the summer of 2014 for cultivation ( Table 1 ). The temperature and pH of the sediment was measured periodically and stayed essentially unchanged throughout the season: 10 °C with a pH of 6.8.

Three cultivation media were used: R2A, 1:100 Nutrient Agar (1:100 NA) and Soil Extract Agar (SE). R2A was made following the manufacturer recommendations (BD, Difco 218263). A 1:100 dilution of Nutrient Agar was made using 0.8 g/L Nutrient Broth (Difco 247940) and Bacto Technical Agar (15 g/L; Difco 281230). Sediment from the lake was mixed with DI water and sterilized at 121°C and 15 PSI for one hour. The solution was allowed to sediment and the supernatant was collected. For SE agar, Bacto Technical Agar (15 g/L) was added to the sediment and autoclaved. The average temperature of the lake throughout the entire sampling campaign was 10°C, however the sediment experiences colder temperatures during other parts of the season (such as 2°C). To simulate the natural conditions of the lake, all cultures were incubated at both 0–2°C and 10°C.

Samples for standard cultivation were collected at three time points between June and July 2014 ( Table 1 ) from both RL1 and RL2. Using a sterile teaspoon, the uppermost 2–3 mm of sediment was transferred to a 50 mL Falcon tube and immediately transported to the laboratory within an hour. RL1 and RL2 samples were combined and vortexed. Serial dilutions through 10^–5 were prepared using phosphate-buffered saline. Each dilution was plated on three different media: R2A, a 1:100 dilution of NA, and SE. To capture facultative anaerobes, a subset of R2A and NA plates were concurrently incubated in anaerobic boxes under 95% nitrogen and 5% carbon dioxide at room temperature.

In situ devices were constructed as described above. Sediment for diffusion chamber inoculum was collected and serial dilutions were prepared in 10mL of 42°C warm agar, and 3 mL was loaded into each diffusion chamber, which was then sealed with a polycarbonate membrane. Traps and FPMTs were filled with sterile agar. Itips were filled with either R2A, a 1:100 dilution of NA, or SE. In situ devices were placed at the sample site 3 times throughout the season ( Table 1 ) and left to incubate for 2–3 weeks ( Supplementary Figures S1, S2 ).

After incubation, all devices (9 DCs, 6 Traps, 1 filter plate, 9 Itips per site) were aseptically disassembled with a sterile blade and agar containing microorganisms was carefully removed using sterile loops. The agar mixture was combined with sterile media, homogenized, and vortexed. The mixture was used as an inoculum for serial dilutions through 10⁻⁵, of which 100 µl of each dilution were spread on three types of solid media: R2A, a 1:100 dilution of nutrient agar, and soil extract agar, and incubated at 0°C or 10°C. Contents of the growth chamber in iPore devices were retrieved with a sterile toothpick and streaked directly on agar plates.

Standard cultivation plates and plates from serial dilutions of in situ devices were incubated at 0°C or 10°C for at least three weeks. Following incubation, plates were individually examined and dilutions resulting in single colonies were selected. Biomass from single colonies was lifted off the plate with a toothpick, restreaked, and incubated on the same media and at the same temperature as the parent plate. To the best of our ability, colonies were picked to encompass as many different representative phenotypes as were distinguishable under a dissecting scope. Denser and less dense plates were also examined under the dissecting scope for additional unique phenotypes, which were isolated as described above.

Sealed petri dishes were transported in an enclosed container with ice packs to Northeastern University, Boston MA, U.S.A. and immediately returned to their original cultivation temperature. Isolates were cultured on either 1% NA, 10% NA, or R2A until determined pure by microscopic visualization. Pure isolates were archived in 20% glycerol at −80°C. Taxonomic identification was performed by sequencing the 16S rRNA gene. Biomass from a colony was picked with a sterile toothpick and homogenized with molecular-grade water for colony PCR. One microliter of homogenate was used as a template for PCR-enabled 16S Sanger Sequencing using the 27F (5’-AGAGTTTGATCCTGGCTCAG-3’) and 1492R (5’-GGTTACCTTGTTAGGACTT-3’) primers (Lane, 1991) and the HotStarTaq system (Qiagen Cat #203445). PCR was performed under the following conditions: 15-minute denaturation at 95°C, followed by 20 cycles of 1 minute at 95°C, 1 minute at 55°C, and 1 minute at 72°C. PCR products were purified and sequenced commercially (at Macrogen or Genewiz) by fluorescent terminator sequencing using the 27F primer. Some isolates were re-sequenced in the case of poor quality with the use of the 1492R primer.

In total 1109 isolates were sequenced (Table 2). The sequences were assessed for quality and manually trimmed. After trimming low-quality bases, the average sequence length was 772 base pairs (standard deviation=129 bp). Sequences were imported into QIIME2 version 2022.8 (Bolyen et al., 2019) and dereplicated using VSEARCH (Rognes et al., 2016). Sequences were clustered at 97% sequence similarity into OTUs, and taxonomy was assigned using full length 16S reference sequences from the SILVA taxonomy database release 138 (Quast et al., 2013). A phylogenetic tree was generated and rooted using the Mafft and fasttree QIIME2 plug-ins (Katoh et al., 2002; Price et al., 2010). Figures were constructed using phyloseq (McMurdie and Holmes, 2013), ggplot2 (Hadley Wickham, 2016) and ggtree (Yu, 2020; Yu et al., 2017) in R (version 4.2.3) (R Core Team, 2021).

To compare cultivation approaches to each other, isolation power was calculated as the number of OTUs captured by a method, divided by the total number of OTUs observed across the study. Isolation efficiency was calculated by dividing the isolation power by the number of isolates from that method.

We isolated a total of 1109 colonies and sequenced these isolates using Sanger sequencing of the 16S rRNA gene ( Table 1 ). After sequencing, quality control, and removal of non-bacterial sequences we retained 1,093 sequences. These sequences clustered into 155 97% rRNA gene sequence similarity based OTUs. OTUs were classified within four bacterial phyla: Proteobacteria, Actinobacteria, Bacteroidota, and Firmicutes (nomenclature based on SILVA release 138). Proteobacteria was the most abundant phylum in our isolate collection, encompassing 58.6% of cultured isolates. Within the Proteobacteria phylum, we cultivated representatives of Alpha- and Gammaproteobacteria. Our sequenced isolates spanned 77 genera, but two genera dominated, with 19.1% of isolates classified as Flavobacterium and 10.5% as Pseudomonas. We also cultivated a high level of intra-generic diversity of Flavobacterium, isolating 19 unique Flavobacterium OTUs.

We used an array of cultivation approaches—standard (aerobic and anoxic), and in situ—to establish a representative culture collection of the lake sediment. Using aerobic standard cultivation, we cultured 319 isolates which clustered into 92 OTUs. Using anoxic standard cultivation, we cultured 20 isolates and 5 OTUs. Among all in situ approaches, we cultured 770 isolates and 104 OTUs.

Standard, anoxic, and in situ cultivation techniques yielded diverse collections of isolates. Both standard and in situ approaches resulted in cultured isolates spanning 4 different phyla, while anoxic approaches resulted in cultured isolates from 2 phyla. In situ approaches resulted in isolates spanning 59 unique genera across 28 families and 18 orders. Standard approaches resulted in isolates from 48 unique genera across 27 families and 18 orders. Anoxic cultivation resulted in isolates from 4 genera, 3 families, and 2 orders.

When comparing the OTU level composition of our isolate collections, we observed minimal overlap of isolates obtained by the different approaches: only 3 of 155 unique OTUs were common among all three broad methodologies ( Figure 1 ). Using in situ cultivation methods, we successfully cultured 62 OTUs that were not recovered through standard cultivation alone. Conversely, traditional methods yielded 49 OTUs that in situ approaches failed to capture. These differences were also apparent when comparing which orders were successfully cultivated by these 3 broad approaches ( Figure 2 ). In particular, there were 4 bacterial orders for which isolates were only cultivated using in situ approaches, and 4 that were only captured using standard cultivation.

We observed low overlap in the OTUs we cultured using each in situ cultivation method, with only one OTU overlapping between all methodologies ( Figure 3 ). The largest number of unique OTUs were cultivated using the diffusion chamber and iPore devices. When comparing the in situ approaches, we observed a similar pattern as when we compare standard and in situ approaches as a whole, in that some orders were only successfully cultivated by some approaches ( Figure 4 ). In particular, use of the iPore was necessary to culture three orders: Reyranalles, Solirubrobacterales, and Staphylococcales. Use of the diffusion chamber was necessary to culture two orders: Rhodobacterales and Aeromonadales.

We tested whether or not anoxic subculture of the colonies initially grown in the diffusion chambers increased the diversity of our culture collection. The majority of isolates initially cultured in the diffusion chamber were successfully subcultured in aerobic conditions, and 13 OTUs were cultivated using both anoxic and aerobic conditions. However, 4 OTUs required anoxic subculture conditions to be successfully isolated ( Figure 5 ).

Taxonomy played a role in which OTUs were successfully cultivated by each method. In particular, OTUs belonging to Thermoleophilia, Reyranalles, Solirubobacteriales and Staphylococcales were only cultivated using the iPore, while Streptomycetales, Xanthomonadales, Chitinophagales, and Cytophagales were only cultivated using standard aerobic approaches ( Table 3 ). However, all three broad approaches recovered a relatively even spread of OTUs across the phylogenetic tree ( Figure 6 ).

The majority of isolates recovered had a 16S rRNA gene sequence that matched between 97–100% to the closest reference sequence in the SILVA database ( Table 4 ). Using standard aerobic cultivation, we isolated four OTUs with less than 96% sequence identity to their closest match in the SILVA database, with the lowest-identity isolate sharing 93% sequence similarity with a Flavobacterium species. Using standard anoxic cultivation, we recovered one isolate that shared 91.5% sequence identity with its nearest match, which belonged to the family Microbacteriaceae in the phylum Actinobacteriota. The two most novel organisms were isolated using the iPore (89%) and diffusion chamber (89%) methods ( Table 4 ). The iPore isolate, which shared 89% sequence identity, was most closely related to an uncultured organism in the family Solirubrobacteraceae (phylum Actinobacteriota). Similarly, the diffusion chamber isolate with 89% sequence identity was most closely related to a member of the family Oxalobacteraceae. Incubating diffusion chamber contents under anoxic conditions also resulted in one isolate with <96% rRNA gene sequence identity to its closest SILVA match, specifically showing 95% similarity to a Mucilaginibacter species. Additionally, one isolate from the filter plate method was identified as relatively novel, sharing 95.8% rRNA gene sequence identity with a member of the family Acetobacteraceae.

To analyze how methods compared to each other, we calculated isolation power to determine how many of the total available OTUs each method could capture and divided by the total number of isolates to get an isolation efficiency (Table 5). Standard cultivation had the largest isolation power (capturing 85% of all OTUs found in study), while the diffusion chamber had the second largest isolation power. All other isolation methods captured <30% of the OTUs detected in the study. However, when considering how this translates into diversity (isolation efficiency) the iPore had the highest isolation efficiency—even though it only captured 29% of the OTUs, it was more efficient than any other method.