Four cleanrooms at MSFC were selected for sampling together with two uncontrolled lab spaces. One cleanroom is not maintained and certified as a cleanroom but still has controlled access. In all cases, air samples were collected into 10 mL sterile water using a Coriolis μ air sampler (Bertin, Montigny-le-Bretonneux, France). Subsequently, field control samples were collected by waving a swab through the air for approximately 5 s. Surface samples of small area (5″ by 3″) using either dry or sterile water-wetted swabs (806WC, Puritan) or large area (27″ by 27″) were collected using 9″ by 9″ sterile water-wetted wipes (TX3211; TexWipe, Kernersville, NC).

Air samples (300 L/min) were collected for 5 min into 10–15 mL sterile water. After completion, collection cones were capped and stored up to 7 days at 4°C until processed. Swabs were collected and then stored dry in sterile tubes at 4°C for up to 7 days until processed. Wipes were dampened in 10 mL sterile water within a sterile petri dish and following collection, stored in sterile glass jars at 4°C up to 27 days until processed.

Tryptic Soy Agar (TSA) (BD Biosciences, Franklin Lakes, NJ) and Tryptic Soy Broth (TSB) (BD Biosciences, Franklin Lakes, NJ) were used as media for cultivation of the bacteria. All swabs were streaked directly onto the surface of a TSA plate and then incubated at 32°C for at least 2 days. Wipes were submerged in 35 mL of sterile water and then sonicated. A volume of 250 μL was then pipetted onto TSA, spread using a sterile spreader and incubated at 32°C for at least 2 days. Last, 100 μL of liquid collected via Coriolis air sampler was spread on TSA using sterile spreaders and incubated at 32°C for at least 2 days. In most cases, if overgrowth was observed, or too numerous colonies to count (TNTC), dilutions were made of the liquid into sterile water and replated until countable colonies were observed.

Strains of Bacillus atrophaeus ATCC 9372 and Deinococcus radiodurans ATCC 13939 purchased from American Type Culture Collection (ATCC) were used as positive controls. Sterile water (n = 3) and DNA purification kit blanks (n = 3) were employed as negative controls.

For all plates, representative colonies for all morphologies observed were selected and given a unique identifier. Selected colonies were all streaked to confirm purity on TSA and incubated at either 25°C or 32°C depending on whether the colony was classified as fungi or bacteria, respectively, until sufficient growth and isolated colonies were observed for subsequent experiments. If multiple colony morphologies were detected, single colonies were selected and streaked again. If molds were observed and single colonies could not be obtained, the isolates were sub-cultured until only a single morphology was detected.

For the physical curation, when possible, strains were subcultured from one colony using a sterile loop into 1.5–3 mL TSB and grown stationary and incubated at either 25°C or 32°C. When turbid growth was observed, the strains were frozen with a 1:1 volume of 25% sterile glycerol in cryotubes and stored at −80°C. If an organism was unable to grow in liquid culture, colonies from the TSA plates were scraped with a sterile loop into a 1:1 mixture of TSB:25% glycerol and frozen. All frozen cultures were then re-streaked on TSA with a sterile loop and grown at 25°C or 32°C until growth was seen. If no growth, a mixed culture, or a culture with unmatched description was observed, the culture was discarded, and renewed attempts were made to obtain a pure culture from the original plates. It was only after a stock culture was frozen and the growth was verified that the original plates were discarded. All strains were classified as either suspected fungi or suspected bacteria based on colony morphology.

DNA extraction was conducted using a Quick-DNA Fungal/Bacterial Miniprep Kit (Zymo Research, D6005, Irvine, CA). To collect pellets, pure cultures were scraped from TSA plates using sterile loops and deposited into a weighed microcentrifuge tube. The tubes were centrifuged at maximum speed for 1 min in an accuSpin Micro 17 Microcentrifuge (13-100-675; Thermo Fisher Scientific, Waltham, MA). The microcentrifuge tubes were weighed, and pellet weights were recorded. Pellet wet weights of 50–100 mg were targeted and multiple plates were used if necessary to get sufficient biomass. Pellets were stored in a −20°C freezer until ready for processing. For DNA extraction, pellets were thawed in a biosafety level 2 (BSL2) hood and the manufacturer’s instructions were followed for isolating genomic DNA. DNA concentration was measured using a Thermo Scientific Nanodrop Lite (Waltham, MA).

Microbial taxonomic characterization was performed using capillary electrophoresis (i.e., Sanger) sequencing of small subunit (SSU; 16S or 18S) ribosomal RNA (rRNA) gene amplicons generated from PCR amplification of genomic DNA using bacterial or fungal primer sets. Off-the-shelf primers for bacterial 16S rRNA gene amplification (1,500 bp amplicon) were purchased from Integrated DNA Technologies (IDT). Fungal primers ITS5F and ITS4R, custom synthesized by IDT (Coralville, IA), were used to amplify internal transcribed spacer (ITS) regions (500 bp amplicon) of suspected fungal isolates (Schoch et al., 2012).

PCR was conducted using an Applied Biosystems SimpliAmp Thermalcycler (A41192; Thermo Fisher Scientific, Waltham, MA). Detailed information can be found in the Supplementary methods.

Amplicons were sequenced by Azenta Life Sciences-Genewiz for Sanger sequencing. FASTQ sequences were downloaded, individually assessed and trimmed of unresolved bases (on average the first 20 to 40 bases and the last 20 to 150 bases depending on the length and quality of the sequence), and annotated using the rRNA/ITS database using the Targeted Loci Project Information from either the 16S ribosomal RNA sequence (Bacteria and Archaea) or ITS from Fungi and type reference material in the National Center for Biotechnology Information (NCBI) Basic Local Alignment Search Tool [BLAST; e.g., (Altschul et al., 1990)]. Genus-level annotation was only assigned when the closest taxonomically annotated reference sequence was at least 97% similar, and species-level annotation was only assigned when reference sequences were >99% similar, and there was a 0.2% difference between the top IDs (Supplementary Table 1).

Media used for culturing strains was TSA, Nutrient Agar with 1% glucose (BD Biosciences, Franklin Lakes, NJ), Lennox Broth (LB) agar, and TSB. Frozen stock cultures were scraped with a sterile loop and streaked across the surface of a TSA plate, a LB agar plate, or a nutrient agar plate, and then incubated at 32°C for at least 2 days. Once adequate growth was observed, cells from a single colony were collected with a sterile loop and inoculated into 3 mL of TSB. Cultures were grown either stationary or shaking at 250 rpm at 32°C for at least 2 days or until adequate turbidity was observed. Optical density was measured at 600 nm wavelength using a Genesys 50 UV/VIS spectrophotometer (Thermo Fisher Scientific, Waltham, MA) immediately following vigorous vortexing. Cultures were centrifuged and resuspended in sterile water to an OD₆₀₀ of 0.1 (or approximately 1 × 10⁶ cells/mL). Positive controls B. atrophaeus ATCC 9372 and D. radiodurans ATCC 13939 were included as needed.

Kapton-HN film was cut into 1-in² coupons and sterilized using a 3870ELP Heidolph Tuttnauer benchtop sterilizer. While each run was performed with independent cultures prepared specially for that experiment, the data from each run informed the next, meaning that the runs were all conducted differently and are outlined individually below and as described in Supplementary Table 2. Diluted cultures were mixed with a vortex and the specified volume was applied as a single droplet to the surface of a labeled Kapton coupon. All work was performed in a biosafety cabinet. Samples were allowed to dry for up to 3 days. Samples were then transferred to the radiation test facility for exposure. In all cases, media controls without bacteria were similarly exposed. Ambient coupons devoid of microbes but spotted with sterile water (negative control) were transferred between buildings along with the experimental coupons but were unexposed to vacuum or radiation. For the final run, a second set of controls, called vacuum controls, were also subjected to the ~1E-6 Torr vacuum, like the test specimens.

In a series of experiments, microbial cultures were deposited on sterile Kapton coupons and subjected to proton and ultraviolet (UV) radiation to assess their resilience under extreme conditions. The coupons were prepared in a biosafety cabinet, where microbial samples were dried before exposure. Proton radiation experiments were conducted at the Combined Environmental Effects Facility (CEEF) at MSFC. It is difficult to measure the dose of radiation received by such small targets (microbes), and pathfinder studies demonstrated the usefulness of relying on fluence for dosing (Cengel et al., 2010). Therefore, samples were exposed to 100 keV protons at fluences ranging from 2 × 10¹⁵ to 4 × 10¹⁵ p+/cm² under high vacuum (~1E-6 Torr). Each run featured minor procedural variations. In the first run, 250 μL of microbial culture was applied, but all samples being irradiated, leaving no controls. Subsequent runs utilized 200 μL to ensure better drying and included ambient unexposed controls. The second, third, and fourth runs proceeded with increased fluences and adjusted exposure times, ranging from 10.7 to 36.5 h, under vacuum durations of 46 to 145 h (Supplementary Table 2).

Additionally, a UV radiation experiment exposed samples to 254 nm light at an intensity of 80 W/m² at a distance of ~18 cm for 5 or 10 min using a Spectroline XL-1500 UV Crosslinker. These samples were processed on the same day to minimize contamination. Instances of contamination were seen only in the first run, and improved sample handling was employed in subsequent runs, improving outcomes.

Following radiation experiments, samples were transferred to a BSL2 lab for further processing. In all cases, sample processing included the transfer of a single Kapton coupon into a sterile 50 mL conical tube. A volume of 5 mL of sterile water was added, and the sample was vortexed vigorously to release cells from the coupon surface. In the case of the first and second runs, 250 μL of eluate was pipetted on appropriate solid media and spread across the surface using a sterile L-shaped spreader. Plates were incubated at 32°C for 2 days. For the last two runs, spot plating was performed with 100 μL. All negative control samples were treated the same way as the experimental samples. An inoculum control was not included for the first run. For the inoculum growth control in the second run, 100 μL was spread plated to verify growth. For the remaining runs, 100 μL of inoculum were pipetted as spots on the agar surface and incubated at 32°C until sufficient growth was observed. Due to the high number of plates, quantitative data was not obtained, rather an arbitrary qualitative system was employed to describe the growth as growth (+), minor growth (minor), no growth (−), or contamination. Growth is defined as colonies observed in three replicate samples per microbe. Minor growth is defined as at least one colony observed in at least one replicate sample. No growth was defined as the lack of observable growth in any sample. Contamination was indicated if microbial colonies differing in appearance from those of the selected isolates were identified. Control coupons were processed identically. Finally, inocula were plated to ensure that adequate starting culture was present and to compare morphology and determine any contamination.

WGS of strains PPS68, PPS72, PPS117, and PPS120 were carried out as previously described (Popovich et al., 2021). DNA quantity and quality were measured using fluorimetry (Qubit 4; ThermoFisher) and electrophoresis (TapeStation4150; Agilent). Library preparation was performed using a Nextera XT kit (#FC-131-1024; Illumina Inc. San Diego, CA) according to the manufacturer’s instructions with 1 ng template input and 12 cycles of indexing PCR. An equal-volume pool of all libraries was then prepared, and deep sequencing was performed using an Illumina NovaSeq X instrument targeting 5 million clusters per sample (paired-end, 2 × 150 base reads). Draft genomes were assembled using the software package CLC Genomics Workbench (v22; Qiagen). Library preparation was performed at the Genomics and Microbiome Core Facility (GMCF) at Rush University, IL, USA, and deep sequencing was performed at the DNA Services Facility at the University of Illinois at Urbana-Champaign.

Bacterial genomes were de novo assembled using the CLC Genomics Workbench software (v24) using shotgun sequence data generated on an Illumina sequencer. Default settings were applied for the assembly process, except for the minimum contig length, which was set to 200 base pairs to ensure the inclusion of smaller but potentially important contigs. Using BLAST analysis (i.e., BLASTn-v2.12.0+) against the NCBI’s 16S ribosomal RNA sequences database, 16S rRNA genes were extracted from assembled genome sequences. In addition, the assembled genomes were annotated by Rapid Annotation using Subsystem Technology (RAST) as described previously (Chander et al., 2017). Data for each annotation was exported from the RAST server and further processed through Pan-ResistomeFinder. The “role” column in the RAST exported annotation data shows gene/protein while other columns assign these genes into larger groups based on their function. There may be multiple entries of a single gene/protein in the column “role” based on their multiple roles. To account for the functional diversity of each gene, uniqueness of genes in this column was ignored.

Strains with experimentally validated D₁₀ values of 2 kGy or higher (for γ-irradiation or equivalent UV-C fluence) were designated as “resistant,” while those below this threshold served as “sensitive” controls (Table 1). The final panel comprises eleven resistant strains Deinococcus radiodurans R1 (Daly, 2009), Deinococcus radiomollis (Callegan et al., 2008), Kocuria rhizophila PT10 (Guesmi et al., 2021b), Kocuria rosea DSM 20447 (Asgarani et al., 2012), Kineococcus radiotolerans SRS30216 (Bagwell et al., 2008), Promicromonospora panici PT9 (Guesmi et al., 2021a), Rubrobacter xylanophilus DSM 9941 (Ferreira et al., 1999), Rubrobacter radiotolerans DSM 5868 (Ferreira et al., 1999), Rubrobacter radiotolerans RSPS-4 (Ferreira et al., 1999), Geodermatophilus obscurus DSM 43160, and Geodermatophilus dictyosporus DSM 44208 (Montero-Calasanz et al., 2015). The sensitive group includes Escherichia coli O157, E. coli K-12 MG1655, and Arthrobacter sp. PAMC 25486 (KOPRI Repository, n.d.). These additional set of strains provide a robust two-class framework for downstream comparative analysis of gene presence patterns related to radiation resistance.

NCBI taxonomy database confirms that, apart from the genus Deinococcus (order Deinococcales, phylum Deinococcota), all other reference strains in the radiation phenotype panel belong to the phylum Actinobacteriota. The resistant group includes Kocuria rhizophila PT10 and Kocuria rosea DSM 20447 from the order Micrococcales, Kineococcus radiotolerans SRS30216 from Kineosporiales, Promicromonospora panici PT-9 from Propionibacteriales, Geodermatophilus obscurus DSM 43160 and Geodermatophilus dictyosporus DSM 44208 from Geodermatophilales, and Rubrobacter radiotolerans DSM 5868, Rubrobacter radiotolerans RSPS-4, and Rubrobacter xylanophilus DSM 9941 from Rubrobacterales. Among the sensitive strains, Arthrobacter sp. PAMC 25486 (order Micrococcales) also belongs to Actinobacteriota, while the two Escherichia coli strains, O157 and K-12 MG1655, are members of the phylum Proteobacteria (order Enterobacterales). This updated composition emphasizes broad phylogenetic coverage while confirming that the majority of strains in the panel, particularly the resistant ones, are taxonomically affiliated with Actinobacteriota.

For extraction and comparison of stress and resistance-associated genes from multiple microbial genomes, we used Pan-ResistomeFinder (https://github.com/atulchander/Pan-ResistomeFinder). This tool supports automated processing of genomic annotation files in .tsv, .csv, .xls, or .xlsx formats, each corresponding to a different strain. It ensures the presence of core functional descriptors (Category, Subcategory, Subsystem, and Role) and constructs a unified gene presence–absence matrix by merging input files on these shared descriptors. Genes were marked with a binary indicator (1 for presence) across strains, and missing annotations were imputed as 0. The merged matrix captures the distribution of functional genes across the strains. To focus specifically on stress and resistance-related functions, Pan-ResistomeFinder performs a keyword-guided resistome search. This produces a filtered output containing genes relevant to resilience, which is basis for subsequent resistomics. To characterize strain-specific resistome overlap, we implemented a focused extraction protocol using Pan-ResistomeFinder. First, the resistome output matrix filtered for stress and resistance-related genes was queried to identify total resistome in the dataset. In this total resistome dataset, a subset of data was created showing presence of genes in A. koreensis PPS68 and simultaneously found in at least one of the confirmed resistant strains. This subset was saved as the “shared resistome” dataset, representing genes potentially conserved in resistance-competent Actinobacteria. From this shared subset, we further excluded any genes that were also present in any known sensitive strains (E. coli O157, E. coli K-12 MG1655, and Arthrobacter sp. PAMC 25486). The resulting exclusive set, comprising genes unique to strain PPS68 and resistant strains only, was saved separately to capture PPS68-associated resistome signatures that may be selectively retained in radiation-adapted taxa.

We used Genome Taxonomy Database (GTDB)-Tk v2.4.0 to compute the Average Nucleotide Identity (ANI) between our query genomes and the reference genomes from the GTDB release r220. This analysis aimed to identify the closest representative genome for each query genome based on ANI. Four query genomes in .fna format were analyzed against the comprehensive GTDB reference dataset. The analysis was conducted using the ani_rep command in GTDB-Tk, with the genome directory, output directory, file extension (.fna). The results included two main output files: gtdbtk.ani_summary.tsv, which provided ANI values for all comparisons, and gtdbtk.ani_closest.tsv, which lists the closest representative genome for each query genome.

Further analysis was performed using the FastANI software with default parameters to calculate the Average Nucleotide Identity (ANI) for the four query genomes: strain PPS68, strain PPS72, strain PPS117, and strain PPS120. Output included ANI values, number of fragments mapped, and total fragments analyzed. FastANI analysis was performed by the Rush Research Bioinformatics Core (RRBC) at Rush University. These genomic comparisons were performed between the assembled draft genomes and reference genomes downloaded from the NCBI database. All available NCBI genomes from genera of interest, including GenBank (GCA) and RefSeq (GCF), were downloaded for comparison. Genus-level taxonomic identification was initially confirmed using 16S ribosomal RNA gene BLAST analysis (Altschul et al., 1997; Jain et al., 2018).

The 16S rRNA analysis identified the closest taxonomic affiliations of the query strains based on sequence similarity to known type strains. Strain PPS117 showed 99.03% similarity to Mycetocola zhadangensis ZD1-4. Strain PPS120 exhibited 98.56% similarity to Erwinia tasmaniensis Et/199. Strain PPS68 shared 99.47% similarity with Arthrobacter luteolus CF-25. Strain PPS72 exhibited 99.93% similarity to Paenarthrobacter nitroguajacolicus JCM 14115.

The GTDB-Tk ani_rep command was used to compare query genomes against a comprehensive reference dataset of 113,104 bacterial genomes, providing detailed assessments of genomic similarity through Average Nucleotide Identity (ANI) and alignment fractions. For strain PPS68, high ANI matches were observed and identified as A. koreensis (GCF_009193255.1) since ANI was 98.7% and 0.9317 alignment fraction. Strain PPS72 shows highest matches (88.65 and 88.26%) with Paenarthrobacter nitroguajacolicus (GCF_001375615.1) and Arthrobacter sp. (GCF_000281065.1). Strain PPS120 showed a closest match to Erwinia sp. (GCF_900068895.1) with an ANI of 94.1% and an alignment fraction of 0.8363, though it did not meet the threshold for a close match. For strain PPS117, ANI analysis found no significant matches, despite a 16S rRNA similarity of 99.0% with M. zhadangensis ZD1-4, indicating the potential for a novel species. However, according to FastANI, the closest match for the strain PPS117 was with M. zhadangensis (82.9%, GCF_014636455.1) (Table 4; Supplementary Table 3).

The functional analysis identified 1,990 genes, observed across four bacterial species, which were classified into various function-based categories according to RAST (Figure 1A; Supplementary Table 4). In this dataset, 522 roles of genes were found to be conserved across four bacterial species, constituting ~26% of the total dataset. These conserved genes form part of the core genome shared by four species analyzed (Figure 1B; Supplementary Table 5). The conserved genes have functional annotations predictive of their role in regulating essential pathways like, carbohydrate metabolism, amino acid biosynthesis, protein processing, and cofactor and vitamin metabolism. Moreover, these genes are also related to the processes, nucleotide synthesis, DNA and RNA metabolism, and respiration which are consistently maintained, underscoring the fundamental requirements for cellular maintenance and energy generation. Additionally, several genes linked to membrane transport, lipid biosynthesis, and stress defense mechanisms were conserved, reflecting their important role in maintaining structural integrity and adaptive potential under dynamic environmental conditions.

Despite the conservation of core genes, variations in specific metabolic pathways were observed. Notably, carbohydrate metabolism genes showed significant diversity in these 4 organisms (Supplementary Table 4). Erwinia sp. PPS120 exhibited a broader set of genes associated with sugar utilization, indicating metabolic flexibility and the ability to exploit diverse carbon sources. Conversely, A. koreensis sp. PPS68 and Paenarthrobacter sp. PPS72 had more specialized genes in some pathways (Supplementary Table 4), which may reflect adaptations to specific environments with limited substrate diversity.

Around 90 genes were directly reported in the stress response category in RAST annotations, with an additional 463 genes identified as possible stress-response related. Out of these 553 functional entries, 21% were conserved across all four species and were mainly associated with stress response, membrane permeases, transporters, efflux pumps, resistance genes, antimicrobial genes, heat shock proteins and DNA repair genes.

All species carry genes related to oxidative stress responses and detoxification, that might help them enhance survival in challenging environments. The presence of glutathione-dependent detoxification pathways in Erwinia sp. PPS120, for instance, may highlight its adaptability to oxidative environments. Compared to the other 3 cleanroom strains, A. koreensis PPS68 also displayed a robust set of stress response genes, including those for oxidative damage management and membrane stability, which may be contributing to its resilience to desiccation, radiation, and extreme temperatures. To cross-validate the genes that might be contributing to the radiation resistance in A. koreensis PPS68, we conducted further comparisons of these cleanroom strains with literature searched radioresistant and radiosensitive strains predominantly belonging to the phylum Actinobacteria.

Building on the above comparative overview of stress-response related functional annotations, a resistomics analysis was conducted by selecting literature searched 11 radioresistant strains of phylum Actinobacteria and 3 probable radiosensitive strains which includes a strain PAMC 25486 of Arthrobacter sp. (Figure 2; Supplementary Table 4). Further, by using Pan-ResistomeFinder, we have identified 220 PPS68 roles present in at least one additional radio-resistant strain (Figure 3). These functional roles include multiple oxidoreductases, membrane-associated efflux pumps, and DNA-repair enzymes, underscoring a conserved molecular machinery that may underpin the high radiation and desiccation tolerance shared across the resistant panel. Conversely, only 26 PPS68 roles/genes remained after all genes found in the sensitive controls were subtracted; this exclusive subset pinpoints candidate determinants that are retained by resistant strains yet absent from radiation-sensitive taxa, and therefore may contribute most directly to the extreme-tolerance phenotype observed in A. koreensis PPS68 (Supplementary Table 6). This analysis supports the prediction that PPS68 possesses an expanded DNA repair capacity, characterized by an augmented base-excision repair (BER) module operating in tandem with non-homologous end-joining (NHEJ) ligases (LigC/LigD) and accessory polymerases. Together, these components constitute a multi-layered strategy for resolving radiation-induced DNA damages. Concurrently, osmoprotection and redox balance in PPS68 appear to be regulated by dual glycine-betaine uptake and synthesis systems, thiamin-precursor importers, and folate-cycling enzymes Supplementary Table 7. In parallel, detoxification and adaptive signaling may be supported by multidrug efflux pumps, a c-di-GMP-modulating diguanylate cyclase/phosphodiesterase, and nitrite/phosphate regulatory elements (Supplementary Table 6).