The study was conducted at the BR2 nuclear reactor at the Belgian Nuclear Research Centre (SCK CEN) in Mol, Belgium. The sampling environment consisted of an open basin surrounding the reactor vessel (Supplementary Figure 1). The reactor successively goes through cycles of operation (e.g., production of radioisotopes for medical and industrial use) and shutdown of approximately 30 days each, which are associated with changes in physico-chemical parameters of the water such as flow rate, temperature, and radioactivity. For this study, two sampling campaigns were performed: in the first one, the basin water was studied for three consecutive cycles interspersed with shutdown periods over a period of 8 months (from September 06, 2016 to April 28, 2017). The second one was conducted 1 year later (from April 16, 2018 to June 11, 2018) where it was only monitored for one cycle. In this case, the cell density was also assessed for each sample.

The study of the reactor basin environment represented a challenge in terms of its low bacterial load combined with the presence of radioactivity. For sample collection, a filtration system was therefore designed to minimize the risk of radioactivity exposure and contamination during sampling in accordance with the ALARA principle (As Low As Reasonably Achievable) and implemented in a sampling glove box, routinely used to take samples from the basin and other watery environments (e.g., the primary and secondary cooling circuits). This also ensured that enough cell material could be collected for downstream analyses. Of note, filtration was chosen as a preferred method over centrifugation, as this resulted in a more effective and safer sample collection.

The system was composed of a stainless steel filter holder containing a 0,2 μm pore sized polyethersulfone (Supor^®) filter membrane (Pall Corporation, Port Washington, NY, United States), connected to the corresponding tap on one side and a flow meter on the other side in order to measure the filtered volume (Supplementary Figure 2). Ten liters of water were filtered per sample, after which the filter membrane was retrieved from the system and stored at −20°C in a controlled area (for a period of 1–9 months) before DNA extraction to allow for radioactive decay.

DNA was extracted from the filter membranes following a previously described protocol (Vilchez-Vargas et al., 2013) and subsequently purified using Amicon^® 30 kDa filter cartridges (Merck, Darmstadt, Germany). DNA concentrations were measured using a Quantus^TM fluorometer (Promega, Madison, WI, United States). For the 16S rRNA amplicon sequencing, the hypervariable V3-V4 region was chosen to specifically target bacteria (as opposed to fungi and/or Archaea) and the resulting amplicons were sequenced in two runs using V3 chemistry (2 × 300 bp) with the Illumina MiSeq platform (sequencing performed by Eurofins Genomics, Ebersberg, Germany). Positive (Cupriavidus metallidurans CH34 cultures filtered over the same filter membranes as previously mentioned) and negative controls (blank filter membranes and demineralized water) were also added as part of the experiment. They underwent the same DNA extraction procedure as the other samples.

The data generated by the sequencing platform were demultiplexed and the datasets consisted of two separate fastq files (forward and reversed). The sequences were subsequently analyzed through the OCToPUS pipeline as described in Mysara et al. (2017): In short, reads went successively through pre-assembly denoising (Bankevich et al., 2012), contig assembly, quality filtering using mothur (Schloss et al., 2009), denoising using IPED (Mysara et al., 2016), chimera removal using CATCh (Mysara et al., 2015) and finally Operational Taxonomic Unit (OTU) clustering with 97% cut-off using UPARSE (Edgar, 2013). All resulting OTUs were taken into account for further analyses without applying a minimum abundance threshold.

The OTUs were then taxonomically assigned using the mothur classify.seqs command with the RDP dataset (version 16) as reference. In order to properly compare the samples from both campaigns combined, they were rarefied to the same read count as the smallest sample (11,683 reads), using the mothur sub.sample command. Although a large number of reads is typically removed during this process, this does not significantly affect the subsequent alpha and beta diversity analyses. Rarefaction curves were obtained by using the mothur rarefaction.single command and subsequently plotting the data in the R software. Sample metadata are included in Supplementary Table 1.

Alpha (Inversed Simpson index) and beta [Theta-YC distance calculation followed by non-metric multi-dimensional scaling (NMDS)] diversity analyses were also performed in mothur using the summary.single, dist.shared, and nmds commands. In order to correlate the OTUs and metadata with the NMDS dimensions, a Spearman rank correlation analysis was performed using the mothur corr.axes command. OTUs and metadata that were significantly correlated (r > | 0.6|, p < 0.05) with the represented NMDS axes are displayed on the NMDS plots. It must be noted that the structure of our data did not allow for the use of mixed models.

During the second sampling campaign, a cell counting method was used to study the population dynamics of the basin water across the two reactor operation conditions (cycle vs. shutdown). The method consisted of heterotrophic plate count, where 100 μl of basin water was directly spread onto R2A agar (Reasoner and Geldreich, 1985) plates in triplicates. The plates were incubated for 2 weeks at room temperature on a laboratory bench before colony counting. Flow cytometry was also considered as a cell counting method, but proved to be unreliable as the cell numbers during cycles dropped significantly, thereby making it impossible to distinguish between bacterial cells and instrument background. This meant that relative OTU abundances could not be corrected for absolute cell numbers.

The physical parameters of the BR2 waters are constantly monitored using on-line measuring instruments. Relevant data for the bacterial community (conductivity, temperature, radiation, and flow rate) were retrieved from an in-house software (BIDASSE) database and organized according to the sampling time points.

Water samples from both cycles and shutdown periods from campaign 1 were collected to perform a chemical analysis of ecologically relevant compounds such as nitrate, phosphate, and carbon. The ions (nitrite, nitrate, phosphate, and acetate) were quantified through ion chromatography using a 930 Compact IC Flex (Metrohm, Herisau, Switzerland) with a Metrosep A Supp 16-250/4.0 column and a Metrosep A Supp 16 guard/4.0 column, using an eluent solution of 7.5 mM Na₂CO₃ and 0.75 mM NaOH. TIC (total inorganic carbon) and TOC (total organic carbon) were measured using a FormacsHT-I TOC analyzer (Skalar Analytical, Breda, Netherlands). O₂, H₂, and ROS were not measured in our system.

To assess the influence of temperature, another critical parameter, on the community, an experiment was designed to test its effect on cell number. Briefly, basin water was collected from the BR2 reactor during a shutdown period in an autoclaved glass container and incubated at 35°C (approximately the same temperature as observed during a cycle) for a period of 24 h. As a control, the same volume of water was kept at room temperature (21 ± 2°C) next to the incubator. After 24 h, a fresh sample was collected from the BR2 reactor (also at room temperature or 21 ± 2°C) in order to be compared with the control. Cell numbers for all three conditions were measured using a flow cytometry procedure: 100 ml of basin water collected during shutdown was centrifuged at 17 000 × g for 30 min, after which the supernatant was discarded and the (invisible) cell pellet re-suspended in 1 ml of bottled Evian^® water filtered over a 0.2 μm Acrodisc syringe filter (Pall corporation). The solution was then stained with SYBR^® green (Life Technologies, Carlsbad, CA, United States), incubated for 20 min at 37°C and run on an Accuri^TM C6 flow cytometer (BD Biosciences, San Jose, CA, United States) in four replicates of 200 μl. Cells were manually gated on green fluorescence (FL1) vs. size (FSC) plots based on a negative control (0.2 μm filtered Evian^® water).

During this study, four relevant physical parameters were monitored in order to potentially correlate those with the microbial community. The first parameter was the flow rate of the basin water cooling circuit (Figure 1A). During shutdown periods, the standard flow rate was 50 m³ h^–1 while during operation it increased to 500 m³ h^–1 to cool down the basin water. A second parameter was the global gamma radioactivity in the basin water (Figure 1B). As expected, the activity during operation significantly increased (up to a maximum of 9,500 cps) in comparison to shutdown periods (with a minimum of 3,500 cps) because of the nuclear fission reactions taking place in the reactor core. When the reactor is running, the radiation is deposited in the pool at a calculated dose rate of 755 Gy h^–1, 98% of which is deposited in a zone within 5 cm around the core and 100 cm above and below the central plane. During shutdown periods, the dose rate lowers to 1 Gy h^–1.

The third parameter was the temperature of the basin water (Figure 1C), which followed the same trend as observed for the other parameters: a significant increase during cycles (up to a maximum of 35°C), followed by a decrease to previous levels during shutdowns (with a minimum of 15°C). Lastly, the conductivity of the basin water also showed some variation across the timeline, although to a much smaller extent than the other parameters, with values ranging from 0.96 to 1.25 μS cm^–1 (Figure 1D). As a side note, the flow rate in the purification circuit where deionization happens through mixed bed ion exchange resins remained constant throughout cycles and shutdown periods with a value of 30 m³ h^–1.

The results of the chemical analysis performed on a selection of cycle and shutdown samples from campaign 1 can be found in Table 1. The concentrations of acetate and nitrate were extremely low, while the values for nitrite, phosphate, formate, TIC, and TOC were below the detection limit of the measuring instruments (10 μg l^–1 for ion chromatography and 500 – 700 μg l^–1 for TIC-TOC analysis, respectively). Furthermore, no significant difference between cycle and shutdown values could be detected. Other elements such as Zn, Pb, Co, Ni, Be, Fe, Mg, Cu, Al, Ba, Li, and F were also measured, but values were all below 20 μg l^–1. pH was maintained at an average value of 6 ± 0.4. Finally, the total gamma activity across campaign 1 and 2 amounted to 1.9 × 10³ Bq l^–1 on average during cycles and <80 Bq l^–1 during shutdowns. 95% of the activity during cycles originates from Na-24, a radioisotope with a 15-h half-life. The remainder comes from other radioisotopes such as Cr-51, Sb-124, Re-186, Np-239, W-188, and Co-60.

In order to explore the effect of the changing physical conditions on the cell number of the bacterial population, heterotrophic plate count was used during the second sampling campaign (Figure 2). The data for this method follow a clear pattern: a steep decline in cell number of 1.2 log₁₀ CFU ml^–1 at the start of the cycle, followed by a stabilization and slow increase of the population by 0.5 log₁₀ CFU ml^–1 during the cycle and finally a recovery to previous cell density levels (up to 3.3 log₁₀ CFU ml^–1) after the cycle.

The observed pattern of cell number can be correlated with the change in physico-chemical parameters described previously. During a cycle, the flow rate of the cooling circuit for the basin water goes up from 50 to 500 m³ h^–1. After passing through this circuit, the water is transferred back into the basin via a conduit following the reactor wall. At this point, the water is exposed to high levels of radiation originating from the nuclear fission taking place in the reactor core (gamma heating of 0.83 W g^–1 = 830 Gy s^–1 for a representative cycle of operation). Due to the high flow rate of the cooling circuit during a cycle, the water is frequently exposed to this radiation, with a turnover time of ca. 104 min.

The summary of the data analysis for the basin water samples given in Table 2 shows that the average number of OTUs for both campaigns did not significantly differ across cycles and shutdowns. Rarefaction curves for all samples are shown in Supplementary Figure 3.

Figure 3 shows the 10 most abundant OTUs in the basin water, representing 90 and 76% of the total reads for campaign 1 and 2, respectively. A clear shift in the community profile across the different cycle and shutdown periods is apparent. During the first sampling campaign, the most abundant OTU in the first two cycles was assigned to an unclassified Gammaproteobacterium (OTU5), accounting for 28% of total amplicon reads. For the third cycle, an OTU assigned to Pelomonas (OTU2) became predominant at 43%, although it could already be observed in lower proportions, at 5%, during the previous cycles. During the shutdown periods in between the cycles, the community profile shifted drastically and became dominated by an OTU assigned to Methylobacterium (OTU3) at 41%, while the previously mentioned OTUs (OTU5 and OTU2) were detected in much lower proportions (3 and 7%, respectively). This change can probably be assigned to the shift in physico-chemical parameters that occurs with the transition from a cycle to a shutdown period.

Although the second sampling campaign took place 1 year after the first one, the same pattern could be observed, indicating that the system is rather stable. Nevertheless, a difference can be noted for the unclassified Gammaproteobacterium and Pelomonas observed in approximately equal proportions (13 and 10%, respectively), whereas the community was either dominated by one or the other during the first sampling campaign.

As can be observed in Figure 4, the cell number significantly increased when exposed to a higher temperature, indicating that temperature has an opposite effect to radiation. Since the overall cell number decreases during a cycle, this shows that the effect of radiation exposure largely outperforms the effect of temperature.

The alpha diversity showed a peak in diversity at the start of each cycle, followed by a subsequent downward trend (Supplementary Figure 4). This was true for both sampling campaigns, which further confirmed the robustness of the system under study.

For the beta diversity analysis, an NMDS was performed on all the samples using the mothur program (Schloss et al., 2009). As can be observed in Figure 5, all shutdown samples from the first campaign clustered together, indicating that the community returns to an equilibrium after each cycle. Cycles 1 and 2 also clustered together in the top right quadrant, whereas cycle 3 was clustered separately in the bottom right quadrant, indicating a shift in the community at this point in time. This was also confirmed in Figure 3. Furthermore, the data points for the second campaign can be observed to blend in with the data points from campaign 1. Taking into consideration that the sampling campaigns were approximately 1 year apart, this further corroborates the fact that the system is very stable. Indeed, the shutdown data points for both campaigns were clustered together in the two left quadrants, whereas the cycle data points were clustered in the two right quadrants. Furthermore, the cycle from campaign 2 intercalates in between the cycles from the previous campaign, which supports a model where the bacterial community oscillates between different states.

In order to evaluate which parameters were responsible for the community shifts, a Spearman rank correlation analysis was performed between the data points on the NMDS axes and the observed OTUs and physico-chemical parameters. The OTUs with the largest impact on the separation between cycle and shutdown samples are represented as blue arrows in Figure 5. Regarding the physico-chemical parameters, it was observed that gamma radiation, flow rate, and temperature showed a significant correlation with the horizontal axis, whereas conductivity was marginally significant. Acetate and nitrate concentrations did not show a significant correlation. This suggests that radiation exposure, flow rate and temperature could be important factors in the community shift observed across cycles and shutdown periods. Correlation coefficients and p-values for the described OTUs and physico-chemical parameters can be found in Table 3.