Introduction

Across the tree of life, there are dramatic differences in ionizing radiation (IR) resistance. Among organisms from the same order and even between species which share a large core of genes and which evolved from a proximal common ancestor, radiation resistance is not predictable using genomics-based bioinformatic approaches1,2,3,4,5,6,7,8,9,10,11,12. Mounting evidence supports that radioresistance is a polyphyletic metabolic trait that evolved as a byproduct of resistance to other more commonly encountered environmental stressors such as desiccation11,13,14,15. Indeed, the etiologic reactive oxygen species (ROS) responsible for metabolism-induced oxidative stress are the same as those generated from water by IR10. It is the relative abundance, distribution and fate of different ROS within and between irradiated species that vary greatly10,11. It follows that evolution in an environment that is prone to either biotic or abiotic ROS production16,17, could (as a byproduct) drive the evolution of IR-resistant organisms.

As a group, fungi are very IR-resistant and can colonize diverse habitats exposed to harsh conditions, including chronic ionizing radiation (CIR): radioactive waste sites, nuclear disaster zones, and space stations18,19,20,21,22,23. However, not enough is known about fungal stress responses and IR resistance mechanisms. It was recently reported that polyextremotolerant fungi accumulate high concentrations of Mn2+ metabolite antioxidant complexes (Mn antioxidants), which very efficiently scavenge IR-induced ROS11,24,25. This is analogous to extremely IR-resistant Deinococcus bacteria26,27. Rationally-designed Deinococcus Mn antioxidants are now used in the development of irradiated vaccines and as in vivo radioprotectors28,29,30. However, very little is known about the nature of fungal Mn antioxidants24. Thus, a study of stress resistance mechanisms in fungi may offer fresh insight into the design of radioprotectors and radiomitigators for humans, and for combatting fungal diseases in humans and in agricultural plants31,32. Historically, stress response mechanisms in fungi to chemicals and elevated temperatures have been actively studied in just a few model species (mainly in Saccharomyces cerevisiae, Schizosaccharomyces pombe and Candida albicans)33,34,35,36. The investigation of radiation responses among fungi is a comparatively neglected topic.

We previously showed that many fungi can grow under intense CIR dose rates of 13–67 Gy/h, with basidiomycete and ascomycete yeasts identified as particularly CIR-resistant groups. Among 145 phylogenetically diverse fungi that we tested, 78 grew under 36 Gy/h18,37. Consequently, this dose rate represents a convenient “dividing line” for separating CIR-sensitive from CIR-resistant strains. We also tested the same strains for resistance to acute IR. Unexpectedly, there was only a weak correlation between resistance to CIR and resistance to acute IR37. We suggested that the weakness of the association is caused by qualitative differences between chronic and acute radiation stresses: whereas the rates of DNA damage production and repair are critical to CIR resistance in replicating cells, the amounts of DNA damage (double strand breaks in particular) limit survival in acutely irradiated, non-replicating cells37,38.

Here we compared CIR exposure with four environmentally-relevant stressors by comparing the abilities of 95 yeast or dimorphic fungal strains from two major phyla (Ascomycota and Basidiomycota) to resist CIR and acute IR; heavy metals; low pH; and elevated growth temperatures. CIR resistance in this study represents the ability to grow (i.e. to remain metabolically active and proliferate) under continuous irradiation, rather than the ability to remain temporarily dormant during exposure and recover afterwards. The 95 strains (Table 1) were chosen to represent a large and diverse sample of wild-type fungi, which was intended to provide detailed information on stress response ranges. All these strains were able to grow on solid rich medium in the absence of CIR, and then were scored for growth on the same medium incubated under CIR.

Table 1 Table of fungal strains and their resistance phenotypes.

We used logistic regression to quantify the correlations between the growth abilities of studied fungal strains under CIR and their tolerance to the other stressors. Ensemble machine learning, which can handle complex non-linear relationships between variables, was used for additional confirmation of the results. The correlations found in this large number of diverse fungal strains provide new insight and opportunities for additional research into the mechanisms utilized by a multitude of wild-type fungal taxa to counteract severe stresses. In particular, they offer a way forward for studying how and to what extent the molecular mechanisms for resisting CIR are partnered with those for resisting other adverse conditions, and for promoting development of IR countermeasures.

Methods

Strains

The 95 basidiomycetous and ascomycetous yeasts in Table 1 were selected from a collection of 145 phylogenetically diverse fungi assembled and reported previously as part of a Department of Energy (DOE) study dedicated to bioremediation of radioactive waste sites18,37. We previously showed that many of these fungi could grow under 36 Gy/h37. The 95 strains were chosen for their ability to grow well on solid YPD medium at room temperature (RT). CIR resistance can be reliably assessed only on solid media, with constant access to atmospheric oxygen from above and nutrients from below. In liquid media the results are poor because the cells locally deplete oxygen (only ~10 ppm available, vs. ~200,000 ppm on the surface of solid media) and nutrients and stop growing, and/or radiation-induced mutants can outgrow the wild-type parental strain27.

Resistance measurements for radiation and other stressors

Resistance levels to CIR, acute IR, and heavy metals were measured and compared in 95 different fungal strains listed in Table 1. The ability to grow under CIR (called CIRgrowth, scored as 1 for growth and 0 for no growth) on YPD plates at pH 7.0 (1% yeast extract; 2% peptone; 2% D-glucose; 2% bacteriological agar), was assessed as shown in Fig. 118. Survival following acute forms of gamma radiation was determined on YPD plates by colony forming unit (CFU) assay as described previously26, and expressed as the dose killing 90% of the population (D10). Acute exposures of liquid fungal cultures were performed in a 60Co irradiator (10 kGy/h) at 0 °C.

Figure 1
figure 1

Scoring growth on nutrient agar plates under CIR (36 Gy/h). For a given strain, a sector (1–8) was inoculated on YPD (yeast medium) (a,c) and TGY (bacterial medium) (b,d) agar plates. The inoculated plates were incubated in a 137Cs gamma irradiator at 22–25 °C. The plates were then photographed and yeast sectors scored as either CIR-resistant (c1,3,5) or CIR-sensitive (c2,4,6). For each YPD test plate, there was an identically inoculated TGY plate that included two bacteria: Deinococcus radiodurans (ATCC BAA-816) (CIR-resistant) and Pseudomonas putida (ATCC 47054) (CIR-sensitive), which served as CIR operational controls. Sectors: 1. EXF-6761; 2. EXF-6219; 3. EXF-5735; 4. EXF-6218; 5-6. Standard laboratory S. cerevisiae strains FY1679 (diploid) and BY4741 (haploid), respectively; 7. D. radiodurans; and 8. P. putida. No irradiation (a,b); 36 Gy/h (c,d).

Throughout this work, chronic exposures (36 Gy/h) were performed in a 137Cs irradiator at RT for 3–6 days. This time was sufficient to clearly assess whether or not a given strain is able to grow and proliferate under 36 Gy/h CIR while nutrients in the medium remain abundant. Binary scoring of growth on agar plates was sufficient to perform correlation analysis and created approximately balanced classes containing strains that either did or did not grow.

Maximum growth temperature (Tmax, °C) was determined on solid YPD by inoculating the strain to single colony, incubating the plates at various temperatures (25–50 °C; temperature maxima) for 7 days, and visually inspecting the plates for colony formation. The highest concentrations of HgCl2 (Sigma, 215465), merbromin (Sigma, M7011), K2Cr2O7 (Sigma, P2588), and CrCl3 (Aldrich, 27096) supporting growth were determined in liquid AM (Acidiphilium Medium); YPD is not suitable for measuring metal toxicity because of its metal-chelating properties18. The overnight cultures were pre-grown at optimal temperatures in YPD medium, washed twice in sterile MQ and used to inoculate fresh AM media supplemented with different concentrations of heavy metals in 96-well plates to a final OD6000.1. The strains were incubated at optimal temperatures. After inoculation, the OD600 was measured in one week. The lowest pH supporting growth (lowpHgrowth, coded as 0 or 1) was determined as described previously18.

The measured variables D10, CrCl3, HgCl2, K2Cr2O7, and MER were log10-transformed to bring their distributions closer to normal. This procedure produced the variables called logD10, logCrCl3, etc. Another binary variable called Ascomycota was added to test for the potential effect of phylum. It was coded as 1 for strains belonging to Ascomycota, and at 0 for strains belonging to Basidiomycota.

Correlation analysis and logistic regression

The associations between the predictor variables (logD10, logCrCl3, logHgCl2, logK2Cr2O7, logMER, Tmax, lowpHgrowth and Ascomycota) and the outcome (CIRgrowth) were assessed using R 3.5.1 software (https://www.r-project.org/). We calculated Pearson correlation coefficients, and performed logistic regression with information theoretic multi-model inference (MMI) based on the Akaike information criterion with sample size correction (AICc)39. The logistic regression with MMI provided a parametric method for assessing the main effects of the predictor variables. As an alternative backup approach that can handle non-linear dependences and is more robust to correlations between predictors40, we also analyzed the data using machine learning by customized generalized boosted regression (GB) with synthetic noise variables as benchmarks of predictor performance41,42. These machine learning analyses are described in Supplementary Methods.

Pearson correlation matrices of the studied variables were generated using the cor and corrplot functions in R 3.5.1. MMI was implemented using the glmulti R package. This approach fitted all possible predictor combinations (model structures) to the data by logistic regression and assessed each combination’s support by AICc. It provided 95% confidence intervals (CIs) for each predictor, taking into account model selection uncertainty (i.e. the variability in which predictors are present and which are absent from a particular fitted model structure). It also provided a relative importance score for each predictor, which was calculated using the sum of Akaike weights for all fitted model structures that contained the given predictor39. The strongest predictors had 95% CIs not overlapping zero and high relative importance scores. They were used to build preferred models, which were then evaluated by constructing receiver operating characteristic (ROC) curves using the pROC package.

Data sets

Logistic regression that included all predictor variables showed that the variable Ascomycota produced a high variance inflation factor (VIF) of 4.2, which may indicate multicollinearity in the data set. Consequently, we split the data set into 2 parts by phylum (Ascomycota, 67 strains, and Basidiomycota, 28 strains) and analyzed these parts separately. Because a substantial number (34) of the studied fungal strains belonged to the species Saccharomyces cerevisiae, we tested their influence on the results by performing the analyses after excluding the S. cerevisiae data, or conversely, by analyzing the S. cerevisiae data alone.

Additional analyses

The analysis approaches described above assumed that the data for each fungal strain are statistically independent. To account for the potential effects of correlations between data for different species belonging to the same genus, we also implemented mixed effects logistic regression models with random intercepts and slopes for each genus, using the lme4 R package. For the fixed effects, we used each predictor (logD10, logCrCl3, logHgCl2, logK2Cr2O7, logMER, Tmax, lowpHgrowth and Ascomycota) one at a time. Several predictors together were not used because such complex mixed effects models did not converge on the data sets used here (yeasts from all phyla combined). Goodness of fit for these mixed effects models was assessed by conditional and marginal R2.

Results

Visual comparisons of how the continuous predictor variables behaved in each fungal data set are shown in Fig. 2. Correlation matrices of the variables in these data sets are shown in Fig. 3. They suggested that for Ascomycota the strongest and most significant correlation of CIRgrowth was with logCrCl3, whereas for Basidiomycota it was with Tmax. Interestingly, acute IR resistance (logD10) did not turn out to be strongly correlated with CIRgrowth (Table 2). For example, several fungi were able to grow under 36 Gy/h CIR despite having logD10 below the 25th percentile (EXF-6463 Candida pseudoloambica, EXF-7107 Geotrichum sp., EXF-5288 Kluyveromyces marxianus, EXF-5871 Saccharomyces cerevisiae, EXF-7167 Saccharomyces paradoxus, EXF-7964 Wickerhamomyces anomalus). In contrast, some others were unable to grow under CIR despite having logD10 above the 75th percentile (EXF-7173 Saccharomyces paradoxus and S. cerevisiae strains EXF-5294, 6246 and 6248, Supplementary Data File 1). These examples illustrate that discordance between acute IR and CIR resistances can occur even among strains within the same species (e.g. S. cerevisiae or S. paradoxus).

Figure 2
figure 2

Box plots that summarize and compare continuous predictor variables for yeasts belonging to phylum Ascomycota and those belonging to phylum Basidiomycota.

Figure 3
figure 3

Visualization of pairwise Pearson correlation matrices of all variables for yeasts belonging to phylum Ascomycota and those belonging to phylum Basidiomycota. Red star symbols indicate statistical significance levels: 3 stars indicate p < 0.001, 2 stars indicate p < 0.01, 1 star indicates p < 0.05, no stars indicates p > 0.05. Due to multiple comparisons, only 3 star significance levels are likely to indicate strong associations. Crossed out boxes represent meaningless correlations of a given variable with itself. Blue ellipses represent positive correlations, and red ones represent negative correlations. Darker color tones and narrower ellipses represent larger correlation coefficient magnitudes.

Table 2 Associations between various predictor variables and CIR resistance (CIRgrowth, i.e. ability to grow under 36 Gy/h on YPD).

While the correlation matrices show only pairwise correlations, the logistic regression and machine learning approaches looked for associations between CIRgrowth and all predictors analyzed together. These methods agreed with each other, and with the correlation matrices, in identifying logCrCl3 as the strongest predictor of CIR growth in Ascomycota and Tmax in Basidiomycota (Table 2). Exclusion of S. cerevisiae did not qualitatively change the results: logCrCl3 and Tmax were still the only predictors that had 95% CIs not overlapping zero in logistic regression MMI analysis and achieved high scores in machine learning analyses (see Supplementary Table 1). Using S. cerevisiae data alone did not produce strong correlations between any of the variables, perhaps due to a roughly 3-fold reduction in data set size (34 S. cerevisiae strains vs 95 in the full data set).

Based on these results, the preferred logistic regression model for Ascomycota contained logCrCl3 as the only predictor with a best-fit coefficient of 3.21 (standard error, SE 0.81, p-value 7.66 × 10−5) and an intercept value of −7.38 (SE 1.99, p-value 2.12 × 10−4). This model achieved good performance in discriminating between strains that were able to tolerate 36 Gy/h from those that were unable to do so: its ROC curve area was 0.795 (95% CI: 0.699, 0.892). The preferred logistic regression model for Basidiomycota contained Tmax as the only predictor with a best-fit coefficient of 0.479 (SE 0.163, p-value 3.26 × 10−3) and an intercept value of −14.41 (SE 4.91, p-value 3.37 × 10−3). This model achieved very good performance: its ROC curve area was 0.949 (95% CI: 0.862, 1.000). These logistic regression model fits for Ascomycota and Basidiomycota are shown in Fig. 4. To illustrate the same patterns in a non-parametric way, we also plotted the relationships between CIRgrowth, logCrCl3 and Tmax in Fig. 5.

Figure 4
figure 4

Visualizations of preferred logistic regression models (curves) and data points (open circles) for Ascomycota and Basidiomycota yeast data sets. Shaded regions around the best-fit model curves represent 95% prediction intervals. The data points had binary values of 0 or 1 on the y-axis (corresponding to no growth or growth under 36 Gy/h, respectively), so to prevent overlap of the data points and improve their visualization we moved them along both the y-axis and the x-axis in these plots by small random increments.

Figure 5
figure 5

Associations between CIRgrowth with logCrCl3 or Tmax for yeast from both phyla combined. Small black circles represent strains that could not grow under 36 Gy/h (CIRgrowth = 0), and large blue circles represent strains that could grow under this dose rate (CIRgrowth = 1). To prevent overlap of the data points and improve their visualization we moved them along both the y-axis and the x-axis by small random increments.

Combined analyses of yeast data from both phyla with the Ascomycota variable removed again pointed to logCrCl3 and Tmax as the best predictors. There was no strong evidence for interactions between logCrCl3 and Tmax: when an interaction term between these variables was introduced into logistic regression models, its MMI-corrected 95% CIs overlapped zero.

To account for the potential effects of correlations between data for different species belonging to the same genus, we also implemented mixed effects logistic regression models with random intercepts and slopes for each genus, using combined data for both Ascomycota and Basidiomycota. For the fixed effects, we used each predictor (logD10, logCrCl3, logHgCl2, logK2Cr2O7, logMER, Tmax, lowpHgrowth and Ascomycota) one at a time. The model with Tmax as the fixed effect predictor outperformed all other tested models and achieved conditional R2 (which represents the variance explained by the entire model) of 0.989 and marginal R2 (which represents the variance explained by the fixed effects only) of 0.416. In this model, the slope coefficients for Tmax were smallest for the genera Cryptococcus and Rhodotorula (1.711 and 1.725, respectively) and largest for the genera Saccharomyces and Schwanniomyces (2.857 and 2.860, respectively). For the model with logCrCl3 as the fixed effect predictor, both the conditional R2 and marginal R2 were 0.312. These results suggest that the correlation between CIRgrowth and Tmax varies considerably by genus, whereas the correlation between CIRgrowth and logCrCl3 does not appear to vary by genus.

Discussion

We compared CIR exposure with other stressors by comparing the abilities of 95 yeasts from two different phyla (Ascomycota and Basidiomycota) (Table 1) to resist the following challenges: (a) Chronic and acute gamma radiation, which generate ROS by water radiolysis10. (b) Transition metals that generate ROS by redox-cycling12. (c) Low pH (2.3), where protons (H+) potentiate the production of hydrogen peroxide (H2O2) from superoxide (O2.−)10. (d) Elevated growth temperatures that increase production of metabolism-induced ROS36. We observed a robust correlation between resistance to CIR and Cr3+ (logCrCl3) in Ascomycota yeasts. The correlation with Cr6+ (logK2Cr2O7) was weaker (Fig. 3, Table 2). While Cr6+ is a known human carcinogen, its reduction to Cr3+ renders the metal less mutagenic and carcinogenic and even essential or beneficial in some situations like diabetes43,44. In basidiomycete yeasts, the strongest predictor of CIRgrowth was the maximum temperature that supported growth (Tmax), rather than heavy metal tolerance (Fig. 3, Table 2). All basidiomycete yeast strains with Tmax below 30 °C were also unable to grow under 36 Gy/h (Fig. 4). Perhaps, this phenomenon is related to antioxidant activity because heat stress is known to enhance ROS production, and counteracting these ROS by antioxidants is involved in yeast thermotolerance36.

Significantly, the correlations between all of the tested stressors (CIR, acute IR, low pH, elevated temperature, chromium and mercury) tended to have a positive sign (Fig. 3). This suggests that in yeast some shared mechanisms may contribute to resisting multiple environmental stressors including gamma radiation. For example, two strains of the common laboratory yeast species S. cerevisiae (EXF-4916 and EXF-5284, Table 2), which is considered an environmentally-robust baker’s yeast45, were resistant to all stressors tested (they were at or above the 80th percentile for tolerance to each stressor).

A feature shared by fungi and prokaryotes shown to possess resistance to high radiation doses is their great ROS-scavenging capacity10,11. In wild-type yeasts and bacteria, the intracellular content of Mn antioxidants is strongly correlated with IR resistance11,24. Indeed, under nutrient-replete conditions, a high intracellular concentration of Mn antioxidants renders antioxidant enzymes such as superoxide dismutase (SOD) dispensable for acute and chronic IR resistance26,38. In the case of Deinococcus spp., the bacteria are typically capable of surviving 10 kGy of acute radiation under nutrient rich conditions46, can grow luxuriantly under CIR at 60 Gy/h, and are resistant to the toxic effects of chromium19. Manganese is unique among redox active transition metals found in cells: Mn redox-cycling favors O2.− scavenging without the release of extremely reactive hydroxyl (HO•) radicals. In contrast, redox-cycling of other transition metals (e.g., Fe and Cr) gives rise to HO• radicals (Fenton and Haber-Weiss reactions)15. Thus, in cells lacking Mn antioxidants, O2.− radicals can become a significant source of HO• radicals, and hence a significant factor in the biochemical mechanism of cellular damage caused by most redox active metals (e.g. Fe, Cr, U)10,12,19,46.

Fungi also accumulate high concentrations of Mn antioxidants and are highly resistant to oxidative stress11,24. However, fungal resistance to acute IR (logD10) did not turn out to be a good predictor of resistance to CIR (CIRgrowth) in any of the data sets analyzed here (Table 2). Acute IR resistance also appeared to be uncorrelated with the majority of resistances to other tested stressors, except for Hg (logHgCl2) in Ascomycota (Fig. 3). Lack of correlation between acute IR resistance and resistance to other stressors (including CIR) was also reported in other studies47,48, including those on directed evolution of bacteria by selection for acute IR resistance49.

Unfortunately, current understanding of chronic radiation resistance and stress responses in general is limited for fungi, outside of a few model species. Our results, based on a large variety of fungi from different phyla, suggest that the mechanisms involved in resistance to acute doses of gamma radiation in fungi may be quite distinct from those involved in resisting the other evaluated stressors, which were chronic by nature. A similar case for bacteria is supported by the finding that the antioxidant enzyme catalase is dispensable for resistance to acute IR, but not CIR37; and similarly, E. coli strains evolved for acute IR resistance by directed evolution can be CIR sensitive37,49. A likely explanation is that different responses are needed for cells to cope with large amounts of simultaneously produced damage from acute doses of gamma radiation, followed by recovery under non-stressful conditions, compared with those needed to proliferate under a continuously elevated rate of damage production (presumably dominated by ROS-related mechanisms) from CIR, heavy metals or elevated temperature37.

Our results are consistent with the concept of core stress response, where relatively large numbers of enzymes are expressed in response to different stresses, and exposure to one stress type can cause cross-protection from other stressors18,33,34. Core stress responses were investigated and found in several fungal species and are probably widespread, although the specific genes involved can differ widely between species18,33,34. Similarly, genetic heterogeneity also is a central characteristic of IR resistance phenotypes in general, which has rendered bioinformatic approaches to gauging ROS stress responses futile38. Thus, further physiological studies are warranted to investigate the mechanistic overlap between resistances to CIR, chromium and elevated temperature in certain fungal groups.