CRISPR-Cas systems provide adaptive immunity against mobile genetic elements for bacteria and archaea. Like the adaptive immune system of jawed vertebrates, the CRISPR-Cas systems can remember previously encountered pathogens, initiate a rapid response to a second invasion, and eliminate the recurrent invader. However, unlike the ubiquitous presence of adaptive immunity in jawed vertebrates (Müller et al., 2018), the CRISPR-Cas systems are only present in about 40% of bacteria (Makarova et al., 2020). The patchy distribution of the CRISPR-Cas systems among bacteria is a recognized mystery (Ledford, 2017; Koonin, 2018). Given the constant horizontal transfers, their absence in more than half of bacterial genomes is unlikely to happen just by chance (Bernheim, 2017). Instead, it should be attributed to a tradeoff between the costs and benefits of the CRISPR-Cas systems. First, the acquirement and maintenance of CRISPR-Cas systems would sequestrate limiting resources such as the building blocks, the energy, and the transcription and translation machines (Lynch and Marinov, 2015; Vale et al., 2015; Frumkin et al., 2017). Second, the autoimmune response and cell death induced by self- and prophage-targeting spacers might be a selective force for the loss of CRISPR-Cas systems (Rollie et al., 2020; Wimmer and Beisel, 2020). In addition, for the viruses with high densities, high mutation rates, high genetic diversities, or carrying anti-CRISPR proteins, the efficiency of CRISPR-Cas systems is limited (Weinberger et al., 2012; Iranzo et al., 2013; Westra et al., 2015; Trasanidou et al., 2019). CRISPR-Cas systems are not favored in conditions with high antibiotic pressures because they inhibit horizontal gene transfer, an efficient way for bacteria to acquire antibiotic resistance (Palmer and Gilmore, 2010).

Just after discovering the CRISPR-Cas structures, it was noticed that they are more prevalent in the thermophilic archaea and the hyperthermophilic bacteria (Jansen et al., 2002; Makarova et al., 2002). Later large-scale analyses confirmed the prevalence of CRISPR-Cas systems in thermophiles and hyperthermophiles and showed a positive correlation between CRISPR abundance and growth temperatures (Anderson et al., 2011; Makarova et al., 2011; Weinberger et al., 2012; Gophna et al., 2015). Recently, Weissman et al. (2019) tried a phylogenetically corrected machine learning approach using models like logistic regression, sparse partial least squares discriminant analysis, and random forest with the data split into different blocked folds by their pairwise distance on the phylogeny. Their results indicate that temperature and oxygen levels might be the most influencing ecological factors determining the distribution of the CRISPR-Cas systems. In this paper, we describe an abrupt increase of bacterial CRISPR-Cas systems at around 45°C and put forward a new hypothesis on the thermal distribution of the CRISPR-Cas systems.

The phylogenetic relationships among the analyzed species were retrieved from Genome Taxonomy Database (GTDB) (Parks et al., 2021). The GTDB group constructed the phylogenetic tree of bacteria and archaea using constantly updated whole-genome sequences. They selected a representative genome for each species. The sample size of their representative genomes is similar to the NCBI genome database.¹ For the accuracy in estimating the CRISPR-Cas abundance, we only retrieved the representative genomes assembled at the levels of “complete genome” or “chromosome” from the GTDB database (accessed: Dec. 24, 2021), 4908 bacterial and 292 archaeal genomes. Then, we downloaded the genome sequences from ftp://ftp.ncbi.nlm.nih.gov/genomes/ and annotated the CRISPR-Cas systems using CRISPRCasFinder v1.3 (Couvin et al., 2018). According to Couvin et al. (2018), the annotated CRISPR arrays were classified into four categories, 1 to 4, according to their evidence levels. The CRISPR arrays with evidence levels 3 and 4 are highly likely candidates, and those with evidence levels 1 and 2 are potentially invalid. Therefore, only the CRISPR arrays with evidence levels 3 and 4 were counted in calculating CRISPR array abundance. We counted the putative CRISPR arrays with evidence levels 1 and 2 as zero. The abundance of an object in a genome was defined as the simple count of the object (CRISPR arrays, CRISPR spacers, cas genes, or cas gene clusters) in the genome.

Based on these 5200 representative genomes, we constructed three datasets.

The first is the smallest but most accurate, including 1351 species (1168 bacteria and 183 archaea, Supplementary Table 1). This dataset was constructed by retrieving the direct links of Genbank IDs and optimal growth temperature of the same strain from the database TEMPURA (Sato et al., 2020) and the reference (Lyubetsky et al., 2020), and then manually matching the strain names of the other records in TEMPURA with the complete genomes deposited in Genbank. In this dataset, the Topt and the genome of each species must come from the same strain. Potential errors resulting from polymorphism of Topt or CRISPR abundance within each species have been eliminated.

The second dataset contains 3154 genomes (2944 bacteria and 210 archaea, Supplementary Table 2). In constructing this dataset, polymorphisms among the strains of a single species were neglected. The strain name of each species was overlooked. The Topts were retrieved from a series of sources. For the conflicting records of Topt between different resources, the preferential rank was from the Topts with GenBank IDs in TEMPURA (Sato et al., 2020) and the reference (Lyubetsky et al., 2020), the Topts in reference (Madin et al., 2020), the Topts not associated with GenBank IDs in TEMPURA, the growth temperatures in reference (Madin et al., 2020), the growth temperatures in the reference (Engqvist, 2018), to the growth temperatures in BacDive (Reimer et al., 2018).

The third dataset contains all the 5200 representative genomes retrieved from GTDB (Supplementary Table 3). The Topt values of these species were predicted using a machine-learning method (Tome, version 1.0.0) reported in reference (Li et al., 2019).

All the temperature values analyzed in this study were first rounded into integers.

These three datasets gave similar column charts for the relationships between Topt and CRISPR-Cas abundance. We presented the charts from the 3154 genomes in the main text, and all further analyses were based on this dataset. The charts obtained from the other two datasets were deposited in Supplementary Tables 4, 5.

The presence or absence of each prokaryote in particular environmental conditions was obtained from the isolation source data of the BacDive database (Reimer et al., 2018). The restriction-modification (RM) enzyme genes in each genome were obtained from the PADS Arsenal database (Zhang et al., 2020). All these data were deposited in Supplementary Table 2.

The phylogenetic signals (λ) of CRISPR array abundance, CRISPR spacer abundance, cas gene abundance, cas gene cluster abundance, and growth temperatures were estimated using the phylosig function of the R (Version 4.0.3) package phytools (Version 0.7-70) (Revell, 2012). The phylogenetic generalized least squares (PGLS) regression was performed using the R (Version 4.0.3) package phylolm (version 2.6.2) (Ho and Ane, 2014). Pagel’s λ model has been applied in the analyses.

The non-linearity of the relationships was estimated using the generalized additive model (GAM) that was integrated into the R package mgcv (version 1.8-33) (Wood, 2017). The derivative of the GAM curve was calculated using the R package gratia (Version 0.6.0) (Simpson, 2022).

Previous studies have found that prokaryotes living in high temperatures are more likely to have CRISPR-Cas systems (Jansen et al., 2002; Makarova et al., 2002, 2011; Anderson et al., 2011; Weinberger et al., 2012; Gophna et al., 2015; Weissman et al., 2019). This study presented a more detailed description of the relationship between CRISPR-Cas and Topt. At low (4 − 34°C) and high temperatures (50 − 85°C), the abundances of bacterial CRISPR-Cas are not significantly correlated with Topt. However, at around 45°C, bacterial CRISPR-Cas abundance increases sharply with Topt. Most significantly, the bacterial type I CRISPR-Cas abundance exhibits an abrupt jump at 45°C. As we see, the evolutionary and mechanical links between growth temperature and CRISPR-Cas abundance previously proposed (Weinberger et al., 2012; Iranzo et al., 2013; Høyland-Kroghsbo et al., 2018) could not explain the abrupt transition at around 45°C. Temperature influences many aspects of cellular processes and the physical features of the environment (Clarke, 2014). No matter linear or non-linear, the biological effects of temperature increase gradually.

From the literature, we noticed that the thermal distribution of eukaryotes has an abrupt decline at around 45°C. Very few eukaryotes can live above 60°C (Tansey and Brock, 1972; Brock, 1985, 2001; Clarke, 2014). Bacterial communities are often heavily consumed by eukaryotic predators (Jousset, 2012). In freshwater and marine environments, viral lysis and predation by ciliated and flagellated protists contributed to most bacterial mortality (Pernthaler, 2005; Takasu et al., 2014). In addition, some bacteria could also kill other bacteria and consume the released nutrients. Except for a few exceptions, most predatory bacteria could not grow at temperatures above 45°C (Reichenbach, 1999; Williams and Chen, 2020). Here, we propose that the accessibility of cellular predators along the temperature axis might indirectly govern the thermal distribution of the CRISPR-Cas systems (Figure 9). Cellular predators are the bacterivorous eukaryotes and predatory bacteria, such as protozoa, nematodes, myxobacteria, and Bdellovibrio.

In birds, it has been shown that predation risk could significantly reduce the allocation of the limiting resources to immune function (Møller and Erritzøe, 2000). The birds captured by cats consistently had smaller spleens than those killed by non-predatory reasons like collisions with windows or cars. When the hosts are exposed to lethal predators, predator-mediated mortality becomes dominant, and the pathogen-mediated mortality decreases relatively. Consequently, the priority to invest in immune function is reduced. Physiological and evolutionary reducing the allocation of the limiting resources to immune function would be favored.

We proposed that the same case might happen in bacteria. For picophytoplankton, grazer-mediated mortality and viral-mediated mortality are inversely correlated (Pasulka et al., 2015; Staniewski and Short, 2018). Indirect interactions among grazers and viruses are destined to occur, provided that bacterial cells have tradeoffs in grazing and virus resistance. When bacterial mortality mostly comes from predator grazing, the benefits of adaptive immunity might be dwarfed by the costs in the maintenance and expression of the CRISPR-Cas systems, like allocating the limiting resources, targeting host or prophage genome, and inhibiting horizontal gene transfer. By contrast, when bacterial mortality mostly comes from viral lysis, some costs of CRISPR-Cas systems would be tolerable because the benefits of adaptive immunity outweigh the costs. At around 45°C, with the increase of environmental temperature, cellular predator abundance decreases sharply (Brock, 1985; Clarke, 2014); thus, the grazing risk of bacteria should be abruptly relieved. Bacteria growing at higher temperatures should die mainly from viral lysis. Antivirus immunity should be favored even if it costs the bacterial cells. Within the temperature ranges inaccessible to grazing predators or within the temperature ranges fully accessible to the grazing predators, temperature changes have little effect on the evolution of the antivirus system (Figure 9).

Besides explaining the thermal distribution, our hypothesis suggests that other environmental factors that severely reduce cellular predator abundance should also affect bacterial mortality and CRISPR-Cas distribution. A generalized prediction of our hypothesis is that bacteria living in extreme environments inaccessible to cellular grazers should carry more CRISPR-Cas systems in their genomes. Weissman et al. (2019) recently found a negative interaction between CRISPR-Cas systems and oxygen availability and hypothesize that oxidative-stress-associated DNA repair processes might interfere with the function of CRISPR-Cas systems. Here, we provide an alternate explanation for their observation by extending our hypothesis. All the well-known cellular predators are aerobic organisms. There are no cellular predators or only a few unknown predators in anoxic environments. Similar to growth temperatures at around 45°C, the transition from an oxygenated to an anoxic environment would substantially reduce the grazing-caused bacterial mortality and indirectly increase the requirement of antivirus immunity.

Besides growth temperature, many other ecological and evolutionary factors that might influence the phylogenetic distribution of CRISPR-Cas systems (Palmer and Gilmore, 2010; Weinberger et al., 2012; Iranzo et al., 2013; Vale et al., 2015; Westra et al., 2015; Trasanidou et al., 2019; Rollie et al., 2020; Wimmer and Beisel, 2020) are beyond the scope of this hypothesis. Even for the thermal distribution of CRISPR-Cas systems, we are open to other possible explanations. A bacteriophage infecting the tropical pathogen Burkholderia pseudomallei was found to be temperate at lower temperatures (25°C) and tends to go through a lytic cycle at higher temperatures (37°C) (Shan et al., 2014). At least for type I CRISPR-Cas systems, targeting temperate phages has been demonstrated to be a driving force for the loss of adaptive immunity (Rollie et al., 2020). Therefore, bacteria living in low temperatures might have fewer CRISPR-Cas systems because of the temperate-phage-induced bacterial autoimmunity. However, the temperature-associated switching of the life cycle of the bacteriophage of B. pseudomallei is just a piece of isolated evidence. At present, we do not know how many phages in nature have similar temperature-associated switching of life cycle as the bacteriophage of B. pseudomallei.

We propose that bacterial cells lose the CRISPR-Cas systems when bacterial mortality mostly comes from predator grazing. However, in natural environments, this does not always happen even below 45°C. Bacterial mortality is destined to be more or less contributed by viral lysis because of the widespread of viruses. In addition, prey bacteria are not entirely passive to be grazed, and they could evolve grazing-resistance capacities, like high motility, large size, and biofilm formation (Matz and Kjelleberg, 2005; Jousset, 2012; Lurling, 2021). In a long-term arms race between prey bacteria and grazing protists/bacteria, the prey bacteria might, in some periods, be free of grazing risk and grazing-caused mortality because of newly evolved grazing-resistance strategies. In this case, viral lysis becomes dominant in bacterial mortality, and the bacteria experience an intense selective pressure to have CRISPR-Cas systems. Therefore, our hypothesis is not exclusive to CRISPR-Cas-rich psychrophilic and mesophilic bacteria.

There is no clear pattern in archaeal CRISPR-Cas abundance along the axis of growth temperature (Figure 7). The difference in CRISPR-Cas distribution between bacteria and archaea might come from the physiological, genomic, or ecological differences between the two domains. It is also possible that random noises resulting from the small sample size have masked the thermal distribution pattern. In a recent analysis on the relationship between growth temperature and GC content using the Topt dataset from the database TEMPURA (Sato et al., 2020), we found that Topt is significantly correlated with bacterial genome GC content (N = 681) but not archaeal genome GC content (N = 155). Then, we randomly drew 155 bacteria from the 681 bacteria 1000 times. In > 95% rounds of resampling, the positive correlations became statistically non-significant (p > 0.05) (Hu et al., 2022). The results suggest that the effective sample size in phylogenetically related data is much smaller than the census number of the analyzed lineages. We hope to replicate the present analysis in archaeal genomes in the future when several hundred or thousands of archaeal species are available.

Our hypothesis states that antivirus immunity is restricted at low temperatures because bacterial mortality mostly comes from grazing. However, sharp increases of antivirus immunity at around 45°C were observed in the CRISPR-Cas systems but not in the RM systems. Here we make further speculation. The intense grazing pressures at low temperatures would favor the acquirement of grazing resistance. Some grazing-resistant genes, like that in type III secretion systems (Matz et al., 2011) and that involved in biofilm formation (Wang et al., 2017), have been found in genomic islands, a widespread tool for horizontal gene transfer (Juhas et al., 2009). In the evolution of grazing resistance, horizontal transfer of the grazing-resistant genes would be selected. Both the CRISPR-Cas systems and the RM systems could depress horizontal gene transfers. However, the RM systems would not stop the transfer among closely related bacteria, especially when they have shared RM systems (Dimitriu et al., 2019). Besides this, the CRISPR-Cas and RM systems differ in many aspects (Dimitriu et al., 2020). We are open to other possible explanations.

The CRISPR-Cas systems are known to be enriched in thermophilic and hyperthermophilic prokaryotes. In this paper, we take a step further by revealing an abrupt increase of bacterial CRISPR-Cas abundance at around 45°C and putting forward a new hypothesis on the thermal distribution of bacterial CRISPR-Cas systems. Grazing of cellular predators and viral lysis are the primary sources of bacterial mortality; their negative interaction largely influences the tradeoffs between the costs and benefits of antivirus strategies and grazing resistance strategies. As cellular predator diversities and grazing risk precipitously decline at around 45°C, viral lysis becomes the dominant source of bacteria mortality, and the requirement of adaptive immunity might be increased indirectly.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

D-KN conceived the study and wrote the manuscript. X-RL and Z-LL performed the analysis. All co-authors have reviewed and approved the manuscript prior to submission.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.