For the sit-and-wait pathogens, proteomes of eight representative bacteria were selected. For the other three categories, only proteomes of five representative bacteria were sourced. All proteomes are downloaded from UniProt (The UniProt Consortium, 2017). In order to make the investigation more persuasive, we included quantitative data, such as external survival time and host mortality, for comparison. For details, please refer to Table 1. All eight sit-and-wait bacteria were selected based on a literature search using key words such as “sit-and-wait,” “pathogenic,” and “bacteria,” in PubMed and Google Scholar databases. Representative bacteria from other categories were also selected using the same method. All absent data were further sourced from government databases, medical and biological textbooks, and also referenced papers. The collected epidemiological data are only representative and should not be considered to be comprehensive due to the large volume of literature available. It only serves to provide preliminary support for the linkage between durability and virulence.

Enzymes directly linked with energy storage metabolisms in bacteria were first collected through a comprehensive up-to-date review of literature (Lopez et al., 1998; Henrissat et al., 2002; Zhang et al., 2002; Ishige et al., 2003; Kalscheuer and Steinbuchel, 2003; Brown and Kornberg, 2004, 2008; Kulaev et al., 2004; Uchino et al., 2007; Rao et al., 2009; Finkelstein et al., 2010; Kalscheuer, 2010; Achbergerova and Nahalka, 2011; Wang and Wise, 2011; Whitehead et al., 2014; Albi and Serrano, 2016). For a complete list of all the enzymes and their distributions in bacterial proteomes, please refer to Supplementary Table S1 of the review by Wang and Wise (2011). All sequences of ESM-related enzymes were downloaded from The UniProt Consortium (2017). HMMs of all enzymes were constructed by using HMMER based on multiple sequence alignments downloaded directly from Pfam database (Eddy, 2003; Finn et al., 2010). The command, hmmsearch, was used to search all proteomes for homologous sequences of ESM-related enzymes with E-value set to 1e-50. Complete gain or loss of all enzymes in the metabolism of an energy storage compound was marked as 1 and 0, respectively. Short Python scripts (Cock et al., 2009) were written and executed during the processes of file handling and text mining in order to integrate data from hmmsearch results into Supplementary Table S2.

Seed sequences of ASR related proteins were selected by using a web-based browser for Gene Ontology (GO) terms and annotations, QuickGo (provided by the UniProt-GOA group at EBI) (Binns et al., 2009). Database was filtered by designating the taxonomic group as bacteria (Taxonomy Identifier = 2). Meanwhile, 35 GO identifiers under the category of response to abiotic stimulus (GO:0009628) were selected for filtering meaningful and important proteins associated with ASR. A complete list of the 35 GO identifiers is presented below in Supplementary Table S1. In total, 1406 bacterial proteins were collected as seed proteins and used for constructing HMMs corresponding to ASR after using Perl script nrdb90.pl to remove the sequences with more than 90% similarity from the selected proteins (Holm and Sander, 1998). After obtaining sequences for all seed proteins, remote BLAST was performed to collect homologous sequences for each seed protein from the NCBI non-redundant database of protein sequences. In addition, since the sequences returned from remote BLAST were in extensible markup language (XML) file format, a Python script for parsing XML file was also written to extract the homologous sequences. The standalone command-line version of MUSCLE was used so the MSAs were done automatically (Edgar, 2004). Heads or tails of multiple sequence alignments tend to be more inconsistent (Wise, 2010). Thus all MSAs were manually edited to remove heads and tails by using JalView, a free program for MSA editing, visualization, and analysis (Waterhouse et al., 2009). HMMER was selected for the construction of HMMs through hmmbuild command by using multiple sequence alignments. Since HMMER only recognizes STOCKHOLM format, all MSAs results were converted from FASTA to STOCKHOLM format. For detailed steps and python scripts of HMM construction from scratch, please refer to Wang (2013). Python scripts and HMM model for bacterial ASR and ESM are both available under request. For searching homologs, routine procedures are performed by following HMMER User’s Guide (Eddy, 2003). In order to avoid the influence of bacterial proteome size, the number of ASR genes was converted into the percentage of ASR genes in bacterial proteome.

VF sequence package includes 2595 non-redundant core bacterial virulence factors sourced from VFDB (downloadable from VFDB website http://www.mgc.ac.cn/VFs/) (Chen et al., 2005). BLAST-equivalent phmmer command from HMMER package was recruited to automatically search VF homologs in all selected proteomes by using E-value of 1e-50. For usage of HMMER package please refer to the protocol document (Eddy, 2003). Results were all under further processed through a set of short python scripts (Cock et al., 2009). In order to avoid the influence of bacterial proteome size, number of VFs was converted into percentages of VFs in the respective bacterial proteomes.

Unless otherwise indicated, Student’s t-test was used for all statistical analyses via a Python script, specifically the scipy statistical computation package. Data used in the comparisons can be found in the Tables. Where a range is given, the midpoint in the range has been used in the statistical analyses.

Eight representative non-vector-borne human pathogens were collected based on literature review and first studied in terms of external survival time and host mortality. All these pathogens have common features, that is, long environmental survival time and high mortality (Table 1) when compared with bacteria in the other three categories (Table 2). According to the sit-and-wait hypothesis, higher durability should facilitate higher virulence. However, linear regression of host mortality (percentage) against persistence in the environment (in days) across the data in Table 1 only yielded a modest, and not statistically significant, correlation (r-value 0.387, P-value 0.344). While this may be due to the small sample size and the use of statistics from non-standardized experiments for the survival and mortality measurements, it is also understandable given the argument that durability is an enabler of virulence, but is not causally linked.

In order to better understand the selected bacteria and their sit-and-wait potentials, we dive into each species for a detailed investigation. Acinetobacter baumannii is a ubiquitous Gram-negative bacterial pathogen in the environment, which is capable of causing a wide range of infections and ranked among the top 10 pathogens causing septicemia (Jawad et al., 1998). Studies have shown that A. baumannii can persist in hospital environments for up to 3 years (Sherertz and Sullivan, 1985). However, a systematic study found out that multi-drug resistant (MDR) and susceptible strains of A. baumannii have similar survival times in distilled water, for a mean time of 27 days (Jawad et al., 1998). As for mortality, MDR A. baumannii shows a higher rate (26%) than the susceptible strain (17.6%) (Sunenshine et al., 2007), which may be caused by more virulence factors in the MDR strains rather than survival time in the environment. Since the mortality rate of uninfected control is 11.2% (Sunenshine et al., 2007), the average mortality rate for A. baumannii is re-adjusted to10.6%. Burkholderia pseudomallei is a Gram-negative bacillus and the causative agent of Melioidosis with varied severity and difficulty of treatment (Shams et al., 2007). It is normally dwelling on inmate surfaces and can survive in soil microcosms for at least 120 days (Shams et al., 2007; Kamjumphol et al., 2015). A recent study of a B. pseudomallei outbreak in Singapore between 2001 and 2010 hospital showed that the bacteria has a high mortality of 16.5% among 170 patients (Chien et al., 2015). Survival of Corynebacterium diphtheriae on dry surface is reported to be up to 6 months (Kramer et al., 2006). As a leading cause of childhood mortality in developing countries, a study of diphtheria spanning 12 years in a hospital found that the mortality rate is as high as 36.5% (Dandinarasaiah et al., 2013). Haemophilus influenzae is capable of surviving on a dry surface for 12 days and of causing a variety of clinical manifestations with a mortality rate worldwide ranging from 10 to 30% without vaccination (Peltola, 2000; Kramer et al., 2006).

Leptospira interrogans is a worldwide zoonotic pathogen and causes the emerging infectious disease Leptospirosis (Levett, 2001). Its survival in a typical New Zealand soil is at least 42 days (Hellstrom and Marshall, 1978). A systematic study showed that the worldwide mortality rate of Leptospirosis ranges from 4 to 13% between 1970 and 1999 (Costa et al., 2015). Mycobacterium tuberculosis is a typical sit-and-wait pathogen. It causes serious public health issues and its estimated overall mortality rate is 12.3%, based on case studies between 2003 and 2007 (Lin et al., 2014). Its viability and virulence in soil as Mycobacterium tuberculosis complex (MTC) lasts more than 12 months (Ghodbane et al., 2014). Pseudomonas aeruginosa is a major nosocomial pathogen with high morbidity and mortality that ranges from 18 to 61% (Kang et al., 2003) and can survive in bottled water for more than 50 days (Khaniki et al., 2014). However, the survival time heavily relies on temperature, with high temperature leading to better viability. Although Yersinia pestis has a vector-borne transmission route and causes endemic bubonic plague, it can also be transmitted from human-to-human via air droplets and causes pneumonic plague (Galimand et al., 2006). It has the ability to survive and remain virulent in the external environment for more than 280 days, which makes it one of most dangerous sit-and-wait bacterial pathogens (Ayyadurai et al., 2008). Without antibiotic treatment, the death rate would be 100% within 24-h infection, while mortality of treated patients can approach 50% (Pechous et al., 2016).

Vector-borne pathogens tend to have a comparatively shorter external survival time or no lifecycle out of hosts/vectors at all, due to their highly developed adaptation to their host environment. For example, Rickettsia rickettsii cannot survive outside hosts (Azad, 2007). However, our review of the literature did not found survival times of Anaplasma phagocytophilum, Borrelia burgdorferi, and Ehrlichia chaffeensis in the environment. The reported external survival time of the three pathogens in Table 2 is actually their viability in in vitro cells or blood products under refrigeration conditions (Badon et al., 1989; Kalantarpour et al., 2000; McKechnie et al., 2000). Thus, these pathogenic bacteria could be very sensitive to environmental stresses and can only survive within host or vector cells. F. tularensis is an exception and is able to survive in the water system for up to 2 months (Mironchuk and Mazepa, 2002), which suggests that F. tularensis may have multiple transmission routes, hence, higher virulence. The vector-borne pathogens normally have high mortality rate (Titball and Sjostedt, 2003; Ryan and Ray, 2004). Although Lyme disease is reported to be rarely fatal (Kugeler et al., 2011), if left untreated, the mortality rate of the disease can be high. A similar situation also applies to A. phagocytophilum and E. chaffeensis infections. Obligate intracellular bacteria (commensal, symbiotic, and pathogenic) are all exclusively associated with hosts, with extremely weak environmental survival ability and do not contribute to host illness and mortality or only cause mild syndromes to the host. The maximal survival time falls into the scale of several hours. Finally, free-living bacteria generally have long survival times but do not have the ability to attack humans, so there is no reported mortality. For representative examples, please refer to Table 2.

Comparative analysis of the four bacterial groups in terms of external survival and mortality shows obvious differences among each other. Sit-and-wait pathogens generally have higher mortality rate and strong durability in the environment. Although vector-borne pathogens seem to be long-lived and highly pathogenic (Table 2), the survival data come from persisting in in vitro cells and blood samples. There are insufficient reports relating to their environmental survival. As for obligate intracellular bacteria, due to their high dependence on hosts, both the durability and mortality are inconsequential. Finally, free-living bacteria display effectively infinite survival times in the environment but no host mortality. The epidemiological data clearly separate the bacteria into four groups according to the durability and mortality factors. In addition, high durability and virulence are apparently associated with sit-and-wait pathogens.

When t-tests were used to compare the counts of durable-energy-reserve pathways in the sit-and-wait species versus the other groups of species, we noted that sit-and-wait species have significantly more than both host-associated (P-value 0.000105) and vector-born (P-value 0.000105) species, but the counts of energy reserve pathways in the sit-and-wait species are similar to those found in free-living species (P-value 0.545) (Figure 2). That is, sit-and-wait pathogens and free-living bacteria normally use more than one storage compound for energy metabolism, while vector-borne and intracellular bacteria seem to almost completely lose the ability to accumulate and utilize energy compounds. This expands the conclusion of Henrissat et al. (2002) that glycogen metabolism loss could be a marker for bacterial parasitic lifestyle.

Among the five energy storage reserves studies, wax ester (WE) is a neutral lipid that is accumulated as intracellular storage compounds by a few bacteria, such as Acinetobacter spp. and Rhodococcus opacus (Ishige et al., 2003). In addition, WE is also reported to be detected in the following bacteria: Moraxella, Micrococcus, Fundibacter, Neisseria, Marinobacter, Pseudomonas, Corynebacterium, Mycobacterium, and Nocardia (Ishige et al., 2003; Waltermann and Steinbuchel, 2005). A recent study shows that the final enzyme responsible for WE synthesis in bacteria is the bifunctional wax ester synthase/acyl-coenzyme A (acyl-CoA): diacylglycerol acyltransferase (WS/DGAT), which exhibits both acyl-CoA: fatty alcohol acyltransferase (WE synthesis) and acyl-CoA: diacylglycerol acyltransferase (TAG synthesis) activity (Kalscheuer and Steinbuchel, 2003). Thus, TAG and WE share the same final synthesis step. Both TAG and WE possess higher calorific value than carbohydrates and proteins, which means that when oxidized, more energy is released. That is why these reserves are able to support the environmental persistence of sit-and-wait bacteria. A recent study by Sirakova et al. (2012) shows that the dormancy of TAG and WE deficient Mtb strains via inactivating fatty acyl-CoA reductase genes is disrupted with increased cell wall permeability, high metabolic activity, and also impaired antibiotic tolerance (Sirakova et al., 2012), which indicates that the two energy storage compounds play essential roles in the environmental persistence of Mtb. Another bacterial energy reserve PHB acts as an osmotically neutral compound for carbon and energy storage (Uchino et al., 2007). Many bacteria accumulate it during stationery phase, especially when nitrogen is limited and a carbon source is abundant, which makes it a more widespread energy reserve than TAG and WE in bacteria. Sporadic studies have already proposed that accumulation of PHB can increase bacterial survival ability (Lopez et al., 1998). For example, B. pseudomallei, which does not store glycogen at all but is able to accumulate and utilize PHB, is rather durable in the external environment (Inglis and Sagripanti, 2006).

PolyP is a linear polymer of 100s of inorganic phosphate residues linked together by higher-energy phosphoanhydride bonds. It is ubiquitous in Nature and has been claimed to be present in all bacteria (Rashid et al., 2000a). In addition, it is labile and rapidly used up. Thus, there is no much literature studying its distribution in bacteria although it has been linked with bacterial mobility, biofilm development, long-term survival, and virulence (Rashid et al., 2000a,b; Kim et al., 2002). Last, but not the least, among the five energy reserves is glycogen, a homopolysaccharide consisting of glucosyl residues only. Currently, glycogen has been reported to exist in more than 50 bacteria across all bacterial species (Preiss, 2009). Theoretical study suggests that more than 300 bacterial proteomes harbor a complete glycogen metabolism pathway (Wang and Wise, 2011). According to numerous experimental studies, glycogen is strongly linked with bacterial lifestyle, persistence, and virulence, etc. (Strange, 1968; Bonafonte et al., 2000; Henrissat et al., 2002; Wilson et al., 2010).

Individually, each of the five energy storage compounds contributes to bacterial durability. With more than one ER in a bacterium, it would be strong advantage for the species to survive harsh conditions. This also explains why sit-and-wait pathogens normally harbor more than one type of energy reserves. However, only glycogen and PolyP metabolism have been shown experimentally to impact bacterial virulence at currently stage, which links durability and virulence together. As for WE, TAG, and PHB, more studies are required and similar tests should be performed in order to check their contributions to the virulence of pathogenic bacteria at molecular level.

Analysis of the second factor, ASR, provided similar groupings. The group of sit-and-wait species has a significantly larger percentage of ASR proteins compared with the host-associated species (P-value 0.0143, versus threshold of 0.05) and compared with the vector-born species (P-value 0.00871). This is understandable, as host-associated species and vector-born species are largely shielded from the external environment. Interestingly, the sit-and-wait species also have a significantly higher percentage of ASR protein when compared with the free-living species (P-value 0.0357). We speculate that this may be because sit-and-wait pathogens have not only to deal with a hostile external environment, but also host immune system attack.

The situation for the third factor, presence of virulence factors, is a little more complicated. It can be argued that the number of virulence factors do not represent the degree of bacterial virulence due to the influences of gene transcription and post-translational modification. This is to some level true. However, the number of VFs is a sufficient indicator to reflect the diverse mechanisms that bacteria are armed with to cause harm in hosts, which is an important aspect of bacterial virulence. Analyzing the percentages of virulence factors in Supplementary Table S2 as a proxy for virulence, we found that sit-and-wait pathogens have a significantly higher percentage than vector-born species (P-value 0.0442), a higher percentage (but not significantly so) than free-living species and a similar percentage to host-associated species (P-value 0.639). The first of these results is likely due to vector-born pathogens needing to evade the vector immune systems of both the vector species and the eventual host, which may constrain the range of virulence factors. The vector-borne pathogen F. tularensis appears to be an outlier in this group with its large percentage of virulence factors, which matches its mortality very well. ESM analysis also shows that F. tularensis has the complete glycogen metabolism pathway. Although it is accepted as a vector-borne pathogen, it is also reported to have long-term survival ability, which makes it a potential pathogen similar to Yersinia pestis for multi-modal transmission (Mironchuk and Mazepa, 2002). Further investigation into the composition of the virulence factors should be performed for a better understanding of this phenomenon.

Comparison of sit-and-wait pathogens and host-associated bacteria in terms of the number of virulence factors did not find significant difference. Such a result is reasonable since host-associated bacteria also need to invade and colonize hosts, which should necessitate them being equipped with a set of VFs. As for non-pathogenic free-living bacteria, it is a little bit confusing to see that they have considerable number of virulence factors in the proteomes. The reason for the paradox is that virulence factors have also been identified in non-pathogenic bacteria (Niu et al., 2013). Keen reported that non-pathogenic bacteria are able to encode VFs like types III and IV secretion systems previously considered unique to pathogens (Keen, 2012). These findings make the classical definition of virulence factors controversial. A recent study classified VFs into two categories: pathogen-specific VFs and general VFs and general VFs are common in non-pathogenic bacteria (Niu et al., 2013). In our preliminary study, we used all VFs as a single group in order to search proteomes thoroughly for all VFs. For future work, we will make our screening subtler by using different categories of VFs, which might differentiate distribution patterns of VFs between bacterial groups.

Genome reduction is the dominant theory for bacteria transiting from free-living to facultative and finally to intracellular bacteria with parasitic or symbiotic lifestyles, and has been widely observed in Nature. Examples include Mycoplasma and Rickettsia (Moran, 2002; Wolf and Koonin, 2013). For the 20 bacteria listed in Supplementary Table S2 separated by transmission modes, the analysis of bacterial proteome sizes shows obvious differences between the four groups. The average proteome sizes for sit-and-wait, vector-borne, host-associated, and free-living bacteria are 3989, 1295, 855, and 3547, respectively. Statistical analysis shows that proteome sizes of sit-and-wait pathogens are significantly different from vector-borne (P-value 0.00351) and host-associated bacteria (P-value 0.00177), but are not significantly different from those of free-living bacteria (P-value 0.543). Considering that both sit-and-wait pathogens and free-living bacteria have a long-term external survival period, they need to adapt to different niches and cope with a variety of stresses. These results are consistent with epidemiological data in terms of bacterial lifestyle. Secondly, distribution patterns of ESMs (Wilkinson, 1963) shown in Figure 2 match with transmission modes and also correlate with bacterial proteome sizes.

Theoretical analysis in this study shed light on the relationship between durability and virulence, although the small sample volume compromises its strength. By incorporating previously experimental studies, we collected several candidates that may support this linkage, which may worth further experimental exploration. For example, urease, an enzyme that catalyzes the hydrolysis of urea into carbon dioxide and ammonia, has been reported to enhance survival of pathogens that experience acidic environments, such as Yersinia pseudotuberculosis (Hu et al., 2009), Helicobacter pylori (Stingl et al., 2002; Schoep et al., 2010), Haemophilus influenzae (Murphy and Brauer, 2011), and Edwardsiella ictaluri (Booth et al., 2009), by neutralization of the acidic environments. Interestingly, urease has also been suggested to be a virulence factor that contributes to the persistence and replication in low pH intracellular compartments of macrophages (Booth et al., 2009; Murphy and Brauer, 2011). Such a linkage shows that bacterial resistance to abiotic stresses may be a contributor to virulence.

Another abiotic stress that bacteria frequently encounter is ROS, such as superoxide anion radical, hydrogen peroxide, and hydroxyl radicals. ROS causes damages to DNA, RNA, protein, and lipids (Cabiscol et al., 2000). For example, intestinal mucosa uses epithelial NADPH oxidases to generate ROS for defending against Campylobacter jejuni infection, impairing bacterial capsule formation and decreasing C. jejuni virulence (Corcionivoschi et al., 2012). Oxidative stress has also been reported to affect the pathogenicity of E. coli by decreasing the expression of all virulence factors (Hegde et al., 2008). In addition, when invading hosts, bacterial pathogens risk being phagocytosed by phagocytes that generate superoxide and other ROS in order to kill these intruders (Hassett and Cohen, 1989). Environmental factors, such as desiccation and near UV radiation, can also generate intracellular oxygen ions (Cabiscol et al., 2000). In order to survive these harsh conditions, bacteria develop a set of mechanisms for sensing ROS acutely and maintaining ROS at a certain steady level. Once bacteria sense high-level ROS, defense mechanisms are initiated for getting rid of ROS by expressing antioxidant enzymes, such as superoxide dismutase and catalase, and repairing oxidative damages, such as initiating DNA-repair systems.

Take B. pseudomallei and Salmonella typhimurium for example. Both of the bacteria are human pathogens that are capable of intracellular survival in human macrophages and can synthesize superoxide dismutase (SOD, EC = 1.15.1.1), an enzyme that changes superoxide, a compound with superoxide anion (O₂^-) generated by passing a single electron to molecular oxygen (Shvinka et al., 1979), into oxygen and hydrogen peroxide (H₂O₂). Comparisons of wild type and SOD-deficient strains of both pathogens revealed that viabilities of SOD-deficient strains decrease and presence of SOD is required for pathogen virulence (De Groote et al., 1997; Vanaporn et al., 2011). However, hydrogen peroxide is still a comparatively strong oxidizer for bacteria. Thus, in order to avoid its damage, many bacteria possess an enzyme known as catalase, which can decompose H₂O₂ into H₂O and O₂. Mandell (1975) studied catalase functions in Staphylococcus aureus and concluded that catalase has a role in protecting intraphagocytic microbes and may be an important virulence factor. Recent investigation confirmed the potential role that catalase plays for the virulence of pathogens by studying Leptospira interrogans (Eshghi et al., 2012). However, this is not a solid relationship. A study of Salmonella typhimurium showed that catalase mutant was resistant to macrophages and retained full virulence (Buchmeier et al., 1995). Another study focusing on Staphylococcus aureus showed that catalase-negative strain retained virulence in a mouse model of chronic granulomatous disease, which contradicted with the result reported previously (Messina et al., 2002). Thus, although antioxidant enzymes are widely accepted as bacterial protectors against oxidative stresses, their role as a virulence factors should be investigated more deeply.

Desiccation is also a common abiotic stress for free-living stage bacteria in drought-like conditions. For example, many nosocomial pathogens have the ability to persist on dry inanimate surfaces, such as scissors, metal plates, and fabrics etc., in order to increase their opportunities to transfer from one patient to another, which is why inanimate surfaces are often considered as the source for outbreaks of nosocomial infections (Kramer et al., 2006). However, bacterial desiccation resistance varies. Helicobacter pylori, a Gram-negative bacterium causing non-lethal gastritis, can only survive for less than 90 min on dry surfaces while Corynebacterium diphtheriae, a pathogenic bacterium that causes diphtheria with fatality rates of 36.5%, is able to persist on dry surfaces for up to 6 months (Kramer et al., 2006). This empirical evidence also indicates a possible linkage between ASR and virulence. Currently, several desiccation resistance mechanisms have been reported for bacteria. The most well-known mechanism is to use non-reducing disaccharides, trehalose and sucrose, as desiccation stress protectants. Trehalose is a disaccharide with a α-1,1-glucosidic bond linking two glucosyl residues. Sucrose is composed of glucose and fructose. Non-reducing disaccharides increase bacterial desiccation tolerance mainly by lowering the temperature of dry membrane phase transition and maintaining general protein structure (Leslie et al., 1995). In addition, these disaccharides are responsible for the formation of vitreous cytoplasmic matrix, a virulence factor for bacterial pathogenicity (Garcia, 2011). Besides, trehalose metabolism is also interlinked with glycogen and maltose pathway, the later contributing to E. coli O157:H7 colonization in the intestine of mice hosts (Jones et al., 2008). The relationship between trehalose, maltose, and glycogen could serve a good example for tackling the sit-and-wait hypothesis experimentally by linking ER, ASR, and virulence all together.