Clostridioides difficile phage biology and application

ABSTRACT

Clostridium difficile, now reclassified as Clostridioides difficile, is the causative agent of C. difficile infections (CDI). CDI is particularly challenging in healthcare settings because highly resistant spores of the bacterium can persist in the environment, making it difficult to curb outbreaks. Dysbiosis of the microbiota caused by the use of antibiotics is the primary factor that allows C. difficile to colonize the gut and cause diarrhea and colitis. For this reason, antibiotics targeting C. difficile can be ineffective at preventing recurrent episodes because they exacerbate and prolong dysbiosis. The emergence of antibiotic resistance in C. difficile also presents a significant threat. The diverse array of bacteriophages (phages) that infect C. difficile could offer new treatment strategies and greater insight into the biology of the pathogen. In this review, we summarize the current knowledge regarding C. difficile phages and discuss what is understood about their lifestyles and genomics. Then, we examine how phage infection modifies bacterial gene expression and pathogenicity. Finally, we discuss the potential clinical applications of C. difficile phages such as whole phage therapy and phage-derived products, and we highlight the most promising strategies for further development.

This review describes the current knowledge of C. difficile phage structure and replication with a focus on how this understanding can be applied to the development of novel therapies against C. difficile infections.

INTRODUCTION

Clostridium difficile, recently reclassified as Clostridioides difficile, is a Gram-positive, spore-forming anaerobic bacterium that represents a significant global health problem (Lawson et al. 2016: 95–9). Most cases of nosocomial acquired diarrhea are caused by C. difficile (Miller et al. 2011: 387–90). Patients with C. difficile infection (CDI) can present with symptoms ranging from mild to severe diarrhea, pseudomembranous colitis and fulminant colitis (Kyne et al. 1999: 107–13; Maroo and Lamont 2006: 1311–6). CDI typically arises following antibiotic usage that destabilizes the gut microbiota, allowing C. difficile to overgrow (Leffler and Lamont 2015: 1539–48). Asymptomatic carriers, animal reservoirs and food sources likely contribute to C. difficile exposure outside healthcare settings (Rodriguez et al. 2017; McLure et al. 2019: 183–7; Songer et al. 2009: 819; Songer and Uzal 2005: 528–36).

The secretion of toxin A and toxin B by C. difficile is the main driver of the symptoms of CDI (Lawson et al. 2016: 95–9; Lyras et al. 2009: 1176–9; Voth and Ballard 2005: 247–63). The pathogenicity locus (PaLoc) harbors the tcdA and tcdB genes coding for the large Clostridial toxins A and B (Voth and Ballard 2005: 247–63). It also encodes toxin regulatory elements such as the alternative sigma factor tcdR for positive transcriptional regulation (Mani and Dupuy 2001: 5844–9), tcdC for an anti-sigma factor that negatively controls transcription of the toxin (Dupuy et al. 2008: 685–9) and tcdE for the control of toxin secretion (Govind and Dupuy 2012: e1002727). Other virulence factors, such as the binary toxin (CDT) and cell surface proteins, are located outside of PaLoc, but they appear to significantly influence the virulence of C. difficile (Geric et al. 2003: 5227–32; Rupnik, Wilcox and Gerding 2009: 526–36).

Typical treatments for CDI involve the use of antibiotics like vancomycin and fidaxomicin (Khanna and Gerding 2019: 95–102). Metronidazole was previously an option for first-line treatment for non-fulminant CDI (McDonald et al. 2018: e1–e48). A review of clinical data found that fidaxomicin is the most effective and specific antibiotic available for the treatment of CDI (Nelson, Suda and Evans 2017), but all current antibiotics fail to permanently clear CDI because they disturb commensal gut microbes, leading to frequent recurrence of the disease (Hedge et al. 2008: 949). Furthermore, antibiotic resistance poses a significant immediate and long-term threat to CDI patients (Freeman et al. 2015: 248. e9–e16; Spigaglia 2016: 23–42). A recent study identified plasmid-mediated metronidazole resistance from internationally sourced strains that is likely horizontally transmissible (Boekhoud et al. 2020: 1–12). The Centers for Disease Control and Prevention (CDC) considers C. difficile as one of the top three urgent antibiotic resistance threats (Prevention 2019). Clostridioides difficile strains less susceptible to fidaxomicin and/or resistant to metronidazole, vancomycin, tetracycline and chloramphenicol have recently emerged (Peng et al. 2017: 1998–2008), demonstrating that new therapies are needed as soon as possible. As we will discuss, C. difficile bacteriophages have demonstrated promise as a novel treatment strategy against CDI. However, there is a lack of understanding regarding C. difficile phage biology and genomics, and there are numerous obstacles and technical challenges to the use of phage-based therapies. This review seeks to clarify such topics to guide future research on novel CDI treatments.

Clostridioides difficile PHAGE BIOLOGY

Morphology

Currently known phages of C. difficile belong to the order Caudovirales

Clostridioides difficile phages infecting human isolates (e.g. Goh et al. 2007: 676–85, 2005: 1079–83; Mahony, Bell and Easterbrook 1985: 251–4; Meessen-Pinard et al. 2012: 7662–70; Nale et al. 2012: e37263; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34; Sell, Schaberg and Fekety 1983: 1148–52), animal isolates (e.g. Sekulovic et al. 2014; Li et al. 2020: 2555–63) and environmental isolates (e.g. Hargreaves et al. 2013: 6236–43; Nale et al. 2016b: 968–81; Rashid et al. 2016: 310; Riedel et al. 2017: 23–8) have all been classified in the order Caudovirales. This classification refers to tailed bacteriophages with an icosahedral capsid, dsDNA and a tail of variable length (Dion, Oechslin and Moineau 2020: 1–14). A portal protein and other associated proteins facilitate the passage of the DNA from the capsid into the tail, which delivers the material into the target cell upon binding of tail fibers to receptors on the host (Iwasaki et al. 2018: 528–36). Clostridioides difficile phages fall either into the Myoviridae or Siphoviridae families of the order Caudovirales. Myoviridae, the most common family of C. difficile phages, have a long, non-flexible tail tube surrounded by a contractile tail sheath (Fokine and Rossmann 2014: e28281; Leiman et al. 2004: 419–29). Siphoviridae phages have long, flexible and non-contractile tails (Plisson et al. 2007: 3720–8). The first C. difficile phage sequenced in 2006 was a Myoviridae phage (phiCD119; Govind, Fralick and Rolfe 2006: 2568–77), while the first Siphoviridae phage was sequenced 4 years later (phiCD6356; Horgan et al. 2010: 34–43).

Although C. difficile phages are classified as either myoviruses or siphoviruses, studies have noted that myoviral and siphoviral particles display significant morphological diversity within their respective families (Hargreaves et al. 2013: 6236–43; Shan et al. 2012: 6027–34). Figure 1 summarizes the variations in size and length of the two families of C. difficile phages. Some studies further differentiate between phages within the Siphoviridae and Myoviridae families by classifying them into smaller groups such as medium myoviruses (MMs), long-tail myoviruses (LTMs), small-tail myoviruses (SMVs) and siphoviruses (SVs). These sub-groups were defined based on the physical dimensions of the phages (Hargreaves et al. 2013: 6236–43; Hargreaves 2013).

Figure 1.

Typical Morphology of C. difficile phages. Diagram of Myoviridae (M) and Siphoviridae (S) physical dimensions in nanometers (nm). All values were calculated from named C. difficile phages. The vertical double-headed arrow indicates how the Myovirus tail sheath can contract along the length of the tail. The horizontal double-headed arrow indicates the relative flexibility of the Siphovirus tail relative to its Myovirus counterpart. Diagrams are not to scale.

Lifestyle

All known C. difficile phages display a temperate lifestyle

In a typical phage replication cycle, a virion first binds to a receptor on the surface of a susceptible host and subsequently injects its genetic material inside the cell (Ofir and Sorek 2018: 1260–70). Lytic phages immediately begin producing virions, which ultimately leads to lysis of the host. Temperate phages, on the other hand, can opt for a lytic cycle or a lysogenic cycle depending on the physiological status of the host. In the latter scenario, lysogeny leads to the integration of the phage genetic material into the host chromosome to form a prophage, or the viral DNA is maintained as a circular or linear plasmid. All characterized C. difficile phages are temperate phages. Goh et al. (2005: 1079–83) hypothesized that the tendency of C. difficile to sporulate in the environment is a selective pressure that favors temperate phages. The strictly anaerobic lifestyle of C. difficile and the unlikely chances that a phage could infect a live vegetative cell outside a mammalian gut could be another reason. A temperate lifestyle thus seems to be the best option for C. difficile phages because they can persist in spores as prophages and then undergo lytic replication when the spores resume growth as vegetative cells. However, strictly lytic phages that infect other sporulating bacteria like Bacillus cereus (Lee et al. 2013: 2101–8) and Clostridium perfringens (Ha et al. 2019: 1002) have been described, so sporulation alone cannot be the only reason for the apparent absence of strictly lytic phages in C. difficile. The lack of naturally occurring, strictly lytic phages thus far has had significant impacts on the development of phage therapy for the growing problem of CDI. Phage engineering could potentially overcome the limitations of the temperate lifestyle typical of C. difficile phages. A recent study reported that removing lysogeny-related genes (integrase and CI repressor) in C. difficile phage phiCD24–2 increased lysis of the host and prevented the formation of lysogens in vitro, but for reasons that remain unclear, lysogens were detected in vivo after challenging mice with the engineered phage (Selle et al. 2020).

The attachment of C. difficile phages to the host remains poorly understood

As previously mentioned, the replication cycle of a phage begins with attachment to a specific receptor on its host, which generally occurs in a three-step fashion (Ofir and Sorek 2018: 1260–70). Phages first randomly ‘scan’ the surface of a cell to find a suitable receptor, followed by reversible binding to a phage receptor on their target host (Uchiyama et al. 2011: e26648) and then irreversible binding to either that same receptor, a different receptor, or both (Dowah and Clokie 2018: 535–42). The rate at which phages adsorb to the surface of their host varies significantly. For example, phage JD032 displayed 80% adsorption after 30 min (Li et al. 2020). In contrast, maximum adsorption for phage 56 took 30 s, while phage 41 adsorbed to 40% over the same period (Mahony, Bell and Easterbrook 1985: 251–4).

There are still gaps in our understanding of the phage attachment process, as both the identity of the receptor(s) on the host required for successful infection and the receptor-binding protein(s) on the phage need to be identified. Yet, a few recent studies provide useful insight into this topic. One study with phiCDHM1, phiCDHM3 and phiCDHM6 underlined the heterogeneity and complexity of phage attachment in C. difficile (Thanki et al. 2018: 411). Although efficient irreversible phage adsorption was required for successful infection, a high total adsorption rate (reversible + irreversible adsorption) was not always indicative of the capacity of a phage to infect a given host. For example, about 30% of phiCDHM6 was found to adsorb to one strain it did not infect, whereas about 20% of phage particles bound to the propagation hosts. Irrespective of total adsorption, at least 80% of the bound phage had to be irreversibly attached to their host for successful infection to occur (Thanki et al. 2018: 411). Interestingly, phiCDHM1 and phiCDHM6 were also able to adsorb at 5–15% to other unsusceptible species such as Escherichia coli, Pseudomonas aeruginosa and Staphylococcus aureus, suggesting that certain phages can reversibly bind to surface components common to many bacteria (Thanki et al. 2018: 411). Other phages like CD140 and JD032 also showed over 80% adsorption to their respective target strains (Ramesh, Fralick and Rolfe 1999; Li et al. 2020: 69–78), and CD140 adsorbed at about 1–30% to non-susceptible strains (Ramesh, Fralick and Rolfe 1999: 69–78). In summary, these results suggest that C. difficile phages may initially bind to a conserved phage receptor on the surface of the cell, but that additional downstream factors determine the host range of specific phages (Thanki et al. 2018: 411).

Electron microscopy imaging (TEM) of phage 56 showed that a significant number of phage particles adsorbed to a lipid-like structure upon addition of MgCl₂ (which was associated with increased lytic activity and higher phage titers) (Mahony, Bell and Easterbrook 1985: 251–4). Another study showed that phage phiHN10 was less effective at infecting susceptible strains lacking pili (Phothichaisri et al. 2018: 1701). However, the phages still adsorbed to a ‘mesh-like structure’ on the surface of bacteria that did not have pili. Further analysis suggested that this material consisted of S-layer proteins. SDS-PAGE experiments demonstrated that treating S-layer proteins with phage slowed the migration of the proteins relative to the untreated sample. The data suggest that phages directly interact with S-layer proteins. The recent work by Kirk et al on Avidocin further supports that the S-layer is a candidate phage receptor (Kirk et al. 2017: eaah6813). The Avidocins (also referred to as Diffocins) are phage tail-like bacteriocins structurally resembling Myoviridae phages (Kirk et al. 2017: eaah6813). While studying the lytic activity of the Avidocins on the epidemic C. difficile strain R20291, two spontaneous Avidocin-resistant mutants were isolated. Both mutants had lost the S-layer protein SlpA from their surface, due to a point mutation in the slpA gene that introduced a premature stop codon. Complementation of the mutant strain with a wild-type slpA gene restored susceptibility to Avidocin lytic activity. Furthermore, the authors have shown that complementation with SlpA types identified in other C. difficile strains conferred susceptibility to new Avidocins. Moreover, the authors could associate the SlpA cassette type of a given strain with susceptibility to Avidocins expressing specific tail fibers that were identified as the receptor-binding proteins (RBPs). The study also provided indirect evidence that SlpA was also used as a receptor by Myoviridae phages. Replacement of original Avidocin RBPs with homologous proteins from known prophages changed the specificity of the Avidocin towards that of the corresponding prophage (Kirk et al. 2017: eaah6813). Ongoing work from the Fortier group confirms these findings (to be published). Altogether, these studies provide evidence that SlpA is a major phage receptor in C. difficile.

Because the Avidocins are structurally very similar to Myoviridae phages, the identification of the RBPs in the Avidocins can serve as a starting point to identify corresponding RBPs in phages. Another recent study identified the potential RBPs in the C. difficile phage phiCDHS1 (Dowah 2018). Antibodies against the tail fiber proteins Gp22 and Gp29 individually prevented phage infection when co-incubated with the phage prior to adsorption, indicating that they are essential for phiCDHS1 phage infection.

Host ranges of C. difficile phages

The mechanisms behind phage-receptor interactions could partially explain why C. difficile phages have a relatively limited host range. That being said, other criteria make certain strains susceptible to phage infection, including the presence of endogenous prophages that provide phage immunity and active antiphage mechanisms such as restriction-modification, abortive infection and CRISPRs (Clustered Regularly Interspaced Short Palindromic Repeat sequences; Boudry et al. 2015). The interpretation of a phage's host range also depends on the number and genetic diversity of C. difficile isolates tested. For example, in one study, phage CD140 was only able to infect its host strain, but only twelve isolates were tested (Ramesh, Fralick and Rolfe 1999: 69–78). In another study, phage phiCD38–2 could infect 99/207 tested isolates from 11 distinct PCR ribotypes, but 79 of the susceptible isolates were of the same PCR ribotype (RT027; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Whether phages have a preference for C. difficile isolates of human or animal origin is unclear (Sekulovic et al. 2014: 2555–63), but a link between PCR ribotype and phage susceptibility has been reported (Fortier and Moineau 2007: 7358–66; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Since certain PCR ribotypes are more often associated with animal strains (e.g. ribotype 078), there could be some preference, although clear evidence is still missing. The lack of a standardized approach for testing host ranges may also contribute to the variability of reports in the literature.

The interpretation of spot test assays can be misleading if inappropriate dilutions are tested since temperate phages often tend to produce turbid lysis zones. The soft agar overlay method that involves incorporating phage dilutions is more reliable but is time-consuming when multiple strains must be tested. Moreover, the presence of a high concentration of divalent cations (e.g. 0.1–0.4 M Mg₂²⁺) in the soft agar also seems to be critical for productive infection by some but not all phages (Sekulović and Fortier 2016: 143–65). More thorough testing of phage host ranges combined with a mechanistic understanding of the molecular determinants responsible for defining host ranges would prove beneficial for developing C. difficile phage therapy. Phage cocktails could therefore be developed to target multiple strains based on the expression of specific receptors, with the possibility to prevent bacteria from developing resistance (Schmidt 2019: 581–7).

Lysogeny and the transition to the lytic cycle

Although all known C. difficile phages display a temperate lifestyle, the processes leading to lysogeny vary between individual phages. When C. difficile phages undergo lysogeny, the prophages typically integrate into the host genome (Goh et al. 2007: 676–85; Govind, Fralick and Rolfe 2006: 2568–77; Meessen-Pinard et al. 2012: 7662–70). However, not all phages behave this way. For instance, phage phiCD38–2 maintains itself as an independent plasmid when in the prophage state (Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34), and a later study found that the prophage expresses parA, a partitioning homolog, which is a gene thought to be involved in plasmid maintenance (Sekulovic and Fortier 2015: 1364–74). A total of two identical phages, phiCD211 and phiCDIF1296T, were simultaneously isolated and sequenced by different research teams, with the latter being found to form an episome in its host (Boudry et al. 2015; Wittmann et al. 2015). Most recently, phiSemix9P1, the first C. difficile phage found to encode the binary toxin gene, was also determined to form a plasmid upon lysogeny (Riedel et al. 2017: 23–8). Phage phiCD6356 also encodes a parA homolog, and although the nature of the prophage DNA molecule has not been experimentally determined, it may be episomal as well (Horgan et al. 2010: 34–43). Of interest, a study by Amy et al. (2018: 25–38) described the identification of large cryptic plasmids of 42–47 kb detected in 4.9% of 451 clinical isolates surveyed. According to the authors, these plasmids seemed to have recombined with phages since multiple phage genes were detected, along with genes generally found in plasmids. These large plasmids could be misidentified prophages, but experimental evidence confirming this hypothesis remains to be obtained.

When a phage is in the lysogenic lifestyle, various conditions can promote the transition to lytic replication depending on the phage. This prophage induction can occur either spontaneously (Mahony, Bell and Easterbrook 1985: 251–4; Meessen-Pinard et al. 2012: 7662–70) or following cellular stressors (Howard-Varona et al. 2017: 1511–20). Spontaneous phage induction from clinical isolates has been reported during in vitro experiments (Mahony, Bell and Easterbrook 1985: 251–4; Meessen-Pinard et al. 2012: 7662–70), as well as in vivo, during an infection since free phage particles were detected in the stools of CDI patients (Meessen-Pinard et al. 2012: 7662–70). Cellular stressors are typically used by researchers to induce the production of high phage titers from lysogenic strains. One study noted that exposing isolates to ultraviolet light (UV) induced a 10-fold greater release of phage particles than spontaneous induction (Mahony, Bell and Easterbrook 1985: 251–4). Most studies that characterized C. difficile phages used either UV (Sekulovic et al. 2014: 2555–63), Mitomycin C (Fortier and Moineau 2007: 7358–66; Nale et al. 2012; Li et al. 2020: e37263), or Norfloxacin (Nale et al. 2012: e37263, 2016: 968–81), and in some cases antibiotic treatment allowed up to a 4-log increase in phage release compared to non-induced cultures (Meessen-Pinard et al. 2012: 7662–70). The methods used to induce and isolate various C. difficile phages are indicated in Tables S1 and S2 (Supporting Information). Clostridioides difficile isolates can harbor multiple prophages (Amy et al. 2018: 25–38; Fortier and Moineau 2007: 7358–66; Goh et al. 2007: 676–85; Nale et al. 2012; Hargreaves 2013: e37263), and the choice of antibiotic and its concentration can influence which phages are induced from a given isolate (Nale et al. 2012: e37263; Sekulović and Fortier 2016: 143–65). Hence, it is always good to test different conditions to maximize phage recovery (Sekulović and Fortier 2016: 143–65). Earlier studies reported that antibiotic induction induced both intact phages as well as many ‘defective’ phages, characterized by malformed or incomplete tails. It now seems that these incomplete tails were in fact phage tail-like bacteriocins (Fortier and Moineau 2007: 7358–66; Goh et al. 2007: 676–85; Nale et al. 2012: e37263).

The stability of C. difficile phages is impacted by temperature and pH conditions

The replication cycle characteristics have been studied for a number of C. difficile phages using one-step growth experiments. In this type of assay, the number of newly formed phage particles is measured over the course of a single replication cycle to determine characteristics such as the latency period, rise period, release period and burst size (Zhao et al. 2016: 1–12). The latency period represents the time elapsed between the initial adsorption and the release of the first phage particles (Abedon et al. 2001: 4233–41), whereas the rise and release periods correspond to the time during which new phages are released by the lysis of the host, and rapidly accumulate in the culture supernatant (Zhao et al. 2016: 1–12). According to available data, the duration of a typical replication cycle varies greatly from one C. difficile phage to another. For example, seven phages (phiC2, phiC5, phiC6, phiCD38–2, 41, 56 and JD032) display latent periods ranging from 30 min to 2 h (Goh et al. 2005: 1079–83; Mahony, Bell and Easterbrook 1985; Li et al. 2020: 251–4; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Upon cell lysis, the typical burst size observed is 5–175 plaque-forming units (PFU) per cell (Goh, Chang and Riley 2005: 129–35; Mahony, Bell and Easterbrook 1985; Li et al. 2020: 251–4). As a general reference, the required time to clear a culture following Mitomycin C induction of prophages is typically 3–5 h with C. difficile phages (Fortier and Moineau 2007: 7358–66; Sekulović and Fortier 2016: 143–65).

Environmental conditions are known to affect the activity of C. difficile phages. In one study, virus particles were stable for 1 month if stored at 4°C, but they degraded at higher temperatures (Sell, Schaberg and Fekety 1983: 1148–52). Phages display unique ranges of tolerable temperatures, but exceeding the tolerance leads to the rapid degradation of the viral particles (Horgan et al. 2010: 34–43; Mahony, Bell and Easterbrook 1985; Li et al. 2020: 251–4; Phothichaisri et al. 2018: 1701; Sell, Schaberg and Fekety 1983: 1148–52). Besides temperature, pH can also greatly affect phage stability (Phothichaisri et al. 2018; Li et al. 2020: 1701; Ramesh, Fralick and Rolfe 1999: 69–78). Acidic conditions (below pH 3) generally inactivate phages (Ramesh, Fralick and Rolfe 1999: 69–78; Vinner et al. 2017: e0186239). Phage therapy experiments with phage CD140 confirmed that neutralizing stomach acidity was required for ensuring that the phage reached the site of infection (Ramesh, Fralick and Rolfe 1999: 69–78). Phage JD032, phiHN10, phiHN16–1, phiHN16–2, phiHN50 and phiCDKM9 demonstrated the highest stability between pH 5–11 (Phothichaisri et al. 2018; Li et al. 2020: 1701; Vinner et al. 2017: e0186239). Understanding the stability of C. difficile phages to pH and temperature fluctuations will be crucial to the development of robust phage therapy and phage-based products.

Prevalence of prophages in C. difficile genomes

Mobile genetic elements, which include prophages, account for at least 11% of the genome of the C. difficile reference strain CD630 (Sebaihia et al. 2006: 779–86). Besides offering insight into genetic diversity and the evolution of bacteria (Canchaya, Fournous and Brüssow 2004: 9–18), analyzing prophages in bacterial genomes is an important first step before inducing, characterizing, and sequencing novel phages. It is preferable for researchers to spend their limited resources only on inducing strains with likely prophages, as functional phages are typically recovered from 2 to 18% of tested isolates (Fortier and Moineau 2007: 7358–66; Goh et al. 2007: 676–85; Horgan et al. 2010: 34–43; Mahony, Bell and Easterbrook 1985: 251–4; Sell, Schaberg and Fekety 1983: 1148–52). Some authors are adapting high-throughput methods to increase efficiency (Phothichaisri et al. 2018: 1701), but these techniques are not able to predict which isolates contain inducible phages. We have to keep in mind that the induction procedure can also greatly influence the recovery rate (Sekulović and Fortier 2016: 143–65).

Typical C. difficile typing methodologies such as PCR ribotyping, multilocus sequence typing (MLST), multilocus variable number of tandem repeat analysis (MLVA), pulsed-field gel electrophoresis (PFGE), and restriction enzymes analysis (REA) do not take the presence of prophages into consideration (Killgore et al. 2008: 431–7), but PCR primers (Hargreaves et al. 2013: 6236–43; Phothichaisri et al. 2018: 1701; Shan et al. 2012: 6027–34) and bioinformatic tools (Sekulovic et al. 2014: 2555–63) have been developed that can reveal the presence of prophages (Sekulović and Fortier 2016: 143–65). Bioinformatic tools in particular are becoming more powerful for identifying putative prophage genomes. Tools such as PHASTER (Arndt et al. 2016: W16–W21), PHAST (Zhou et al. 2011: W347–W52), PhiSpy (Akhter, Aziz and Edwards 2012: e126) and several others as described in a recent review (Arndt et al. 2019: 1560–7) can be used to identify potential prophage regions (Hargreaves et al. 2015: 1842–55) or to annotate phage genomes using the corresponding phage protein databases (Garneau et al. 2018). Table S3 (Supporting Information) shows the prevalence of putative prophage regions identified using PHASTER in representative C. difficile ribotypes and toxinotypes including RT012 (Groß et al. 2018: 1), RT027 (Groß et al. 2018: 1; Guerrero-Araya et al. 2020: mgen000355; Stabler et al. 2006: 7297–305; Suzuki et al. 2017: 70; Williamson et al. 2019), RT078 (He et al. 2010: 7527–32; Monot et al. 2015; Li et al. 2020: 15 023; Riedel et al. 2017: 23–8; Sangster, Hegarty and Stewart 2015: 96–103; Wu et al. 2019: 1–13), RT106 (Carlson et al. 2020: 102 142; Kociolek et al. 2018: 1222–9; Ozer et al. 2017), toxin A⁻B⁺ (Du et al. 2014: 3264–70; He et al. 2010: 7527–32; Stabler et al. 2006: 7297–305) and non-toxigenic strains (Brouwer et al. 2012; Darling et al. 2014: 1–4; Janssen et al. 2016: 652–6; Pereira et al. 2016: 1–7; Riedel et al. 2017: 23–8; Zhang et al. 2015: 49–52). Varying ribotypes within a group may contribute to different patterns of prophage carriage. The results of this analysis show that there are a variety of prophage carriage patterns in different strains of C. difficile.

The large number of incomplete regions relative to complete prophages in some strains demonstrates that C. difficile strains contain many phage-derived genes even in the absence of functional prophages. In most cases, incomplete prophages were smaller than the 30 kb minimum length typically observed with C. difficile prophages (see Table S2, Supporting Information). Not all complete prophage regions called by PHASTER are necessarily functional. The built-in BLAST results and draft annotation allow researchers to determine which complete prophage regions contain an integrase (denoted as C_int in Table S3, Supporting Information). Typically, studies use the presence of an integrase to exclude false-positive search results (Fan et al. 2016: e2012; Fan et al. 2014: 243).

Characterized and sequenced C. difficile phages

Sell, Schaberg and Fekety (1983: 1148–52), published one of the first studies on the characterization of C. difficile phages and proposed that phage susceptibility patterns could be used as a typing scheme to study CDI outbreaks. This study and subsequent ones by other groups reported the induction of different phages from C. difficile isolates and tested their host range as part of a phage typing program (Dei 1989: 351–4; Mahony, Bell and Easterbrook 1985: 251–4; Mahony et al. 1991: 1873–9; Sell, Schaberg and Fekety 1983: 1148–52). It was eventually concluded that phage typing of C. difficile isolates was not a suitable strategy due to the variability of the results, bacterial resistance and narrow phage host ranges. Although these early studies established most of the methods for studying C. difficile phage and prophage induction, the isolated phages have not been characterized in detail. More recent studies have greatly expanded the catalogue of characterized C. difficile phages while providing a far more in-depth analysis of their characteristics. Yet, only a limited number of C. difficile phages have been sequenced and made available in public repositories since the first genome was released in 2006 (Govind, Fralick and Rolfe 2006: 2568–77). In the absence of a dedicated database for C. difficile phages, a library of reported and characterized C. difficile phages is given in Table S1 (Supporting Information), while characterized phages that have been sequenced are reported separately in Table S2 (Supporting Information).

Clostridioides difficile phage genomes

Basic genome characteristics

There is a limited but growing catalogue of sequenced genomes from phages infecting human isolates (e.g. Goh et al. 2007: 676–85; Meessen-Pinard et al. 2012: 7662–70; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34), animal isolates (e.g. Sekulovic et al. 2014; Li et al. 2020: 2555–63) and environmental isolates (e.g. Nale et al. 2016b: 968–81; Rashid et al. 2016: 310; Riedel et al. 2017: 23–8) of C. difficile, for example. A phylogenetic tree of sequenced C. difficile phages is illustrated in Fig. 2A. All of the sequenced phages that are catalogued in Table S2 (Supporting Information) have a dsDNA genome with a GC from 28.7 to 29.9% (Horgan et al. 2010: 34–43; Mayer, Narbad and Gasson 2008; Li et al. 2020: 6734–40; Riedel et al. 2017: 23–8), which is similar to the average GC content of C. difficile (29.06%; Sebaihia et al. 2006: 779–86). The typical length of the C. difficile phage genome varies from 32 000 bp (phiCDHM11; Hargreaves and Clokie 2015: 2534–41) to 56 606 bp (phiSemix9P1; Riedel et al. 2017: 23–8) for example, but phages with much larger genomes have recently been characterized. The first of such phages uncovered was phiCD211/phiCDIF1296T, with a 131 326 bp genome (Boudry et al. 2015; Wittmann et al. 2015; Garneau et al. 2018). Large phages may be more common than previously thought, as 5% of isolates analyzed in one study displayed significant identity to phiCD211 (Garneau et al. 2018), and more large phages have been characterized recently (Table S2, Supporting Information; Ramírez-Vargas, Goh and Rodríguez 2018: 26).

Figure 2.

Phylogenetic tree of sequenced phages and conserved modular layout of C. difficile phage genomes. Phage genome sequences were aligned using ClustalW before constructing a Maximum likelihood tree in MegaX software (A). Genome alignments of representative phages from each cluster in A, including a group of singletons (dotted black box) were performed using Artemis Comparison Tool (ACT) with a minimum score cutoff of 215 and a minimum percentage identity cutoff of 76%. (B). Red bars indicate strong sequence alignment, whereas blue implies weaker alignment. Colored regions on the genomes represent the most common functional modules for DNA packaging (yellow), head packaging (red), tail structure (blue), lysis (grey), lysogeny control (orange) and DNA replication, transcription and recombination (green). Some phages also show modules for the binary toxin locus (black) and integration and excision (purple). Areas with two colors indicate modules with combined functions. Sequences for each phage were retrieved from GenBank (see accession numbers in Table S2, Supporting Information). Module classifications can be found in the source publications provided for each phage in Table S2 (Supporting Information), or they were determined by visualizing the GenBank genome annotation in Artemis Software (specifically for phiCD24-1). The annotation of phage JD032 was reported with the terminase gene as the last gene in the genome rather than the first gene, so the sequence was reverse complemented in ACT to correct the order of the modules. The analysis demonstrates that C. difficile phages show conserved synteny even with low sequence identity between individual phages. For more closely related phages, genomes tend to show the alignment between similar or adjacent modules.

Clostridioides difficile phage genomes are modular

A unifying feature of all C. difficile phages, like for phages of other species as well, is a modular layout, meaning that different regions of the genome encode proteins that have related functions. The order and contents of the genomes provide insights into phage biology. In the case of phiCD119, the first C. difficile phage genome sequenced and analyzed, the authors classified the genes into modules for head structural components and DNA packaging, tail structural components, cell lysis, lysogeny control and DNA replication and modification (Govind, Fralick and Rolfe 2006: 2568–77). Subsequent studies have also highlighted the modular layout of phage genomes, but sometimes with different module nomenclature. Based on the available studies, six basic categories have been described; DNA packaging, head/capsid structure, tail structure, cell lysis, lysogeny control and DNA replication, transcription, recombination and modification. Figure 2B provides examples of the genome synteny among C. difficile phages and even distantly related phages have a conserved modular layout of their genomes. Modular layouts are not unique to C. difficile phages. For example, phiCD6356 has a similar genome structure to Siphoviridae phages of other Gram-positive bacteria with low GC contents (Desiere et al. 2001: 240–52; Horgan et al. 2010: 34–43; Lucchini, Desiere and Brüssow 1999: 427–35).

It is important to note that modular layouts do not necessarily correlate with sequence similarity. The large (over 100 kb) C. difficile phages such as phiCD211, phiCD5763 and phiCD5774 and others showed a consistent modular order, but they shared only 53% sequence identity (Ramírez-Vargas, Goh and Rodríguez 2018: 26). In addition, these large phages have a slightly different genome organization, with lysogeny-related genes inserted between capsid and tail genes, whereas in most other C. difficile phages, these genes are located immediately downstream of the lysis module (Ramírez-Vargas, Goh and Rodríguez 2018: 26). A total of two obstacles to further understanding C. difficile phage genomics are the frequency of genes with unassigned functions and the lack of experimental evidence to confirm the function of annotated genes. It is common to report less than half of predicted ORFs with a designated function (e.g. Goh et al. 2007: 676–85; Govind, Fralick and Rolfe 2006: 2568–77; Horgan et al. 2010: 34–43). When functions are called, they are typically assigned based mainly on amino acid homology with proteins and/or protein domains available in databases. There is little experimental evidence to confirm the function of most predicted proteins in each module. Despite these limitations, here we will examine trends in the genomic structure of various sequenced C. difficile phages.

DNA packaging

By convention, phage genomic maps and annotations typically begin with the terminase (Ramírez-Vargas, Goh and Rodríguez 2018: 26), whose gene products are involved in cutting and packaging the DNA into the capsid during phage assembly (Catalano and CMLS 2000: 128–48; Rentas and Rao 2003: 37–52). Most phage genomes contain two terminase genes encoding a small subunit (terS) followed by a large (terL; Goh et al. 2007: 676–85; Govind, Fralick and Rolfe 2006: 2568–77; Hargreaves, Kropinski and Clokie 2014: e85131; Horgan et al. 2010: 34–43; Mayer, Narbad and Gasson 2008: 6734–40; Meessen-Pinard et al. 2012: 7662–70; Rashid et al. 2016: 310; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34), but some phages have only one annotated terminase gene (Horgan et al. 2010: 34–43; Li et al. 2020). Despite the varying nomenclature between phages with separate modules for DNA packaging and structural genes (e.g. phiCD38–2; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34) and phages that combine these modules into one category (e.g. phiCD119; Govind, Fralick and Rolfe 2006: 2568–77), C. difficile phages generally encode a portal gene and capsid-related genes following the terminase subunits. Figure 3 illustrates the phylogenetic relationships between the TerL proteins of the phages listed in Table S2 (Supporting Information). Previous studies have demonstrated that phages with closely related TerL tend to have similar packing strategies (Casjens and Gilcrease 2009: 91–111; Rashid et al. 2016: 310). The portal protein also plays a role in DNA packaging, as it allows the phage DNA to exit the capsid during the infection process and later to allow the newly synthesized DNA to enter a new capsid of progeny phage (Dröge et al. 2000: 117–32).

Figure 3.

Phylogeny of C. difficile phage terminase large subunits (TerL). The amino acid sequence of the terL gene from each phage in Table S2 (Supporting Information) were aligned using MUSCLE in MegaX software, which was then used to generate a maximum-likelihood (ML) tree. All sequences were retrieved from GenBank (see Table S2, Supporting Information for accession numbers). Phage names are color-coded to indicate their DNA packaging strategy as predicted by Rashid et al. (2016: 310); P22-like headful (blue), 3′ extended cohesive ends (COS, red) and unknown (black). Phages with experimental evidence in support of the predicted packaging mechanism are indicated by a single asterisk (*). Phages that have been reported since the analysis performed by Rashid et al. (2016: 310) are indicated by two asterisks (**). Phage phiCD24-1 (***) was predicted by Rashid et al. (2016: 310) to have a P22-like headful strategy, but a more recent study by Garneau et al. (2017: 1–10), was unable to determine its packaging mechanism. Closely-related phages as evidenced by TerL protein comparison are likely to have the same DNA packaging strategy, which suggests that JD032 may use a 3′ extended COS mechanism while CDSH1 may use a P22-like headful strategy, for example.

The mode of phage DNA encapsidation and the type of genomic DNA termini has not been systematically determined experimentally for all C. difficile phages. Depending on the type of terminase and the mechanism of phage DNA packaging, distinct genome extremities are created during the encapsidation process. Most phages fall within one of four main categories: (i) cos-type (phages with 3′ or 5′ single-stranded cohesive termini), (ii) pac-type or ‘headful’, (iii) DTR (direct short or long terminal repeats) or (iv) Mu-like (random encapsidation of phage DNA with host DNA fragments at the extremities). Until now, C. difficile phages have been classified into the cos-type, pac-type, or DTR (short and long) categories (Garneau et al. 2017: 1–10; Horgan et al. 2010: 34–43; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34), and some phages could not be assigned to any of the known categories. Experimental determination of the type of genomic termini and hence, mode of DNA encapsidation, can be labor-intensive and bioinformatic tools have recently been developed to facilitate this process. One of them, called PhageTerm (Garneau et al. 2017: 1–10), uses raw Illumina sequencing reads to calculate biases in sequencing coverage to determine the type of physical genomic termini. Of note, a phylogenetic analysis of terL, the gene encoding the large terminase subunit, was used to predict the mode of encapsidation of a number of C. difficile phage genomes, and the results were generally in agreement with those later predicted with PhageTerm (Garneau et al. 2017: 1–10; Rashid et al. 2016: 310).

Structural module

Downstream of the DNA packaging genes, C. difficile phage genomes encode numerous proteins that make up the head and tail structures. Annotation and nomenclature of these genes vary from one phage to another and gene functions such as portal protein, major capsid protein, head protein, morphogenesis proteins, scaffold proteins, head-tail adaptors and/or head-tail joining proteins have been reported (Goh et al. 2007: 676–85; Horgan et al. 2010: 34–43; Mayer, Narbad and Gasson 2008; Li et al. 2020: 6734–40; Meessen-Pinard et al. 2012: 7662–70; Rashid et al. 2016: 310; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Adaptor proteins in tailed phages interact with the portal protein to prevent genome loss from the capsid (Chaban et al. 2015: 7009–14). Scaffold proteins aid in phage capsid assembly during the construction of progeny phage (Dokland et al. 1997: 308–13; Moore and Prevelige Jr 2002: 10245–55).

After the capsid-related genes, C. difficile phages encode tail structural proteins. While some phages like phiCDHM1 have their head and tail genes reported in the same module (Hargreaves, Kropinski and Clokie 2014: e85131; Ramírez-Vargas, Goh and Rodríguez 2018: 26), most annotations represent tail-related genes under a separate category (Goh et al. 2007: 676–85; Horgan et al. 2010: 34–43; Mayer, Narbad and Gasson 2008; Li et al. 2020: 6734–40; Meessen-Pinard et al. 2012: 7662–70; Rashid et al. 2016: 310; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Because many gene products have no homologous counterparts in databases, some genes have been assigned arbitrarily to a functional module based on their location. For example, a gene of unknown function located between two genes predicted to code for tail proteins will generally be included as a tail-related gene, and so on (Ramírez-Vargas, Goh and Rodríguez 2018: 26). However, in absence of a predicted function or homologous gene in databases, it cannot be excluded that a gene located within a cluster of structural genes could code for another unrelated function. Such gene could have been acquired by horizontal transfer, for example (Taylor et al. 2019: 1–31).

Clostridioides difficile phages, like other tailed phages, encode a tape measure protein whose gene length is directly related to the size of the tail (Katsura 1987: 73–5). It is often one of the easiest genes to identify in a Siphoviridae or Myoviridae phage genome, because of its size and location. Myoviridae phages in particular can sometimes be distinguished by the presence of a tail sheath protein that forms the contractile structure around the tail core (Fokine and Rossmann 2014: e28281). PhiC2, phiMMP02, phiMMP04, phiCDHM1, phiCDKM9 and phiCDKM15 (Goh et al. 2007: 676–85; Hargreaves, Kropinski and Clokie 2014: e85131; Meessen-Pinard et al. 2012: 7662–70; Rashid et al. 2016: 310) are all Myoviridae phages with annotated tail sheaths, while phiCD119 and phiCD27 (Govind, Fralick and Rolfe 2006: 2568–77; Mayer, Narbad and Gasson 2008: 6734–40) are known Myoviridae phages that do not have annotated tail sheath genes. However, considering the frequent lack of homologous genes in public databases, this is not surprising. Myoviridae phages of C. difficile often encode tail genes that are related to those of the Bacillus subtillis defective prophage PBSX (Myoviridae) such as xkdM (tail tube protein (Jin et al. 2014: 739–52)), xkdK (tail sheath (Jin et al. 2014: 739–52)), xkdP (lysin (Buist et al. 2008: 838–47)) and xkdQ (hydrolase (Meessen-Pinard et al. 2012: 7662–70)) among others.

Lysis module

The lysis module, as the name implies, comprises genes whose products allow lysis of the host cell membrane at the end of the replication cycle to release progeny phages. Nonfilamentous dsDNA bacteriophages generally utilize a holin-endolysin pair to lyse their host (Young, Wang and Roof 2000: 120–8; Young and Reviews 1992: 430–81) and most C. difficile phage genomes encode these genes close to one another in their lysis module. Typically, phages encode a holin, a small liposoluble protein that helps permeabilize the host membrane, followed by an endolysin, whose function is to degrade the peptidoglycan cell wall (Young and Bläsi 1995: 191–205; Young and Reviews 1992: 430–81). All known endolysin genes from C. difficile phages contain an N-acetylmuramoyl-L-alanine amidase domain. A phylogenetic analysis allowed for the clustering of phages based on the similarity of their endolysin, and Siphoviridae phages, Myoviridae phages and large phages (e.g. phiCD211-like) grouped in distinct clusters (Rashid et al. 2016: 310). The large phages phiCD211, phiCD5763 and phiCD5774 were determined to encode a holin, endolysin and amidases in their lysis module, and the holin was generally located after the endolysin (Ramírez-Vargas, Goh and Rodríguez 2018: 26). In the case of phiCD211, a protein with a mannosyl-glycoprotein endo-beta-N-acetylglucosaminidase-like domain was found to be encoded between the endolysin and holin genes (Garneau et al. 2018).

Lysogeny control module

Clostridioides difficile phages, like all temperate phages, encode specific genes to maintain lysogeny and mediate their integration and excision from the host genome. A review by Feiner et al. (2015: 641–50) provides more details on the regulatory mechanisms of phage lysogeny. To summarize, phages infect a host and use an integrase to incorporate their genome within the host's chromosome. To transition from the lysogenic to the lytic cycle, a phage typically uses an integrase/recombinase and an excisionase or host integration factors to extricate its genome from that of the host. Phage integration occurs at a particular attachment site denoted as attB in the bacterial genome and attP on the phage genome. In the case of C. difficile phages like phiCD119 and phiC2 (Goh et al. 2007: 676–85; Govind, Fralick and Rolfe 2006: 2568–77), the integrase is located close to the attP site, which is a common layout for temperate phages (Zimmer, Scherer and Loessner 2002: 4359–68). Integrase/recombinase genes may be located in the DNA module rather than the lysogeny control module as in the case of phage JD032 (Li et al. 2020). Of note, a few prophages like phiCD38–2 (Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34) and phiCD211/phiCDIF1296T (Garneau et al. 2018) are maintained as independent plasmids, even though they encode predicted integrases in their genome. These integrases could be non-functional, forcing these prophages to replicate as episomes, but this remains to be investigated.

In the prototypical temperate phage lambda, the CI repressor prevents the initiation of the lytic cycle while the Cro repressor inhibits lysogeny as part of a complex genetic switch that regulates the phage lifestyle (Lee, Lewis and Adhya 2018: 58–68). The lytic/lysogenic switch in C. difficile phages has not been investigated to the same level, but genes coding for typical phage repressors, including CI and Cro have been identified in several phages. That being said, based on homology searches only, genes coding for CI and Cro repressors sometimes cannot be clearly identified. Moreover, when identified, they are not systematically colocalized with other regulators (e.g. phiCD38–2; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Other phages like phiC2, phiMMP02 and phiCD27 lack predicted CI or Cro repressors (Goh et al. 2007: 676–85; Mayer, Narbad and Gasson 2008: 6734–40; Meessen-Pinard et al. 2012: 7662–70). Instead, they encode multiple repressors and antirepressors with unknown or highly speculative functions (Goh et al. 2007: 676–85; Mayer, Narbad and Gasson 2008: 6734–40; Meessen-Pinard et al. 2012: 7662–70).

Previous studies have noted that genes such as ParA, ParB and ParS help stabilize DNA during cell division (Abeles, Friedman and Austin 1985: 261–72; Yamaichi and Niki 2000: 14 656–61). Phages that form extrachromosomal plasmids like phage P1 use ParA/ParB to maintain their plasmid during lysogeny (Howard-Varona et al. 2017: 1511–20). PhiSemix9P1 and phiCD38–2 prophages are maintained as plasmids, and the authors suggest that ParA is an important protein that facilitates this lifestyle (Riedel et al. 2017: 23–8; Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). However, phiC2 contains ORFs with sequence homology to ParA and ParB, yet the prophage is integrated (Goh et al. 2007: 676–85). Therefore, it is impossible to infer the mode of prophage maintenance based only on the presence/absence of a partition gene homolog, like we cannot presume that a prophage integrates the chromosome of its host simply because it encodes an integrase.

DNA replication, transcription, recombination and modification module

Due to the broad scope of the functions encoded in this module, diverse sets of genes have been identified in this region that are predicted to participate in the phage replication process, transcription control, recombination and DNA modification. To simplify reading, we will simply refer here to the DNA replication module. In phage phiCD119, phiCD211 and phiCD27, a DNA methylase is predicted to participate in DNA modification (Govind, Fralick and Rolfe 2006; Garneau et al. 2018: 2568–77; Mayer, Narbad and Gasson 2008: 6734–40; Wittmann et al. 2015). In phages, DNA methylases are most likely involved in the modification of the phage DNA in order to avoid enzyme restriction by the host (Adams and Burdon 1985: 73–87). Resolvases are other genes encoded by some phages, such as phiCD119 (Govind, Fralick and Rolfe 2006: 2568–77), that are thought to carry out roles related to phage DNA integration (Christiansen et al. 1996: 5164–73; Mauritzen et al. 2020: 730) and packaging the DNA into the capsid (Mashal, Koontz and Sklar 1995: 177–83). Phages encode a variety of proteins related to DNA recombination (Mayer, Narbad and Gasson 2008: 6734–40). For example, phiC2 encodes a RusA-like protein (Goh et al. 2007: 676–85), which is hypothesized to be a phage-derived DNA endonuclease/recombinase that can modify DNA at Holliday junctions (Mahdi et al. 1996: 561–73; Sharples, Bolt and Lloyd 2002: 549–59). This phage also encodes ERFs (essential recombination function proteins) thought to be involved with the circularization of the phage genome (Iyer, Koonin and Aravind 2002: 8). To alter the transcription of the host, phages like JD032 encode RNA polymerase sigma factors as well as ssDNA-binding proteins, all of which aid the virus in taking over host gene expression in the early stages of infection (Li et al. 2020).

It remains to be investigated how these various proteins of the DNA replication module impact the course of infection. Studies on Klebsiella phages demonstrated that the prevalence of homing endonucleases, for example, partially explained a higher resistance to DNA restriction digestion and a broader host range relative to phages that lacked such enzymes (Maciejewska et al. 2017: 673–84). Performing similar analysis with C. difficile phages could reveal novel insights about phage infection processes and suggest which phages are best equipped to avoid host antiphage defenses in the context of therapeutic applications.

THE ROLE OF PHAGES IN THE VIRULENCE AND BIOLOGY OF C. difficile

Phage-mediated alteration of host gene expression

Prophages in C. difficile can have a diverse impact on host gene expression at both the mRNA and protein levels. These effects alter the virulence and biology of the host. Despite the limited number of studies that examined prophage-mediated gene regulation in C. difficile, it seems clear that phages have the potential to influence their respective hosts in unique ways. Phages can even interact with existing prophages, as phiCD38–2 upregulated several genes involved in DNA replication and gene regulation located in phi027, another prophage conserved across ribotype 027 strains (Sekulovic and Fortier 2015: 1364–74). Some of the known and predicted effects of prophages on the virulence of C. difficile are discussed below and a summary is provided in Table 1. Given the extensive analysis performed by some authors, this list is non-exhaustive.

Table 1.

Effects of C. difficile phages on host gene expression.

Phage	Genes upregulated	Genes downregulated	Differential regulation	Reference
phiCD119	-	Transcription-level: PaLoc: tcdA, tcdB, tcdR, tcdE, tcdC (via phage-encoded repressor repR)	-	Govind et al. (2009); Revathi, Fralick and Rolfe (2011)
phiC2, phiC5, phiC6, phiC8	Protein-level: PaLoc: tcdB*	-	Protein-level: PaLoc: tcdA*	Goh, Chang and Riley (2005)
phiCD27	-	Protein-level: PaLoc: tcdA, tcdB	-	Meader et al. (2010, 2013)
phiCD38–2	mRNA-level*: PaLoc: tcdA, tcdB, tcdC, tcdE, tcdR, Antiphage: cwpV Metabolism: PTS glucose	mRNA-level: Metabolism: PTS fructose, PTS, sorbitol	-	Sekulovic, Meessen-Pinard and Fortier (2011); Sekulovic and Fortier (2015)
	Protein-level* PaloC: tcdA, tcdB	Other: transcriptional regulators, ABC transporters, holin, hypothetical protein, two- component system
phiCDHM1	-	-	mRNA-level: Quorum sensing: AgrB, AgrC (qualitative)

Differential regulation refers to cases in which genes are upregulated or downregulated depending on the phage and strain of C. difficile tested. All strain-specific effects are labelled with an asterisk (*).

Influence of prophages on toxin production and PaLoc gene expression

As previously mentioned, the secretion of toxin A and toxin B by C. difficile is the primary driver of the clinical symptoms associated with CDI (Lawson et al. 2016: 95–9; Lyras et al. 2009: 1176–9; Voth and Ballard 2005: 247–63). Hypervirulent strains of C. difficile sometimes also encode a binary toxin (CDT) (Gerding et al. 2014: 15–27). An evolutionary relationship between phages and C. difficile toxins has been proposed. For example, the tcdA gene is homologous to ORFs of Clostridium tetani phage CT2 (Canchaya et al. 2003: 238–76) and ORFs of the Lactobacillus casei phage A2 (Proux et al. 2002: 6026–36), suggesting that toxin A may have phage origins (Goh, Chang and Riley 2005: 129–35). Some authors suggest that the close relationship between sequences of tcdE, tcdA and tcdB suggests that phages are the source of the PaLoc in C. difficile (Goh, Chang and Riley 2005: 129–35).

Although a prior review noted that multiple Clostridium species undergo toxigenic conversion via temperate phages (Fortier 2017: 169–200), this has not been demonstrated in C. difficile so far and no toxin A or toxin B genes have been reported in the genomes of C. difficile phages. Phage phiC2 displayed low levels of homology to tcdB (unpublished data), however, a tcdB probe failed to hybridize with the DNA (Goh, Chang and Riley 2005: 129–35). Still, we cannot exclude the possibility that a phage could carry tcdA or tcdB, considering that the C. difficile phage phiSempix9P1 was recently found to encode a complete binary toxin locus comprising cdtR, cdtA and cdtB (Riedel et al. 2017: 23–8). Only the expression of cdtR was confirmed by RT-PCR so the functionality of this binary toxin remains to be demonstrated experimentally. However, the potential of phages to encode and regulate toxins suggests that toxigenic conversion is possible. Figure 4 provides a visual representation of instances in which phage lysogeny has been associated with changes in toxin production at the mRNA level, the protein level and/or during secretion.

Figure 4.

Effect of lysogeny on toxin production by C. difficile. Uncertain changes of toxin regulation during transcription (txn), translation (tsl) and secretion are noted with a question mark. Upregulation is represented by an upward arrow, downregulation is represented as a downward arrow and no change is shown as a horizontal line. These effects are further described in Table 1. (A) phiCD119 downregulates toxin production through the transcriptional repressor RepR. (B) In selected strains, phiC8 upregulates toxin A transcription (double-dotted arrow) while toxin B expression (not shown) is unaffected. (C) In other strains, phiC8, phiC6 and phiC2 lysogeny is associated with higher levels of secreted toxin B, while toxin A expression (not shown) is unaffected. (D) phiC6 downregulates toxin A transcription, but not protein levels. Toxin B protein levels are increased outside the cell while transcription remains unchanged. (E) phiCD27 lysogeny is associated with less extracellular toxins. (F) phiCD38-2 lysogeny is associated with an increase in transcription and secretion of both toxins.

In C. difficile, toxins have been shown to be upregulated or downregulated following stable infection with certain phages, with different outcomes depending on the specific phage and host (Goh, Chang and Riley 2005: 129–35). Some reports showed that phages can downregulate toxin A and/or toxin B. For example, phiCD119 was found to reduce toxin production at the protein level both in vitro (Govind et al. 2009: 12037–45) and in vivo (Revathi, Fralick and Rolfe 2011: 125–9). The phage accomplishes this by encoding a repressor called RepR, which binds to a DNA region upstream of tcdR, leading to reduced transcription of the PaLoc genes tcdR, tcdB, tcdE, tcdA and tcdC (Govind et al. 2009: 12037–45). Phage therapy in a batch fermentation model with phiCD27 was found to reduce toxin A and toxin B levels (Meader et al. 2010: 549–54). A later experiment with phiCD27 in a human colon model determined that this effect was due to the direct effects of the phiCD27 prophage in lysogens that formed during the experiment (Meader et al. 2013: 25–30).

Other phages were shown to upregulate toxin genes. PhiC2, phiC5, phiC6 and phiC8 upregulated toxin B production at the protein level in most strains tested, while toxin A protein levels were unaffected (Goh, Chang and Riley 2005: 129–35). Although tcdB and tcdA transcription are typically co-regulated in C. difficile, the results showed that, in two cases, the transcription of either toxin A or toxin B was altered while the other remained unaffected. Also, toxin B levels increased at the protein level in certain lysogens of all phages despite no change in tcdB transcription. This suggests that phages have a greater effect on toxin production and/or secretion than transcription alone. PhiCD38–2 upregulated toxin A and B, both at the mRNA and the protein levels in RT027 isolates (Sekulovic, Meessen-Pinard and Fortier 2011: 2726–34). Moreover, the transcription of all five PaLoc genes was upregulated. JD032 was shown to alter the expression of four cell surface proteins, among which cwp2 (Li et al. 2020) was found to interfere with toxin A release and to promote cell adhesion in the reference strain CD630 (Bradshaw et al. 2017: 2886–98).

The PaLoc also encodes tcdE, whose product was shown to facilitate toxin secretion (Govind and Dupuy 2012: e1002727; Govind et al. 2009: 12037–45; Tan, Wee and Song 2001: 613–9). TcdE shows significant similarity to phage holins (Tan, Wee and Song 2001: 613–9), which create holes in the plasma membrane of the host at the end of the lytic cycle to allow endolysins to reach the peptidoglycan layer from within (Wang, Smith and Young 2000: 799–825; Young, Wang and Roof 2000: 120–8). The interactions between holins and holin inhibitors are known to influence the timing of when such holes form, which may influence how effectively a phage can lyse its host (Young, Wang and Roof 2000: 120–8). However, the relationship between holins, endolysins and cell lysis has not been investigated in C. difficile. The activity of the holin-like TcdE was shown to be required for efficient toxin secretion (Govind and Dupuy 2012: e1002727). PCR and Southern blotting revealed the presence of tcdE-like genes in the genomes of phiC2, phiC5 and to a lesser extent in phiC8 (Goh, Chang and Riley 2005: 129–35). Interestingly, only the Myoviridae phages contained the tcdE-like genes. It is unclear whether C. difficile phages can encode TcdE and/or manipulate holin expression in host cells, thereby affecting toxin secretion (Goh, Chang and Riley 2005: 129–35). For example, phiCD38–2 downregulated a holin but upregulated toxin transcription in R20291 lysogens (Sekulovic and Fortier 2015: 1364–74).

Effects on surface proteins

CwpV is a phase-variable protein expressed in about 5% of C. difficile cells under normal in vitro growth conditions (Emerson et al. 2009: 541–56). Although the protein sequence varies between strains, in particular in the C-terminal region, its core structure and characteristics are conserved (Reynolds et al. 2011: e1002024). In one study, phiCD38–2 was found to upregulate the expression of CwpV in R20291 lysogens (Sekulovic and Fortier 2015: 1364–74). More specifically, the expression pattern in the lysogen was reversed relative to normal conditions; about 95% of cells expressed CwpV while 5% did not. The observed change in transcription as evidenced by RNAseq was therefore caused by a greater proportion of cells expressing CwpV rather than an increase in transcription of the gene per se (Sekulovic and Fortier 2015: 1364–74).

The RecV recombinase was required to catalyze the inversion of the genetic switch upstream of the cwpV gene (Reynolds et al. 2011: e1002024; Sekulovic and Fortier 2015: 1364–74). The C-terminal domain of the protein CwpV was also found to promote bacterial aggregation (Reynolds et al. 2011: e1002024), but was later found to also confer antiphage protection when expressed at the surface of C. difficile cells (Sekulovic et al. 2015: 329–42). The exact mechanism by which phiCD38–2 alters the frequency of recombination of this genetic switch remains to be elucidated. Nevertheless, the hijacking of a bacterial surface protein by a prophage to provide superinfection exclusion represents a novel defense mechanism. Yet, another C. difficile phage, JD032, also altered the expression of cwp2 and cwp66 (Li et al. 2020), which are cell wall proteins involved in bacterial adhesion (Péchiné, Denève-Larrazet and Collignon 2016: 91–101; Waligora et al. 2001: 2144–53). Although no experiments have tested bacterial adhesion in response to phage infection, these examples suggest that C. difficile phages can manipulate the surface of their host cell in several ways.

Prophages and quorum sensing

Phage phiCDHM1 was the first C. difficile phage known to encode quorum sensing genes, which the authors hypothesized underwent horizontal transfer from the host genome (Hargreaves, Kropinski and Clokie 2014: e85131). A region of the phage genome identified as the lysogeny module was found to encode AgrD (a pre-peptide of autoinducing peptide), AgrB (an enzyme that converts the pre-peptide into the autoinducing peptide) and AgrC (a histidine kinase that activates the response regulator). Because phiCDHM1 does not encode the response regulator AgrA, the exact role of these genes in quorum sensing in C. difficile remains to be clarified. Nevertheless, the agrB and agrC genes were shown to be expressed, suggesting that the resulting proteins could possibly complement the endogenous Agr system previously identified in C. difficile (Martin et al. 2013: 3672–81). Another important virulence mechanism also known to promote antibiotic resistance is the formation of biofilms, and prophages of C. difficile were found to play a role in this process (Slater et al. 2019: 1–15). One of the genes that can mediate biofilm formation is luxS, whose product induces the production of the autoinducer AI2. In LuxS-deficient strains, 18 downregulated genes were observed, with all being located in prophage regions. Additional experiments demonstrated that LuxS-deficient bacteria produced less biofilm with a lower concentration of extracellular DNA (eDNA) and less phage DNA than wild-type bacteria. The study proposed a mechanism by which the host induces prophages to disperse cellular contents involved in biofilm formation, which has been observed in other species (Rossmann et al. 2015: e1004653; Secor et al. 2016: 49). Further research will be necessary to understand the impact of quorum sensing on phage biology and vice versa. Nevertheless, these findings underline the risks of temperate phage therapy, which could enhance the virulence of C. difficile, for example through biofilm formation.

Effects on antibiotic resistance

Phages and phage-like genetic elements are known to be capable of moving antibiotic resistance genes between many species of bacteria (Brown-Jaque, Calero-Cáceres and Muniesa 2015: 1–7). In C. difficile, phage phiC2 was found to promote the transfer of the transposon Tn6215 between strains, which encodes erythromycin resistance (Goh et al. 2013). This study was the first to document phage-mediated transduction in C. difficile. Interestingly, only one of the four phiC2-susceptible strains received the transposon despite all having intact transposon integration sites. This suggests that there are specific requirements for successful transduction. Given that other transposons encode resistance to chloramphenicol (Lyras et al. 1998: 1563–7), MLS_B (Macrolide-lincosamide-streptogramin B; Dingle et al. 2014: 36–52), tetracycline (Jasni et al. 2010: 4924–6; Mullany et al. 2012: 2147–53) and possibly vancomycin (Knight et al. 2016) in C. difficile, careful examination of how phages interact with these elements is needed for the development of safe phage therapy. Since quinolone treatment was found to increase prophage induction in vitro (e.g. phiMMP04), antibiotic usage could increase the risk of phage-mediated horizontal transfer of resistance genes in vivo (Meessen-Pinard et al. 2012: 7662–70). Better characterizing the mode of encapsidation used by phages could also help predict the risk of transduction (Garneau et al. 2017: 1–10).

Effects on CRISPR

CRISPR (Clustered Regularly Interspaced Short Palindromic Repeat sequences) and associated Cas proteins form an adaptive immune system in prokaryotes that consists of repeated sequences of host DNA in alternance with short DNA fragments taken from mobile genetic elements (such as plasmids and phages; Westra et al. 2012: 311–39). Upon transcription of these arrays, small RNAs are produced that guide the Cas protein complex to break down invading phages (Bhaya, Davison and Barrangou 2011: 273–97). The types of CRISPR elements as well as a detailed look at CRISPR mechanisms can be found in prior reviews by Westra et al. (Westra et al. 2012: 311–39) and Bhaya et al. (Bhaya, Davison and Barrangou 2011: 273–97). Clostridioides difficile is known to contain active CRISPR elements, and strain CD630 contains 12 possible CRISPR arrays while R20291 contains 13 such regions (Soutourina et al. 2013; Boudry et al. 2015: e1003493). Moreover, CRISPR arrays were also found in prophage regions. For example, the two prophages of CD630, phiCD630–1 and phiCD630–2, each contains a unique pair of identical CRISPR elements that are co-transcribed (Soutourina et al. 2013; Boudry et al. 2015: e1003493).

The location of CRISPR elements in phage genomes suggests that these could be transferred by transduction, thus conferring protection against phage infection to new lysogens (Hargreaves, Kropinski and Clokie 2014: e85131; Sorek, Lawrence and Wiedenheft 2013: 237–66; Soutourina et al. 2013: e1003493). Evidence from the analysis of CRISPR elements suggests that C. difficile interacts frequently with phages, as more than one in three CRISPR spacers analyzed in one study matched to phage sequences (Boudry et al. 2015). Another study found matches between CRISPR spacers and the genomes of 31 C. difficile phages and prophages (Hargreaves et al. 2014). The number and quality of matches between CRISPR spacers and phage genomes should, in principle, predict whether a given phage can successfully infect a strain of C. difficile. In one study, the prediction of the susceptibility of C. difficile strains to phage infection based on the presence or absence of CRISPR spacers gave a rather good correlation (Boudry et al. 2015). In particular, strain R20291 had more mismatches between its CRISPR spacers and the genomes of several phages compared to strain CD630, which correlated with greater susceptibility to the phage panel used (Boudry et al. 2015).

Few studies have discussed how C. difficile phages manipulate CRISPR mechanisms during the infection process. CDKM15 has a CRISPR array that can target sequences on multiple C. difficile phages, however, no CRISPR-associated (Cas) genes were found in CDKM15 (Rashid et al. 2016: 310). Although the authors suggest that these arrays may protect from superinfection, how this could occur remains unexplored. The large episomal phage phiCD211 was found to encode a putative CRISPR-associated Cas3-HD endonuclease protein (ORF154), close to two short CRISPR arrays (Garneau et al. 2018). In type I-B CRISPR systems, the Cas3 nuclease, as part of the Cascade complex, nicks and degrades the target DNA. The functionality of this CRISPR-Cas locus in phiCD211 remains to be demonstrated. Phage JD032 was also shown to downregulate CRISPR-Cas type I-B, which may help the phage infect its host more effectively (Li et al. 2020). To our knowledge, this is the only study as of the writing of this review that has examined how phages can alter the transcription of CRISPR elements in C. difficile.

A better characterization and comprehension of C. difficile CRISPR systems could help improve the effectiveness of future phage therapy. Some phages seem to represent better targets for CRISPR elements. One study found that phiCD38–2 and phiMMP04 were the top two targeted phages with 54 and 42 corresponding spacers, respectively, while phiCD24–1 was the least targeted with only two spacers in analyzed CRISPR arrays (Boudry et al. 2015). Currently, CRISPR matches with phage DNA are not sufficient to predict phage susceptibility. Mismatches between spacers and phage DNA, depending on the position of the mismatch, likely decreases the susceptibility to CRISPR inactivation. Furthermore, strains lacking CRISPR spacers towards certain phages can still be resistant (Boudry et al. 2015). This is because multiple factors can also affect susceptibility to phage infection, like the presence of a suitable host receptor, the presence of other similar prophages, or other antiphage mechanisms. There are yet more complex relationships between CRISPRs and toxin-antitoxin system components, which both were found in prophages. It has been proposed that these toxin-antitoxin systems could contribute to antiphage protection (Soutourina 2019: 253). Further research into C. difficile anti-phage defenses will therefore be critical to help predict the effectiveness of phage therapy.

Clostridioides difficile PHAGE APPLICATIONS

Earlier studies have explored the potential of phages as a typing method, which was eventually abandoned because of the poor reliability of this approach compared to molecular methods (Dei 1989: 351–4; Mahony, Bell and Easterbrook 1985: 251–4; Mahony et al. 1991: 1873–9; Sell, Schaberg and Fekety 1983: 1148–52). The primary application of C. difficile phages in recent days is the development of novel CDI therapies. As of the writing of this review, research describing single-phage therapy, phage cocktails, phage-derived products (endolysins, PT-LPs and fusion proteins) have been reported. Here we will consider the advantages and disadvantages of each approach. A summary of phage therapy and phage-based therapy experiments is provided in Table 2.

Table 2.

Clostridioides difficile phage-based therapies.

Reference	Treatment agent	Model	Findings
Ramesh, Fralick and Rolfe (1999)	CD140	In vivo hamster model	• Stomach acid destroyed phage
			• Phage treatment improved hamster survival
			• Phage resistance observed in one hamster
			• Phage treatment did not protect from a second infection
Mayer, Narbad and Gasson (2008)	phiCD27 endolysin (CD27L)	In vitro	• Specifically targeted C. difficile
			• Active against 30/30 strains tested (including ribotype 027)
			• Biologically active over wide pH range
Meader et al. (2010)	phiCD27	In vitro batch fermentation	• Lysogens showed reduced toxin production
			• Treatment reduced C. difficile while leaving commensal species unharmed
			• Phage resistance observed
			• C. difficile not fully cleared by phage treatment
Mayer et al. (2011)	Endolysin catalytic domain CD27L1–179	In vitro	• Modified endolysin demonstrated greater effectiveness than CD27L
			• Commensal species unharmed
			• Endolysin could be modified to kill other pathogenic species
Meader et al. (2013)	phiCD27	In vitro human colon model	• Treatment reduced vegetative cells
			• Spore production increased in lysogens
			• Toxin production was reduced
			• Commensal species unharmed
			• Phage lysogeny and resistance observed
			• However, there was no significant difference in C. difficile levels relative to the control in one replicate
Sangster, Hegarty and Stewart (2015)	PTLPs derived from RT078 C. difficile isolate (HMC114)	In vitro	• Out of 25 samples of ribotype 27 bacteria,21 were susceptible to PTLPs while 2 were partially susceptible.
			• Other ribotypes of C. difficile were unharmed
			• Other species were unaffected by treatment
			• PTLPs do not kill their strain of origin
Wang et al. (2015)	Recombinantly expressed catalytic domain of lysin PlyCD (PlyCD1–174)	Ex vivo treatment mouse colon model	• Treatment worked in the presence of intestinal contents
			• C. difficile colonization was reduced relative to controls
			• This method could be used alone or with other therapy
			• Rectal delivery of endolysin domains produced inconsistent results in vivo
Mehta et al. (2016)	CD11 and CDG endolysins	In silico endolysin identification and in vitro testing	• Two endolysins were identified from the genomic sequences of C. difficile strains
Nale et al. (2016a) "Bacteriophage…’	phiCDHM1, phiCDHM2, phiCDHM3, phiCDHM4, phiCDHM5, phiCDHM6, phiCDHS1 single-phage therapy and phage cocktails	•In vitro	• Single-phage therapy led to phage resistance
		• In vivo	• Lysogens of one phage could be infected by another
		hamster model	• Phage cocktails killed C. difficile in vitro and did not lead to phage resistance
			• C. difficile colonization, sporulation was reduced upon phage cocktail therapy in vivo
Nale et al. (2016b) ‘Get…’	phiCDHM1, phiCDHM2, phiCDHM5, phiCDHM6 single-phage therapy and phage cocktails	G. mellonella larvae CDI model	• Biofilm formation was prevented in vitro
			• Phages could reduce existing biofilms
			• Phage cocktails were more effective than single phages at clearing/preventing biofilms
			• Vancomycin pre-treatment improved phage efficacy
			• No phage resistance was observed with phage cocktails
Vinner et al. (2017)	phiCDKM9 encapsulated in a polymer (Eudragit® S100 with and without alginate)	In vitro spot tests of encapsulated phages after simulated pH changes	•Encapsulation protected phage from degradation in the conditions of the GI tract
Nale et al. (2018)	phiCDHM1, phiCDHM2, phiCDHM5, phiCDHM6 phage cocktails	In vitro batch fermentation model	• Phage therapy eliminated C. difficile in the model
			• Commensal microbes were unharmed
			• Enterobacteria, lactobacilli and total anaerobe abundance increased in phage-treated groups, suggesting that phage therapy may protect from future infections
Shan et al. (2018)	CDSH1 single-phage therapy	In vitro HT-29 tumorigenic colon cell model	• Planktonic and adhered C. difficile cells on HT-29 cells decreased
			• Phage treatment was more effective in the presence of HT-29 cells
			• Phage-mediated lysis of bacteria did not release toxins or damage HT-29 cells.
Peng et al. (2019)	Recombinant protein made of phiC2 lysin (PlyCD) and human defensin protein HD5	• In vitro culture	• MIC of fusion protein was lower than either component alone and typical antibiotic therapies
		• In vivo mouse model	• In vivo treatment reduced symptoms and reduced death
			• Toxin production and sporulation was reduced in vivo
Selle et al. (2020)	Wild-type phiCD24–2 single-phage therapy, engineered phiCD24–2 single-phage therapy (carrying CRISPR-Cas3 components)	• In vitro	• Modified phage caused a larger drop in C. difficile and delayed the rebound of the culture more effectively than wild-type phage
		• In vivo mice model	• Modified phage reduced the bacterial load in feces and reduced vegetative cells
			• Wild-type phage reduced vegetative cells but did not lower bacterial load
			• Phage modification increased lytic activity, but phages remainfed temperate

Single-phage therapy

Single-phage therapy was the first phage-based strategy applied to the treatment of CDI. Hamsters challenged with C. difficile were administered phage CD140 before assessing their health outcomes (Ramesh, Fralick and Rolfe 1999: 69–78). The trial was successful at improving hamster survival from a challenge with C. difficile, suggesting the potential therapeutic application of phages. Single-phage therapy experiments with phiCD27 in both an in vitro batch fermentation (Meader et al. 2010: 549–54) and a human colon model (Meader et al. 2013: 25–30) found that phage therapy reduced levels of C. difficile and toxin production while leaving commensal species unharmed. These results show that phage therapy can avoid arguably the biggest pitfall of antibiotic use against CDI; microbiome dysbiosis. By leaving the natural gut communities intact, phage therapy should be able to treat CDI without increasing the risk of recurrent episodes.

It is important to note that the experimental model used may influence the observed lytic activity of phages. A single-phage therapy experiment with phage CDSH1 found that the virus was more effective at clearing planktonic and adhered C. difficile cells in culture when in the presence of HT-29 human colon cells (Shan et al. 2018: 1–8). The authors proposed that this was due to a strong interaction between phages and colonic cells, which is also consistent with the bacteriophage adhering to mucus (BAM) model, which describes how phages can bind to the glycoproteins in the mucus layer to facilitate infection (Van Belleghem et al. 2019: 10). Testing C. difficile phage therapy using multiple models, such as in vitro batch cultures, in vitro gut models and in vivo assays using mammalian hosts (e.g. mice, hamsters) and other models such as the Galleria mellonella wax worm will be essential to accurately predict how phage therapy will perform under real-world conditions (Nale et al. 2016a: 1383, b:968–81, 2018: 13).

Although the single-phage therapy experiments discussed so far demonstrated the potential to clear C. difficile in vitro and in vivo, there are obstacles to using single-phage therapy in clinical settings. As previously discussed, environmental conditions like pH can reduce the stability and activity of phages. Consistent with these observations, CD140 phages were inactivated by stomach acid shortly after being administered to hamsters with CDI, so neutralization was necessary to carry out the treatment (Ramesh, Fralick and Rolfe 1999: 69–78). At the time, this was a significant practical issue for phage therapy. More recently, synthetic polymers have been developed to protect phages from the extreme conditions of the stomach (Vinner et al. 2017: e0186239), suggesting that delivery is no longer a challenge for single-phage therapy or other kinds of phage-based therapy.

However, obstacles encountered in the CD140 phage therapy study have yet to be fully resolved; i.e. the short-term only protection offered by phage therapy, the limited host range of individual phages and phage resistance. Phage therapy with CD140 did not protect from a second infection upon a second challenge with the bacteria (Ramesh, Fralick and Rolfe 1999: 69–78). No other single-phage therapies report long-term protection, which is unfortunate given the potential for patients to be re-exposed to C. difficile spores in healthcare settings or the environment. However, this is expected since phages will eventually be cleared from the organism if no bacterial host is present to support replication and propagation. This challenge highlights the need for other treatments to be developed for use after the clearing of the initial infection. Successfully clearing CDI with phage therapy in the first place is complicated by phage resistance and the limited host range of C. difficile phages. In the first C. difficile phage therapy experiment, phage CD140 was chosen because it demonstrated lytic activity against the selected strain of C. difficile. The authors noted that CD140 had a limited host range. In a clinical setting, screening phages for activity against the C. difficile strain(s) infecting the patient would be inefficient. Over the course of the CD140 phage therapy, one hamster died from CDI because phage CD140 integrated into the bacteria, forming a resistant lysogen. Although all other hamsters survived, this example clearly demonstrates that phage-therapy cannot be applied until the problem of lysogeny is overcome. In summary, the findings indicate that lysogeny can undermine the efficacy of phage therapy. Since no strictly lytic phages have been isolated so far against C. difficile, and thus naturally occurring phages all have the potential to lead to lysogeny, other options must be found.

Phage therapy with genetically modified phages could provide a solution thanks to recent advances in phage engineering. A modified version of the C. difficile phage phiCD24-2 carrying a mini CRISPR-Cas3 array displayed greater lytic activity than its unmodified counterpart both in vitro and in vivo (Selle et al. 2020). This was the first study that used a genetically engineered phage to treat C. difficile. The authors reported an increase in C. difficile burden after a few days of treatment, which was due to the emergence of phage-resistant lysogen. Removing lysogen-related genes coding for the CI repressor and integrase (phiCD24-2 Δlys) greatly improved the efficacy of the wild-type and CRISPR-engineered phage in batch cultures, since no lysogens were observed 22 h after infection. However, for reasons that still need to be identified, lysogeny occurred in vivo, suggesting that something unexpected occurred. One hypothesis that was proposed is that recombination with an endogenous prophage or prophage remnant sharing sufficient homology with phiCD24-2 could have occurred, thus rescuing the lysogenic potential of the engineered phage (Selle et al. 2020).

Despite this obstacle, these results demonstrate that phage engineering is likely the best approach for making single-phage therapy a viable option for the treatment of CDI. Phage engineering has proven successful when applied in other species. For example, a strictly lytic phage was engineered from the temperate Mycobacterium smegmatis phage ZoeJ by removing its repressor gene (Dedrick et al. 2019: 730–3). Nevertheless, more work is needed to improve the stability and safety of engineered phages. Given that phage resistance can occur due to a variety of factors like phage receptor mutations, CRISPR systems, or through superinfection exclusion by temperate phage (Sekulovic et al. 2015: 329–42), using a single phage for therapy may still prove too risky.

Phage cocktails

The use of a combination of phages may be a superior strategy for C. difficile phage therapy, even if strictly lytic phages were available. Nale and colleagues have published a series of articles that describe the optimization of a multiple-phage therapy to overcome issues with temperate phages, phage resistance and narrow host ranges (Nale et al. 2016a: 1383; Nale et al. 2018: 13; Nale et al. 2016b: 968–81). The first of these studies compared the effectiveness of five temperate phages used individually or as a cocktail against C. difficile (Nale et al. 2016b: 968–81). The host range and activity of several phage combinations were measured using a 24-h bacterial culture model. Although the host range and lytic capabilities of each phage varied, phage resistance occurred using all individual phages. Meanwhile, multiple-phage cocktails could infect a broader range of isolates than single phage and were more effective at clearing C. difficile than single-phage therapy. This improvement occurred in part because lysogens formed with one phage were still susceptible to infection by another phage from the cocktail. This finding is significant because it demonstrates the possibility to improve the efficacy of naturally occurring temperate phages while limiting phage resistance due to lysogeny. However, considering the well-known problems that can arise during therapy with temperate phages, this strategy should be considered with great care. That being said, cocktails of strictly lytic phages can overcome the problem of phage resistance due to receptor mutation or CRISPR resistance for instance. Hence, if strictly lytic phages become available one day, developing cocktails with these would be the best strategy.

Nale and colleagues also used a batch fermentation model composed of healthy human fecal slurry inoculated with a C. difficile strain of ribotype 014/020. They found that an optimized 4-phage cocktail targeting C. difficile could completely eliminate the bacterium after 24 h (Nale et al. 2018: 13). Most importantly, the simulated microbiome was unharmed, and the abundance of commensals such as enterobacteria, lactobacilli and total anaerobes was not affected by phage treatment. This suggests that phage therapy could prevent recurrence while eliminating C. difficile, but additional evidence would be useful to confirm the protective potential of phage cocktails for therapy. In addition, the cocktail used was specific to ribotype 014/020 strains only, thus limiting its applicability in the clinic. Considering the narrow host range of most C. difficile phages, several cocktails would need to be developed to target the most clinically relevant strains, which represents a challenge considering the current phages and data available.

Another study by the same group of researchers compared the most effective phage cocktail with single-phage therapy in a wax moth model (Nale et al. 2016a: 1383). The authors were interested in determining how effective phage therapy was on C. difficile biofilms, which contribute to the virulence of the pathogen. While both phage therapies reduced and prevented biofilm formation, phage cocktails were more effective than single phages at this task while also preventing phage resistance. Interestingly, the best treatment regimen consisted of treating worms with vancomycin prophylactically before phage application (Nale et al. 2016a: 1383), which is in conflict with previous reports on phage therapy. It has been shown that phage therapy can re-sensitize multidrug-resistant bacteria such as Acinetobacter baumannii and P. aeruginosa to antibiotics (Chan et al. 2016: 26717; Schooley et al. 2017). Some suggest that phages could even serve as a ‘pre-biotic’ to improve the efficacy of existing therapies (Moelling and Broecker 2016: e1251380). The results with the wax moth model experiments highlight a facet of phage therapy that needs further investigation, i.e. the relationship between phages, antibiotics, C. difficile and the timing of therapy.

Even though phage cocktails appear to be a potential alternative CDI therapy, one must keep in mind that experimental applications of phage cocktails in a lab setting with known C. difficile strains are totally different from the challenges of using phages in a clinical setting. In the latter case, the C. difficile strain would need to be isolated first, cultured and characterized minimally before assessing the activity of phage cocktails to target the infecting C. difficile strain. There are two strategies of phage cocktail design that could be considered for this challenge; a one-size-fits-all cocktail made of a few phages that could be readily used against the most prevalent C. difficile strains, or a personalized cocktail if specific strains are involved (Schmidt 2019: 581–7). Designing cocktails that target different phage receptors would be an effective strategy, but the lack of data on receptor candidates in C. difficile stands as an obstacle to developing such a treatment. A recent review by Schmidt (2019: 581–7) noted that a poor understanding of phage receptor dynamics was a contributing factor to the failure of a twelve-phage cocktail in treating burn patients with P. aeruginosa infections. Ultimately, phage cocktails present a potential for customizable treatment that could avoid the issue of phage resistance, but this strategy is not ready for widespread application to treat CDI until more is understood about the interactions between phages, bacteria and antibiotics.

Phage tail-like particles (PT-LPs)

A study by Nagy and Foldes (NAGY and Földes 1991: 321–6) first described defective phage structures that were later reported by Sell, Schaberg and Fekety (1983: 1148–52), who showed that filtrates from Mitomycin-C treated cultures of C. difficile could inhibit C. difficile lawn formation without producing whole phages. The authors suggested that bacteriocins were responsible, which are protein structures produced by bacteria during a stress-response to kill competing bacteria (Gebhart et al. 2012: 6240–7). These bacteriocins, described in these early papers and then later in the paper by Fortier and Moineau (2007: 7358–66), represent phage tail-like particles (PT-LPs). They were later characterized in more detail by a group from a biotech company from California, Avidbiotics (Gebhart et al. 2012: 6240–7). Other studies have demonstrated that bacteriocins can be induced from C. difficile using antibiotics (Fortier and Moineau 2007: 7358–66; NAGY and Földes 1991; Hargreaves 2013: 321–6; Nale et al. 2012: e37263; Shan et al. 2012: 6027–34). In one early study on C. difficile PT-LPs, plasmid carriage did not correlate with PT-LP production, which led the authors to hypothesize that they were encoded on the chromosome (Mahony et al. 1991: 1873–9). PT-LPs appear to be abundant in C. difficile; out of 92 RT027 isolates examined for the presence of prophages in one study, all contained PT-LPs whereas 65 contained an inducible prophage (Nale 2013). Hence, PT-LPs can be found in C. difficile strains regardless of phage carriage (Hegarty et al. 2016: 167–75).

The two types of bacteriocins, known as R-type and F-type (Michel-Briand and Baysse 2002: 499–510; Nakayama et al. 2000: 213–31), share certain structural components with phages. R-type bacteriocins are similar to Myoviridae phages, as they consist of a non-flexible tail tube wrapped with a contractile sheath. F-type bacteriocins consist of a non-contractile structure that resembles Siphoviridae phage tails. Despite the similarities between bacteriocins and phages, bacteriocins can be distinguished by the absence of a capsid and nucleic acid (Sangster, Hegarty and Stewart 2015: 96–103). R-type bacteriocins act in a two-step process; they bind to their target strain via specific receptors and then the contraction of the sheath leads to the puncturing of the cell membrane and the lysis of the bacteria (Gebhart et al. 2012: 6240–7). F-type bacteriocins may also compromise the membrane of the target bacteria, but the mechanism is not well understood (Scholl 2017: 453–67).

PT-LPs that demonstrate specific lytic activity against C. difficile are known as ‘diffocins’ (Gebhart et al. 2012: 6240–7). Diffocins isolated from an RT078 strain were observed to lyse a variety of RT027 isolates, which led the authors of the study to suggest that diffocins could serve as a specific and cheap alternative antimicrobial against C. difficile (Sangster, Hegarty and Stewart 2015: 96–103). Since PT-LPs do not replicate in the host (Schwemmlein et al. 2018: 1750), they do not have the potential to increase the virulence of target strains. This is a significant advantage over phage therapy, which can affect host virulence as previously discussed. Also, just as the host range of phages is a design consideration for phage therapy, the host specificities of diffocins pose certain advantages and disadvantages. Diffocins can lyse more strains of C. difficile than whole phages because, as opposed to the latter, they are not subject to interference due to CRISPRs and prophage-mediated repression for instance, which could make diffocins a more practical treatment option (Sell, Schaberg and Fekety 1983: 1148–52). On the other hand, some strains of C. difficile may be unaffected by a given diffocin (Mahony et al. 1991: 1873–9). Ample evidence has established that bacteriocins such as diffocins do not act on the strain producing them (Gebhart et al. 2012: 6240–7; Sangster, Hegarty and Stewart 2015: 96–103; Scholl et al. 2009: 3074–80). Also, diffocins from a particular ribotype did not display activity against other strains of the same ribotype (Sangster, Hegarty and Stewart 2015: 96–103). Overall, diffocins have a narrower host range than endolysins (see below) but broader than phages. The ORFs that influence PT-LP host specificity are different from those of phages (Gebhart et al. 2012: 6240–7). For example, PT-LPs encode both a truncated tail fiber protein relative to phages and a unique ORF that could bind to a bacterial receptor (Gebhart et al. 2012: 6240–7). Such variations may contribute to the observed differences in host range between phages and PT-LPs. As a result of the host range limitations of PT-LPs, a cocktail of diffocins may be a more effective approach to target a variety of C. difficile strains.

The expression of diffocins in a heterologous host like Bacillus subtilis as demonstrated by Gebhart et al. (2012: 6240–7) opens the possibility of using engineered commensal species to deliver diffocins in the gut. This negates the need to neutralize stomach acid prior to administration as noted in a different study (Gebhart et al. 2015). Moreover, diffocins can be genetically engineered to extend their host range. A total of two studies have demonstrated that the gene encoding the receptor-binding protein (RBP) in diffocins can be replaced by another RBP from another diffocin, or even from a Myoviridae phage (Gebhart et al. 2015; Kirk et al. 2017: eaah6813). Exchanging the RBP with that of another diffocin or phage confers the host range of the diffocin or phage from which it originated. In vivo experiments in mice revealed that diffocins can prevent CDI without impacting the gut microbiota (Gebhart et al. 2015). The analysis of diffocins was pushed even further in a recent study that showed that different chimeric diffocins can be generated to target various C. difficile strains. These modified diffocins were named ‘Avidocin-CD’, and they recognize C. difficile through a specific interaction with the surface-layer protein SlpA. This protein was shown to be the receptor recognized by RBP from diffocins (or Avidocin-CD), with specific RBP targeting specific SlpA isoforms (Kirk et al. 2017: eaah6813). This important development on diffocins suggests that a panel of recombinant Avidocin-CD could be designed and produced in the form of cocktails to target multiple strains of C. difficile. However, the study noted that receptor-mediated resistance could arise. A greater challenge with diffocins or Avidocin-CD is the large-scale production of these particles in heterologous systems. Since these are non-replicating biological weapons, they would also need to be administered to patients in multiple doses over several days (especially if used in prophylaxis), therefore making this strategy economically challenging as well.

Endolysins and fusion proteins

Phage-based therapies present the possibility to harness the targeted lytic activity of phages without the complexity of having to overcome phage lysogeny and resistance. Phage endolysins, which are proteins encoded by the lytic phages for the hydrolysis of the host cell wall, are one of the phage-based therapies that show promises for treating CDI (Mayer, Narbad and Gasson 2008: 6734–40). Phage proteins like endolysins have also been investigated for their use in detection assays for multiple bacteria (Meile et al. 2020: 944), but this has not been attempted to our knowledge in C. difficile. Endolysins are composed of a cell wall-binding domain that determines host specificity and a hydrolase domain that carries out the lysis (Loessner 2005: 480–7). An in-depth review of endolysins in C. difficile has recently been published (Mondal et al. 2020: 1813533). Mayer, Narbad and Gasson (2008: 6734–40) were the first to demonstrate the potential utility of phage endolysins for CDI therapy. The phiCD27 endolysin, like the whole phage, was lethal against C. difficile while leaving commensal species unharmed. Unlike the whole phage, however, the endolysin was effective against all strains of C. difficile tested and was active over a wider range of pH conditions. The study also reported the expression of the endolysin in a modified Lactococcus lactis strain, showing that engineered commensal microbes could be utilized to deliver endolysin treatments (Mayer, Narbad and Gasson 2008: 6734–40). In theory, the growth of the engineered commensal bacteria could provide lasting protection from recurrence, but the advantages and disadvantages of this approach have not been investigated. A follow up to the phiCD27 endolysin study investigated whether an engineered endolysin could provide more potent antimicrobial activity. A truncated endolysin CD27L_1–179 demonstrated enhanced lytic activity relative to the full-length endolysin yet still spared commensal species (Mayer et al. 2011: 5477–86). Other modifications of the CD27L_1–179 catalytic domain allowed the researchers to give the endolysin activity against other pathogens such as Listeria monocytogenes, further showing the adaptability of such agents to different challenges.

CD27L_1–179 was not the only modified endolysin to show improved capabilities relative to its natural counterpart. A study on the endolysin PlyCD, derived from a prophage from strain CD630, found that the catalytic domain (PlyCD_1–174) displayed greater lytic activity than the whole endolysin itself (Wang et al. 2015: 7447–57). Both endolysins were active against all tested C. difficile strains with no activity against other bacterial species, and both maintained activity in a range of pH from 5 to 8. In vitro tests demonstrated that PlyCD_1–174 did not significantly reduce C. difficile growth in culture but combining the agent with a low vancomycin treatment showed strong synergy. Ex vivo tests showed that PlyCD_1–174 reduced C. difficile colonization significantly on its own. A later study used a different approach by expressing a recombinant protein made of the catalytic domain of the endolysin from phage phiC2 and the functional domain of the human defensin protein HD5 (Peng et al. 2019: 3234). The fusion protein was more effective than either of the proteins alone, showing a lower MIC (0.78 µg/ml) than its counterparts and typical antibiotics in vitro. Like the aforementioned studies, the chimeric enzyme functioned under variable pH conditions ranging from 6 to 8. In vivo treatment also reduced toxin production, sporulation, CDI symptoms and death.

These studies on endolysins and fusion proteins suggest that modified endolysins could serve both as a standalone therapy or a combination therapy to extend the usefulness of existing antibiotics. These approaches to CDI therapy provide many of the advantages of phage therapy (e.g. specificity) while reducing drawbacks (e.g. narrow host range, lysogeny and resistance). Given that endolysins can be identified in silico and then tested through recombinant expression (Fujimoto et al. 2020: 380–9. e9; Mehta et al. 2016: 2568–76), developing novel therapeutics with existing technology is feasible. Endolysins and their derivatives can kill more strains of C. difficile than a typical C. difficile phage and function under a wide pH range. Most importantly, genetic engineering allows designing combinations of endolysins with other proteins (e.g. human defensin) and reduces the risk of bacterial resistance, which is low for this type of therapy (Mondal et al. 2020: 1813533). In contrast with diffocins or Avidocin-CD, large-scale production of single purified proteins such as endolysins is technically easier in terms of industrial production, but the dosing in patients will remain as challenging as with larger non-replicating bacteriocins. Effectively delivering protein-based therapies may also be challenging given the abundance of proteases in the gut environment (Biancheri et al. 2013: 269–80).

CONCLUDING REMARKS

Due to the inherent limitations of antibiotic therapy for preventing recurrent CDI, coupled with the growing threat of antibiotic resistance, there has been a renewed interest in phages as a source of novel therapeutics. While studies have begun to elucidate how C. difficile phages infect and alter their hosts, the scope and depth of information available for known phages are highly variable. Many details of the phage replication cycle and lifestyle remain unknown. Understanding C. difficile phage biology and genetics is necessary to provide a foundation for developing phage therapy and phage-based therapy for CDI.

When considering what future directions these therapies might take, there are a variety of options each with advantages and drawbacks. Strictly lytic phages incapable of lysogeny represent the simplest solution. However, the unavailability of such phages is a major concern. Several studies have shown that wild-type temperate phages, when used in single therapy, quickly lose their effectiveness due to lysogeny. Moreover, lysogeny with certain phages is known to increase the virulence of the bacterial host. Hence, this precludes the use of single and even cocktails of temperate phages until these drawbacks are addressed, for example through the development of engineered phages.

Fortunately, phage-based therapeutics have the potential to keep the specificity of phages while alleviating the issues inherent to whole phages. PT- LPs such as wild-type diffocins or recombinant Avidocin-CD are the most readily available phage-based therapeutics. However, very little work has been done to test their therapeutic capability in vivo. Likewise, engineered endolysins and fusion proteins could be developed into potent off-the-shelf treatments for most C. difficile strains. Despite these promising strategies, the complexity and cost of the large-scale production of Avidocin-CD and recombinant endolysins, as well as the complexity of their application in a clinical setting will represent significant challenges.

Major advances have been made on many fronts in recent years. The number of fully characterized phages has increased, though it is still limited. Our understanding of phage-host interactions has also improved, but many questions remain. The development of new genetic tools to manipulate C. difficile and its phages also brings interesting perspectives for the future development of engineered phages and phage-derived therapeutics.

ACKNOWLEDGMENTS

We thank lab members for helping proofread the manuscript.

FUNDING

This work is in part supported by a grant from the National Institutes of Health (R01-AI132711) to X. Sun, , and from the Natural Sciences and Engineering Research Council of Canada (NSERC RGPIN-2020-05776) to LC Fortier.

Conflicts of Interest

None declared.