The anaerobic, rod-shaped, Gram-positive bacterium C. perfringens (Cp), belonging to the Clostridiaceae family, resides in the top six of bacteria causing food poisoning in the United States (Scallan et al., 2015). It is widely distributed in the intestines of animals, especially poultry, as well as in diverse other environments. This is in part due to its ability to form spores that are highly resilient to stressors, such as toxic chemicals, heat and radiation (Paredes-Sabja et al., 2008). Cp is a versatile pathogen of humans and husbandry animals causing wound infections, enteritis/enterocolitis and enterotoxaemia (for recent review see Mehdizadeh Gohari et al., 2021).

The virulence of Cp can largely be ascribed to its vast arsenal of toxins. These toxins are used to classify isolates to one of seven toxin types (A–G) (Rood et al., 2018). Specific toxin types, in turn, correspond with different disease manifestations. Toxin type F (producing alpha toxin and C. perfringens enterotoxin), for example, is associated with human food poisoning, antibiotic-associated diarrhea, and sporadic diarrhea. Type G [producing alpha toxin and Necrotic enterocolitis B-like toxin (NetB)] is linked to necrotic enteritis of poultry (Cooper et al., 2013; Rood et al., 2018; Mehdizadeh Gohari et al., 2021). In addition to NetB, a recent genomic analysis positively associated the toxin gene tpeL, and other virulence genes pfoA, cpb2 and cna to necrotic enteritis-linked isolates (Kiu et al., 2019). Lepp et al. (2021) furthermore showed the important role of adhesive pilus of toxin type G strains in causing necrotic enteritis in poultry. Toxin type C causes severe and lethal necrotic enteritis in newborn piglets (Posthaus et al., 2020).

Cp strains st13 and SM101 have fully sequenced genomes and have been the most extensively studied with respect to C. perfringens pathogenicity (Sekiya et al., 2021a). Aside from these fully sequenced strains, further comparative genome analysis of 206 publicly available Cp strains’ sequence data indicated five closely-related phylogenetic groups (phylogroups) reflecting different disease involvements (Abdel-Glil et al., 2021), thus demonstrating the large repertoire of potential virulence factors. Particularly phylogroup 1 contained the human food poisoning strains producing enterotoxin and the Darmbrand strain. Kiu and Hall (2018) schematically depicted the disease-linked virulence factors of Cp in the context of intestinal infections. They also reviewed the capacity of abundant histotoxic gas production alongside rapid growth of Cp known as clostridial myonecrosis (gas gangrene). Excessive hydrogen sulfide produced by Cp is toxic to human cells and is associated with intestinal inflammatory diseases such as ulcerative colitis (Rowan et al., 2009) and NEC in preterm infants with severe intestinal tissue necrosis and gas production (Neu and Walker, 2011).

Infections with Cp cause severe disease in extremely low birth weight preterm infants, with an incidence that varies from 2 to 13% (Alsaied et al., 2020) and a mortality rate of 30–50% with surgical NEC (Kiu and Hall, 2018). To prevent NEC in preterm infants various strategies have been explored, including the introduction of probiotics (Alfaleh and Anabrees, 2014), breast milk feeding and probiotics (Jin et al., 2019) and human milk oligosaccharides to support bifidobacteria and lactic acid bacteria (Bode, 2018). Phage introduction via fecal transplantation was discussed as a solution (Hofer, 2021). Despite safety concerns related to the possible unintended transfer of pathogens via fecal transplantation, phage therapy via this route has been suggested as a good option to explore (Hofer, 2021).

Since the European ban in 2006 on the use of antibiotics as growth promotors in animal feed, alternative strategies to reduce diseases in animals have been explored also to ensure food safety. The preventive strategies against necrotic enteritis in poultry, caused by Cp, has gained a lot of attention. In the poultry industry, natural feed additives were shown effective in reducing or even abolishing Cp infections (Caly et al., 2015). Typical naturals explored were live bacteria and yeasts (probiotics), plants and plant molecules (such as essential oils), organic acids, enzymes, lysozymes, antimicrobial peptides, bacteriocins (i.e., antimicrobial peptides produced by bacteria), vaccines and bacteriophages. Antimicrobial molecules, such as bacteriocins or phage endolysins, were considered good candidates for new antimicrobials (Caly et al., 2015).

In addition to the health burden delineated above, Cp infections are accompanied with extremely high costs. The care for NEC-affected infants amounts to $500 million to $1 billion in the United States (Neu and Walker, 2011). The financial loss for the poultry sector has been estimated close to $6 billion in the United States in 2015 (Wade and Keyburn, 2015).

Clostridioides difficile encodes genetic variants of three well characterized toxins at two distinct loci: toxin A and B in the pathogenicity locus (PaLoc) and Cd binary toxin in the correspondingly named locus (CdtLoc). While the former two toxins represent primary virulence factors for Cd infection, Cd binary toxin has been associated with increased severity of Cd infection. Cd binary toxin-induced actin depolymerization produces microtubule-based membrane protrusions, which form a network on epithelial cells and increase bacterial adherence (Gerding et al., 2013). The production of the three toxins is primarily controlled by regulators specific to each locus (Majumdar and Govind, 2022), and has been described to be strain-specific. For example, the regulator of Cd binary toxin production (CtdR) increased the production of toxin A, toxin B and Cd binary toxin in two epidemic ribotype 027 human strains, but not in a ribotype 078 animal strain (Lyon et al., 2016). Toxin A and toxin B production in the PaLoc can be influenced by some prophages, e.g., by phiCD38-2 (Supplementary Table 2). PaLoc has some features in common with phages. In particular tcdE, a gene encoding a membrane protein similar to a phage holin, was shown to promote toxin secretion. This observation shows the linkage between phages and Cd virulence (Fortier, 2017).

Cd can be classified based on ribotype (Lawson et al., 2016) but other classification schemes have been proposed as well. Based on the fact that neither ribotyping nor sequence typing could predict the toxin genotype of 478 whole genome sequenced isolates, it was proposed to also include toxin variant type for comprehensive Cd strain classification (Li Z. et al., 2020).

Clostridioides difficile was first recognized as a pathogen in 1978 (Gerding, 2009), and, soon thereafter, Cd infection (CDI) became the most frequently observed cause of antibiotic-associated diarrhea in hospitals, causing 15–25% of all cases (Bartlett and Gerding, 2008). Toxigenic strains are part of the infant microbiota, and indeed Cd is frequently encountered in the gut microbiota of infants without any symptoms of disease (Adlerberth et al., 2014). However, disruption of microbiota composition by malnutrition or antibiotic use can initiate CDI with increasing Cd toxin concentrations, Cd colonization and biofilm formation in the gut, resulting in a severe disease status (Vedantam et al., 2012). Especially broad-spectrum antibiotics, such as ampicillin, clindamycin and third-generation cephalosporins, are considered important risk factors for CDI (Oldfield et al., 2014). This millennium, more epidemic types of Cd have increased the incidence of CDI and changed epidemiology and severity of disease particularly for elderly in hospital and nursing homes. Those more epidemic types are represented by ribotype 027 (Denève et al., 2009) and ribotype 078 (Goorhuis et al., 2008). Dietary trehalose further enhances virulence of epidemic Cd strains (Collins et al., 2018). On top of this, Cd is now increasingly community acquired (Smits et al., 2016). An important reason for this is the fact that Cd has an extensive repertoire of properties to survive harsh conditions. Spores of this bacterium can survive high temperature, disinfectants, and antibiotics, thereby promoting the spreading of this bacterium. This phenomenon, combined with a compromised immune system of elderly and hospitalized patients, explains their role in nosocomial infections upon antibiotic treatment. Cd colonization can be observed in up to 73% in inpatients above 65 years old (Denève et al., 2009).

Standard treatment of patients with mild CDI relies on withdrawing broad spectrum antibiotics and correcting fluid loss due to the antibiotic-associated diarrhea. Treatment of severe and recurrent CDI currently relies on the use of antibiotics vancomycin and fidaxomicin (Debast et al., 2014). Also fecal microbiota transplantation, first used in the 1950s, has become an established treatment option (Meighani et al., 2017). Fecal microbiota transplantation appeared successful with a cure rate of approximately 93% in patients with recurrent CDI and was therefore accepted as a valid alternative to failing antibiotic treatment (Kassam et al., 2012). Despite the somewhat successful treatments currently used, novel therapeutic strategies are urgently needed.

Preventive strategies for CDI are more important than curing and could be established via reduction of patient susceptibility, prevention of transmission of Cd in health care institutions and rapid diagnosis of Cd patients (Surawicz et al., 2013). Antimicrobial stewardship is also highly recommended in view of prevention. Other options for prevention are being explored, including vaccines that are currently under clinical trials (Henderson et al., 2017; Legenza et al., 2017). A number of studies showed the potential benefit of probiotics in preventing CDI upon timely use in hospitalized adults receiving antibiotics, although the efficacy of probiotics was reduced by half when probiotics were administered more than 2 days after antibiotic treatment was started (Shen et al., 2017). A new potentially preventive measure studied in vitro was the use of the human milk oligosaccharides 2′-O-fucosyllactose and lacto-N-neotetraose. These human milk oligosaccharides or their mixture decreased the presence of Cd in human gut microbiota versus untreated conditions. This decrease coincided with increased acetate and Bifidobacteriaceae levels, thereby suggesting a microbiota-modulated anti-Cd effect, rather than a direct effect of the human milk oligosaccharides (Vigsnaes et al., 2021). These findings suggest that CDI can possibly be prevented by modulating microbiota composition and activity.

An issue in the treatment of CDI is the frequency of recurrence. This is observed in about 15–30% of the cases, making CDI an important clinical and life-threatening issue (Song and Kim, 2019). As second recurrence rates increased to about 40 and to 65% for next recurrences, recurrent CDI is very difficult to treat and health care costs are dramatically increased by CDI patients (Ghantoji et al., 2010). Primary and recurrent CDI were calculated to cost up to $5,000 and $18,000 per case, respectively. Ghantoji et al. (2010) suggested therefore that the enormous increase in the occurrence of CDI with impressive numbers of patients and related high costs for health-care justifies an investment in CDI prevention and control.

In the presence of an already established Clostridioides infection, when antibiotics have proven ineffective (for example because of the emergence of resistance, or because of the risk of side-effects that may exacerbate the disease phenotype), alternative therapeutic options need to be explored. Bacteriophages represent a possible solution in this scenario (Zuo et al., 2018).

Phages of C. perfringens belong to the order of Caudovirales. Table 1 lists the Cp phages described to date in the literature. The best characterized phages belong to the Podoviridae family and Picovirianae subfamily, and the majority of Cp phages originate from poultry. All phages described here selectively target Cp, with no known activity toward other bacteria. Nevertheless, differences in host range of the phages have been observed within the Cp species. For example, phage 3626 was lytic against 11 out of 51 strains examined (21.6%), while phage 8533 was able to lyse only four strains (7.8%) which overlapped with phage 3626 sensitive strains (Zimmer et al., 2002a). The high specificity of phages provides potential therapeutic advantages. By only targeting the infection-causing pathogen, the gut microbiome—necessary for gut homeostasis—may remain intact. Phage engineering, focusing either on optimal physicochemical properties or host range, may further enhance the applicability of phages and enable therapeutic application as described below.

A limited number of in vivo studies has investigated the use of Cp–specific phages, alone or as cocktail, to prevent or control Cp infection-associated morbidity and mortality in broiler chickens. The readout for treatment efficacy concerned parameters related to the extent of pathogen-induced intestinal mucosa damage in the animals, i.e., decreased digestion and absorption, reduced weight gain and increased feed conversion ratio (FCR), as well as mortality rates and lesion scores.

Application of the phage cocktail “INT-401” aimed to prevent or control necrotic enteritis induced in birds challenged with Cp strain C6. This cocktail—composed of five bacteriophages: CPAS-7, CPAS-12, CPAS-15, CPAS-16, and CPLV-42 (Table 1)—was chosen for its broadened host range for Cp and non-infectious nature for non–Cp strains. Orally administered INT-401 was shown to reduce necrotic enteritis-associated mortality of infected broiler chickens by 92%, and significantly improved weight gain and FCR. The bacteriophage preparation could successfully be administered via the feed or water to broilers raised in a floor pen, demonstrating its utility in a commercial setting. A vaccine-derived approach also showed efficacy. However, oral administration of the bacteriophage cocktail resulted in higher reduction of mortality caused by necrotic enteritis compared with alternative methods of administration of either the vaccine or the cocktail preparation (Miller et al., 2010).

The application of a single bacteriophage as food supplement was explored in a subclinical model of necrotic enteritis induced by Cp strain CJ17, harboring netB coding for toxin involved in necrotic enteritis, in broiler chickens. The Siphoviridae phage phiCJ22 (Table 1), isolated from a fecal sample at a chicken farm in South Korea, demonstrated broad antimicrobial activity, with 41/45 of the strains evaluated showing sensitivity (Bae et al., 2021). Bacteriophage treatment effectively controlled necrotic enteritis resulting in reduced mortality, growth performance compensation, reduced numbers of Cp strains in the gut, and improved intestinal integrity. As phiCJ22 demonstrated high stability under varying pH (between 5 and 9), temperatures (up to 60°C) and drying conditions, its use in poultry farms or food processing plants seems feasible (Bae et al., 2021). Similar beneficial results were obtained for a podovirus Cp phage used to treat necrotic enteritis induced by a netB-negative Cp type A strain. Cp re-isolated from the phage-treated group did not display phage resistance (Hosny et al., 2021).

Although none of the studies above observed the development of phage resistant Cp variants, Oechslin (2018) reported that the emergence of phage-resistant bacterial variants was observed in 9 out of 11 experimental studies in which various intestinal bacteria were targeted. Only one of these studies involved Gram-positive bacteria (Enterococcus faecalis). As such it was hypothesized that phage resistance may be a more acute problem for Gram-negative bacteria with respect to resistance selection. However, as reported in this review and Supplementary Table 2, phage resistance was potentially observed in various studies involving Cd, e.g., in the case of Cd phages phiCDHM2 (Nale et al., 2016b) and CD140 (Ramesh et al., 1999). Whether or not the intestinal environment, as suggested by Oechslin, may be more susceptible to the evolution of phage resistance as a result of its complexity remains to be investigated. From the reviewed studies on emerging phage resistance it is clear that, although at a fitness cost for the bacteria, phage resistance is not necessarily accompanied by decreased infectivity in the intestinal milieu (Oechslin, 2018).

In addition to phages, phage-derived proteins have also attracted interest as novel antimicrobials. Endolysins, in particular, have gained traction in recent years due to their broad lytic activities against Gram-negative and Gram-positive bacterial cells (Love et al., 2020; Abdelrahman et al., 2021), including Cp. Other advantages of this strategy over whole phages include the absence of toxins or virulence genes and potentially a higher stability and therefore broader application range. In general, lytic enzymes are composed of two distinct domains connected via a linker. These include a N-terminal catalytic domain that hydrolyzes specific sites within peptidoglycan and a C-terminal cell wall-binding domain.

In 2002, the first endolysin from Cp phage phi3626 was characterized (Zimmer et al., 2002b). Since then, other phages and associated endolysins have been described, as shown in Supplementary Table 1. Virtually all endolysins originating from Cp phages exclusively showed lytic activity against Cp. A recent study on the endolysin Psa and its catalytic domain (Psa-catalytic domain) shed light on the governing factors of this specificity (Sekiya et al., 2021a,b). The species specificity of Psa lytic activity primarily depended on the binding specificity of the cell wall-binding domain, as also observed for other endolysins (Zimmer et al., 2002b). Based on X-ray and modeling studies, the Psa-catalytic domain was also found to exhibit species-specific enzymatic activity. This was apparently due to the unique Cp peptide bridges that are close to the bond that is cleaved by the endolysin. In Cp, this bridge involves the cross-linking by L,L-diaminopimelic acid, while directly cross-linked meso-diaminopimelic acid is found in most other clostridia. L,L-diaminopimelic acid is accommodated especially in binding groove-2 of the Psa-catalytic domain. This is in line with the observed lytic activity of Ply3626 toward C. fallax, implicated in gas gangrene. Some C. fallax strains share the same type of peptidoglycan (A3γ) as Cp, i.e., an L,L-diaminopimelic acid-glycine bridge to the terminal alanine of the opposite peptide chain (Zimmer et al., 2002b).

The applicability of Cp endolysins have so far only been studied in vitro. Lysins ZP173, ZP278, and Psm (Supplementary Table 1) were investigated for their utility to control food contamination (Kazanavičiūtė et al., 2018). The endolysins examined were lytic in vitro against various strains. Under conditions of improperly kept food, the lysins protected cooked turkey from Cp NCTC8237 spoilage. Some of the phage endolysins (Psm and ZP173) conferred better protection than nisin—the only bacteriocin approved as a food preservative. In contrast to nisin, these endolysins were able to completely eradicate the bacteria with no colonies after 18 or 43 h of incubation (Kazanavičiūtė et al., 2018). The fact that the lysins were expressed by Nicotiana benthamiana leaves posed an advantage over Escherichia coli expression when it concerns food applications. In addition to high protein yields, plant-based expression systems using Generally Recognized as Safe hosts (food species) are anticipated to be safe food additives or processing aids even if not completely purified. Therefore, production costs and process complexity may be reduced (Kazanavičiūtė et al., 2018). In addition to the lysins described by Kazanavičiūtė et al. (2018), endolysin PlyCP41 was also successfully produced in plant tissue with suggested use of crude plant sap as potential Cp antimicrobial agent (Hammond et al., 2019).

An interesting premise for increased antimicrobial activity of endolysins is the use of a “cocktail,” in analogy to phage cocktails. To our knowledge, this strategy has to date only been tested in vitro. In the case of Cp, the combination of endolysins Psa and Psm (Supplementary Table 1) showed a synergistic effect on the lysis of the bacterium. This was due to the fact that both endolysins target different regions in the peptidoglycans (Sekiya et al., 2021a). A synergistic killing effect was already demonstrated for two phage-encoded lysins for Streptococcus pneumoniae, which similarly target distinct regions in the cell wall of the host bacterium (Loeffler and Fischetti, 2003). However, the in vivo efficacy of such a combination approach remains to be confirmed. Suggested applications of the enzyme mixture included the spraying of areas to treat drug-resistant Cp, pre-mixing into poultry feed against foodborne Cp infections, and enteric capsule preparations for precision targeting to fight intestinal diseases (Sekiya et al., 2021a).

An alternate combination therapy showing initial promise is the expression of a biologically active anti-Cp endolysin by a probiotic bacterium that has the capability to exclude Cp from the gastrointestinal tract of poultry birds. Lactobacillus johnsonii FI9785, shown to reduce colonization and shedding of Cp in chickens after a single oral dose, was engineered to express CP25L (Supplementary Table 1). The nisRK two-component regulatory system from the Lactococcus lactis nisin A biosynthesis operon was integrated into the bacterial chromosome to allow constitutive expression of the endolysin under the control of the nisA promoter. Secretion of the active endolysin to L. johnsonii culture supernatants was achieved using a signal peptide (SLPmod) and subsequent lytic activity was demonstrated in vitro by growing colonies of L. johnsonii on plates incorporating autoclaved Cp cells. Zones of lysis developed around colonies expressing the lysin (Gervasi et al., 2014a). However, survival of the L. Johnsonii strain was shown to be cumbersome in follow-up co-culture studies, which mirrored the complex microbial communities and varying conditions of the gastrointestinal tract. It was therefore concluded that the delivery of endolysin from a species that is able to better cope with conditions in selected areas of the gastrointestinal tract might be more suitable (Gervasi et al., 2014b).

In addition to the application of wild type endolysins, fused endolysins have been investigated for introducing either improved or novel properties. These studies are described in the engineering section.

The first study to characterize Cd-specific phages as a lead up to assessing their therapeutic merit involved four clinically relevant phages (Goh et al., 2005). These phages—isolated from patients with Cd-associated diarrhea—were found to be non-lytic and belong to the Myoviridae (phiC2, phiC5, phiC8) and Siphoviridae (phiC6) subfamilies of Caudovirales. Cd phages described before and since (Supplementary Table 2) have also been classified to these two subfamilies and are of a similar lysogenic lifestyle. The lysogenic nature of Cd phages has been suggested to be due to high integrase concentrations, favoring the lysogenic rather than the lytic cycle, as observed for phage JD032 (Li T. et al., 2020). However, several of the phages identified are not integrated in the host genome, but are episomal instead (Smits et al., 2022). To date, this has mainly been observed for large phages of around 130 kbp (i.e., phiCD5763, phiCD5774, phiCD211), although phiCD38-2 (41 kbp) is also presumed to be a plasmid prophage (Sekulovic et al., 2011).

As illustrated in Supplementary Table 2, several of the phages have broad in vitro activity, covering a multitude of Cd ribotypes and strains. For example, phiCDHM3 was able to lyse 31 isolates representing 12 ribotypes (Nale et al., 2016b). Another myovirus, CDKM9, was found able to infect ribotype 027, an activity generally restricted to siphoviruses (Rashid et al., 2016).

In terms of the molecular interactions between host and infecting phage, little is known. However, binding studies showed that phiHN10 (Supplementary Table 2) can specifically bind to S-layer proteins, indicating that these proteins may act as host receptors (Phothichaisri et al., 2018). From the host perspective, the conserved cell wall protein CwpV from Cd was shown to confer resistance to infection by different phages, including members of the Siphoviridae and Myoviridae families (Sekulovic et al., 2015). Cell wall protein CwpV is thought to prevent phage DNA from entering the cell, a mechanism reminiscent of superinfection exclusion systems encoded by other lysogenic phages (Sekulovic et al., 2015).

Whole genome sequencing of phiCDHM1 (Supplementary Table 2) showed that this phage contains homologs of accessory gene regulator (agr) genes, which make up the host quorum sensing system (Hargreaves et al., 2014). The phage agr genes were proposed to be of a host origin—acquired via horizontal gene transfer or phage infection—and either evolving within the phage genome or representing a subtype. As the genes were transcribed and retained during replication, the genes are likely of benefit in promoting the survival and replication of phiCDHM1 and its host. This may involve lysogen allelopathy (i.e., enhancing the fitness of their hosts through killing of susceptible competitors) or inhibition of secondary phage infection (Hargreaves et al., 2014).

Phage phiSemix9P1 (Supplementary Table 2), isolated from soil, has a functional binary toxin locus (CdtLoc) in its genome. Although unique to this phage, it demonstrates that lysogenic phages may play an important role in the spread of the toxin gene to pathogenic Cd strains. Similar chromosomal incorporations of virulence factors have not been observed for other clostridia (Riedel et al., 2017).

The fact that all characterized Cd phages are lysogenic poses some additional challenges compared to, for example, lytic Cp phages when considering therapeutic applications. As described in various studies, lysogenization of the Cd host may result in resistance by multiple mechanisms, such as immunity to superinfection. The lysogen may also show increased toxin production, as observed for CD274/phiCD38-2 (Supplementary Table 2). Various strategies have been explored to overcome these hurdles, including phage cocktails, phage-derived proteins, and engineering approaches.

The efficacy of phage monotherapy has been investigated in various studies. In an in vitro batch fermentation model, prophylactic phiCD27 (Supplementary Table 2) treatment resulted in significantly reduced Cd cell numbers and prevented TcdA and TcdB production. Phage phiCD27 treatment after 6 h of Cd growth was less effective and resulted in re-growth of the pathogen due to lysogenization of phiCD27 with its host strain NCTC 1120 (Meader et al., 2010). Phage resistance was also observed in one of three replicate experiments during prophylactic phiCD27 therapy in a weir cascading system simulating the proximal-to-distal gut environment. In two other experiments, Cd cell numbers were significantly reduced in this in vitro human colon model (Meader et al., 2013). In both studies on the effectiveness of phiCD27, commensal microbiota was unaffected by the treatment.

Single-phage treatment of Cd strains CD105LC2 and CD105HE1 resulted in limited clearing of the cultures in an in vitro study by Nale et al. (2016b). The most effective phage (phiCDHM2; Supplementary Table 2) showed undetectable bacterial counts by 5 h, but regrowth by 24 h, which may be due to emerging phage resistance. Combining multiple phages prevented the appearance of resistant/lysogenic clones. With phage cocktails, a lytic activity on 18 of the 21 clinically relevant ribotypes tested was observed, including ribotypes 002, 005, 014/020, 015, and 078 (Nale et al., 2016b). A four-phage cocktail (phiCDHM1-phiCDHM2-phiCDHM5-phiCDHM6) was also more effective in reducing CD105LC2 biofilms in vitro (Nale et al., 2016a) and eradicated cultures of a clinical ribotype 014/020 strain in fermentation vessels spiked with combined fecal slurries from four healthy volunteers (Nale et al., 2018). Consistent with other studies, a prophylactic regimen was more effective at clearing Cd than a remedial regimen, where the phage cocktail was added after 5, 24, 36, and 48 h after bacterial inoculum was added. Nevertheless, Cd was fully lysed during both. Neither treatment regimens had a significant negative impact on the viability of five common major bacterial groups in the human gut (enterococci, bifidobacteria, lactobacilli, total anaerobes, and enterobacteria), concurring with previous studies by Nale using this four-phage cocktail. On the contrary, a significant abundance increase was observed in the phage-treated vessels for enterobacteria, lactobacilli, and total anaerobes, which was suggested to add to the prevention of Cd colonization (Nale et al., 2018).

In the first in vivo study examining single-phage treatment efficacy, phage CD140 (Supplementary Table 2) was orally administered to hamsters challenged with Cd strain 135I (Ramesh et al., 1999). One of the challenges accompanying oral administration is phage viability, as most of the phages succumb under the harsh acidic environment of the stomach. Acid sensitivity of CD140 presented an issue that was overcome by bicarbonate neutralization of gastric acid prior to phage treatment (Ramesh et al., 1999). This approach has also been applied in other studies. Although effective, dilution of phages may impact their effectiveness. In the study by Ramesh, a single phage CD140 dose at the time of Cd challenge was sufficient for all treated animals to survive, in contrast to untreated animals. However, the treated animals were not resistant against Cd re-challenge two weeks after bacteriophage treatment discontinuation. The short-term effectiveness of CD140 indicates that bacteriophages are not retained in the gastrointestinal tract. This is to be expected: if host bacteria are killed bacteriophages will disappear. In the study it was noted that one of the animals receiving multiple bacteriophage doses still developed CDI. The recovered strain was not lysed by CD140, indicating preexisting resistance or the development of resistance to CD140 (Ramesh et al., 1999).

In another in vivo study, phage cocktails of two or four phages were effective in reducing CD105HE1 colonization in a hamster model of CDI, as measured by bacterial counts (Nale et al., 2016b). All animals were sensitized to Cd by orally administering clindamycin 5 days prior to challenge with bacterial spores. The first treatment with single or combination phages was given at the time of Cd challenge, with subsequent doses administered every 8 h until the endpoint of 36 h, when toxin production in the cecum and colon in untreated animals was found to be limited. Subsequently, total bacterial load (spores and vegetative cells), viable counts of recovered bacteria and numbers of spores in the cecum, colon and cecal contents were determined. The reduction in levels of bacterial counts were at least 4 log units (lumen) or approximately 2 log units (tissues) for the two-phage combinations. The four-phage combination (phiCDHM1-phiCDHM2-phiCDHM5-phiCDHM6; Supplementary Table 2) showed similar results. Despite reduced colonization, Cd was still detectable in the cecum and colon of most animals. However, analysis of individual isolates from treated animals showed that the recovered bacteria were still sensitive to the phages used in the treatment. It was therefore suggested that either the multiplicity of infection was too low, or the bacteria were physiological insensitive to infection of the phage encountered. In addition to colonization, phage treatment also affected sporulation. Especially in the colon, the number of spores recovered in treated animals was substantially lower than in controls, dropping 2 log units. For analysis of disease onset involving CD105HE1, treatment with phages at the time of Cd challenge was continued every 8 h until the experimental endpoint (i.e., a core body temperature drop to 35°C at which recovery is no longer possible). The four-phage treatment delayed the CDI-associated core body temperature drop on average with 33 h compared to untreated hamsters (Nale et al., 2016b). Thus, phage cocktail treatment could delay—but not prevent—irreversible disease onset. The effectiveness of the four-phage cocktail was also compared to monotherapy in a Galleria mellonella larva CDI model induced with strain CD105LC2 (Nale et al., 2016a). Again, the cocktail was more effective in killing the pathogen. As observed in other studies, prophylactic treatment showed better results than remedial treatment. In the latter set-up, phage treatment was successful as an adjunct to the first-line antibiotic vancomycin (Nale et al., 2016a).

Moving toward a more human translatable model of phage intervention, Shan et al. (2018) studied the in situ dynamics between phiCDHS1 (Supplementary Table 2), C. difficile CD105LC1 (ribotype 076) and HT-29 cells. The choice of HT-29 cells for simulating colonic CDI was based on the pathogen’s ability to adhere to this human colon tumorigenic cell line (Drudy et al., 2001) and the cell line’s ability to withstand anaerobic conditions (Shan et al., 2018). The study showed that free Cd numbers were more effectively reduced in the presence than absence of HT-29 cells by phiCDHS1. Adhered Cd was also significantly reduced compared to non-phage-infected controls, whereas phage production was increased in the three-component system of phage, pathogen, and human colonic cells. No cytotoxic effect of the phage on HT-29 cells was observed, indicating that toxin release was not occurring in this study. However, the timing of intervention may be key in this respect. To prevent toxin release, Cd cells need to be killed before they can reach the toxin-producing stationary phase (Hofmann et al., 2018). As both Cd (60%) and phiCDHS1 (70%) were found to adhere to HT-29 cells, it was proposed that the increased phage activity in the presence of the human cells is due to facilitated interaction between the two entities on the HT-29 monolayer. Phage phiCDHM6 (40%), but not phiCDHM3, also adhered to the human cell line (Shan et al., 2018). The observation that phages colonize the gut and protect it against bacterial infections supports the benefits of prophylactic use observed in other studies.

To overcome the emergence of lysogens and the potential release of toxins, the application of phage-derived endolysins targeting Cd has been explored in a limited number of studies (Table 2). Especially the benefits of using a catalytic domain-only protein has been a major focus of research.

Bacteria produce a variety of particles that resemble phage tails and possess antibacterial activity despite the absence of a capsid and nucleic acid. The use of these phage tail-like particles (PTLPs) has attracted attention as an alternate means of combating Cd (Sangster et al., 2015). PTLPs have been identified in both Gram-positive and -negative bacteria and are potentially remnants of lysogenic phages. Several Cd strains produce PTLPs upon induction of the SOS response, thereby killing competing, non-self Cd strains (Gebhart et al., 2012). The homology of PTLPs to phage tail proteins may open avenues to rational design of such bacteriocins in the future. However, for the purpose of this review, the focus will be on the application of endolysins. The current status of PTLPs has recently been reviewed by Dams et al. (2019) and Heuler et al. (2021).

The first endolysin to be characterized was phiCD27-derived CD27L (Table 2). The enzyme was successfully overexpressed in E. coli and showed broad activity: all 30 Cd strains tested were sensitive, including two strains of the hypervirulent ribotype 027. Non-Cd strains, including some commensal gastrointestinal bacteria, were insensitive to the enzyme, indicating that CD27L is not expected to disturb the gut microbiota. For envisaged gastrointestinal applications, expression of CD27L by a heterologous L. lactis system was established (Mayer et al., 2008). Despite the observed lytic specificity in the first study, CD27L later was found able to lyse various less closely related species, i.e., Bacillus amyloliquefaciens, Bacillus cereus, and Bacillus subtilis (Mayer et al., 2011). Notably, all sensitive species shared the same peptidoglycan type A1γ. Truncation of the endolysin to its mere N-terminal catalytic domain (CD27L_1–179), enhanced lytic activity and somewhat broadened the host range compared to the full-length enzyme. However, specificity toward peptidoglycan type A1γ species remained and was attributed to the configuration of the substrate binding site (Mayer et al., 2011).

The PlyCD catalytic domain (PlyCD_1–174) similarly possessed superior lytic activity, with a wider Cd strain spectrum than the full-length structure (Table 2). PlyCD_1–174 was not active against other potential intestinal Firmicutes, except B. subtilis. When following vancomycin pre-treatment, PlyCD_1–174 significantly reduced the titer of Cd in vitro at doses at which both interventions were ineffective as monotherapy. The catalytic domain of PlyCD also demonstrated ex vivo efficacy against Cd vegetative cells infecting the murine large intestine in the presence of its contents. Colonization by Cd was significantly reduced compared to controls. Rectal delivery of PlyCD_1–174 via an enema to test its efficacy in an in vivo CDI mouse model was hindered by blocked colons, leading to confounding results (Wang et al., 2015).

To circumvent difficult culturing conditions inherent to anaerobic pathogens, Fujimoto et al. (2020) developed a metagenome analysis pipeline for discovering novel prophages from overlapping bacterial and viral genomes found in the gut of 101 healthy individuals. As a proof of concept focusing on Cd, 10 open reading frames homologous to known endolysins were identified from prophages from the healthy individuals as well as 17 clinical isolates. The endolysins were subsequently all shown to be effective in vitro against strain ATCC 43255. Tested endolysins also demonstrated in vivo activity in a murine CDI model (Fujimoto et al., 2020).

Recently, an endolysin (CWH) from phiMMP01 was characterized that differs from other published endolysins and contains two catalytic domains (Table 2). Truncation studies showed that the full-length protein and the protein consisting of the two catalytic domains (CWH_351–656) had a similar host range, with the latter showing faster lysis. Apart from Cd (Table 2), the constructs were also able to lyse some B. cereus and Clostridium scindens strains. CWH_351–656 additionally showed low activity against E. faecalis and Listeria innocua (Mondal et al., 2021).

The studies conducted so far concur in a higher lytic activity and broader host range for truncated endolysins consisting of only the catalytic domain(s). This can be explained by the autoinhibitory role that has been attributed to the cell wall-binding domain, obscuring the catalytic site in absence of a cognate cell wall (Low et al., 2005). This presumed allosteric inhibition is lifted upon binding of the cell wall-binding domain to the pathogen cell wall. Alternatively, structures on the outer surface of the Cd cell (e.g., S-layer proteins) or other non-steric cell wall factors (e.g., charge complementarity) could simply hinder peptidoglycan accessibility to exogenously added full-length lysin (Wang et al., 2015).

To overcome narrow host range and bacterial resistance, various engineering strategies have been pursued for phages and their endolysins. These include random and site-directed mutagenesis, the construction of fusion proteins, and the use of endogenous CRISPR-Cas systems as a payload to circumvent lysogenization. In order to guide engineering efforts, a molecular understanding of the multiple steps involved in phage—host interactions (Bertozzi Silva et al., 2016) is needed. For example, key to the host range is the binding of the phage’s receptor binding proteins (RBPs) to host receptors. To date, information of RBPs and corresponding host receptors is limited in general (Dunne et al., 2021), and specifically for Cp and Cd targeting phages, as described below.

A random mutagenesis study of C. perfringens ATCC 13124 demonstrated that this strain was resistant against phage CPS1 (Table 1) when a gene, encoding UDP-glucose 4-epimerase (GalE), was mutated. In conjunction with a reduced mutant host production of capsular polysaccharides and reduced host absorption by competitive inhibition by glucosamine and galactosamine, it was concluded that capsular polysaccharides act as a receptor to CPS1 (Ha et al., 2019). However, the corresponding receptor binding protein on CPS1 was not identified.

Mutating the conserved catalytic site of an endolysin to modify its host range may negatively impact its stability and consequently its applicability as antimicrobial agent. Ritter and Hackel (2019) therefore explored the application of molecular modeling techniques combined with yeast surface display and high-throughput sequencing for identifying thermally stabilizing mutations of the LysCP2 (Supplementary Table 1) catalytic domain. Using sequence homology-guided generative models, stabilizing mutations distal from the catalytic site were predicted. Stability information of multi-mutants was used to extract stabilizing single-mutants. From the in silico analyses, five designed proteins, incorporating 5–6 point mutations each, were selected, produced in E. coli and evaluated for thermal stability and activity. Four of the five proteins displayed improved thermal stability (max. 4°C). Moreover, three of the proteins exhibited a 4- to 5-fold greater activity than the wild type LysCP2 catalytic domain following thermal treatment (42°C for 30 min) in a Cp strain ATCC 12916 cell wall degradation assay (Ritter and Hackel, 2019).

Mayer et al. (2011) demonstrated that molecular modeling-supported analysis of an endolysin active site can guide host specificity modification. By comparing the catalytic sites of Cd endolysin CD27L_1–179 and Listeria endolysin PlyPSA, rational site-directed mutagenesis was applied to broaden the CD27L_1–179 host range to include Listeria species (Table 2; Mayer et al., 2011).

In an attempt to modulate efficacy, the catalytic domain of LCD (Table 2), a phiC2-derived endolysin homologous to PlyCD, was fused to the functional domain of human alpha-defensin 5, resulting in the fusion protein LHD (Peng et al., 2019). Human alpha-defensin 5 had demonstrated activity against hypervirulent Cd strains by mechanisms of membrane depolarization and bacterial fragmentation specific to this pathogen (Furci et al., 2015). LHD, showed higher lytic activity than LCD toward strains R20291 (ribotype 027) and M120 (ribotype 078) and similar activity toward VPI 10463 (ribotype 087), LC693, 1377 (ribotype 012), and CD630 (ribotype 012). The minimum inhibitory concentration of LHD was about 2-fold lower that LCD. Oral delivery of LHD to R20291-challenged mice decreased diarrhea, abolished mortality and reduced fecal levels of Cd spores and toxin (Peng et al., 2019).

The lysogenic nature of Cd phages is a limiting factor in their therapeutic application. With infection potentially leading to lysogeny, phage resistance can arise through a variety of mechanisms (Bondy-Denomy et al., 2016), as reported in multiple studies. To overcome this hurdle, Selle et al. (2020) engineered phage phiCD24-2 (crPhage) to carry a lethal genome-targeting CRISPR array, redirecting the endogenous CRISPR-Cas3 activity against the bacterial chromosome. Both host range and morphology were unaffected by the genomic modification. However, the maximum reduction of the numbers of colony forming units and the suppression of Cd culture growth (24 h time point) was higher for crPhage. Nevertheless, Cd treatment with both the wildtype (wtPhage) and crPhage resulted in the occurrence of lysogens. By knocking out key lysogeny genes (Δlys), the resistance against wtPhage and crPhage could be eliminated (Selle et al., 2020). As opposed to wtPhage, mice challenged with C. difficile CD19 spores showed significantly reduced fecal Cd burden when treated with crPhage. However, as observed in vitro, Cd vegetative cells were not eradicated by phage treatment and lysogens were observed. Several of the crPhage lysogens had lost part of the inserted genome. Thus, lysogens with an intact genome-targeting CRISPR are likely unstable (Selle et al., 2020). Treatment of mice with crPhage Δlys showed lower fecal C. difficile burden than the wtPhage Δlys. crPhage Δlys treatment was also devoid of tissue damage in the host, which was observed for crPhage and wtPhage Δlys. Lysogeny also occurred in vivo for wtPhage Δlys and crPhageΔlys, potentially because other Cd prophage genes are able to functionally complement the excised genes. Further elucidation of the mechanisms of lysogeny and its impacts on host cell physiology was deemed necessary (Selle et al., 2020).

Apart from host specificity and lysogeny, delivering the phage at its target at the right dose while retaining its activity has been investigated. Phage infectiousness is subject to various physical and environmental stressors. Thus, relevant considerations for phage engineering include the routes of delivery and conditions encountered, stability of the phage during storage and formulation, and properties necessary for its envisaged application (Jończyk-Matysiak et al., 2019).

Thermal stability was conferred to endolysins targeting Cp by producing chimeric proteins (Swift et al., 2015, 2019). The cell wall-binding domain derived from an endolysin of a Cp bacteriophage targeting strain Cp39 (including PlyCP10, PlyCP41, and PlyCP26F; Supplementary Table 1) was fused with the catalytic domain derived from an endolysin of a thermophilic bacteriophage of Geobacillus strains. All fifteen chimeric proteins examined showed higher thermostability than the native Cp endolysins, and were able to kill swine and poultry disease-associated strains of Cp (Swift et al., 2019). Enhanced thermostability is of importance as animal feed is often heat-treated during production of feed pellets.

Several engineering steps have been taken to ensure survival of orally delivered phages at low pH. In a study by Vinner et al. (2017), Cd phage CDKM9 (Supplementary Table 2) was encapsulated in a pH responsive polymer using a flow focusing glass microcapillary device. The microcapsule was designed to protect phages in their transit through the stomach, and to disintegrate in the infected colon when encountering the typical pH of 7, thereby releasing the phages. Indeed, the encapsulated phages were stable during a 3 h exposure to simulated gastric fluid at pH 2 and underwent a burst release triggered at pH 7 (Vinner et al., 2017). This would technically support the oral ingestion of phage particles against CDI.

The isolation of phages from environmental samples and the subsequent characterization of their host specificity is a time-consuming task, often accompanied by trial and error. Machine learning models have been generated to facilitate the host spectrum characterization of phages and are largely based on genomic information. For example, the web application HostPhinder predicts hosts based on similarities (co-occurring k-mers) between the query phage genome and the genomes of reference phages with known hosts (Villarroel et al., 2016). Recently, a conceptual framework was described for the construction of digital phagograms, a 3-layered modeling approach that can potentially predict whether phage—bacteria interactions result in productive infections. In this machine learning approach, the phage infection process is split into multiple consecutive steps that reflect the underlying microbiology at the species level. The first layer captures host recognition and adsorption via bacterial surface receptors, e.g., by using data regarding RBPs. The second layer covers the evolutionary dynamics between host and phage, e.g., genetic defense mechanisms such as CRISPR-Cas. The last layer incorporates metabolic conversion of the host by the virus, e.g., by using transcriptomics data (Lood et al., 2022).

A model described by Boeckaerts et al. (2021) focused on clinically relevant pathogens of the ESKAPE group. This group includes C. difficile, although it represented only 2% of the data set. The latter consisted of 1170 phage RBPs with known hosts. By transforming RBP calculated properties (e.g., protein sequence, DNA composition, structural features) into a numeric vector, machine learning models were trained to predict phage-pathogen infection ability. Notably, it was observed that highly dissimilar RBPs could still target the same host. The final model outperformed BLASTp host predictions for RBPs for which the similarity to annotated RBPs in the model fell below 75% (Boeckaerts et al., 2021). BLAST constitutes the gold standard protein sequence similarity tool. Recently, the BLAST extension Phirbo was described making the tool more suitable for predicting phage—host interactions (Zielezinski et al., 2021). Increased performance was obtained by incorporating sequence similarities of both phages and bacteria to a reference data set, expressed as relatedness.

A limitation of the RBP-based (Boeckaerts et al., 2021) and other models (e.g., Villarroel et al., 2016) is that predictions are not on strain, but species level. Strain-level predictions are currently hindered by lack of data but appear feasible in the near future giving the attention phages are currently fostering. A model that does predict phage—pathogenic bacterial strain interactions is one described by Leite et al. (2019). However, Clostridiaceae (0.3%) were underrepresented in this model, which was based on primary protein structure information and domain-domain interaction scores (Leite et al., 2019). Therefore, the phages and pathogens used for the training of a model, translating to an applicability domain, should be considered when applying in silico models.

A limitation inherent to all current in silico models is that not all aspects of host specificity are represented in the models. Factors, such as immunity to CRIPSR-Cas systems, superinfection resistance, host receptors etc., are not accounted for. Another limitation of models is the availability of data regarding strains to which phages are inactive, despite being an essential part of model training sets. Although encountered in some of the in vitro studies reviewed here, such data is generally difficult to obtain.