Personalized bacteriophage therapy outcomes for 100 consecutive cases: a multicentre, multinational, retrospective observational study

1Antimicrobial resistance (AMR) is a prominent global health threat with an estimated 1.27 million attributable deaths in 2019 and there is an urgent need to seek alternative antimicrobial strategies. Bacteriophage therapy (BT), the use of bacteriophages—the viruses of bacteria—to treat bacterial infections, was first applied by Félix d’Hérelle in 1919, and further developed and applied in the former Soviet Union.

2A recent systematic review confirmed that BT can generally be considered as safe, with a low incidence of adverse events, and could be a promising strategy against AMR. However, high-quality trials are required to make useful predictions on the outcome of bacteriophage treatments. A number of companies are currently attempting to develop and market defined broad-spectrum BT products in compliance with contemporary requirements, which involves good manufacturing practices (GMP) certification, preclinical research (toxicity and pharmacology) and conducting randomized controlled trials (RCTs). However, the handful of bacteriophage RCTs that have been performed so far have not brought the expected results in terms of effectiveness. A commonly reported reason for these disappointing results is the use of invariable one-size-fits-all bacteriophage products.

3In contrast, an increasing number of successful BT cases are reported in the scientific literature. Irrespective of an obvious positive-result publication bias, most of these successful cases used tailored bacteriophage products. In addition, these personalized bacteriophage preparations, which were shown to target the infecting bacteria in vitro before their clinical application, were often used in combination with antibiotics. When appropriate, bacteriophage preparations were adapted to counter bacterial resistance that had emerged against the applied bacteriophages during BT, or bacteriophages were pre-adapted (‘trained’) or engineered to be more effective.

4Here we report the retrospective, observational analysis of the first 100 consecutive BT cases of difficult-to-treat infections, enabled by a Belgian consortium. Because all BT cases were included in this study, not only successful, interesting, or challenging cases, we were able to (1) evaluate how often personalized BT produced a positive clinical outcome (general efficacy) and (2) identify functional relationships that are general in all cases.

5Considering the relatively high number of combined categorical and numerical variables in the analysed data, the majority of patients were unique cases in most of the variables. As a result, on this dataset, no inferential statistics could be applied because these data were neither a random nor a representative sample of a population of BT-treated patients. As such, any data analysis can only be interpreted as information pertaining to the analysed patient population.

6Nevertheless, the knowledge gained from these cases is likely to help physicians to select effective treatment protocols and design future clinical trials.

7A Belgian BT consortium, consisting of the Queen Astrid Military Hospital (QAMH), KU Leuven and Sciensano (formerly known as the Scientific Institute of Public Health), facilitated BT in about 140 difficult-to-treat infections in patients in Belgium and abroad (as of July 2023), not taking into account the patients treated in the context of prospective clinical trials. The selection of patients was largely based on clinical need, regulatory approval and the availability of well-characterized bacteriophages targeting the infecting bacteria (Extended Data Fig. 1). Personalized bacteriophage preparations were produced at the QAMH in accordance with the rules in force in the territory at the time of their use in clinical practice. Of note, most selected cases concerned personalized BT as salvage therapy after standard antibiotic treatments had failed. Quality and safety of the bacteriophage preparations were verified by Sciensano according to the specifications of the Belgian bacteriophage active pharmaceutical ingredient (API) monograph, that is, the genomic analysis of the bacteriophage and its bacterial production host (with an emphasis on safety), and the determination of lytic activity (titre), pH, bioburden (total viable aerobic count), bacterial endotoxin level and genome sequence (identity and purity) of each bacteriophage API batch. The BT protocols that were suggested to the treating physicians were based on the experiences of the George Eliava Institute of Bacteriophages, Microbiology and Virology (Eliava Institute) in Tbilisi, Georgia (personal communications), and on the application instructions of the Ministries of Health and of Medical and Microbiology Industry of the former Union of Soviet Socialist Republics (USSR)–.

8During the study period (1 January 2008 to 30 April 2022), 1,066 BT requests were submitted to the QAMH. These requests resulted in 100 BT cases (9.4%). Two hundred and sixty BT requests addressed to the QAMH between April 2013 and April 2018 were analysed in detail. Only 15 (5.8%) of these 260 requests resulted in actual BT. Two hundred and forty-five requests were rejected for diverse reasons: 70 applicants (26.9%) did not respond to requests for additional information; 124 requests (47.7%) concerned bacterial species against which no bacteriophages were available at the QAMH; for 46 requests (17.7%), other therapeutic options were considered more opportune; and in 5 cases (1.9%) the available bacteriophages did not target the patients’ infecting bacterial strains. Rejected applications were usually referred to BT centres abroad. We consider these percentages as representative of the present patient cohort, minding an increase in the percentage of requests that resulted in BT (9.4% versus 5.8%), which is due to the increasing number of therapeutic bacteriophages in the QAMH collection. Time to treatment was dependent on whether suitable quality-controlled bacteriophages were available on hand (these could be provided immediately), or whether bacteriophages needed to be produced at the QAMH and quality and safety tests performed by Sciensano (this would take on average of 3 weeks in non-emergency cases).

9A retrospective analysis of a de-identified BT database containing demographic, bacteriophage product and clinical data showed that personalized BT of 100 consecutive patients targeted 114 difficult-to-treat infections (as diagnosed by the treating physicians), including 14 second-site infections. Baseline characteristics of the patients are presented in Supplementary Table 1 and Extended Data Table 1, and provide an overview of these BT cases, which were performed by a total of 63 Bacteriophage Therapy Providers in 35 hospitals, 29 cities and 12 countries (Fig. 1a). Twenty-seven of the 100 BT cases/patients were previously reported,–. Since 2008, the number of BT cases performed under the umbrella of different regulatory frameworks and facilitated by the Belgian consortium has increased steadily (Fig. 1b). The prevalence of the main infection types is shown in Fig. 1c. The most common indications for BT include lower respiratory tract infections (LRTI; 25.4% (29/114 infections)), skin and soft tissue infections (SSTI; 22.8% (26/114)), bone infections (BoneI; 14.0% (16/114)) and upper respiratory tract infections (URTI; 11.4% (13/114)). Fourteen patients presented with a second-site infection, more specifically a bloodstream infection (BSI; n = 10), a urinary tract infection (UTI; n = 2), an SSTI (n = 1) or a URTI (n = 1). Age and gender distribution are shown in Fig. 1d. The median age of the patients was 53 years (1–91 years), and 56.7% of the patients were male. Of note, 5 patients were 1 year or younger. Fourteen bacterial species were targeted (Fig. 1e), with the highest prevalence for Pseudomonas aeruginosa (49/100 patients) and Staphylococcus aureus (39/100 patients).

10Twenty-six individual bacteriophages (Supplementary Table 3) and six defined bacteriophage cocktails (Supplementary Table 4), including two commercially available cocktails (PyoPhage and IntestiPhage) produced by the George Eliava Institute of Bacteriophages, Microbiology and Virology (Eliava Institute) in Tbilisi (Georgia), were used. Bacteriophages were provided by the QAMH and 16 Bacteriophage Donors affiliated to 10 institutes in 7 countries.

11Most BT providers adhered to BT protocols proposed by QAMH physicians, which resulted in a surprisingly small variation in BT protocols within a given indication. Table 1 provides a general overview of these protocols, while the individual protocols of the 100 cases are listed in Supplementary Table 1. Bacteriophages were administered intravenously to 20 patients (Supplementary Table 1); in 10 of them as stand-alone BT, in 10 concomitantly with intralesional (n = 4), nebulized (n = 3), topical (n = 2) or generalized (multiple application routes; n = 1) bacteriophage application. In 10 patients, intravenous bacteriophages were used to treat or prevent bloodstream infections. In 69.3% (79/114) of targeted infections, bacteriophages were administered in combination with standard-of-care antibiotics.

12The most frequently used bacteriophages, that is, Staphylococcus bacteriophage ISP (33 patients) and P. aeruginosa bacteriophages 14-1 (22 patients), PNM (21 patients) and PT07 (18 patients) (Supplementary Table 3), were regularly (one to two times per year) adapted using a selection of three to five recent bacterial strains of concern. In addition, 13 bacteriophages were specifically pre-adapted to lyse the patient’s bacteria in a therapeutically relevant manner (Methods), that is, to produce stable lysis (without emergence of bacteriophage-insensitive mutants) in liquid culture for typically 24–48 h at a multiplicity of infection (MOI) ≤ 1 (Extended Data Table 2). The genomes of the pre-adapted bacteriophages were sequenced, analysed and compared to those of their original precursors as part of the Sciensano coordinated SAPHETY project (https://www.sciensano.be/en/control-and-safety-assessment/safety-therapeutic-bacteriophage-preparations), which focuses on setting new standards for the quality and safety of therapeutic bacteriophage products. One pre-adaptation effort increased the activity of S. aureus bacteriophage ISP against an S. epidermidis clinical isolate in view of personalized BT. The pre-adaptation process (four serial passages) resulted in missense mutations in three genes, including a carbohydrate-binding domain protein and a uracil-DNA glycosylase (Extended Data Fig. 2), which are closely related (closest BLAST hits) to two previously identified receptor binding proteins. However, the increased virulence and resistance suppression of the pre-adapted ISP variant was accompanied by a decreased host range. Where the original ISP clone showed a moderate activity (efficiency of plating (EOP) ≤ 0.01) against 3/16 S. epidermidis strains, the adapted variant showed a therapeutically acceptable activity against the patient’s strain only.

13For 21 patients, sufficient and adequate consecutive bacterial samples and/or serum samples were provided, allowing assessment of (1) the potential in vivo emergence of resistance against the applied bacteriophages, (2) in vitro bacteriophage–antibiotic interactions and/or (3) the emergence of bacteriophage immune neutralization (Table 2).

14For 16 patients, sufficient bacterial samples (isolated before, during and after BT) were available to evaluate the possible emergence of bacterial bacteriophage resistance. Whether adequate samples were available was not directly linked to the clinical indications for BT but depended mainly on the treatment centres and their bacteriological monitoring routines. For 5 patients in Table 2, no adequate sample sets were available (indicated with ‘NSA, no samples available’ in Table 2). We observed the in vivo selection of bacterial strains exhibiting a bacteriophage-insensitive phenotype, and the possible underlying phenotype–genotype associations in 7 of these 16 (43.8%) patients (patients 16, 20, 30, 54, 64, 82 and 91 in Table 2). Whole-genome single nucleotide polymorphism (SNP) analysis was performed for bacterial isolates from the patients where bacteriophage insensitivity emerged. In two patients (64 and 82 in Table 2), sequential bacteriophage-susceptible and bacteriophage-insensitive P. aeruginosa isolates were determined not to be clonal. Phylogenetic comparison showed that for patient 82, bacteriophage-susceptible strains belonged to an emerging rare sequence type (ST)235, whereas bacteriophage-resistant strains belonged to the more prevalent multidrug-resistant ST357 (Table 2 and Fig. 2a),. For patient 64, the susceptible strain was ST1233 (same ST as the strains from patient 91), while the resistant strain was determined to be ST549 (Table 2 and Fig. 2a). In these two patients, BT probably selected for P. aeruginosa strains that were not a suitable host for the applied bacteriophages. Clinical improvement was reported in both patients.

15SNPs or deletions in genes related to the bacteriophage receptor were assumed to be the basis of the resistance phenotype in five patients (16, 20, 30, 54 and 91 in Table 2). In three of them (patients 30, 54 and 91), the targeted P. aeruginosa strains were not eradicated. The selection of bacteriophage-resistant mutants in two patients (16 and 20) was previously described,. In patient 16, an isolate of the targeted Achromobacter xylosoxidans strain emerged to harbour a missense mutation in the gene coding for the putative colicin I receptor Cir, which was identified as a bacteriophage receptor. In patient 20, a missense mutation occurred in the pilB gene of the targeted P. aeruginosa strain, while the pilM gene was inactivated by the insertion of IS5 transposase. Both pilB and pilM are involved in the biosynthesis of Type IV pilus (T4P), the receptor for the applied P. aeruginosa bacteriophage PNM.

16Bacteriophage-resistant P. aeruginosa mutants were also isolated from patients 30, 54 and 91 infected with P. aeruginosa (Table 2 and Fig. 2b–d). Among these mutations, SNPs were identified that corresponded to regions related to T4P in all three patients. In one patient (54), this mutation was in the pilR gene, coding for the transcriptional activator of a two-component system that regulates expression of the major pilin subunit PilA (Fig. 2b). In another patient (30) isolate, this mutation was in a gene coding for an inner membrane component, PilC, essential for T4P biogenesis (Fig. 2c). For patient 91, a premature stop codon was introduced producing a truncated gene variant of the gene fimV, which expresses a part of the inner membrane assembly of T4P in P. aeruginosa (Fig. 2d). In addition, this patient was shown to harbour bacteria that exhibited simultaneous resistance to all three unique P. aeruginosa bacteriophages from treatment: PNM, PT07 and 14-1 (Table 2 and Fig. 2d). Interestingly, we observed two distinct bacteriophage-resistant variants of the initially targeted P. aeruginosa strain, each showing resistance to the three applied bacteriophages, which all had different bacterial receptors. P. aeruginosa bacteriophage 14-1 infects via a lipopolysaccharide (LPS) receptor. Unsurprisingly, SNPs were identified in genes in the outer core of the P. aeruginosa LPS membrane, that is, wapH, galU, wbpR (gene products truncated in these three mutant variants) and wapR. Although the receptor for P. aeruginosa bacteriophage PT07 is not known, sequence similarity to PAK-P1-like bacteriophages (98.26% identity to bacteriophage PaP1) suggests that this bacteriophage is dependent on the P. aeruginosa MexAB-OprM multidrug efflux pump. A resistance mutant of PT07 was identified with an SNP in the gene mexB. Two P. aeruginosa isolates (Is 4 and 5 in Table 2) from patient 91 had both the mexB mutation and another mutation in DNA gyrase subunit A (gyrA), part of the bacterial DNA topoisomerase. This mutation (H87A) is within the GyrA quinolone-resistance determining region (QRDR)–. The interplay of the MexAB-OprM efflux pump and a DNA gyrase mutation has been associated with high-level fluoroquinolone resistance in P. aeruginosa. Interestingly, the bacteriophage-insensitive P. aeruginosa isolates retrieved from patient 91 carrying the double mutation in mexB and gyrA showed a re-sensitization to fluoroquinolones while displaying unaltered growth kinetics, illustrated by a decrease in the minimum inhibitory concentration (MIC) from ≥4 to 0.5 µg ml−1 for ciprofloxacin and from ≥8 to 1 µg ml−1 for levofloxacin. Of note, patient 91 was treated concomitantly with bacteriophages and the antibiotics meropenem, colimycin and vancomycin.

17Since these mutations in isolates from patients 30, 54 and 91 are in encoded virulence factors (Type IV pili, lipopolysaccharide), we implemented a Galleria mellonella infection model to readily assess the virulence of bacteriophage-susceptible versus bacteriophage-resistant variants of these P. aeruginosa strains. Larvae infected with original, bacteriophage-susceptible isolates showed rapid and significant mortality within 48 h (100% death) (Extended Data Fig. 3). The groups infected with bacteriophage-resistant mutants from patient 91 with multiple mutations (>2) in genes encoding for different regions (LPS, MexAB-OprM and/or T4P and/or DNA gyrase) showed significantly higher survival rates (P < 0.0001) compared with the larvae infected with the original isolate in this model system. Significantly higher survival rates were also observed for the larvae infected with the bacteriophage-resistant variant of the P. aeruginosa strain isolated from patient 54 as compared with the original bacteriophage-susceptible variant (P = 0.01). However, all larvae from these two groups died in 18 h. The larvae infected with the bacteriophage-resistant isolate from patient 30 showed no difference in survival compared with those infected with the original isolate. Consequently, in patient 91 we saw a combination of antibiotic re-sensitization and reduced virulence of bacteriophage-resistant isolates, which may have contributed to an eventual favourable treatment outcome.

18Bacteriophage–antibiotic–bacteria interactions were analysed for suboptimal ratios of bacteriophages to bacteria (MOI ≤ 1) and subMIC levels (0.5 × MIC) of antibiotics. These suboptimal conditions were necessary to enable the observation of these interactions. If either bacteriophages or antibiotics were applied under optimal concentrations, this would have led to the efficient killing of the bacterial strains by either antibiotics or bacteriophages, making it impossible to demonstrate possible synergistic, additive, or antagonistic interactions. In vitro bacteriophage–antibiotic–bacteria interaction experiments revealed a synergistic or additive effect of bacteriophages and concomitantly applied antibiotics in 9 out of 10 evaluated patients (9, 20, 21, 27, 43, 71, 82, 91 and 92). An overview of the test results is presented in Table 2. The results of the experiments concerning the first 5 patients (9, 20, 21, 27 and 43) were reported previously,,,,. The detailed results (OmniLog growth curves) for the 5 most recent patients (54, 71, 82, 91 and 92) are presented in Fig. 3. In vitro synergy with bacteriophages was observed for 9 antibiotics (aztreonam (patient 20), ceftaroline (92), ceftazidime/avibactam (9), clindamycin (21 and 92), colistin (20 and 91), gentamicin (20), levofloxacin (82), meropenem (9 and 91) and vancomycin (43 and 92)), and an additive effect for three antibiotics (ceftazidime/avibactam (27), ceftazidime (71) and ciprofloxacin (21)). For one patient (54), no significant in vitro interactions between colistin and P. aeruginosa bacteriophages PNM (Fig. 3a) or 14-1 (Fig. 3b) were observed. Bacteriophages 4P and DP1 acted in synergy with levofloxacin (Fig. 3d,e), but showed no clear interaction with tobramycin (Fig. 3f,g), when tested in vitro against the P. aeruginosa strain of patient 82. Importantly, a moderate antagonism was observed for S. aureus bacteriophage ISP with rifampicin (patient 21 in Table 2) in one of our previously published BT cases. Of note, when most of these tests were performed, BT had already started and test results did not influence patient treatment. However, today, on the basis of these results and the overall observation that pathogen eradication is more likely when phages are applied in combination with antibiotics, we strongly advise physicians to have these tests performed before treatment, if time permits.

19For 13 patients, sufficient serum samples (obtained before, during and after BT) were available to allow for an adequate bacteriophage immune neutralization screening. The applied serum concentration (0.9%) and incubation time (30 min) conform to the standard technique developed by M. H. Adams in 1959 to specifically detect bacteriophage neutralization activity. Bacteriophage immune neutralization was observed between 6 and 35 days after initiation of BT in 5 of 13 (38.5%) screened patients (9, 13, 20, 66 and 92 in Table 2 and Fig. 4a–d). Bacteriophage immune neutralization always involved invasive (intravenous and/or intralesional) bacteriophage administrations. In 4 of these 5 cases (patients 9, 13, 20 and 92), clinical improvement and eradication of the targeted bacterial pathogen were nevertheless observed. In a liver transplant patient (43 in Table 2), the intravenous administration of bacteriophages did not elicit any immune neutralization. In another liver transplant patient (20), bacteriophage immune neutralization emerged, but only after 5 weeks, and it concerned 1 of the 3 bacteriophages that had been applied (Table 2 and Fig. 4c).

20Clinical improvement was reported in 77.2% (88/114) of targeted infections and eradication of the targeted bacteria was observed in 61.3% (65/106) of infections for which relevant bacteriological follow-up data were available (Supplementary Table 1). For 8 targeted infections, in 8 patients, no adequate post BT bacteriological data were available (Supplementary Table 1). For 7 of these cases, the treatment centre did not collect the necessary bacteriological data as part of their routine follow up of patients and was not allowed to collect the data prospectively. For the remaining case, it is not clear why the bacteriological data were not available. The treatment centre either did not collect these data, failed to extract these data from the medical files, or was not able or willing to transfer these data to the Phage Therapy Coordination Centre (PTCC).

21BT resulted in clinical improvement without bacterial eradication in 18 of 106 (17.0%) targeted and bacteriologically monitored infections (Supplementary Table 1). Conversely, in 2 patients (44 and 93 in Supplementary Table 1), eradication of the targeted pathogens was observed without clinical improvement. In patient 44, an infection with an additional (non-BT-targeted) bacterial species (Acinetobacter baumannii) surfaced during BT, which ultimately resulted in an A. baumannii pulmonary septic shock and the patient’s death, despite intravenous administration of tigecycline. Patient 93 succumbed to tumour progression and palliative care.

22For 21 of the 92 (22.8%) patients for which bacteriological follow-up data were available, neither clinical improvement nor bacterial eradication could be observed. Five of these patients (3, 36, 40, 69 and 96 in Supplementary Table 1) died. The causes of death were septic shock (n = 2), cardiogenic shock (n = 1), multi-organ failure (n = 1) and COVID-19 infection (n = 1). In 69.3% (79/114) of targeted infections, concomitant standard-of-care antibiotics were administered (Supplementary Table 1).

23Fisher’s exact test for count data showed univariate significant effects on eradication for the following categorical variables: concomitant use of antibiotics (yes or no), antibiotic resistance profile of the targeted bacteria (multidrug resistant or not) and the clinical setting (ambulatory or hospitalized). No effects of patient age or gender on eradication of the targeted bacteria were observed using univariate logistic regression considering solely age or gender, respectively. A stepwise, forward selection logistic regression analysis of eradication on all independent variables determined that the concomitant use of antibiotics (variable ABCONCOM) was the most informative variable in the reduced dataset (Supplementary Table 2). In our dataset of 100 consecutive cases, eradication of the targeted bacteria (variable ERADIC) was 70% less probable when no concomitant antibiotics were used (odds ratio = 0.3; 95% confidence interval = 0.127–0.749). The P value for Fisher’s exact test of independence between ABCONCOM and ERADIC was 0.01488. The contingency table shows that our logistic regression model is right 65% (40 + 20/92) of the time. The antibiotic resistance profile of the target bacteria (ABRPROF) and the clinical setting (CLINSETT) as well as their interactions with the concomitant use of antibiotics (ABCONCOM) were not selected in the overall logistic regression model. This could be attributed to confounding relations between these three variables within this dataset. A significant association was found between clinical improvement and bacterial eradication. Of the 23 patients with no clinical improvement, only 2 patients expressed eradication. Of the 69 patients with clinical improvement, 53 had full eradication. Intravenous BT, as stand-alone or concomitant therapy, was not shown to significantly impact clinical outcome, as also found for patient age or gender, the persistence of the bacterial infection (chronic or acute), or the use of more than one targeting bacteriophage per bacterial strain. Clinical improvement or bacterial eradication was not significantly correlated with the presence of either P. aeruginosa or S. aureus, where other species were not considered separately, as their prevalence in this study population was too low, or with any individual bacteriophage or bacteriophage cocktail.

24Fifteen adverse events were reported, including seven non-serious adverse drug reactions suspected to be linked to BT (Extended Data Table 3). All suspected adverse drug reactions resolved. No correlation between adverse events and a certain bacteriophage product or administration route could be made.

25In this overview of the first 100 consecutive real-world cases of personalized BT treatment, we show that (1) we were able to produce more than 40 batches of personalized bacteriophage APIs, some of them pre-adapted, which were subsequently certified for use in pharmaceutical preparations; (2) when used in the treatment of 114 difficult-to-treat infections of various types and aetiology, in combination with antibiotics in 69.3% of cases, these preparations led to clinical improvement in 77.2% and eradication of the targeted bacteria in 61.3% of cases; (3) seven non-serious suspected adverse drug reactions were reported.

26The overwhelming representation of P. aeruginosa (49% of patients) and S. aureus (39% of patients) as targeted bacterial species is because these are overall major causes of severe nosocomial infections, but are also the main microorganisms causing invasive burn wound infection, which is historically a major focus of attention of the infectiologists of the QAMH, where the first bacteriophage treatments were carried out.

27Of note, all bacteriophage preparations were offered free of charge. However, this endeavour—providing 43 batches of 26 bacteriophages for the treatment of 100 patients—would not have been possible if one had to comply with the large body of costly and time-consuming requirements and procedures for GMP manufacturing and licensing of biological medicinal products. Companies focusing on defined bacteriophage preparations for use in commercially viable indications might be able to deal with the demanding requirements of the conventional medicinal product (drug) licensing pathway, including GMP certification, preclinical testing and clinical trials. However, for a BT centre, these requirements form an insurmountable barrier in terms of timelines and cost. We experienced first-hand how elaborate and logistically complex personalized BT concepts are, compared with one-size-fits-all approaches, with bacterial strains and matching bacteriophages being exchanged between dozens of institutes in 12 countries. As a result, we are focusing on the development of an instant and on-site production system for bacteriophages based on artificial intelligence (AI) and synthetic biology approaches.

28Our BT protocols prescribe relatively low bacteriophage doses, usually ~107 plaque forming units (p.f.u.s) ml−1. In the United States, the Antibacterial Resistance Leadership Group (ARLG) Phage Taskforce suggests using the highest safe and tolerated dose of a bacteriophage product with endotoxin levels below the acceptable limits set by the Food and Drug Administration to maximize bacteriophage concentrations at the site of infection and infect as many host cells as possible with the first dose. The ARLG Phage Taskforce, however, acknowledges that clinical outcomes are not always improved with higher doses, reflecting the complexity of effective bacteriophage dosing. We observed an increase in in vitro bacteriophage efficiency (lytic activity) with increasing MOI up to a certain MOI, after which regrowth can be observed more frequently and at an earlier point in time (Extended Data Fig. 4). The effective bacteriophage doses in the body are also determined by the route of bacteriophage administration. Most established BT protocols presented here are based on the principle that bacteriophages are best administered directly into the site of infection. Oral administrations were not used because no gastrointestinal infections were treated.

29In 17% (18/106) of targeted infections for which bacteriological follow-up data were available, clinical improvement was reported even though the targeted bacteria were not eradicated.

30In 1943, the emergence of bacteriophage-resistant bacterial mutants in liquid cultures was reported. Recently, parallel evolution of bacteriophage resistance and virulence loss in P. aeruginosa response to bacteriophage treatment (in one patient), in vivo and in vitro was reported. In vivo selected resistance was associated with reduced growth rates, whereas in vitro isolates evolved greater biofilm production. Reference showed that when bacteriophage infection risk is high, constitutive resistance mechanisms, such as a mutation of the bacteriophage receptor, are selected by the bacterial hosts, rather than inducible resistance mechanisms, such as a clustered regularly interspaced short palindromic repeats (CRISPR) system. In the present study, we observed in vivo selection of a bacteriophage resistance phenotype in 43.8% (7/16) of patients for which adequate follow-up bacterial samples were available for testing. However, there is a caveat; patients for whom the possible emergence of bacterial bacteriophage resistance could be analysed were treated in only a few hospitals where routine bacterial monitoring generated sufficient suitable samples. This means that the presented bacteriophage resistance data are not generalizable. In addition, due to the limited number of patients in which resistance was demonstrated, it was not possible to statistically show its potential impact on bacterial target eradication. Regardless, failure of eradication was observed in 43% (3/7) of patients with bacteriophage resistance selection and in 22% (2/9) of patients without bacteriophage resistance. Cases where bacteriophage resistance arose were predominantly P. aeruginosa respiratory tract infections. It can be that bacteriophage resistance is more common in this scenario, but it may also be because respiratory tract infections and P. aeruginosa infections are the most represented. Non-synonymous SNPs or deletions in genes affecting the bacteriophage receptor or coding for a DNA gyrase were assumed to be at the basis of the resistance phenotype in five cases. In two patients, bacterial strains that were not hosts for the applied bacteriophages were selected. In some cases, the in vivo selected bacteriophage-resistant mutants were shown to exhibit re-sensitization to certain antibiotics and reduced virulence in a G. mellonella larvae model. The selection of bacteriophage-insensitive bacteria did not prevent the ultimate eradication of the targeted bacterial strains and clinical improvement in four patients.

31So far, all BT RCTs have evaluated defined bacteriophage products as stand-alone therapies, while bacteriophage–antibiotic synergy is increasingly reported in the literature–. On the basis of BT clinical data generated in compassionate use settings, in combination with antibiotic therapy, the ARLG Phage Taskforce recently suggested that BT should be used in conjunction with conventional antibiotics. Correspondingly, here we observed a statistically significant correlation between the eradication of the targeted bacteria and adjunctive standard-of-care antibiotic therapy. In addition, in several of the present 100 cases, it was assumed that the clinical resolution of multidrug-resistant infections was due to the additive or synergistic effect of various bacteriophage–antibiotic combinations,,–,–. It was hypothesized, on the basis of in vitro experiments, that the therapeutic use of bacteriophages binding to P. aeruginosa efflux pumps could select bacteriophage-resistant isolates with changes in the efflux pump mechanism, causing increased sensitivity to certain chemical antibiotics. In the present study, we demonstrated that the therapeutic use of bacteriophage PT07, predicted to bind to the MexAB-OprM multidrug efflux pump, indeed selected (in vivo) bacteriophage-resistant mutants with changes to the efflux pump mechanism, resulting in increased sensitivity to fluoroquinolones. The use of specifically chosen bacteriophages (for example, targeting drug efflux pumps) could therefore re-sensitize bacteria towards antibiotic activity, increasing bacterial killing when used in combination with these antibiotics, and potentially decreasing selection of antibiotic- or bacteriophage-resistant clones. However, caution is warranted, as certain antibiotics can interfere with bacteriophage lytic activity,. It might thus be advisable to measure potential synergy or antagonism for the proposed combinations of bacteriophages and antibiotics before their clinical application.

32A considerable body of experimental data has accumulated showing that bacteriophages can substantially affect immune system cells, and it has been assumed that anti-bacteriophage antibodies appearing over the course of BT could decrease the lytic activity of bacteriophages and cause therapeutic failure. Consequently, the use of the same bacteriophage(s) for several weeks was discouraged in the former Soviet Union. More recently, ref. reported on the development of neutralizing antibodies after 2 months of intravenous BT, which led to treatment failure in an immunocompetent patient with Mycobacterium abscessus pulmonary infection. The ARLG Phage Taskforce advised considering measurement of neutralizing antibodies during prolonged courses of BT. In the present study, we observed bacteriophage immune neutralization emerging 6–35 days after initiation of invasive bacteriophage administration.

33We acknowledge that our analysis, involving 100 severely ill patients for whom BT was a salvage therapy and our primary aim was to help these patients, has intrinsic limitations. No control groups, blinding or randomization were put in place and different medical specialties and infection types were involved. Evaluation of safety and efficacy was not based on pre-defined standardized tests but on the judgement of the treating physicians, and although they were all experienced, this introduces a certain subjectivity. However, we consider that this case series provides key insights that are not only valuable for the treatment of last resort patients, but also for the design of prospective clinical trials such as the PHAGEFORCE study. We confirmed the safety profile of BT and the advantages of combining BT with standard-of-care antibiotic therapy. Statistical analysis showed a significantly higher probability of microbial eradication when BT was combined with standard-of-care antibiotics, and in vitro bacteriophage–antibiotic synergy was demonstrated in 9 of 10 analysed cases. Samples allowing supportive tests in 21 patients, in view of better treatment management, shed more light on some BT issues such as the in vivo selection of bacteriophage resistance, bacteriophage–antibiotics synergy and bacteriophage immune neutralization.

34In conclusion, we present evidence that the use of bacteriophages in addition to standard-of-care antibiotics can significantly improve the eradication rate of targeted bacteria in this patient population. These data can be useful for designing future controlled clinical trials that are urgently needed to assist the BT field.

35We reviewed the first 100 consecutive BT cases facilitated by a Belgian consortium between 1 January 2008 and 30 April 2022. Within this consortium, the QAMH coordinated most BT cases, selecting and producing bacteriophages, and suggesting BT protocols, while KU Leuven performed supporting genomic analyses of bacteriophages under consideration and of bacterial genomes, and Sciensano controlled the quality and safety of individual bacteriophage preparations. The choice for 100 patients is arbitrary and not linked to any prospective sample size determination.

36Physicians requesting BT with QAMH bacteriophage preparations for their patients submitted a BT request to the Phage Therapy Coordination Centre (PTCC) of the QAMH. The PTCC procedure for selecting patients for BT is depicted in Extended Data Fig. 1 and is largely determined by clinical need, regulatory approval and the availability of bacteriophages targeting the infecting bacteria. Clinical applications were performed by, and under the responsibility of, Bacteriophage Therapy Providers in several hospitals in Belgium and abroad. No blinding, masking or randomization were implemented, and investigators and patients were aware of the bacteriophage treatment. Demographic and clinical data were collected through the patients’ treating physicians. Clinical improvement (or not), eradication of the targeted bacterium (or not), and the advent, seriousness and duration of suspected adverse drug reactions and events were assessed by the treating physicians.

37Written informed consent for BT was obtained from the involved patients or their legal representatives according to local provisions. Where warranted, local ethics committee approval for BT was obtained. According to EU Regulation No 536/2014 (Clinical Trials Regulation), its transposition to Belgian Law, and following advice of the Leading Ethical Committee of the ‘Universitair Ziekenhuis Antwerpen’ and the ‘Universiteit Antwerpen’ (ID 3644), which approved the observational study protocol, the present retrospective non-interventional analysis of an existing and de-identified BT database was not considered as an experiment on the human person and did not require a dedicated informed consent. There was no patient compensation for participation in this study. The observational study protocol was registered on ClinicalTrials.gov (Study BT100, ID: NCT05498363).

38Bacteriophages were isolated and characterized by QAMH or were sourced from Bacteriophage Donors. Bacteriophage suspensions were produced in accordance with the guidelines provided by the bacteriophage API monograph, and the methods described in ref., with some modifications. Bacteriophage stocks were prepared using the double agar overlay method with minor modifications. Three to six millilitres of bacteriophage lysate containing 103–105 plaque-forming units (p.f.u.) of bacteriophages were added to a sterile 15 ml Falcon tube (Greiner Bio-One) and complemented with 0.2 ml of a bacteriophage-sensitive bacterial suspension (end concentration of 108 c.f.u.s ml−1) and lukewarm medium (Select Alternative Protein Source (APS) lysogeny broth (LB), tryptic soy broth (TSB) or TSB + 0.5% glycerol (all purchased from Becton Dickinson)) with 0.6% top agar (VWR International), to a total volume of 12 ml. This mixture was plated onto a square (12 ×12 cm) Petri dish (Greiner Bio-One) filled with a bottom layer of APS LB, TSB or TSB medium + 0.5% glycerol (all Becton Dickinson) and 1.5% agar (VWR International), and incubated at 32 °C (for E. coli, K. pneumoniae and P. aeruginosa) or 37 °C (for all the other bacterial species) for 16 h or 48 h (for M. abscessus). The top agar layer was scraped off using a sterile L-shaped rod (Sigma Aldrich), transferred to a sterile 50 ml sterile Falcon tube (Greiner Bio-One) and centrifuged for 20 min at 6,000 g using a Sorvall Legend centrifuge (Thermo Fisher). The supernatant was aspirated using a sterile 30 ml syringe (BD Plastipak, Becton Dickinson) with an 18G sterile needle (BD microlance 3, Becton Dickinson) and filtered sequentially using a 0.45 µm and a 0.22 µm polyethersulfone (PES) Millex-Gp membrane syringe filter (Merck) or using a vacuum filter system (Nalgene, Thermo Fisher). The bacteriophage suspension was centrifuged for 90 min at 35,000 g (40,000 g for podoviruses) using a Sorvall Legend centrifuge (Thermo Fisher). The resulting bacteriophage pellet was diluted in ten times less Dulbecco’s phosphate buffered saline without calcium and magnesium (DPBS, Lonza) than the initial bacteriophage suspension and the pellet was left to dissolve overnight at 4 °C. The bacteriophage suspension was further diluted to a final concentration of generally 109–1010 p.f.u.s ml−1 using DPBS (Lonza) and a volume of 150–250 ml. The diluted bacteriophage suspension was filtered using a 0.22 µm PES Millex-Gp membrane syringe filter (Merck) and subsequently purified from endotoxins using the commercially available kits EndoTrap Blue (Lonza) or EndoTrap HD (Lionex), according to manufacturer instructions. One column was utilized per 50 ml of bacteriophage suspension. Endotoxin-purified bacteriophage suspensions were filtered using medical-grade 0.22 µm polyvinylidene difluoride (PVDF) Millex-Gp syringe filters (Merck) and collected into sterile 125 or 500 ml PETG Nalgene bottles (Thermo Fisher). The final titre of each thus obtained bacteriophage API was 109–1010 p.f.u.s ml−1.

39Sciensano controlled the quality and safety of the bacteriophages. In accordance with the bacteriophage API monograph, this control was implemented on two levels (https://www.sciensano.be/en/control-and-safety-assessment/safety-therapeutic-bacteriophage-preparations). First, a genetic control was performed to check the safety of the bacteriophage to be used in human therapy. For this purpose, genomic DNA of the bacteriophages and their bacterial hosts were isolated and purified, respectively using a MagCore Viral Nucleic Acid and an MgC Bacterial DNA kit with a 60 μl elution volume (Atrida), following manufacturer instructions. Sequencing libraries were constructed using the Illumina Nextera XT DNA sample preparation kit and sequenced on an Illumina MiSeq instrument with a 250 bp paired-end protocol (MiSeq v3 chemistry, Illumina). Trimming of short reads was performed with Trimmomatic (v.0.32). In addition, for bacterial production strains, long-read sequencing was performed using Oxford Nanopore Technologies (ONT)’s rapid barcoding kit SQK-RBK004 and a MinION flow cell (v.9.4.1), according to manufacturer instructions. Super high accuracy base calling was performed using Guppy (v.6.0.1) (ONT) and hybrid assemblies were generated using Unicycler (v.0.4.7). For bacteriophages, genome assembly was performed using SPAdes (Galaxy v.3.15.4+), after which the genome was annotated using Prokka (Galaxy v.1.14.6) with assistance of the PHROGS v.3 database (https://phrogs.lmge.uca.fr/). To detect undesired genes associated with antibiotic resistance or virulence, the complete bacteriophage genome was submitted to the NCBI (National Center for Biotechnology) blastn web interface (https://blast.ncbi.nlm.nih.gov/Blast.cgi) for a similarity search in different databases: ARG-ANNOT (ARG-ANNOT NT v.6 July 2019), CARD (v.3.1.4 to 3.2.5), ResFinder (https://bitbucket.org/genomicepidemiology/resfinder_db) and VFDB full (downloaded on 20 April 2022). Prophage induction was searched by mapping sequencing reads of the production batch to the bacterial production host genome using Bowtie2 (Galaxy v.2.5.0), and looking for significantly increased coverage in predicted prophage positions using PHASTER (https://phaster.ca/) and Prophage Hunter (https://pro-hunter.bgi.com/).

40Second, Sciensano analysed various parameters of each production lot of each bacteriophage API. Bacteriophage identity and purity (scored by the percentage of bacteriophage sequence reads) was determined using DNA extraction and genome sequencing as described above. The potency of the lot was verified using classical double agar dilutions in triplicate. The bioburden (total viable aerobic count) of each bacteriophage API lot was assessed using a validated membrane filtration method based on European Pharmacopoeia (Ph. Eur.) chapter 2.6.12. Briefly, 4 ml of the 150–250 ml bacteriophage API batches (1.6–2.6%) was added to 36 ml of NaCl peptone, after which 10 ml was membrane filtered (Nalgene membrane filter, 0.45 µm). The membrane was then incubated on trypto-casein-soy (TCS) agar at 30–34 °C for at least 72 h and SCG (Sabouraud dextrose agar + chloramphenicol + gentamicin) at 20–24 °C for at least 5 days. After incubation, the number of c.f.u.s per ml of bacteriophage API was determined. Several bacterial and yeast strains were used as positive controls.

41Bacterial endotoxin content of 1 ml samples (0.4–0.6%) was determined using a validated Limulus Amebocyte Lysate (LAL) test, according to Ph. Eur. chapter 2.6.14. Bacterial endotoxin levels were expressed in endotoxin units (EU) per ml (1 EU is equal to 1 international unit (IU) of endotoxin). The acceptance criterion (endotoxin limit) for the final bacteriophage magistral preparations (diluted bacteriophage APIs) was 5 EU kg−1 body mass h−1, irrespective of the administration route.

42A certificate allowing the bacteriophage API to be used in pharmaceutical (magistral) preparations is provided by Sciensano upon successful completion of this two-tiered procedure.

43Sciensano controlled the quality and safety of 43 batches of individual bacteriophage APIs produced by QAMH to treat the first 100 patients. These batches exhibited an average bacteriophage titre of 8.34 × 109 p.f.u.s ml−1 (s.d. 1.16 × 1010), a pH of 7.32 (s.d. 0.037), a bioburden of 0 colony-forming units (c.f.u.) ml−1 (s.d. 0) and a median endotoxin level of 5 EU ml−1 (s.d. 89.14). The bacteriophage APIs, active ingredients of magistral preparations, were diluted in, and/or combined with, the necessary excipients in a hospital pharmacy ‘officina’ immediately before use on a named-patient basis. The endotoxin limit for the bacteriophage magistral preparations was defined on the basis of dosage and the patient’s weight. The administered endotoxin doses were, irrespective of the administration route, always well below the threshold pyrogenic dose for intravenous administration, that is, <5.0 EU endotoxin kg−1 body mass h−1. Bacteriophage genomes contained no genetic determinants known to confer lysogeny, toxicity, virulence or antibiotic resistance. Host bacteria used in the manufacturing process were as safe (or least pathogenic) as possible. Some production hosts were shown to contain prophages. Bacteriophage productions with >5% of sequencing reads derived from actively replicating prophages were not used in therapy. Bacteriophage cocktails produced by the Eliava Institute (PyoPhage and IntestiPhage) were not quality-controlled by Sciensano. These products probably have higher endotoxin content and an unknown prophage content. Hence, they were never administered intravenously.

44The patients’ infecting bacteria were sent to the PTCC and their bacteriophage susceptibility was determined. Susceptibility of bacterial strains towards the available bacteriophage cocktails or APIs was tested using the spot test as described in ref.. Fresh overnight cultures of the patient’s bacterial strains were added to lukewarm (46 °C) media containing 0.6% agar (top agar) and poured onto square (12 ×12 cm) Petri dishes (Greiner Bio-One) containing media with 1.5% agar (bottom agar). Different culture media were used, according to the considered bacterial species. Media were purchased from Becton Dickinson and agar from VWR International. Droplets (10 µl) of serial dilutions of each of the considered bacteriophage solutions were spotted on the top agar layer. Petri dishes were incubated overnight at 32 or 37 °C, according to the considered bacterial species. The next day, the lysis zones produced by active bacteriophages in the bacterial lawn were examined and classified as confluent lysis (4+), semi-confluent lysis (3+), opaque lysis (2+), separate plaques (+) or no activity (−). Next, for bacteriophages producing clear lysis zones, EOP was defined as previously described. The EOP for the patient’s bacterial strain was calculated by comparison with a highly susceptible reference host and defined as the observed number of p.f.u.s on the patient’s bacterial strain (as determined by the above-described spot test) divided by the observed number of p.f.u.s on the reference bacterial strain. The EOP value obtained with the highly susceptible production host strain was considered as EOP = 1.0. In case the picture was unclear (for example, opaque lysis zones) and the results difficult or un-interpretable, the double agar overlay method was used to determine the p.f.u.s on the patient’s strains and the bacteriophage production host, as described above, to define EOP more precisely. When the activity of the bacteriophages was still difficult to assess using the above-mentioned methods based on solid media, liquid broth cultures were used to assess bacteriophage activity, using the OmniLog system (Biolog). Bacterial respiration was measured without and with bacteriophages. Experiments were performed in 96-well plates (Thermo Fisher) in a final volume of 200 µl of LB or TSB medium (Becton Dickinson), supplemented with 100-fold diluted tetrazolium dye mix A or H (Biolog). Bacterial cells were inoculated at a concentration of 105 c.f.u.s per well, calculated on the basis of optical density (OD) at 600 nm and validated using a classical plate culture method. Bacteriophages were added at an MOI range of 100–0.0001, as calculated on the propagation host. Plates were incubated at a bacterial species-specific temperature (32 or 37 °C) for 72 h, and the colour change caused by reduction of the tetrazolium dye due to bacterial respiration (during growth) was recorded every 15 min by the OmniLog system. The results were analysed with Biolog Data Analysis software (v.1.7) and data were exported to Microsoft Excel files.

45We considered the relative EOP as a relative measure of lysis efficiency, which, in this context, is defined as the lytic activity (titre) of the bacteriophage on the patient’s bacterial strain, divided by the titre observed in a reference bacterial host known to be highly susceptible to the bacteriophage. We considered an EOP ≥ 0.1 on the patient’s bacterial strain as therapeutically acceptable on the basis of the expertise from the Eliava Institute. All bacteriophage cocktails were composed of bacteriophages with compatible activities. Since April 2022, when more than one bacteriophage showed adequate in vitro activity, the overall activity of the bacteriophage combinations was analysed using the OmniLog system, as described above. When synergistic or additive effects were observed, the concerned bacteriophage combinations were recommended for clinical use.

46When the observed bacteriophage susceptibility was deemed too low for therapeutic application, and if time and resources permitted, bacteriophages were pre-adapted to increase pathogen clearance and to reduce bacteriophage resistance evolution–. According to the guidelines of the Ministry of Health of the USSR and the empirical experience of the Eliava Institute, adequate bacteriophage cocktails (not individual bacteriophages) should cause stable lysis, that is, without the emergence of bacteriophage-insensitive bacterial mutants, of the target bacteria in liquid medium for a prolonged period (typically 24–48 h), and at an MOI of 0.0001–0.00001 and bacterial concentrations of 106 c.f.u.s ml−1 (refs. –). For individual bacteriophages, MOIs ≤ 1.0 were deemed appropriate. To obtain these bacteriophage virulence and bacterial regrowth suppression thresholds, the (modified) Appelmans method was applied for the pre-adaptation of bacteriophages on bacterial strains, as previously described. To a 15 ml Falcon tube (Greiner Bio-One) were added: 4.5 ml of LB or TSB medium (Becton Dickinson), 0.5 ml of tenfold dilutions of the considered bacteriophage and a volume of either the patient’s bacterial strain or a pre-production panel of collected ‘problematic’ bacterial strains, to obtain a final concentration of 106 c.f.u.s ml−1. The tubes were incubated at a bacterial species-specific temperature (32 or 37 °C) for 48 h. Bacterial growth and bacteriophage activity were monitored by OD measurement at 600 nm using a Lambda 12 UV/VIS spectrometer (Perkin Elmer) after 24 and 48 h of incubation and compared to two negative controls (bacteriophage only and LB or TSB medium only) and a positive control (bacteria only). The tube with the highest bacteriophage dilution showing an OD600 value similar to the negative controls was selected and chloroform was added to a final concentration of 2.0% (v/v). The tube was shaken and incubated for at least 2 h at 2–8 °C. After incubation, the upper phase (without chloroform) was aspirated using a sterile 30 ml syringe (BD Plastipak, Becton Dickinson) with an 18G sterile needle (BD microlance 3, Becton Dickinson) and filtered using a 0.45 µm or a 0.22 µm PES Millex-Gp membrane syringe filter (Merck). The obtained bacteriophage lysate underwent several (at least three) of the above-described passages until adequate virulence and resistance suppression levels were obtained.

47The comparison between a bacteriophage and its patient-adapted version was recently published. However, the genetic comparison of pre-adapted phages with their unadapted ancestors falls outside the scope of this study.

48The stability of the bacteriophage APIs was monitored by determining their titre at 2–8 °C monthly. Bacteriophage APIs with titres of 109–1010 p.f.u.s ml−1 retained their activity for at least 1 year. One or more bacteriophage APIs can be diluted and/or mixed with a carrier (for example, an isotonic intravenous solution or a hydrogel) into a magistral preparation under the supervision of a hospital pharmacist and according to the provisions of a medical prescription provided by the patient’s treating physician. Diluting and mixing various bacteriophages are events that can compromise their stability,, and experiments showed that, in general, magistral preparations are best used within 1 week after their manufacture.

49The PTCC suggested BT protocols on the basis of the application instructions of the Ministry of Health of the USSR– and the Eliava Institute, some of which can be found in the leaflets of their BT products. These documents (in Russian) do not mention any (published) data. One of them states that ‘bacteriophage neutralization can emerge between 10 and 15 days after intravenous application’. We have not been able to determine whether these 30–40-year-old guidelines and instructions may be based on systematic studies, or if they are largely based on empirical experience. Therefore, we prefer to catalogue them as Centre for Evidence-Based Medicine (CEBM) evidence level 5, that is, recommendations formulated by experts on the basis of their own professional experiences. This evidence is probably also based on the review of data from case reports and non-systematic studies.

50Bacteriophage administration intervals were largely influenced by clinical indications and administration routes. For instance, it is more straightforward to apply bacteriophages several times per day to the infected lungs of an intubated patient (nebulization) than to infected burn wounds (topical), which are generally unpacked and treated only once a day.

51For nebulization of bacteriophage preparations, vibrating mesh type nebulizers were advised because they were shown to induce less titre loss due morphological damage than air-jet nebulizers,. For bone and orthopaedic prosthesis infections, we advised the use of a pigtail catheter or another draining device for rinsing the wound cavities before bacteriophage application and for the actual administration of bacteriophages. For topical application, we advised mixing of the bacteriophages with an adequate hydrogel. In general, our protocols prescribed relatively low bacteriophage doses, usually ~107 p.f.u.s ml−1, and ranging from 106–107 p.f.u.s ml−1 for continuous intravenous BT to 109 p.f.u.s ml−1 for topical BT in a few SSTI cases. In contrast, some clinics prefer the administration of considerably higher doses, for instance, up to 1010–1011 p.f.u.s ml−1 for intravenous BT,.

52In addition to bacteriophage susceptibility testing, three BT supportive tests were offered without obligation to the Bacteriophage Therapy Providers to allow for improved BT management: (1) monitoring of the in vivo emergence of bacteriophage resistance using sequential bacterial samples isolated during BT, (2) analysis of the in vitro bacteriophage–antibiotic interactions before the start of BT and (3) evaluation of bacteriophage immune neutralization, or the ability of the patient’s serum to neutralize therapeutic bacteriophages.

53The in vivo selection of bacteriophage resistance was monitored using sequential bacterial samples isolated during BT. Bacteriophage susceptibility was evaluated using the methods described earlier. When decreased bacteriophage sensitivity was observed, the isolate’s genome was sequenced and analysed to determine the clonality of the isolate (compared with the pre-BT isolate) and to investigate the genetic background for the observed bacteriophage resistance phenotype. For genome sequencing, the method described in ref. was followed with some deviations: for nanopore processing, Guppy (v.6.3.8) (ONT) (base calling, demultiplexing) and Porechop (v.0.2.4) (barcode clipping) (https://github.com/rrwick/Porechop) were used. Genomes were assembled with Unicycler (v.0.4.8) and SNP variants were called using Snippy (v.4.6.0) (https://github.com/tseemann/snippy). For genome annotation and visualization, EggNOG-mapper (v.2.1.8), mobileOG-db (v.1.1.2), Phigaro (v.2.3.0), Circos (v.0.69.8) and GC-profile were used. A pan-genome analysis using Roary (v.3.13.0) from annotated genomes (Prokka v.1.14.6) was performed to create a maximum likelihood phylogenetic tree using core alignment in fasttree (v.2.1.10) visualized with iTOL (itol.embl.de). For multilocus sequence typing (MLST), genomes were scanned against PubMLST (https://pubmlst.org/) schemes, including ST111 (O12-1709), ST357 (B14130), ST235 (NCGM2), ST1233 (PcyII-10) PAO1 (ST549) and ATCC 27853 (ST155) as representative genomes/STs. The programs Porechop, Unicycler, Snippy, EggNOG-mapper, Roary, Prokka, Fasttree and MLST were accessed through the Galaxy server (https://usegalaxy.eu/).

54Ten P. aeruginosa isolates (Pa30 (Is 1), Pa30 (Is 3), Pa54 (Is 1), Pa54 (Is 4) and Pa91 (IS 1–6)) were grown in LB broth (Becton Dickinson) to an OD600 of 0.25–0.35. One millilitre of the bacterial cultures was centrifuged and resuspended in sterile DPBS (Lonza). G. mellonella larvae were grouped in batches of 10 (standardized for weight) and then injected in the hindmost proleg with a 10 µl aliquot of 10−5 dilutions (±10 c.f.u.s) of the washed bacterial cultures. After infection, the larvae were incubated in the dark at 37 °C. Activity scores were monitored every 6 h and compared to DPBS-injected controls. Activity scores ranged from 0 to 9, based on activity level (with and without stimulation), melanization and survival.

55Bacteriophage–antibiotic–bacteria growth kinetics were analysed upon request of the treating physicians using the bacterial and bacteriophage isolates obtained before the start of BT. For patients treated before October 2021, these evaluations were performed retrospectively on bacterial and bacteriophage isolates stored at −20 °C in LB + 20% glycerol (Becton Dickinson). Bacterial respiration was measured using the OmniLog system (Biolog). The growth kinetics of the targeted bacterial pathogens were assessed in the presence of the bacteriophages only, the relevant antibiotics (to be used concomitantly) only and bacteriophage–antibiotic combinations. Experiments were performed in triplicate (biological replicates) in 96-well plates (Thermo Fisher) in a final volume of 200 µl of LB or TSB medium (Becton Dickinson) supplemented with 100-fold diluted tetrazolium dye mix A or H (Biolog). Bacterial cells were inoculated at a concentration of 105 c.f.u.s per well, calculated on the basis of OD600 measurements and validated using a classical plate culture method. Antibiotics and bacteriophages were added at subMIC (0.5 × MIC) levels and MOIs ≤ 1.0 (calculated on the propagation host), respectively. The titres of the bacteriophages were confirmed after each experiment using the classical double agar overlay method. Plates were incubated at 37 °C for 72 h and the colour change caused by reduction of the tetrazolium dye due to bacterial respiration (during growth) was recorded every 15 min by the OmniLog system. The results were analysed with Biolog Data Analysis software (v.1.7) and data were exported to Microsoft excel files. We defined bacteriophage–antibiotic combinations as synergistic when the bacterial growth suppression period produced by the addition of both the bacteriophage and the antibiotic is clearly longer than the simple sum of the suppression periods induced by the bacteriophage and the antibiotic separately.

56The possible emergence of bacteriophage immune neutralization, or the ability of the patient’s serum to neutralize therapeutic bacteriophages, was evaluated according to ref., with some modifications. Whole blood samples were collected before BT initiation and at various time points during and after bacteriophage application. Blood was allowed to clot for at least 30 min in a vertical position and then centrifuged in a swinging bucket rotor for 10 min at 2,000 g at room temperature. The obtained serum samples were stored at −80 °C ± 5 °C. To assess the effect of the serum samples on bacteriophage lytic activity, 0.9 ml of 1:100 diluted sera was mixed with 0.1 ml of the bacteriophage suspension at a concentration of 2 × 107 p.f.u.s ml−1 and incubated for 30 min at 37 °C. Bacteriophage lytic activity (titre) was determined before and after incubation with the patient’s serum samples using the double agar overlay plaque assay (as previously described). Comparison of pre- and post-incubation lytic activity allowed for the determination of the proportion of neutralized bacteriophages. Each serum sample was tested in triplicate.

57Clinical improvement, eradication of the targeted bacterium and the advent, seriousness and duration of suspected adverse drug reactions or events were assessed by the treating physicians. Neither safety data were prospectively collected, nor were the descriptors defined in advance to clinicians.

58Before BT, demographic and clinical data were collected through a medical form, which was completed by the Bacteriophage Therapy Providers. The medical doctor’s BT prescription, information regarding the applied bacteriophage product and its administration route, dosage, duration and information regarding possible concomitant (antibiotic) treatments were also recorded. The ‘phagograms’ reporting on the evaluation of the bacteriophage susceptibility of the patient’s bacterial isolates sampled before and sometimes during treatment were also archived. If the bacteriophage treatment was performed in a hospital, a clinical follow-up form requesting information about the clinical outcome (including suspected adverse drug reactions and events) was completed by the treating physician and the nursing team and sent to the PTCC. In case of ambulatory BT, clinical follow-up information was collected directly from the patients. All demographic, bacteriophage product and clinical data were recorded in a Research Electronic Data Capture (REDCap) designed database. Data collection and analysis were not performed blind to the conditions of the experiments.

59In accordance with the guidelines of an international expert proposal for interim standard definitions for acquired resistance, multidrug resistance (MDR) was defined as acquired non-susceptibility to at least one agent in three or more antimicrobial categories, extensive drug resistance (XDR) as non-susceptibility to at least one agent in all but two or fewer antimicrobial categories, and pandrug resistance (PDR) as non-susceptibility to all agents in all antimicrobial categories. The term ‘usual drug resistance’ (UDR) was used to describe isolates that are not fully susceptible, but could nonetheless be readily treated (at least on the basis of the in vitro susceptibility assays) using standard therapies. If an infection persisted for more than 6 months, it was considered a ‘chronic infection’. Clinical improvement was defined as the improvement of at least one symptom associated with the bacterial infection, as assessed by the treating physician. No clinical metrics were applied (for example, illness severity scores). The influence of other (medical/surgical) interventions was not determined. Eradication of the targeted bacterium was defined as the absence of the originally targeted causative agent of the bacterial infection in culture, or when the patient’s treating physician concluded, on the basis of a follow-up survey, that the patient was freed of the targeted bacterial pathogen. Microbiological eradication was not prospectively or systematically evaluated. The period between the start of BT and the evaluation of the clinical outcome varied according to the treating physician and the indication, and ranged from 1 month to 1 year, the latter for difficult-to-treat bone infections.

60The following variables were analysed for 92 of the 100 patients (for which a complete dataset was available): eradication of the targeted bacteria, clinical improvement, concomitant use of antibiotics, antibiotic resistance profile of the target bacteria, suspected adverse drug reactions and the clinical setting (ambulatory treatment or hospitalized). All these variables were binary categorical. In addition, the 14 infection types and 21 bacterial species targeted by BT were monitored on nominal categorical scales. Age and gender were analysed on numeric scales. The statistical analysis was conducted using the statistical software environment SAS (v.9.4). We used a stepwise, forward selection procedure on a reduced dataset (Supplementary Table 2) to determine the most informative variable in the dataset, with the variable ‘Eradication (ERADIC)’ as response variable for our logistic regression model. The probability modelled is ERADIC = ‘Yes’ (that is, successful eradication). A sketch (left) and the contingency table (right) of the logistic regression model used to analyse the reduced dataset (Supplementary Table 2) are depicted below.

61Fisher’s exact test was performed using R (v.4.3.0) (https://www.R-project.org/), and the R Stats Package (v.4.3.2) was used to search for significant correlations between variables. The data presented in Fig. 1 (patient population characteristics) and Fig. 4 (bacteriophage immune neutralization) were analysed using R (v.4.3.0) and visualized with the following packages: tidyverse (v.2.0.0), UpSetR (v.1.4.0), ggmap (v.3.0.2) and rnaturalearth (R package version 0.3.2.9000). The log-rank test with Bonferroni correction for multiple comparisons (GraphPad v.0.5.1) was used for G. mellonella survival curve comparisons.

62Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.