Cluster analysis provided information on similarities of E. coli isolates from humans and different animal populations based on their resistance combinations.
Human isolates from ICU and general ward always clustered together in cluster 1. Isolates from outpatient care were the next closest link in the full model and in two of the four reduced models. This finding supports the hypotheses that most ICU isolates are related to isolates from other parts of the hospital and from outpatients (34). Studies on transmission within health-care-network and patient transfers have also supported this idea (35–37). The slightly larger distance of the outpatient populations in comparison to the inpatient populations (general ward and ICU) might be explained by the specific situation in hospitals, with dominant hospital strains that differs from the outpatient setting (8).
Human isolates from outpatient care clustered closely with clinical isolates from cattle in the full model and two of the four reduced models. The other two reduced models suggested nearest relative frequencies of resistance combinations with clinical isolates from small animals and turkeys. A close similarity between clinical human isolates from outpatient care and clinical isolates from cattle has not been described previously. The reason for this proximity is not clear. Clinical isolates are not likely to be transmitted via food as food is to be harvested from healthy animals. Contact to bovines is also very limited in the general population. Since we do not have any information on the resistance genes from the populations, we are still uncertain whether the pathogens that are involved carry similar particular resistance mechanisms. Therefore, further molecular analyses would be required to better interpret our study result.
In the full model, isolates from the three human clinical populations clustered with clinical isolates from most (6/11) animal populations; i.e. cattle, piglets, sows, turkeys, broilers and small animals. The human populations clustered very close to the clinical isolates from cattle and piglets and to one healthy pig population; weaners on farm. Again, the reason for these closest similarities to clinical animal isolates remains unclear as transmission of clinical isolates by contact or food is unlikely. Nonetheless, infections with E. coli in humans and animals are treated with similar antibiotic groups, e.g. aminopenicillins, cephalosporins, quinolones and aminoglycosides (38). This might lead to a similar selection pressure, which results in similar resistance patterns.
Isolates from young pigs clustered together with the human clinical isolates. This included clinical and non-clinical isolates from pigs. Prevalence of AR in pigs is associated with overall country-specific antimicrobial usage in livestock (39). Penicillins are among the most frequently used antibiotics in pigs in Germany (38, 40). This is in line with our findings, which demonstrate high proportions of resistance solely to ampicillin. The highest proportion of that was found in weaners (fattening piglets, up to 30 kg body weight) from farms (42%) (Fig. 2). Higher single ampicillin resistance in weaners in comparison to other pig populations, foremost growers < 50 kg (fattening pigs), might have been caused by a higher total treatment frequency in fattening piglets (up to 30 kg body weight) in comparison to older fattening pigs (38).
As for healthy pigs, the transmission of bacteria from pigs to humans could be explained via food consumption. Pork meat is occasionally consumed raw in Germany. It is in line with our study findings, which found isolates from pork and human clinical isolates in the same cluster. However, like cattle, the clinical isolates are not likely to be transmitted via food as food is harvested from healthy animals. Another possible explanation is the similar antimicrobial usage (AMU)-pattern between humans and pigs for the antimicrobials included which may create similar resistance patterns, as penicillins are also frequently used in humans.
Our study indicates separate clusters for clinical human isolates (cluster 1) and isolates from most healthy broilers (except broilers from organic farms), and turkey populations and their meat (cluster 3) and for laying hens (cluster 2). It has been reported that extended-spectrum cephalosporin-resistant E. coli from healthy poultry are unlikely to be the causative agents of human UTI (41). Another study revealed low similarities of ESBL/AmpC genes between broilers and the general human population with the exception of the broiler farming communities (8). In line with that, our study indicates a lack of similarities in resistance to the four antimicrobials of E. coli from human and healthy broiler and turkey populations and laying hens.
In the third cluster, healthy broilers and turkeys along with their meats clustered together. Three non-clinical poultry populations were not included in this cluster: broilers from organic farms, laying hens and breeder chickens (Fig. 3). AR in non-clinical E. coli isolates from broilers is associated with the antimicrobial use in poultry production. Resistance proportions in E. coli against penicillins and fluoroquinolones are reported to be 40% higher in countries which have allowed the use of these two antibiotics in poultry than countries which have not (42). In Germany, ampicillin and enrofloxacin, a fluoroquinolone with a similar chemical structures as ciprofloxacin, are authorized antibiotics for the treatment of poultry (43). The total treatment frequencies of penicillins and fluoroquinolones in fattening turkeys and chickens are higher compared to pigs and cattle (38). This might be the reason for higher individual resistance proportions against ampicillin and ciprofloxacin and the higher relative frequencies of the combinations of resistance to both substances compared to other populations (43).
Broilers from organic farms, laying hens and breeder chickens have lower individual resistance proportions against the studied antimicrobials compared to the other healthy poultry populations. This is in line with earlier work on lower resistance proportions in broilers from organic farms (44–46). Lower antibiotic resistance rates might be caused by lower antibiotic usage in organic farming. EU legislation governing organic farming (Reg. (EC) No. 834/2007) foresees the use of antibiotics solely for diseased animals, if phytotherapeutic drugs, homeopathy and other products are not working. This includes the restriction on number of treatments and longer duration of withdrawal periods (47, 48). This may contribute to a lower use of antibiotics in organic broiler farming compared to conventional farming. However, valid specific use data from organic poultry farms are not available for Germany.
For breeder chickens and laying hens, low relative frequencies of resistance combinations were detected with resistance in laying hens even lower than in breeder chickens. Low single resistant proportions to the four chosen antibiotics in these two populations have been previously reported (49, 50). Laying hens and breeder chickens received less antibiotic treatment than broilers, with the lowest antibiotic treatment in laying hens (51). We, therefore, assume that the low relative frequencies of resistance combinations are associated with less antibiotic treatments received in laying hens and breeder chickens compared to broilers. Breeder chickens, i.e. parents and grand-parent flocks of broilers, and laying hens live longer than broilers (approx. 4–6 weeks). It seems reasonable that the microbiome of breeder chickens and laying hens has matured (52, 53). These microbiomes may be more competitive and resilient than those in young broilers contributing to less disease and therefore less treatments. Moreover, the housing conditions of breeder chickens are strictly controlled (54). A controlled housing management might reduce the prevalence of pathogens and their transmission, which also results in fewer antibiotic treatments.
Isolates from wild animals; i.e. wild boars, wild roe deer and venison; clustered closely together with bulk tank milk both from conventional and organic farms. Isolates from these five populations showed the lowest individual resistance proportions and relative frequency of resistance combinations of all populations. Wild animals receive no antibiotic treatment, and therefore are not directly exposed to antimicrobials. However, wild animals were reported to carry AR commensal E. coli (non-clinical E. coli isolates) and play a role as sentinels of environmental transmission of AR (55, 56). The presence of AR in wild animals has been associated to geographical distance to AR sources, such as wastes of antibiotic treated animals or humans (55), and also to human population density (57).
E. coli from bulk tank milk from both conventional and organic farms had low resistance rates and relative frequencies of resistance combinations. Low presence of AR in commensal E. coli (non-clinical E. coli isolates) from bulk tank milk has been previously reported (58, 59). Low systemic use of antibiotics in dairy cattle (51, 60) might result in low AR in the bacteria in milk. However, as E. coli is not part of the healthy milk microbiota and milk from E. coli mastitis is as a rule discarded, the most common source of E. coli in bulk tank milk is environmental, i.e. fecal contamination, mostly originating from the dairy herd (61). Improper milking-system hygiene also plays a role in milk contamination with coliform bacteria from the environment (62).
Clinical isolates from bovines < 1 year had the highest individual proportions of AR for all four antibiotics as well as the highest relative frequency of the resistance combinations (Table 3, Fig. 2). This resulted in higher proportions of resistance combinations in comparison to other populations. Many of the isolates originated from young calves with enteritis. Use of waste milk may have contributed to the high resistance rates (63–65), given that penicillins and cephalosporins are frequently used in the treatment of mastitis of dairy cows (66, 67). Waste milk is likely to contain residues of antimicrobials especially after intramammary treatment of dairy cows. This however cannot explain the comparatively high resistance rates to gentamicin and ciprofloxacin, as these substances are not frequently used in intramammary treatment. Further research into the dynamics of AR in calves is needed to improve the understanding of our study results.
Clinical animal isolates frequently clustered separately from their healthy animal counterparts. Our animal samples originated from two different independent datasets. There is no information whether they originated from the same farms. However, given the large number of farms and the limited number of isolates a large overlap of the source is unlikely. The separation might be caused by differences in selection pressure between the clinical and non-clinical isolates, although they originated from the same animal species and type of population. Non-clinical food-producing animal incl. food isolates were randomly sampled from each federal state in Germany. Clinical food-producing and companion animal isolates were particular isolates from ill animals that might form a specific subpopulation of E. coli strains. The GERM-Vet study protocol states that the animals of origin should not have been treated with the antibiotics within a month prior to sampling. However, it seems possible that these pathogenic isolates had prior specific antibiotic selection pressure in the animal population before the sampling time. An earlier study found the same tetracycline and aminoglycosides resistance genes in commensal (non-clinical isolates) and clinical E. coli (68). Further research into the two different bacterial populations is necessary to better understand the reasons for the differences in AR.
With the sensitivity analysis we aimed to look into consistency of clusters built from the complete model (Fig. 3). Some populations, i.e. human isolates from inpatient care (ICU and general ward) and isolates from wild animals and bovine milk from organic farm; remained in the same sub clusters consistently. This underlines their very close similarity with respect to resistance to the four antimicrobials and a distance to isolates from the other populations.
Removal of individual antimicrobials from the analysis also resulted in changes in cluster distributions compared to the complete model. The removal of one of the three antimicrobials - ampicillin, cefotaxime and ciprofloxacin - at a time made human clinical isolates from outpatient care change their position and nearest neighbors. This indicates a certain distance to the inpatient isolates. On the other hand, the change in the closest neighbor depending on the antimicrobial that was removed, indicates that there was no clear relation to any individual from other population. Removal of one antibiotic influenced the relative frequency proportions of resistance combinations. Resistance rates to ampicillin and ciprofloxacin were high in our study populations. Therefore, the removal of these two antibiotics substantially influenced the cluster order. In contrast, removal of gentamicin did not influence the clusters much. While a full analysis of these findings is outside the scope of this paper, we propose further analyses including additional antibiotics in order to understand the importance of different antibiotic usages in human and animal sectors.
There are a number of limitations to this study that must be acknowledged. Due to differences in the antimicrobials tested in the three systems, we had to choose four common antibiotics that overlapped between the three systems and for which sufficient data were available in ARS. Inclusion of further antimicrobials (e.g. tetracycline), would have reduced the number of available isolates in ARS substantially and would have excluded data from several laboratories, as those did not test E. coli for tetracycline resistance routinely. In ZoMo trimethoprim and sulfonamides are tested as individual substances, while in GERM-Vet and human clinical isolates frequently co-trimoxazol (i.e. a combination of a sulfonamide and trimethoprim) is tested. Colistin and carbapenems have also not been taken into consideration. Colistin is used as a last resort antibiotic in the human sector. However, for methodological reasons resistance data to colistin generated with automated methods are not considered reliable. Regarding carbapenems, different substances were used for animal clinical (imipenem) and non-clinical isolates (meropenem) and therefore data were not considered comparable. Moreover, resistance to carbapenem is extremely rare in animals (69) and also rare in humans in Germany (70).
We used SIR results based on evaluation criteria for humans from EUCAST, as we could re-evaluate the quantitative data from the animal monitoring systems based on these breakpoints. As for the human data, either no quantitative data were available or the tested range was so narrow that a re-evaluation according to ECOFFs was not possible.
This study highlights substantial differences between the three monitoring and surveillance systems (Table 1). Differences in data collection (surveillance versus monitoring), participation system (mandatory versus voluntarily), observed populations (humans versus different animal populations), AST (panel, methods and results) and evaluation criteria (clinical breakpoints and epidemiological cut-off values) should be carefully considered for comparative analysis. For the purpose of comparing resistance proportions, it would be desirable that the One Health community strives towards harmonized evaluation criteria for each antimicrobial in isolates from humans, food-producing animals and food. Alternatively, quantitative data, such as MIC values, need to be collected for allowing the interpretation using different standards based on any required analysis. Rational criteria should be shaped based on various purposes, such as for therapy treatment decisions and comparative analysis of different resistant proportions across different sectors. Joint harmonized MIC value ranges for comparative analyses of human and animal data would better fit for the analysis.
Since routine standardized diagnostics differ between human and animal sectors, it needs to be investigated whether the different laboratory methods yield comparable results. Routine methods are always a compromise between scientific accuracy and economic needs. Increasing costs might discourage widespread use of costly and laborious AST methods in routine laboratories, an aspect that is less relevant in monitoring programs with limited numbers of isolates.
To best of our knowledge, this is the first study that systematically compares the routine laboratory surveillance and monitoring systems for AR in humans with different animal populations and food of animal origin in Germany using cluster analysis. Within the limitations noted above, our results indicate that patterns of resistance combinations are able to provide insights in similarities and discrepancies between isolates from different human and animal populations. Given the current situation on surveillance and monitoring for AR in Germany, we considered it the best approach to compare the national data on AR in E. coli from humans, different animal populations and food based on their phenotypical resistance combinations. Although phenotypic datasets are able to promote the study on resistance combinations, the findings of this study suggest a number of directions, which future studies on molecular level on AR might profitably take. Integration of whole genome sequencing (WGS) into surveillance might help further research into resistance genes similarities. Initiatives on implementation of WGS in AR monitoring system for animals have been already started (71, 72). As genomic information provides better insights into resistance mechanisms, mobile genetic elements, chromosomal mutations and intrinsic resistance, its inclusion in the comparative analysis should be further promoted.