Evaluating biological plausibility in supporting evidence for action through systematic reviews in public health

Abstract

Objectives:

The objective of this research was to develop and test methods for accessing and evaluating information on the biological plausibility of observed associations between exposures or interventions and outcomes in order to generate scientific evidence for action consistent with practice in systematic reviews.

Study design:

To undertake this research, we utilized the example of the observed associations between antimicrobial use in food animals and increased risks of human exposures to antimicrobial resistant pathogens of zoonotic origin.

Methods:

We conducted a scoping search using terms related to biological plausibility or mechanism to identify key references. As recommended by these references we also used expert consultation with researchers and a public health informationist. We used their recommendations, which included expert consultation, to identify mechanisms relevant to biological plausibility of the association we selected to test. We utilized the reviews conducted by the WHO Guidelines Development Group in support of reducing antimicrobial use in food animal production to populate our model for assessing biological plausibility.

Results:

We were able to develop a transparent model for biological plausibiility based the adverse outcomes pathway utilized in toxicology and ecology. We were also able to populate this model using the WHO reviews.

Conclusions:

This analysis of biological plausibility used transparent and validated methods to assess the evidence used in systematic reviews based on observational studies accessed through searches of the scientific literature. Given the importance of this topic in systematic reviews and evidence based decision making, further research is needed to define and test methodological approaches to access and properly evaluate information from the scientific literature.

Background

Evidence based methods in medicine and other health related fields have emphasized biological plausibility as an important element in assessing the strength of evidence since the work of Bradford Hill^1,2. As noted in a recent review of cancer risks, information on biological plausibility is particularly important as a complement to associations observed in epidemiological studies³. For observational studies, the quality of evidence is often judged weaker than evidence based on randomized controlled studies. These study designs, which are necessary given the ethical ramifications of interventions in public health, are considered to be less able to eliminate the effects of residual bias. As a consequence, evaluating biological plausibility or mechanisms may be of particular value in assessing the strength of evidence from this literature. This has been recognized by several regulatory agencies, including the US Environmental Protection Agency and the European Food Safety Agency, as well as by the WHO and CODEX^4,5.

However, despite the importance of the topic, there are no generally accepted methods for evaluating biological plausibility, and many reviews discussing these mechanisms include only general statements on relatively nonspecific physiological events or target organs with no supporting references.

Our research question concerned the biological plausibility of observed associations between antimicrobial use in agriculture and increased risks of human exposures to drug resistant zoonotic pathogens. There are many reviews of this topic, including two recent systematic reviews. One of these systematic reviews was undertaken by a WHO Guidelines Development Group to support its task to develop evidence-based recommendations and guidelines to reduce antimicrobial resistance related to agricultural use⁵. An additional systematic review was published independently⁶. The WHO systematic review utilized the GRADE methodology to assess quality of the evidence, and following GRADE criteria, the evidence was rated of low confidence⁷.

The other systematic review⁶ used a modified GRADE approach for evaluating evidence in which the “sufficient component” causal model proposed by Rothman was incorporated⁸.

Assessments utilizing GRADE can cause confusion among users of guidance based on these reviews. A statement issued by the USDA shortly after publication of the WHO guideline referred to this “low-quality evidence” as effectively disqualifying any WHO recommendations, despite the surrounding analyses and expert opinion⁹. To provide additional support for this evidence, we undertook an assessment of the biological plausibility of the observed associations between antimicrobial use in food animal production and increased risks of human exposures to and infections by antimicrobial resistant zoonotic pathogens¹⁰.

Methods:

We utilized scoping reviews and expert consultation to identify two papers with general discussions of methods related to biological plausibility ^11,12. From these papers, we identified the following search terms “methods”[Subheading] OR “methods”[All Fields] OR “methods”[MeSH Terms]) AND (“research design”[MeSH Terms] OR (“research”[All Fields] AND “design”[All Fields]) OR “research design”[All Fields] OR “test”[All Fields]) AND (“biology”[MeSH Terms].OR “biology”[All Fields] OR “biological”[All Fields]) AND plausibility[All Fields] to access papers from the biomedical literature with more detailed methods for defining causal pathways in terms of molecular and genetic mechanisms^3,13,14. With further expert consultation, we further accessed papers from the toxicology and ecology literature that defined mechanisms as causal pathways in the context of adverse outcome analytic methods^15–17. We utilized the adverse outcome pathway model, as it more closely represents the research question we sought to investigate, that is, a series of discrete mechanistic events not as strictly limited to one molecular pathway as in Lewis et al³. This methodology utilizes schematics to represent pathways, as shown in an example in Figure 1.

Figure 1.

An adverse outcome pathway as used in toxicology to define events in a causal sequence connecting exposures to outcomes at the population level ¹⁵.

To apply this model, we utilized a scoping review approach, including reviews, to identify sources of information on the biological plausibility of observed associations between antimicrobial use in agriculture and increased risks of human exposure to and infection by antimicrobial resistant pathogens from food animals. We developed and populated a similar structure for this review based on a conceptual structure that represents a sequence of mechanisms involved in the emergence and dissemination of antimicrobial resistance^18–21To this model we added the routes that connect these events in agriculture to human exposure. Consistent with WHO practice in guideline development, we sought a global sampling of papers.

Our conceptual model is shown below (Figure 2).

Fig. 2.

A conceptual model of the mechanisms by which use of antimicrobials in food animal production increases risks of antimicrobial resistance as well as exposure of human populations to pathogenic bacteria.

In this model, antimicrobial pressure includes the following variables: volume of antimicrobial use, concentrations of antimicrobials encountered by pathogens in animal guts, duration of antimicrobial use, and use of >1 antimicrobial at a time. Selection for resistance includes both natural selection through evolutionary mechanisms and horizontal gene transfer of one or multiple resistance genes. Resistance dissemination includes clonal expansion of resistant organisms as well as gene flow among organisms through horizontal gene transfer involving mobile genetic elements, conjugation, and other mechanisms. Reservoirs include the resistome (defined as microbial resources of resistance genes) and the mobilome (defined as microbial resources for enabling intercellular transfers of resistance genes) that are available within microbiomes in hosts and the external environment²². We defined human exposure pathways to include direct and indirect animal:human contact; releases from animal confinement houses; waste disposal; and consumption of food products derived from animals^23,24.

Results:

STEP 1 Antimicrobial pressure → selection for resistance

Fundamental to our understanding of mechanisms involved in the emergence of antimicrobial resistance is the fact that antimicrobial resistance is inherent within microbial populations. For billions of years, microbes have produced almost all currently used antimicrobial molecules in response to intensive competition for resources and survival within the microbiome²⁵. In this context, AMR evolved as an evolutionary mechanism by which microbes survived through natural selection by random gene mutation that encoded traits that conferred resistance to these natural biotoxins.

In contrast, human uses of antimicrobials are very recent, beginning in the early 1940s. Yet due to this prehistory, resistance mechanisms were already present within bacterial populations²⁶. During the first years of experimentation by Fleming and others, resistance was recognized as a consequence of exposure. Evolutionary theory explained the emergence of antimicrobial resistance as a process of random genetic mutations that conferred biological resistance to drugs²⁷. This theory also supported the assumption that each instance of resistance required either vertical transmission from the replication of a resistant organism or a separate evolutionary event. At first, little was known of the specific mutations or molecular mechanisms of AMR, but with the rapid development of molecular genetics, these altered proteins were identified²⁸.

Evolutionary theory also supported the assumption that there was a cost of resistance involving a trade-off between resistance and growth rate (the rK selection theory). Without this cost, bacteria would be equally likely to be resistant or susceptible in the absence of AM pressure and with the removal of AM pressure the prevalence of resistant strains would decrease. However, experimental observations contradicted theory, which was amended to include more complex evolutionary responses, such as “bet hedging,” by which microbial populations under AM pressure could acquire additional mutations to compensate for the cost of resistance²⁹.

Over the past 50 years, a substantial revolution has occurred in our understanding of the mechanisms by which AMR emerges and is disseminated. Current research now supports the hypothesis that horizontal gene transfer (HGT), rather than mutation, is the major mode by which bacteria (and other microbes) respond to antimicrobial pressure³⁰. Horizontal or lateral gene transfer among live cells was observed, although not understood mechanistically, as early as 1928³¹. Bacteria utilize several mechanisms to share resistance genes, including conjugation or exchange through direct cell:cell contact, transformation or incorporation of naked DNA from disrupted organisms in the extracellular environment, and transduction involving transfer of genetic material by transposable genetic elements^27,32. Later experiments demonstrated mechanisms by which donor cells initiate plasmid mediated gene transfer and how antimicrobials stimulate intercellular signaling between susceptible and resistant bacterial strains to initiate events including gene transcription that facilitate horizontal gene transfer from chromosomal DNA within the donor cell and responses such as swarming within the susceptible recipient organisms^32–34. The mechanisms by which resistance genes that are transferred among cells can be incorporated into the chromosomal genome of the recipient cell and expressed are also understood³⁵.

Concentrations of antimicrobials

The conditions of AM use also affect resistance emergence as well as dissemination. The most significant overall risk factor driving AMR emergence in any setting is the volume of drug use. Associations between overall drug use and prevalence of AMR has been shown by cross sectional comparisons of national drug use data³⁶ and longitudinally after bans on use of certain drugs in agriculture ³⁷. In addition, the concentrations of AMs to which microbes are exposed is also significant. Exposures to subtherapeutic concentrations of AMs (defined by bioassay at concentrations below the minimum inhibitory concentration, or MIC) are particularly effective as drivers of selection for AMR. This seemingly paradoxical observation reflects the Nietzschean aspects of bacteria: that which does not kill them makes them strong. Higher concentrations of AMs (greater than or equal to the MIC) kill bacteria whereas sublethal exposures stress but spare bacteria. As a consequence, these stressful but nonlethal conditions are particularly effective as drivers of selection for AMR through two mechanisms: increased growth and mutation rates and enhanced transfer of resistance plasmids and conjugative transposons³⁸.The survivors acquire resistance through these mechanisms as well as increased incorporation of resistance genes into chromosomal DNA. Continuous or prolonged low level AM use also expands the resistome and enhances the role of mobile genetic elements (MGEs), including plasmids, in mediating the dissemination of resistance within hosts and the environment within the microbiome^22,39

Use of multiple drugs

Repeated exposure to multiple AMs affects the emergence and dissemination of multidrug resistance through HGT of mobile genetic elements containing multiple resistance genes encoding resistance to several drugs. This results in both cross resistance and co-selection. These mechanisms were first demonstrated in 1989, with experiments showing that cross resistance among antimicrobials can be selected by one drug represented in the multi-drug resistant cassette⁴⁰. Through HGT, bacteria not only exchange individual resistance genes but also cassettes of multiple resistance genes, which encode for co-resistance to multiple antimicrobials. In other words, both pathogenic and nonpathogenic bacteria can easily share an entire cookbook of avoidance tactics rather than a single recipe. In response to repeated exposures to multiple AMs, bacteria acquire “genetic capital” in the form of sequential acquisition of resistance genes that can be transferred as a package through transposons within the mobilome⁴¹. These cassettes may be highly complex. A salmonella strain resistant to 13 antimicrobials was isolated from a child living on a farm who presented with ceftriazone resistance; all but one of the genes encoding multidrug resistance were on the same plasmid⁴². These multigene cassettes can include metal resistance genes such that co-selection and cross resistance can) also be driven by metals such as copper, cadmium, nickel. mercury, arsenic, and zinc^43,44.

These conditions -- use of concentrations of antimicrobials that result in subtherapeutic microbial exposures and use of multiple drugs in feeds – are common in the use of antimicrobials in poultry and livestock production. Another agricultural us→e is the long duration of repeated exposures for so-called prophylaxis or metaphylaxis (preventive treatment in the expectation of but absence of diagnosed disease). This may also involve sublethal concentrations of antimicrobials⁴⁵. These low dose and extended exposures to single or multiple antimicrobials condition networks of gene flow within the microbiome such that HGT is facilitated and role of mobile genetic elements in mediating resistance gene flow is enhanced within the gut microbiomes in animal hosts as well as in the environment⁴⁶.

STEP 2 Selection → Dissemination of resistance

HGT enables the rapid and efficient dissemination of resistance among bacteria (and other microbes) through highly efficient community signaling within the microbiome. This is in contrast to evolutionary mechanisms dependent upon random mutation or clonal expansion. At low concentrations, horizontal transfers of resistance genes among microbes rather than vertical transmission or de novo mutations are now recognized as the most important mechanism and explanation for the rapid and far ranging dissemination of resistance within and among microbial populations within hosts and the environment⁴⁷. These mechanisms support highly efficient mobilization of community resources of resistance. As a consequence, these resources are available to microbial networks that can be geographically distant as well as phylogenetically distinct.

Within and among microbial communities, HGT moves individual resistance genes as well as cassettes of multiple genes that encode for co-resistance and co-selection of resistance^22,30. These mechanisms underlie the complexities and underscore the facility with which bacteria respond to antimicrobial pressure with both emergence and dissemination. Once a new resistance trait and gene emerges, it spreads rapidly among microbial communities. This dissemination is further facilitated by movement of bacteria through air and water, changes in methods of food animal production, and human behavior including food consumption patterns, global travel, and international trade in animals and food.

These mechanisms of dissemination are exemplified by the rapidity and global range of resistance of β-lactams as evidenced in the emergence of extended β-lactamases in response to the introduction of new cephalosporins ^48,49. Since the isolation of the first of these drugs in 1948, there are now five generations of cephalosporins. Bacteria have rapidly responded to each generation of new cephalosporins with increasing numbers of distinct β-lactamase genes, now exceeding 1000⁴⁸. Both resistant bacteria and resistance genes encoding extended spectrum β-lactamase (ESBL) have spread rapidly and globally⁵⁰. Moreover, ESBL resistance genes are frequently bundled with other resistance determinants in transposable gene cassettes⁵¹_. Co-selection has been suggested as the mechanisms for the rapidity of selection for resistance to novel cephalosporins such as carbapenem as well as colistin⁵².

STEP 3 Dissemination → Reservoirs of resistance

Resistance reservoirs include the resistome (defined as the biological resources for responding to antimicrobial pressure) and the mobilome (defined as all the biological resources for transferring genes in response to pressure) ²².

These reservoirs exist within microbes and as naked DNA within physiological niches such as the gut and ecological niches in the external environment. The increasing use of antimicrobials has enlarged the resistome and increased the activity of the mobilome ^22,53. Increases in antimicrobial resistance genes and class 1 integrons have been reported in animals fed antimicrobials and have been documented in studies of soils treated with animal wastes or veterinary antimicrobials^47,54,55.

The environmental reservoirs of resistance may constitute the largest resources of these functions and are of specific concern in the context of agricultural uses through the release of untreated animal wastes containing resistance genes and antimicrobials that augment selection pressures within environmental microbiomes³⁹.

The environmental resistome has been a source of resistance in pathogenic bacteria isolated from humans²⁵. Since agriculture is situated directly within the physical and biotic environment, with numerous porosities from farm to fork, gene flow within and from food animal production contributes significantly to the environmental resistome⁵⁶. This involves both release of antimicrobials as well as resistance genes. Several studies have reported concentrations of antimicrobials in sediments impacted by aquaculture that are many fold greater than the minimal inhibitory concentrations for many drugs and pathogens⁵⁷. In addition, multiple mobile genetic elements have also been measured in soils and sediments⁵⁴. Empirical assessments of gene flow from agriculture into environmental microbiomes in soils and sediments have been published⁵⁸.

STEP 4 Reservoirs → Exposure pathways

To evaluate the last step in this conceptual sequence, exposure of human populations to drug resistant pathogens from food animal production, we considered the role of the mechanisms discussed above within the conditions and context of food animal production. Many of the conditions in food animal production resemble those risk factors that are conducive to the mechanisms of AMR emergence and dissemination first identified in healthcare settings, and for which interventions and guidance programs have been developed and implemented in many countries⁵⁹. They are exacerbated by animal stress and crowding during growth stages and transport^60,61.

In Figure 2, we summarize the evidence for the role of mechanisms listed in Figure 1 within the context of antimicrobial use in food animal production. We also indicate evidence supporting routes of exposure to these zoonotic pathogens from food animal production to human populations.

The food supply is the most significant pathway for human exposure to AMR pathogens from agriculture in terms of numbers of persons exposed, followed by multiple pathways of release to the environment. These two pathways operate both separately and in combination. In addition to consumption of food products from animals, there is an underappreciated and overlooked pathway of food-borne dissemination from the environment to crops consumed by humans. This is of particular risk when crops are grown with animal wastes (as in organic production) or with irrigation by surface water sources contaminated by run off from land disposal of animal^62,63.

The food and environmental pathways of exposure blur distinctions between healthcare and agriculture. Common sources of food are eaten inside and outside of healthcare facilities, and hospitals are located in environments where ambient air and water that may be contaminated by agricultural releases. Moreover, people – patients, visitors, and healthcare personnel -- move in and out of healthcare settings⁶⁴. For this reason, there are no real barriers between the presence of AMR in agriculture and the entrance of these same AMR pathogens into healthcare settings. These factors make it impossible to identify sources of resistance or to allocate burdens of disease between clinical and agricultural uses. This circularity is shown in Figure 5.

Figure 5.

Illustration depicting complex relationships among and between multiple sources of AMR.

Regardless of the original source of AMR, in most cases it is not possible to separate agricultural and clinical sources of genetic determinants of resistance in pathogens isolated from human populations, since genes and pathogens originating in agriculture quickly become sources of exposures and infections in human communities and eventually move into healthcare settings, and strains in humans can be transferred to animal populations. This gene flow goes both ways. There is a well annotated history of the cross transmission of so-called “livestock” strains of MRSA (ST398) from humans to animals and from animals to humans⁶⁵. Some studies of ESBL+ genes in E coli isolates from animals, including carbapenemase, suggest that this may represent contamination of the agricultural environment by human wastes⁶⁶.

Discussion

We undertook this study to improve the evaluation of evidence related to biological plausibility of associations observed in non RCT studies relevant to public health. The development of a transparent method for assessing the quality of these types of associations in observational studies is of high importance. Current assessment methods based on GRADE are not appropriate because of the inherent limitations of public health studies. Moreover, the use of GRADE, as in the systematic reviews conducted by WHO, may lead to underestimation of important findings. The USDA issued a statement shortly after publication of the WHO guideline, which referred to this “low-quality evidence” as effectively disqualifying any WHO recommendations despite the surrounding analyses and expert opinion ⁹. We selected the adverse outcome pathway approach based on our interest in the application of these methods for supporting the evidence derived from observational studies.

With expert consultation, we accessed papers describing general and detailed methods for organizing structural models representing biological plausibility through mechanisms that link exposures to health outcomes. One of these methods uses a comprehensive information set based on the molecular biology of cancer (Lewis et al), and the other utilizes the more generalizable concept of adverse outcome pathways (Ankley et al) ^3,15. We selected this latter model because of its applicability to observational studies and the substantial record of use in toxicology and ecology to support evidence based decisions related to risk assessment^4,67,68. We populated our framework of adverse outcomes pathway analysis, using the literature on mechanisms of antimicrobial resistance and assigned mechanistic evidence to a sequential pathway linking antimicrobial exposure of microbial communities to human exposure to drug resistant pathogens.

We focused on mechanisms that drive microbial response to antimicrobial stress through the emergence and dissemination of resistance as well as accumulation of resistance genes and organisms in reservoirs. To this model, we added evidence on the major pathways of human exposure to AMR pathogens from agricultural sources.

The conditions of agricultural use facilitate many of the mechanisms in AMR emergence and transmission, such as horizontal gene transmission and the frequency of multidrug resistant phenotypes. By including a further focus on agricultural use, this assessment also supported the importance of the microbiome perspective. Moreover, it illuminated the role of agricultural use in expanding environmental repositories or resistomes through the direct contribution of agriculture to multiple pathways of release, and from which AMR genes can be transferred to bacteria in human populations.

Conclusions:

It is recognized that all uses of antimicrobials contribute to the emergence and dissemination of resistance⁶⁹. In the context of increasing global threats of antimicrobial resistance, we need evidence to support effective interventions to control uses of antimicrobials in both healthcare and agriculture. The evidence has been summarized in recent systematic reviews^5,6,70, which reported associations observed between agricultural use of antimicrobials for all purposes and increased risks of AMR exposure of human populations. This paper adds an analysis in support of the biological plausibility of these observations, using published methods based on a mechanistic approach. We conclude that this approach may be applicable to evaluating the evidence for biological plausibility as part of an overall assessment of evidence for action based systematic reviews on topics in which associations have been observed based on observational studies. This first application requires validation by application to other systematic reviews where the criterion of biological plausibility is of value.

Figure 3.

Conceptual model with explanatory text to describe the biological plausibility between agricultural AM use and risk to human population.

Fig 4.

The relationships within the global microbiome and its pangenome including the resistome and the mobilome that support horizontal gene transfer in response to antimicrobial pressure including those genes encoding resistance to clinically important antimicrobials. The panproteome includes the gene products of the microbiome, including the parvome which includes clinically important antimicrobial molecules produced by humans²².

HIGHLIGHTS.

How can we be confident in our findings in public health research?

• Public health policies are often controversial and expensive to implement

• To increase our confidence, we utilize tested criteria to reduce bias in our findings and to support action

• These criteria include evidence of biological plausibility or mechanistic information that connect exposures and outcomes

• Biological plausibility is an important criterion in evidence based decision making especially in public health because our evidence usually comes from observational rather than controlled studies.

• Using methods from ecology and toxicology, we tested the utility of the adverse outcome pathway model to access evidence supporting a model that connects agricultural sources of antimicrobial resistance, pathways of human exposure, and health outcomes.

• We show this schematically below. The model may be useful for other applications.

List of Abbreviations:

AMR

Antimicrobial resistance

ESBL

Extended spectrum beta lactamase

GRADE

Grades of Recommendation Assessment, Development and Evaluation

HGT

Horizontal Gene Transfer

MGE

Mobile genetic elements

RCT

Randomized control trial

WHO

World Health Organization