5

Experimental Strategies for Addressing Risk Issues

INTRODUCTION

As described in Chapter 2, the application of classical gene transfer and genome editing to food animals yields lines of animals that possess heritable genetic modifications (HGMs) aimed at improved production efficiency, greater resistance to disease or environmental stress, novel product qualities, control of reproduction, or other traits. Concerns about possible novel harms posed by HGM animals due to altered food composition, ectopic expression of bioactive, allergenic, or toxic compounds are addressed in Chapter 3. A well-established risk analysis framework may be applied to estimate the likelihood of harm to the animal or to humans becoming realized, and to manage possible risks by mitigating exposure to hazards posed by HGMs (Chapter 4). Against this background, this chapter considers the experimental strategies and methodologies that may be applied to evaluate the safety of HGMs to animals and to consumers of food products derived from them. The chapter presents methods for the identification and analysis of intended and unintended genetic alterations; approaches to identify potentially hazardous compositional changes in HGM animal-derived food products, bioactivity of hormones and other bioactive compounds, toxins, or allergenic components; methods for assessing the risk posed by disease-resistant HGM animals to other animal and human populations; and challenges to studying the potential risks of HGM food animals. It concludes by discussing approaches to risk assessment in the face of uncertainty of the likelihood and magnitude of health hazards that may be posed by HGM animal-derived products.

APPROACHES TO ADDRESS CHARACTERIZATION OF INHERITED SEQUENCE CHANGES

An important early step in the process of developing HGM animals is the characterization of inherited DNA sequences (Figure 4-1). Whether changes that may have occurred in genomic DNA following gene transfer or genome editing can be detected partially depends upon (1) the type(s) and complexity of mutations that have occurred; (2) the process used to create the HGM animal; and (3) the filial generation used as the de facto founder animal from which the HGM is propagated (Burger et al., 2024).

Chromosome segments are constantly rearranging as the result of meiotic recombination and also in response to environmental damage (e.g., sunlight). While scientists refer to these changes as “mutations,” they are not necessarily hazardous but represent normal variation within the genome. Alterations to the genomic DNA sequence of

an animal induced by natural mutation, gene transfer, or genome editing presents potential hazards. It is important to distinguish between exogenous DNA insertions, such as template insertions and viral vector integrations, and endogenous DNA insertions (Hunt et al., 2023). Prior to development and commercialization of an HGM animal line, the animal’s genome should be carefully assessed for the presence of any exogenous DNA insertion events. With this important proviso, not all endogenous DNA alterations can be attributed to the process of achieving HGM due to the occurrence of naturally occurring mutations (Chapter 2). This statement does not necessarily imply that all such replication errors are likely to be hazards that will result in a harm as discussed in Chapter 3, but rather should be taken as part of a holistic risk mitigation strategy as discussed in Chapter 4.

Further, because of the many layers of risk mitigation present in the use of this technology for agricultural applications (Figure 4-1), extensive and comprehensive attempts to find all genome edits in regions with no known functional elements (e.g., unannotated intergenic and intronic regions) are unnecessary. This conclusion stems from the axiom that risk assessment cannot directly factor in results for which there are no known specific harms. In such situation, the guiding metrics for risk assessment should be based upon the impact of such HGMs, if any, on the most relevant measurable outcome, that is, the phenotype of the animal. The animal’s phenotype is the ultimate indication of the outcome of all genetic changes, either natural or induced. However, if the initial phenotypic assessment identifies unanticipated characteristics, additional assessment of the animal is required, which could include determining whether other mutations occurred that are causative of the phenotype. In such cases, it is useful to decipher whether these mutations are from de novo or unintended DNA modifications from the editing tool.

DNA alterations that have occurred as a result of clustered regularly interspersed short palindromic repeats (CRISPR)-Cas editing in human cell lines include megabase-scale deletions, insertions, chromosomal truncations, copy-neutral loss of heterozygosity, translocations, and chromosomal rearrangements (Hunt et al., 2023). An unintended foreign vector sequence was introduced into cow genomes using transcription activator-like effector nucleases (TALENs) by Carlson et al. (2016). Although TALENs are less-used than CRISPR technology, TALENs are more flexible than CRISPR-Cas9 because they do not require protospacer-adjacent motif sites at the intended target site. Because current or future technology could unintentionally incorporate a foreign DNA sequence into an HGM animal, methods are needed to ensure the detection of such an outcome in any genomic region.

Documenting What Has Changed: Reference Genomes, Databases, and Comparators

Reference genomes and pangenomes

In general, across food-animal species, more diversity exists in animal genomes than was recognized before the widespread use of whole-genome sequencing. Initial genome sequencing efforts focused on developing so-called “reference genomes” for each species. A reference genome is a genomic sequence that is assembled and annotated as a representation of the species’ nucleotide sequence. In some cases, a reference genome may be based on genomic sequences from multiple individuals, while in other cases, they are developed from a single individual’s genome (in many cases, an inbred individual) (Crittenden et al., 1993; Clark et al., 2020). While reference genomes provide a key foundational tool to support genomics research, these reference genomes represent either a single individual or an “average” sequence, and often obscure important individual or even breed- and line-related variations. Assessment of the results of gene editing requires accurate knowledge of precisely where single nucleotide variants (SNVs), indels, and structural variants (SVs) may occur in alternative populations across breeds of the same species. As genome sequencing methods became more accessible, alternative genome sequence assemblies were developed to capture different SVs and to better represent variation within and between different lines and breeds of animals.

The inclusion of newly discovered indels and SVs into reference or breed-specific genomes, however, is problematic because upon each release of the next updated version of a reference genome, all the coordinates (reference positions) on the chromosome map change, making it difficult to compare current variants with preexisting variants, that is, those that were discovered within an earlier version of the reference genome. This problem is magnified due to the development of breed- or line-specific genomes, which currently exist for sheep, cattle, pigs, and chickens. Further, if the genome of a crossbred animal is edited, the decision of which genome should

be used as a reference would impact which variants are known to preexist. Comparison of a crossbred animal’s genome to either parent’s breed-specific reference genome would be inaccurate as each breed has specific genome variants such as indels and SVs that contribute to its unique nature. If an edited animal’s genome contains a variant (SNV, indel, or SV) that is not present in one reference genome (but perhaps is in the other), it may be incorrectly identified as an unintended alteration, or the variant sequence may be discarded during the genome assembly and analysis. This lost information results in poor accuracy and reduces the ability to detect preexisting indels and SVs.

An alternative to analyses based on simple species-specific reference genomes is the development and use of a pangenome (Box 5-1).

BOX 5-1
Pangenomes

Pangenomes differ from reference genomes in that they can capture all variation within and between populations. With a pangenome, DNA sequences are mapped using graph theory. Graph theory is a mathematical representation consisting of nodes, edges, and paths, which was developed as a mathematical tool in the 1700s to test theories involving paths, but has since been used in many other applications (Sachs et al., 1988). When applied to the human genome, the pangenome approach for DNA fragment alignment resulted in improvements in read mapping, small-variant genotyping, novel variant discovery, structural variant (SV) genotyping, and representation of complex variants (Liao et al., 2023). In the pangenome representation, nodes represent sequences or segments of DNA that occur in one or more genomes. Each node can correspond to a specific genetic variant or a conserved sequence across different genomes. Edges connect nodes and represent the relationships or transitions between these sequences. Each genome is represented via a unique path through the graph, capturing its specific genetic composition (Figure 5-1). Pangenomes visualize variations such as single nucleotide polymorphisms (SNPs), insertions, deletions, and SVs more effectively than conventional comparative genome browsers or reference genomes.

Currently, pangenomes are available for chickens (Rice et al., 2023) and pigs (Li et al., 2017; Tian et al., 2020), with an ongoing effort to produce them for ruminants. These pangenomes are being assembled using telomere-to-telomere (T2T) sequencing (Kalbfleisch et al., 2024) to ensure complete genome sequences. There is a great need for high-quality T2T genomes for all food-animal species to address proper curation of existing SVs and to document new SVs as they are discovered. T2T genomes are important because evidence is emerging that many regulatory elements exist in intergenic regions associated with highly repetitive sequences (Soto et al., 2023). As noted in Chapter 2, in human populations, only 5 percent of disease-associated single nucleotide polymorphisms (SNPs) are located in gene coding sequences, with the remaining 95 percent of disease-associated SNPs located in non-coding regions, which make up 98 percent of the genome (Orozco et al., 2022). Improved food-animal genomes will support assessment of what has changed in HGM animals and provide a clearer understanding of possible phenotypic impacts of unintended edits to HGM animals. Addressing this gap is considered “nice to know” but unlikely in the United States; however, scientists in the European Union and United Kingdom are making advances in this area of research which may benefit U.S. scientists.

Comparators

The method used for creating a founder HGM animal determines which comparator is most appropriate for assessment of any changes that might have occurred in the DNA due to the gene transfer or editing procedure. For assessment of intended and unintended HGMs of a founder individual, the best comparator is genomic DNA from the dam and sire. Likewise, for offspring of a founder, the grandsire and granddam are the ideal genomic comparators for assessing inheritance of intended and unintended DNA insertions or edits. However, a primary approach for generating founder mammalian livestock, particularly pigs and cattle, is the use of oocytes collected from abattoir-sourced ovaries for in vitro embryo production. The zygote-stage embryo is then subjected to gene transfer or editing and transferred into a recipient female for establishment of pregnancy or developed in vitro to the blastocyst stage before transfer into a surrogate dam (Park et al., 2017; Miao et al., 2019; Ciccarelli et al., 2020). Because the ovaries used for sourcing oocytes usually come from multiple different female donors, precise identification of the dam of a founder can be challenging. As such, direct comparison of the genomic DNA sequence from a founder and offspring of the founder to both the sire and dam may not be possible. In this case, databases are available to support identification of preexisting SNPs and small indels (Cezard et al., 2022; Martin et al., 2023; Sherry et al., 2001) within species-specific databases (Shamimuzzaman et al., 2020). Unfortunately, such databases cannot include all variants due to ever-arising natural mutations, and no existing database includes large indels or SVs.

Another means for generating HGM food animals is through somatic cell nuclear transfer (SCNT). With this method, a donor cell line is first edited or transformed, following which a nucleus from the cell line is used to replace that of an enucleated surrogate embryo to produce a whole HGM animal (Whitworth et al., 2014). For animals that are created using an edited cell line, the closest comparator is the cell line prior to its being edited.

In some species, such as poultry, neither SCNT nor direct embryo editing is possible because of inaccessibility of the zygote at the time of fertilization. In the case of poultry, HGM via gene transfer or genome editing is accomplished by harvesting primordial germ cells (PGCs) from the embryo soon after the egg is laid. These PGCs are then subjected to gene transfer or gene editing and injected back into a surrogate embryo. The comparator in this case is the PGC cell line pre-editing (Chapter 2). The resulting chick will be mosaic, but some of the edited PGCs may have become incorporated into the gonadal tissue (Idoko-Akoh et al., 2018, 2023; Lee et al., 2022). Offspring from the mosaic founder will then have to be screened for presence of the transgene or edit, although methods have been developed such that only genetically modified offspring are produced (Ballantyne et al., 2021).

Mosaic Animals and Optimal Generation to Target Individual as the De Facto “Founder”

A limitation common to all current methods for modifying the DNA of animal genomes is that changes detected in the HGM animal do not necessarily reflect what will be inherited in future generations due to (1) possible heterozygosity of the targeted locus such that the edits are not identical on each chromosome, resulting in alterna-

tive alleles being transmitted to offspring, and (2) the possibility that the founder is mosaic for the gene insertion or edit and thus partially reflects the somatic tissue of the parent rather than the germ cells. A mosaic individual results when the DNA edit is not fully completed or permanently integrated into the genome before the first cell division occurs. In this case, not all cells will contain modified genomes, including some cells that will eventually give rise to the germinal tissue (sperm and ova). As a result, the germinal tissue will be heterogenous and will not consistently produce gametes that include the edit or transgene. The genotype of an individual gamete ultimately determines what can be transmitted to the next generation. Detection of mosaic founders is not trivial, as samples taken from the gonadal tissue of an edited animal may include contamination from surrounding somatic cells. Direct genotyping of individual gametes would address this problem, but that method is not currently possible. An alternative approach would be to use a chosen descendant of the original edited animal as the de facto founder and the focus of genome characterization. Because offspring inherit only one of the alleles at each locus from the parent, mosaicism is eliminated as a confounding factor. This approach would also simplify identification of what has changed in the DNA, particularly SVs (discussed below).

Additional Approaches for Identification of Genomic Modifications

As noted above and in Chapter 3, the use of DNA modification methods for generating food animals can bring about unanticipated changes to DNA. However, the pipeline of integrating HGM animals into a production system includes several layers of assessment that ensure safety of the end product for human consumption (Chapter 4). In many cases, these additional steps make it unecessary to apply stringent criteria for detailed characterization of intended and unintended editing at the DNA level. Nevertheless, a reasonable examination of the range of possible HGMs can be achieved with existing laboratory and bioinformatics analyses.

Methods to explore the intended HGM are straightforward, but the results can be misinterpreted if possible alternative explanations of results are not explored (see Burgio and Teboul, 2020, for extensive review). The first method is polymerase chain reaction (PCR) amplification of the intended target site followed by amplicon sequencing. While this method is well tested and relatively inexpensive, there are limitations that reduce its utility as a verification tool. For example, if the genotyping strategy is such that the genomic DNA sequences for which the primers are designed to anneal are mutated, there will be a failure to amplify. This limitation can be overcome by designing multiple sets of primers that amplify beyond the target region, but there are circumstances in which an indel mutation may be so large that accurate detection by PCR is impossible. Therefore, a multipronged assessment should be considered to fully characterize both intended and unintended editing events.

Whole-genome sequencing is the most comprehensive tool available for examination of alterations in genomic DNA. Currently, genomes are sequenced from a random sample of DNA fragments. Depending upon the approach utilized, these fragments may either be short fragments, ranging up to 600 base pairs (bp), or long fragments, ranging upward to over a million base pairs. Short fragments may be sequenced as either single (~100 bp) or paired end (2×150 bp) reads. When sequencing a paired-end fragment, only the ends are sequenced and there may be an insert up to 300 bp between the sequenced ends. The strength of genome sequencing is that the DNA fragments that were sequenced and used to characterize changes in the genome reflect the actual DNA sequence of that animal across the full genome. Each sequencing technology has its own strengths. The challenge with genome sequencing is to reconstruct the reads into a genome assembly that exactly matches the genome of the individual from which it was taken. The current standard method of reconstruction is to align, or map, the sequence reads to a reference genome, usually developed from a single individual of that species (Crittenden et al., 1993; Warr et al., 2020). In contrast, population genomics analyses within and across species show that every individual has a unique genome sequence due to a variety of mechanisms that create DNA sequence variations, such as meiotic recombination, DNA replication errors, SNVs, and SVs.

Unfortunately, mapping reads to a reference genome may not allow inclusion, and thus identification, of a newly mutated SV, depending upon the sequencing technology utilized and the read length. If the DNA sequence from a given fragment differs from the reference by more than a given percentage, depending upon the bioinformatic software parameters, those reads will be discarded. This loss of reads biases the reconstruction into a genome that resembles the reference genome and is known as a “reference bias” (Martiniano et al., 2020; Oliva

et al., 2021; Hickey et al., 2024). This bias is particularly relevant for exogenous DNA (virus, plasmids, and other foreign DNA) and moderate- to large-sized endogenous SVs because an insertion, deletion, or inversion can cause the read to be grossly different from the reference genome in that region. Because DNA edits generated by currently employed CRISPR-Cas9 technologies are designed to cause one or more double-stranded breaks, the intent of the editing method is to create an SNV, indel, or SV; the latter (SV) is the type of mutation that is most difficult to resolve with genome sequencing and assembly.

The resolution of an SV type is difficult due to the combination of reference bias and haplotyping. A haplotype is a sequence of DNA that is transmitted intact from the parent to the offspring, without recombination. “Phasing” is a related term indicating the ability to identify the parent from which alternative haplotypes came. Phasing is critical to determining which haplotypes combine to form an SV. Reference bias poses a challenge because the intended or unintended edit may not exist in the reference genome (or possibly even in a pangenome for that species) because it is unique to the edited animal. As a result, the sequence reads cannot be mapped and the reads will be discarded, thereby producing a blind spot in the genome assembly. Moreover, even if the sequences could be mapped, most haplotypes cannot be resolved because the break points of the respective alleles on each chromosome are different. Without knowing which haplotypes are associated with which parents (i.e., are phased), most SVs cannot be detected or defined. This result was shown in a study in which a majority of SVs could be identified and correctly defined separately in two haplotype hydatidiform moles, but could not be detected or defined when the two hydatidiform mole’s genomes were combined in silico to make a pseudo diploid (Huddleston et al., 2017).

Some genome sequence limitations can be overcome by recognizing and addressing the causes. However, a more general approach to analyzing intended alterations is needed to find all changes regardless of reference bias. In addition, the use of computer programs and workflows removes the human element from the analysis, which has both positive and negative consequences. Best-practice approaches should include a human element in the analysis and include visual examination of the target area. Such examination could be accomplished via a genome viewer software tool in which “broken” paired reads can be identified and unaligned read ends visualized, as detailed by Mahdi et al. (2025). If broken paired reads are identified or multiple unaligned ends are observed, an indel, SV, or exogenous DNA is likely present and requires further examination.

Unmatched sequence pairs along with their matching mates should be extracted from the original raw sequence data. The mate pairs should then be aligned using a de novo assembler, thereby eliminating the reference bias. The resulting contiguous DNA sequence (contig) alignments can then be identified by a basic local alignment search tool (BLAST) search against all archived DNA sequences to determine whether exogenous DNA is present. For example, this method was used by Norris et al. (2020) in a study revealing that sequences from a DNA cloning vector had been unintentionally integrated into a gene-edited animal. Norris et al. (2020) made this finding by using prior knowledge, submitted by the researchers, as to what cloning vector had been used. They then combined the reference genome of the cloning vector along with the species’ reference genome to align the sequence reads. Results showed that some reads mapped to the cloning vector reference, thereby revealing that the vector had been integrated into the genome of the edited animals.

Alternatively, when the method of Mahdi et al. (2025) was used, unmatched mate pairs were observed. The matching pairs were extracted from the raw unaligned sequences and realigned using a de novo assembler. Following realignment, the resulting contigs were BLAST searched against all archived DNA sequences. The search confirmed the result, that is, presence of DNA sequence from the cloning vector, but without prior knowledge that a cloning vector had been used. This result demonstrates that exogenous DNA can be detected without knowledge of what exogenous DNA might have been integrated.

The example discussed in that case was for a “knock-in,” edit, in which a new DNA sequence was inserted into the genome of an animal. If a “knockout” (when a DNA sequence is removed from the genome of an animal) is the intent, then the same genome sequencing approach can be used, but an additional critical step is necessary, that is, reconnecting unaligned ends (Mahdi et al., 2025). Because the edit in such cases is a deletion, two intended DNA strand breaks are expected, and identification of haplotypes extending across the breaks is necessary such that reads of the same haplotype could be extracted and reassembled by de novo alignment. The resulting contigs are then searched against the species’ reference genome to find what part of the chromosome is deleted. All current software requires haplotype-resolved genomes to reconstruct SVs accurately (Hickey et al., 2024). The

method of Mahdi et al. (2025) was similar in that regard but requires incorporating the human element to resolve the haplotypes by visual inspection.

While specialized programs for detection of SNPs and SVs have been developed, such as Manta (Chen et al., 2016), and DELLY (Rausch et al., 2012), along with commercial software such as CLC Genomics Workbench (Qiagen¹), those approaches do not address the general problems of reference bias or haplotype resolution. As discussed above, a more general solution is to map DNA sequence reads using a graphical format (Garrison et al., 2018; Hickey et al., 2020; Martiniano et al., 2020). A graphical alignment to a reference genome allows a unique genome to be represented for each individual, thereby incorporating all DNA-sequence fragments into a genome that matches the individual from which it was sampled. This approach has been shown to be more accurate than the specialized approaches of DELLY and others (Hickey et al., 2020). However, current software programs used to find SVs by mapping to a graph are accurate only for species that have defined haplotypes (i.e., humans). Nevertheless, haplotypes can be defined for any edited animals at specific sites by visual examination using overhang reads, as demonstrated by Mahdi et al. (2025).

Design of High-Confidence Guide RNAs to Minimize Unintended Editing

Many unintended edits from CRISPR-Cas9 machinery are expected to result from promiscuous hybridization of guide RNAs outside of a targeted site. The guide RNA portion with sequence complementarity to the targeted DNA site, referred to as CRISPR RNA or crRNA, is designed to be 17-24 nucleotides in length with the standard being 20 nucleotides. Research with cell lines and model organisms has shown that a 3-nucleotide mismatch between the crRNA and DNA reduces activity for Cas9 association and strand break formation by 40-fold and that a mismatch of >3 nucleotides all but eliminates Cas9 association and the potential for editing (Bravo et al., 2022). Multiple algorithms have been designed to identify potential unintended interactions by guide RNAs including CRISPOR and Cas-OFFinder (Bae et al., 2014). These algorithms can be used during a CRISPR-Cas9 editing design phase to choose guide RNAs that will have little to no potential for off-target hybridization.

Other Methods to Query Putative Unintended Alterations

Molecular-level detection of unintended changes can be accomplished through several methods, including sequencing of likely potential off-target sites through the use of in silico analytical tools, in vitro cell-free methods, whole-genome sequencing, and whole-exome sequencing. In silico tools can predict genomic DNA sites where unintended CRISPR-Cas editing could occur via the use of prediction algorithms based on single guide RNA (sgRNA) sequences. These in silico tools, however, are sometimes insufficient when considering complex intranuclear microenvironments involving epigenetic and chromatin organization states. Cell-free detection methods that use a reconstituted nuclease reaction to digest DNA or chromatin extracted from cells to identify genomic cleavages (e.g., digenome-seq using cell-free chromatin DNA) offer higher accuracy for suggesting off-target sites for further investigation (Guo et al., 2023). Other more rigorous techniques also can be used; for example, genome-wide unbiased identification of double-strand breaks (DSBs) enabled by sequencing relies on delivery of double-stranded oligo deoxynucleotides (dsODNs) with known sequences that can integrate into DSBs during non-homologous end joining. The integrated dsODNs provide templates for targeted PCR amplification and sequencing of the tagged DNA fragments. For example, linear amplification-mediated high-throughput genomic-wide sequencing better detects DSBs and chromosomal translocations (Guo et al., 2023). Circularization for high-throughput analysis of nuclease genome-wide effects by sequencing is another method (Lazzarotto et al., 2020) that is simplified, sensitive, and scalable in helping to understand the specificity of action of genome editors.

A key issue for detecting unintended genome alterations is the challenge of distinguishing them from natural mutations. Whole-genome sequencing is the standard method for detecting unintended alterations. As noted in Chapter 2, mutation rates vary among species. A recent review of estimates derived from trio sequencing of parents and progeny of 150 species (Bergeron et al., 2023) showed that de novo mutations were more frequent than previously recognized. Whole-genome sequencing can be used to compare the genome sequence before and after

___________________

¹ https://digitalinsights.qiagen.com/clc-genomics-workbench-features

CRISPR-Cas editing and for identification of possible unintended mutations. However, when editing is applied to cultured cells, dividing cells also accumulate spontaneous mutations that are not attributable to genome editing (Carroll, 2019). The sensitivity of whole-genome sequencing for detecting where off-target edits have occurred is determined by sequencing depth (i.e., the number of times that a particular genomic region was sequenced), while the accuracy of determining what has changed at that location is dependent upon the genome-assembly software’s ability to reconstruct the actual indel or SV that occurred. One challenge is to distinguish mutations attributable to genome editing from those of background mutation (Box 5-2). De novo mutations, whether due to the editing process or not, can be detected only by trio sequencing, that is, if the mutation is not in either parent, but present in the offspring, the mutation may be a de novo mutation. It is currently impossible to determine the cause of a new mutation because the editing process does not leave any signature.

Assessment of unintended editing in founder animals

If the HGM founder is intended to be a prototype animal for initial phenotyping, experimental rigor requires that the potential for unintended guide RNA binding be examined to determine whether an acquired trait is due specifically to an intended or unintended editing event. However, as discussed above regarding mosaic animals, the founder animal may not ultimately be used as a source of germplasm for integrating edited genetics into production populations. Thus, the value of assessing founders for unintended editing may generally be limited to the research- and-design phases of genome-editing application development and should be considered on a case-by-case basis.

APPROACHES TO IDENTIFY PHENOTYPIC CHANGES IN HGM ANIMALS

While genome sequencing can provide information about loci where an individual varies from another and the nature of the DNA sequence change, this sequence information cannot identify whether the variation is due to natural variation or genome editing (either intended or unintended). Moreover, the genome sequence itself does not necessarily indicate the biological, physiological, or phenotypic consequences of the alteration. For gene edits that target a specific gene, the function of that gene’s products can be assessed to examine expected biological and phenotypic changes. However, genome editing may also include changes to intergenic regions that may alter regulatory sequences, inducing potential downstream effects that may alter unrelated phenotypes. Alterations to a genome—whether intentional or unintentional—can be broadly categorized as (1) alterations to a well-studied functional element (e.g., gene) that has clearly defined phenotype effects; (2) alterations to a known functional element that is not well characterized and will have poorly defined phenotype effects; and (3) alterations to the genome in a presumed intergenic region whose disruption may result in unintended phenotypic effects. The first category can easily be assessed using defined assays and techniques. The second category is unlikely to arise in a commercial setting, although this approach has been used in model organisms to determine the gene function of poorly characterized genes (e.g., gene knockouts). The third category is more difficult to assess, as some intergenic regions contain important regulatory elements that are poorly annotated across all food animals.

Clark et al. (2020) note that precise annotation of food-animal genomes will help identify optimal gene-editing target sites, offer insights into the effects of genome edits, and connect high-resolution molecular genotypes to resultant phenotypes. Genomic annotation of functional elements in food-animal genomes is being undertaken by the Functional Annotation of Animal Genomes (FAANG) Consortium (Andersson et al., 2015), a global effort engaged in developing assays and collecting information on transcribed loci, methylation, chromatin accessibility and architecture, transcription factor binding sites, and histone modification marks for multiple food-animal species (Andersson et al., 2015). The goal of this project is to provide detailed information about how regulatory elements and chromosomal structure contribute to gene expression and molecular phenotypes. There has been no dedicated U.S. funding program for this annotation project or adequate funding to support understanding of the functional consequences on phenotypes resulting from sequence variation in these regions.

Although the FAANG project will provide essential structural information that informs understanding of gene regulation and genome structure in some food animals, there are some limitations associated with this effort. First, the FAANG project is necessarily limited to “high-quality” genomes and includes only limited work on aquaculture species (particularly invertebrate species), although a European Union-funded consortium is studying six fish species (Johnston et al., 2024). Second, many of the assays being applied are dependent upon tissue type, developmental stage, and environmental factors (e.g., assays such as methylation and chromatin accessibility) present at a specific moment in time, so that a complete annotation would require data collection from a large number of tissues, ages, and husbandry conditions. Third, new -omics technologies are being continually developed that could improve the knowledge of genome function and structure (e.g., Hi-C sequencing to identify open regions of chromatin and single-cell sequencing techniques are now being applied to food animals).

However, the identification of functional elements with food-animal genomes is only the first step toward assessing the impact of any genetic changes to an animal’s well-being and production capability. The data collected by FAANG must also be integrated, curated, and provided in a format that gene editors are able to easily access, readily use, and correctly interpret. Since food animals do not have model-organism databases to provide highly curated, standardized datasets, the FAANG Data Coordination Center ensures high-quality and rich supporting metadata for food-producing and other animals. While this approach provides access to the primary data, more work will be required to disseminate this information to genome resources used by food-animal researchers since no single central database exists. The question of how to integrate, interpret, access, and use data from the FAANG project remains unresolved.

The potential impact of poor or incomplete annotation on the ability to fully assess food safety or other risks posed to HGM animals or from the foods derived from them has not been extensively discussed in the published literature. There are several ways that incomplete genome annotation can hinder the ability to assess how changes to the genome may result in altered phenotypes:

1. Many edits are made in genomic regions currently considered to be non-coding, but because annotation of these regions is quite limited, true non-coding areas from unannotated elements cannot be distinguished, creating a structural annotation problem.
2. A second challenge relates to defining normal gene expression for a tissue and developmental stage. Many research groups collect expression data, but gaps exist, and these data are largely limited to transcriptome information, rather than expressed proteins.
3. The consideration of non-coding RNAs (e.g., microRNAs, long non-coding RNAs, and anti-sense transcripts) adds further complexity, as these regulatory RNAs have been identified but little is understood about their specific functions. Lack of knowledge about non-coding RNAs is another gap in current knowledge, and further limits understanding of the relationship between transcriptome and phenotype.
4. A related issue is the lack of knowledge about whether a small change in gene expression causes a phenotypic effect. This gap creates a functional annotation limitation necessitating better understanding of biochemical pathways and biological processes.
5. The genetic variation within and among lines and breeds of livestock is highly important yet sometimes poorly understood. For example, the differentiation between layer and broiler chicken lines (which have a 97 percent DNA homology across the respective genomes) is much greater than the differentiation among inbred laboratory mouse lines, likely because farm animals have been selected for many more generations than mice and inbreeding was avoided. The implications of natural genetic variation for functional outcomes and phenotypes remains poorly understood and contributes to the complexity of understanding how genetic differences translate into phenotypes and biological relevance.

Assessment of Impacts of Changed Metabolism in HGM Animals

Assessment of potential harm to an HGM animal requires the verification of proper function of the transgene or edit and consideration of any unintended effects caused by expression of the HGM. This assessment may include the application of targeted -omics approaches to assess outcomes. Approaches to verify or assess phenotypes usually emphasize genomics applications; however, linking genetic changes to subsequence expression of phenotypes is not always straightforward. Instead, approaches such as proteomics and metabolomics may be more directly relatable to direct measurement of phenotypes and their use should be considered. The application and utility of omics-level approaches (e.g., transcriptome, proteome, and metabolome evaluation) to analysis of food-animal systems is demonstrated by the results of a small body of empirical studies in transgenic food animals. For example, Clark et al. (2014) applied methodologies aimed at detecting pleiotropic effects at the whole-animal level for a line of transgenic goats producing antimicrobial human lysozyme (hLZ) in their milk, an experiment with the ultimate goal of using the milk to treat childhood diarrhea. Metabolomics was used to determine the serum metabolite profile of both the host (lactating does) and non-target organism (kid goats raised on control or hLZ milk) prior to weaning (60 days), at weaning (90 days), and at 1 month post-weaning (120 days). Serum metabolomics showed differences over time but revealed no significant differences in metabolites between control and hLZ-fed kids after correction for false discovery rate. Serum metabolomics of control or hLZ lactating does showed only one significant difference (q = 0.04) in an unknown metabolite. Histological analysis of kid intestines showed that consumption of hLZ milk resulted in positive or insignificant intestinal morphology and metabolic changes and supported the safety and durability of the hLZ mammary-specific transgene. This work demonstrates the application of a targeted multi-omics approach to assessing both intended and unintended effects upon an HGM food animal.

Another example of using a targeted multi-omics approach is the analysis of transgenic cows expressing human lactalbumin alpha (LALBA) (Wang et al., 2008). Overexpression of human LALBA in cows did not affect milk composition (Zhang et al., 2012). A subsequent study used proteomics to compare colostrum and milk between transgenic and conventionally bred cows (Sui et al., 2014). Proteins expressed only in the transgenic cloned animals were identified and analyzed functionally, and it was noted that these proteins were expressed at levels that would not have been detected by conventional protein detection methods (e.g., a silver-stained Western blot). The hypothesis that heterologous LALBA might abnormally regulate gene expression via endogenous microRNAs or transcription factors and thereby change the metabolomic profile of cow serum was tested by Wang et al. (2017).

Metabolomic profiling of serum showed that LALBA-transgenic cows had a different metabolomic signature compared with non-LALBA cows, although standard serum biochemistry testing indicated that signatures of both groups were within the normal ranges of healthy cows. While the sensitivity of multi-omics approaches can identify differences between conventional and HGM food animals, the proper interpretation of these data is crucial. As noted in Chapter 4, statistical significance does not indicate biological relevance.

As noted in Chapter 4, both hazard identification and exposure are required to estimate any risks posed by food products derived from HGM animals. If a novel product is expressed in HGM animals, then protein characterization and hazard identification encompassing toxicity and allergenicity would be performed on the resulting food products as a part of the core protein characterization studies for food safety assessment. Waters et al. (2021) proposed that additional studies should be required only when a food safety hazard is identified in the first set of core molecular and protein characterization studies (Brune et al., 2021) and that problem formulation should be employed to design any hypothesis-driven supplementary studies to be performed to characterize hazard and exposure only when a hazard is identified.

The committee considered a conceptual approach to testing the safety of foods derived from genetically engineered (GE) crops (NASEM, 2016) for its application to HGM animal lines. Within this approach, the triggers for requiring regulatory testing are the physical and biological characteristics of new crop varieties and foods, which can be measured using a variety of modern molecular technologies. That is, the -omics characteristics for a new plant variety would be evaluated in the context of appropriate comparators. This type of assessment results in the GE crop being classified into one of four categories based upon an -omics-wide comparison to an appropriate non-GE crop. New plant lines assigned to categories 1 (showing no differences) and 2 (showing understood differences with no expected health effects) would be exempt from safety testing, whereas those in categories 3 (understood differences with potential for health effects) and 4 (differences that cannot be interpreted) would be tested. Showing application of metabolomic analyses, Drapal et al. (2023) employed high-resolution mass spectrophotometry techniques to compare the metabolomes of genome-edited and non-edited tobacco and tomato lines, in this case showing no significant differences among the metabolomes of edited and non-edited lines.

While multi-omics approaches have been used to great benefit to understand differences between samples, these methods should be applied with due consideration to the expected effects of the genome editing and should be hypothesis-driven rather than using an open-ended search strategy. Likewise, due consideration should be given to the difference between a statistically significant change and a biologically relevant effect, with comparators used to determine the current ranges measured from foods already present in the human food supply (Chapter 4). Consider, for example, the limited use of -omics-type data in the context of the AquaBounty salmon, in which an aspect of the proteome was used to determine allergenic potential (Chapter 4). Interpretation of those results was difficult because “normal,” or even the range of normal, was not known. There are no reported examples in which the use of the entire transcriptome, proteome, or metabolome was sufficient to support prediction of whether an animal’s phenotype is normal.

APPROACHES TO ADDRESS ASSESSMENT OF ANIMAL WELFARE

As noted in Chapter 3, HGMs can affect food-animal welfare. The concerns at issue are not different in kind from those posed by production of conventionally bred food animals. Federal laws exist with the intention to protect food-animal welfare (Box 5-3). Moreover, in response to consumer concerns about the treatment of food animals, most livestock, poultry, and aquaculture production industries in the United States have developed and implemented certification programs with science-based animal care guidelines (Box 5-4).

Voluntary third-party welfare audits were started to verify claims that animals are being raised according to welfare guidelines. Third-party audits and assessments are completed by independent auditors, who have no stake in, or are impartial to, the farm they are auditing. Third-party audits can ensure that a certification program’s auditors remain consistent and accurate with their inspections as well as provide assurance that the program’s welfare guidelines retain high standards. If a farmer or rancher is not part of a certification program, they may still hire a third-party auditing company to perform a welfare assessment of their farm or operation. Examples of third-party auditing companies are shown in Box 5-3. Existing animal welfare guidelines and certification and auditing programs are appropriate for HGM food animals.

Additionally, quality assurance (QA) programs currently offered in the United States provide nationally coordinated and state-implemented certification programs for those handling livestock. These programs include animal handling and welfare practices designed to result in production of safe and high-quality meat products. Industries offering these programs in the form of certification include but are not limited to the National Cattleman’s Beef Association (Beef Quality Assurance), the National Pork Council (Pork QA) and the American Sheep Industry Association (Sheep Safety and Quality Assurance). Quality assurance programs also assist the food industry in conveying positive messages to consumers about where their meat comes from and how animals were handled. For example, in 2019, Wendy’s and Tyson Foods first required their supplier beef producers and haulers (drivers transporting beef cattle) to be Beef Quality Assurance-certified. Other fast-food chains soon added animal handling and welfare requirements. These actions led to additional certification in Transportation Quality Assurance for food-animal transport.

The multi-stage process for developing HGM food animals (Figure 4-1) could easily incorporate existing welfare regulations, voluntary third-party welfare assessment programs (with independent auditors), and industry certification programs that include training in welfare oversight. Existing laws and programs can be applied to ensure that the welfare of food animals is maintained at a high level irrespective of whether they are derived from conventional breeding programs or as a result of gene transfer or genome editing.

APPROACHES FOR IDENTIFICATION AND ANALYSIS OF CHANGES IN FOOD-ANIMAL PRODUCTS

For HGM animals produced as sources of food, safety assessments must consider whether the modifications are likely to result in altered human food products, including compositional changes or potential for introduced hazards. This consideration is incorporated in risk mitigation Level 5 (Chapter 4). Humans have been consuming animal-derived foods for at least 2.4 million years, and these foods are particularly important as complete protein sources, supplying all essential amino acids needed for human survival, along with key vitamins and minerals.

Composition

Methods for characterizing the composition of foods are well established (Chapter 3). However, with scientific and technical advancements, updates to approaches for assessing the safety of food products from HGM animals may be warranted. As noted above, Brune et al. (2021) proposed that compositional assessment of HGM organisms follow a stepwise approach to determine what, if any, compositional data generation is necessary. If HGMs are designed to affect or regulate biochemical pathways and cascades, hypothesis-driven compositional studies looking at affected pathways might well be warranted, as has been suggested for crops (Herman et al., 2009). The list of compositional analyses that pose a potential human food safety hazard would be based on a hypothesis generated by the nature of the trait being introduced (Waters et al., 2021). Experiments then can be designed to measure the appropriate key nutrients to test that specific hypothesis.

Food Allergies and Intolerances

As noted in Chapter 3, food allergies affect 3-4 percent of the adult population of the United States. Nine foods, four of which—milk, eggs, fish, and shellfish—are derived from animals, commonly elicit allergic response through type I IgE-mediated hypersensitivity. Assessing the potential allergenicity of foods, including those derived from HGM animals, is not a simple problem, because allergens may arise from complex potential hazards (proteins in foods) interacting with a complex target (the human immune system). The immune response will vary across individuals and within a single individual’s lifespan, and may be subject to changes in food composition that occur following processing and food preparation. Moreover, most allergen tests focus on elicitation of an IgE response, not on sensitization, and allergenic proteins tend to be very abundant (e.g., seed storage and major milk components), so altered gene expression may play a role in food allergy responses. Based upon the Codex Alimentarius (FAO and WHO, 2008), the workflow for assessing the risk of allergenicity in HGM animals is to (1) consider the source of the transgene; (2) perform bioinformatic screening against allergenic proteins; (3) perform skin-prick tests; (4) assay resistance to pepsinolysis; and (5) perform serum testing if warranted (Johnson, 2024). However, there are several approaches that would result in improved methods for characterizing the potential allergenicity of foods in general and of foods that are the product of HGM animals in particular. Alternatives to the current sequence comparison procedure include the following: (1) use of multi-feature fusion techniques (e.g., amino acid composition, dipeptide composition, and composition of k-spaced amino acid pairs; (Liu et al., 2023); (2) assessment of primary and tertiary protein structures (e.g., AllerCatPro, Nguyen et al., 2022); (3) assessment of primary protein structure utilizing machine-learning techniques (Nedyalkova et al., 2023); (4) quantifying similarity of proteins to known IgE epitopes (e.g., AlgPred 2.0, Sharma et al., 2021); (5) analysis of primary protein sequence utilizing Restricted Bolzmann Machines (e.g., ALLERDET, Garcia-Moreno and Gutiérrez-Naranjo, 2022); and (6) application of Random Forest approaches including assessment of 29 variables derived from protein sequence and database information (Westerhout et al., 2019). There has been controversy over whether resistance to pepsinolysis is reliable as a predictor of allergenicity. Many food allergens are susceptible to pepsinolysis and many non-allergenic compounds are resistant to it (EFSA GMO Panel, 2021).

HOST-RANGE EXPANSION INTO LIVESTOCK AND HUMAN POPULATIONS

As noted in Chapter 2, gene transfer and genome editing have been applied to produce disease-resistant or disease-resilient food animals, including livestock, chickens, and fishes. While development of such animals provides the opportunity to reduce losses of livestock to disease, it may also pose the hazard of driving pathogen evolution, creating a risk of harm to animals and humans (Chapter 3). One approach to address this risk would be to use gene-editing technology to target multiple genes involved in separate pathways of viral replication (e.g., for viruses, this could mean targeting attachment, uncoating, packaging, and polymerase function), reducing the chance that the pathogen would evolve to overcome a single barrier.

Potential Consequences of HGM for Pathogen Evolution

Genome editing presents a demonstrated opportunity to generate animals with improved pathogen resistance, but with that possibility also comes the risk of creating a pathogen with altered host range or virulence. As with any efforts to reduce the effects of disease (e.g., vaccination, treatment), transgenic or gene-edited animals with resistance or resilience to pathogens may often result in eventual adaptation of the pathogen. Given that viruses and bacteria evolve at a rate many orders of magnitude faster than their host, natural selection can overcome many barriers if the pathogen persists such that it can persistently test new mutations to overcome host barriers to infection and progression of disease.

Approaches for Reducing and Identifying Pathogen Evolution

While pathogen evolution is constant and unremitting, there are several practical approaches that can be employed to reduce the risk that pathogens will evolve to evade host resistance mechanisms. Current approaches for developing disease-resistant or disease-resilient HGM animals have focused on viral pathogens, in part because viruses predominantly rely on a host surface molecule for viral entry into the host cell and because viruses cannot replicate outside of a host cell. However, non-receptor-mediated viral entry has been demonstrated as a (minor) method for host cell entry for some viruses (Chu and Whittaker, 2004; Londrigan et al., 2011; de Vries et al., 2012). In the case of HGM food animals that are expected to be virus-resistant, this can be tested by confirming resistance against newly isolated field isolates either in vitro (using resistant cell lines or organoids, if available) or in vivo (by testing on whole HGM animals and performing virus isolation and/or serology assays). Further approaches for surveillance within the HGM population might include post-slaughter testing for clinical signs of disease or seroconversion. Since disease-resilient animals are expected to have reduced viral loads and possible reduced transmission, standard, existing tests for viral replication, shedding, and transmission can be combined with serology and pathology testing of the HGM lines to confirm the continued disease-resistance efficacy of these animals, particularly when challenged with new field strains.

Virulence is a complex trait, and several virulence factors may act alone or orchestrate with one another to drive pathogenesis. For example, in Marek’s disease virus, polymorphisms in the oncoprotein Meq have a substantial impact on the evolution of the virus toward greater virulence (Conradie et al., 2020). Moreover, pathogen virulence and pathogen transmission have a complex interrelationship that is influenced by the type of pathogen, the way the pathogen is spread (transmission routes), accessibility of susceptible hosts (e.g., disease reservoirs), and host immune defenses and host behaviors (Kennedy, 1995; Lipsitch and Moxon, 1997; Hawley et al., 2023). The critical genes for pathogenesis and transmission, however, vary among pathogens and must be identified and their variants recognized and monitored. In many cases, further research is required to identify and characterize key determinants of disease to improve the determination of which field isolates or variants may pose a threat to human and animal health.

A possible solution to managing the risk that pathogens will evolve to circumvent intervention strategies has been developed and successfully applied in HGM crops (to manage resistance of herbivorous insects to toxins; Zhao et al., 2003; and to promote resistance to disease; Shehryar et al., 2020; Luo et al., 2021) and in treatment of human disease (e.g., human immunodeficiency virus antiviral drug regimens; Thompson et al., 2012). This method, called “stacking,” involves targeting multiple genes or pathways simultaneously, rather than sequentially introducing them as new strains of pathogens emerge. The theory behind this approach is based on how evolution drives pathogens to overcomes the host’s immune-system barriers, that is, via selection upon new random mutations, with the number of chances to create the new mutation being a function of the size of the pathogen reservoir population. Should there be sequential introduction of disease-resistance genes into a food animal, pathogen evolution can occur in a stepwise manner. One or more new mutations may be needed to overcome an edit to a single gene, with approximately 10^-5 probability times the number of pathogens trying to overcome the barrier. As pathogen numbers increase into the millions, the probability approaches 1. Once a host immune barrier is overcome, the pathogen reservoir can greatly expand, increasing the odds that a new mutation will again occur and overcome a second barrier, and so forth. With multiple disease-resistance factors introduced into a food animal, however, all

pathways would have to be overcome simultaneously before the pathogen could expand. With just two disease-resistance factors, the probability decreases by five orders of magnitude (the mutation rate), but more importantly the reservoir is kept from multiplying by the action of both factors. As such, the reservoir population size would have to increase into the trillions before a dual mutation would have a sufficiently high probability of occurring to overcome both host immune barriers. The lesson from this is to apply genome edits or disease treatments simultaneously by targeting different biological pathways that are essential for pathogen infection and replication. Hence, to minimize the risk of a pathogen evolving to overcome a particular disease-resistance edit in a food animal, it may be beneficial to create and simultaneously deploy multiple genes with resistance alleles. This can be achieved by introducing multiple distinct edits within the same gene or by targeting multiple different genes involved in various aspects of the virus life cycle. This could potentially provide a more robust defense against pathogen evolution compared to a single-gene, single-edit strategy. However, current knowledge of genetic targets for HGMs to create novel disease resistance is limited, making the application of the stacking strategy currently infeasible. In addition, gene stacking in HGM crops or combinations of treatments of human patients has not always proven successful. While a five-transgene cassette did confer broad-spectrum resistance to a fungal rust pathogen of wheat (Luo et al., 2021), the appearance of a new rust isolate resistant to several of the transgenes suggests that the gene stacks will need to be deployed strategically in the field to maintain their effectiveness. In a human clinical context, the evidence for the utility of antibiotic combination therapy for reducing the development of microbial resistance is scarce (Siedentop et al., 2024).

HGM-based approaches to disease management are not meant to be stand-alone, but rather to be combined with best-practice biosecurity methods to reduce the occurrence of infectious disease in animal production settings (Stull et al., 2018) and with surveillance of food animals, wildlife, and their vectors. Mathematical models might prove useful for exploring combinations of means for managing disease risk, as was done for modeling deployment of pathogen-resistant HGM crops (Rimbaud et al., 2021). Noting the commercialization of HGM crops resistant to the herbicides 2,4-D and Dicamba, Gould et al. (2018) suggested that landscape-level experiments could be used to evaluate the utility of different combinations of herbicide application, crop rotation, and production practices. In the context of HGM animals, it will be necessary to evaluate the utility of disease-resistant HGM animals in the face of evolving pathogen pressure.

SURVEILLANCE

Pathogen surveillance in animals is the responsibility of multiple government agencies and departments, including the USDA, the Department of Health and Human Services, the U.S. Geological Survey, the U.S. Fish and Wildlife Service, the National Park Service, and state departments of agriculture. In addition, since pathogens are not known to respect national borders and animals, and animal-derived products are important in global trade, international surveillance for disease and pathogens is also crucial. Internationally, pathogens that affect international trade, including many important zoonoses, are reported to the World Organization for Animal Health by its member countries. The organization’s Animal Health Information System includes an International Early Warning System through which member countries have agreed to immediately report occurrences of six categories of animal disease. An International Monitoring System reports the absence, presence, or changes in status of key diseases every six months, and additional information is reported annually.

The quality of surveillance varies greatly among countries and typically does not include wildlife species (Kuiken et al., 2005). For several reasons, the current system has been criticized as not providing a sufficient level of vigilance (Kuiken et al., 2005). First, pathogen surveillance in domestic animals is generally confined to pathogens with known economic impacts. Second, although there are multiple examples of zoonotic pathogens emerging from wildlife, the amount of effort and resources dedicated to surveilling wildlife is considerably lower than the amount dedicated to surveilling domestic animals or livestock (or nonexistent) and these efforts are often passive in nature. Although some programs for wildlife surveillance exist, the number of species and geographical areas covered are small compared with the many wild animal species and their distributions. Most importantly, there is lack of integration among pathogen surveillance systems for humans, domestic animals, and wildlife. The separation of the mandates of organizations charged with human and animal health management discourages integration of their operations.

Recent examples in which pathogen surveillance in animals or its integration with public health has faltered include experiences with severe acute respiratory syndrome and high-pathogenicity avian influenza. As noted in Chapter 6, there is a need for integrating and harmonizing disease and population monitoring tools for wildlife, an approach termed integrated wildlife monitoring. Integrated wildlife monitoring should include passive disease surveillance to improve the likelihood of early detection of emerging diseases, with active surveillance and population monitoring to assess epidemiological dynamics, disease-free states, and the outcome of interventions (Cardoso et al., 2022). Cardoso et al. (2022) recommended that health-related public agencies cooperate and combine resources following the One Health approach.

Elucidating the drivers of pathogen adaptation from animals to humans is an active field of research (Johnson et al., 2020; Drew et al., 2021; Chapman and Chi, 2022; Larsson and Flach, 2022; Maher et al., 2022; Tan et al., 2024). As discussed in Chapter 6, it will be important to conduct research focused on the risk of pathogen evolution within HGM food animals and expansion into other animal and human populations. Systems of rapid disease detection produce data that can be utilized in decision support systems to predict when and where disease is likely to emerge; sources of data to support predictive models include Internet-based and environmental data (Astill et al., 2018). For example, a decision support framework developed for prediction of avian influenza outbreaks (Yousefinaghani et al., 2021) can support health care authorities by providing an opportunity for early control of emergency situations. Big data and big data analytics will be required to integrate voluminous and variable data into predictive models that function in near real-time.

CHALLENGES TO THE STUDY OF RISKS OF HGM FOOD ANIMALS AND APPROACHES TO ADDRESSING CHALLENGES IN STUDY DESIGN

Assurance of safe food production relies on numerous interventions to prevent foodborne illnesses caused by consumption of pathogens, foreign objects, and contaminants. Current practices required by the U.S. Department of Agriculture Food Safety and Inspection Service (USDA-FSIS) reflect the process initiated by the “War on Pathogens” during the early 1990s (Murano et al., 2018). In the United States, implementation of Hazard Analysis and Critical Control Points (HACCP) plans for risk mitigation was mandated by USDA-FSIS (1996) and currently applies to all federally inspected meat and poultry harvest and processing plants. The U.S. Food and Drug Administration (FDA), through the Code of Federal Regulations Title 21, oversees safety of milk, milk products, shell eggs, and other meats. FDA’s guidance was updated in the Food Safety Modernization Act of 2011 (FDA, 2017). The use of HACCP plans, coupled with inspections of animals and facilities, additional state regulations, and periodic testing of hazard reduction efficacy, help ensure a safe U.S. food supply.

Any HGM animals entering the food supply would be subject to established federal requirements, in addition to the FDA assessment of the likelihood that an allergen or hazard has been introduced. With rare exceptions, any likelihood of harm from food products derived from approved animals is expected to approach zero. At present, after two decades of human consumption of two types of proteins in genetically modified plants, no evidence exists for newly introduced allergens or sensitivities (Ladics, 2019). Similarly, an unintended modification of genes associated with allergenic potential or adverse sensitivity in an HGM animal would be identified during the developmental process prior to entry into the food chain and reaching consumers. While adverse effects are not expected from food derived from HGM animals approved to enter the food chain, any potential negative effects could potentially be reported in a system such as the FDA Adverse Event Reporting System (FAERS, 2023) as an additional safety measure.

The study of risks posed by HGM food animals faces several challenges. First, enormous diversity exists across animal breeds, strains, and lines. While improvements have been made in the accuracy and efficiency of methods for making targeted edits (Chen et al., 2024; Klermund et al., 2024; Singh et al., 2024) and in identifying unintended alterations, more research is required to clearly distinguish between mutational changes derived from meiotic reproduction and natural variation (including spontaneous mutation) versus intended alterations. As noted above, while high-quality reference genomes are available for nearly all vertebrate food animals, the process of genome annotation is incomplete, and understanding the effects of both targeted alterations and natural variation on heritable phenotypes is a field of study that is continually changing and updating. These challenges highlight

areas where clear risk assessment and food safety assessment procedures are either absent or lacking sufficient detail (Chapters 3 and 4). The following sections address these areas in more detail.

Assessing Significance in Experimental Design

A point raised during the AquAdvantage salmon food safety risk assessment (FDA-CVM-VMAC, 2010) was that not enough samples were utilized to demonstrate the statistical significance of differences in composition of transgenic versus conventional fish products. This criticism raises the issue of what phenotypic difference between HGM and non-HGM animals is biologically meaningful (significant in terms of composition, nutrition, etc.). Biological significance should not be confused with statistical significance. Statistical significance measures the probability of the null hypothesis being true compared to the acceptable level of uncertainty regarding the true answer. The null hypothesis represents no change, or that the observed effect is due to chance alone. In contrast, biological significance refers to an effect that has a noteworthy impact on the biological system. The assessment of biological significance is determined by the risk/benefit ratio, which can be set at different levels based upon the situation and personal tolerance for the potential risk. Therefore, any experiments to determine differences between conventional and HGM food products must start with a clearly defined biological difference and then show that the actual difference is less than would be considered biologically important (i.e., statistically significant at a given type-1 error rate). This approach requires that researchers and regulators have a clearly defined standard of what difference is biologically important. Only then can scientists determine the appropriate statistical power of a proposed study and the sample size required.

Foods derived from animals have been consumed since before recorded history, and it is only relatively recently that the nutritional composition of food products has been determined analytically. Despite being able to determine the nutritional composition of food at a molecular level, animal products are not defined by nutritional guidelines, but rather by their source (e.g., the cut of meat, the breed). Consumers and food distributors may distinguish between species (e.g., Chinook, coho, or sockeye salmon), production methods (e.g., farmed or wild-caught salmon), or even regional products (e.g., Pacific or Atlantic salmon); likewise, consumers may have further preferences based upon the intended mode of preparation, taste, or nutritional qualities such as the level of omega-3 fatty acids. However, the differences between these products are not measured, and there are no nutritional standards for them. It is expected that even closely related animal products (e.g., the same cut of meat from the same breed of cattle) would have a broad range of variation, and this nutritional composition would be further altered by the food preparation method used.

Another approach for assessing biological changes to food products is sensory testing using consumer (untrained) and analytical (trained) panelists. Sensory evaluation of foods involves scientific methods for testing the appearance, texture, aroma, and taste of a product. It is likely that sensory testing will be completed as part of any typical marketing strategy for new animal-derived products that may be developed from HGM animals. However, since sensory testing focuses on perception of food quality and likability, rather than biological changes that influence those traits, it is not relevant to assessing risk.

HGM animal products for which it can be reasonably assumed that the alteration would impact the consumer food product (e.g., meat cuts, milk) should fall within the expected range of nutritional composition as the genetically unaltered products. Assessing biological differences for food safety assessment should focus on clearly defined phenotypes that have measurable traits and a body of literature to support the determination of expected variance and ranges. Experimental designs should clearly elaborate both the statistical significance and how that would relate to biologically relevant change in the food products. An exception to this would be biofortified or nutritionally enhanced products, which would need to meet current standards for demonstrating the specified nutritional enrichment. The National Academies of Sciences, Engineering, and Medicine (NASEM, 2016) report regarding genetically engineered crops noted that before a test is conducted, it is important to justify the size of a difference between treatments in each measurement that will be considered biologically relevant; this stipulation is also relevant to the context of HGM animals.

Challenges with Current Models for Testing Food Safety

As noted in Chapter 4, the U.S. food supply is very safe. Multiple testing and intervention steps result in few contaminants or toxins reaching consumers, and a rapid and robust food recall system reduces the risks to consumers when issues occur. Current models for ensuring food safety and testing the effectiveness of interventions focus on identification of microbial (biological), chemical, and physical hazards. In addition, animal models are used to test for allergenicity and to determine allergenicity endpoints (Dunn et al., 2017; Ladics, 2019). Human food allergens are well documented and models for testing allergenicity are also well established. However, one potential challenge with food allergen testing is necessarily short term, testing immunoglobulin E (IgE) antibodies produced in response to a food allergen. This type of testing is problematic for assessing food intolerances and for studying longer-term effects of food intolerances. Compared to food allergies, which are estimated to affect 4-6 percent of the U.S. population, the impact of food intolerances may be experienced by as much as 20 percent of the U.S. population (Tuck et al., 2019). Moreover, food intolerances can develop or increase over time with changes in digestion (e.g., due to microbial imbalances in the small intestine or decreases in digestive enzymes as people age), stress, anxiety, and medications. Relatively little is understood about the long-term impact of food intolerances, and studies are confounded by self-diagnosis based on misleading health advice, people incorrectly attributing medical symptoms to food, and the wide range of symptoms. Without improved methods for assessing the long-term impact of food intolerance, it is difficult to determine the impact of changed nutritional composition due to HGM of animal food products. However, food intolerances are currently managed by labeling practices and by changes to the diet.

The potential integration of antibiotic-resistance genes into the genomes of food animals has been suggested as a hazard to animal and public health (Chapter 3); however, newer HGM technologies have moved away from the use of plasmids with antibiotic-resistance genes. Moreover, utilization of appropriate molecular screening methods enables detection of plasmid integration and integration of multiple template copies (Norris et al., 2020). As noted above, the alignment of sequencing data should include both the reference genome and plasmid DNA sequences. PCR genotyping should incorporate plasmid-specific primers. Methods used to detect increased numbers of copies of the template and unintended integration of the template plasmid would include long-range PCR conditions, quantitative PCR (e.g., digital droplet PCR), Southern blot, and long-read sequencing (e.g., using Oxford Nanopore or PacBio technology).

APPROACHES TO RISK ASSESSMENT IN THE FACE OF UNCERTAINTY OF THE LIKELIHOOD AND MAGNITUDE OF HEALTH HAZARDS

Assessing allergenicity Risk for Foods Derived from Genetically Modified Food Animals

Among the potential hazards of greatest concern for food products derived from HGM animals is the unintended increase in a food intolerance or allergenic response in the consumer. Food allergens are well characterized, and food products must be clearly labeled to alert consumers of potential allergens. Since these allergens are well characterized, food products from HGM animals can be tested for allergenic potential. Determining the potential for inadvertently increasing allergenic potential should be based upon consideration of the source animal and the intended modification of that animal. The committee distinguishes four scenarios:

1. Alteration to a gene expressed in a tissue that is not typically consumed where the alteration could arise though conventional breeding. Examples include edits to create polled cattle or NANOS2 gene knockout models to develop surrogate recipients for the spermatogonial stem cells. The risk of consuming food products from these animals is minimal.
2. Alteration to a gene expressed in a tissue that is consumed where the alteration could arise though conventional breeding. Examples include knocking out myostatin expression to produce more muscle mass or alterations for improved disease resistance. The risk from consuming food products from these animals is minimal.

3. Alteration to a gene expressed in a tissue that is not typically consumed where the alteration could not arise though conventional breeding. Examples include boars with a nonfunctional copy of the kisspeptin receptor to eliminate boar taint (Flórez et al., 2023) or expressing bacterial phytase in pigs to reduce phosphorous excretion (Golovan et al., 2001). The risk of consuming food products from these animals is minimal.
4. Alteration to a gene expressed in a tissue that is consumed where the alteration could not arise though conventional breeding. Examples include Atlantic salmon that has a growth hormone gene from Chinook salmon or transgenic cattle that express human enzymes in milk. In this case, the risk of consuming food products from these animals could need further testing to determine whether it is reasonable to test for known allergens.

Risk assessment for foods derived from genetically modified animals should focus on assessing changes to known allergens or triggers for intolerance. Food intolerances, also known as food sensitivities, tend not to be life-threatening, but should also be considered. There is a need for studies to develop models for understanding the long-term effects of food sensitivities.

FUTURE STUDIES TO SUPPORT RISK ASSESSMENT OF LIKELIHOOD AND MAGNITUDE OF HEALTH RISKS (HARMS GIVEN EXPOSURE TO THE HAZARD)

Comparisons of the genomes of HGM food animals with their non-HGM counterparts have shown that most of the genetic differences are due to natural mutations occurring during reproduction rather than to unintended edits or insertions. Complicating the assessment of genetic changes is the fact that HGM food animals are not produced directly, but rather from genome-edited founders that are unlikely to end up in a food production setting. This has two implications: first, the altered animal must survive embryonic development and produce viable animals that are at least as productive as their non-HGM counterparts in order to progress through the process of line development (Figure 4-1); and second, any determination of changes to the germline should assess which genetic changes are inherited by the derived production populations.

Understanding the biological relevance of HGMs also will inform risk assessment of derived food animals. Determining the biological relevance of any genetic changes needs to be based upon phenotypes, as populations have genetic variance without detrimental effects. Related to this is the challenge of understanding the contribution of heritable variation to complex phenotypes and the challenge of understanding how functional genome elements (including regulatory elements) interact to produce complex phenotypes. Mandatory use of HACCP plans (USDA-FSIS, 1996; FDA, 2022) and inspections (of animals and facilities), along with periodic testing of hazard reduction efficacy, has resulted in a safe U.S. food production system. Therefore, evaluation of the differences between conventional and HGM animal-derived food products must start with a clearly defined sense of what would comprise biological significance, followed by evaluation of evidence of whether that difference has been observed. This approach requires that researchers and regulators agree upon a clearly defined standard of what differences are biologically important.

Efforts to prove that the nutritive composition of an HGM-derived animal is not different from its non-HGM-derived equivalent may not be informative and may be difficult and burdensome in the absence of data on the range of composition of food currently being consumed. The defined normal range of nutrient composition of animal-derived foods is generally lacking, may be based on a small number of unrelated samples, or may vary so much as to be unhelpful. There is a fundamental lack of information regarding the full range of nutrient compositions for animal-sourced food products and how these values are altered by natural variation and environment. Efforts focusing on assessing food composition run into the challenge of determining what changes would be biologically relevant, particularly since consumers are offered many different food products and the human diet is complex. Instead, FDA focuses on determining that the product from edited animals is effective (i.e., that GalSafe pig products have reduced alpha-gal sugar content; FDA, 2020).

KEY FINDINGS

1. The characterization of inherited changes to a DNA sequence is an important early step in the process of developing HGM animals. However, risk assessment should focus on phenotypes because DNA alterations can arise from genome editing (through either intended or unintended alterations) or from naturally occurring mutations, and it is not currently possible to definitively differentiate between these two processes. Risk assessments should consider the comparator used to determine changes to the genome, whether these changes are identified in production or in pre-commercialization HGM animals, and the effect (if any) on the phenotype of the HGM animals. Careful consideration should be given to the assessment of phenotype to determine biologically relevant, rather than statistically significant, differences.
2. The implications of natural genetic variation for functional outcomes and animal phenotypes are poorly understood, which is a gap in current knowledge that hinders assessment of potential harms related to HGM food animals. Research is needed to enhance the ability to distinguish between mutational changes that are derived from spontaneous mutation and meiotic processes that give rise to natural variation from changes that are due to biotechnological manipulations.
3. There is a crucial lack of fundamental information about normal variation in food composition, with no single resource to access this information (and no imperative to make this information public) and a lack of agreed-upon best practices and standardizations for such data.
4. Current approaches for determining allergenicity require updating to incorporate new approaches and advances in understanding of allergens. Bioinformatic screening against allergenic proteins should be revised to include multi-feature approaches rather than simply sequence comparisons, including comparison of protein structures and the development and application of artificial intelligence approaches (including machine learning and deep learning). Recent studies have indicated that the assay used to determine resistance to pepsinolysis is not a reliable predictor of allergenicity, suggesting that the use of this test should be reconsidered. Testing IgE antibodies produced in response to a food allergen measures short-term allergenic responses and does not address food intolerances, which can develop or increase over time with changes in digestion. There is a lack of methods for assessing the long-term impact of food intolerance and for determining the impact of food intolerance among the many confounding factors related to this subject.
5. Existing welfare laws apply equally to HGM and non-HGM animals and the text and intent of these laws and HACCP planning for food production should be applied to HGM animals. Any HGM disease-resistant animals should be periodically tested against newly isolated field isolates (using in vitro or in vivo testing) to ensure continued efficacy, and their production should be combined with best-practice biosecurity measures. Surveillance within an HGM food-animal population is relatively straightforward and could include post-slaughter testing for disease or seroconversion of HGM animals.

REFERENCES