CRISPR-Cas9 human gene replacement and phenomic characterization in Caenorhabditis elegans to understand the functional conservation of human genes and decipher variants of uncertain significance

The rapid development and application of whole-exome and whole-genome sequencing technology has dramatically increased the pace at which we associate genetic variation with a particular disease (Auton et al., 2015; Bamshad et al., 2011; Gonzaga-Jauregui et al., 2012; Lek et al., 2016; Metzker, 2010; Need et al., 2012; Ng et al., 2010). However, our ability to sequence genomes has vastly surpassed our ability to interpret the clinical implications of the genetic variants we discover. The majority of genetic variants identified in clinical populations are currently classified as ‘variants of uncertain significance’, meaning that their potential role as a causative agent in the disease in question, or their pathogenicity, is unknown (Richards et al., 2015). Many variants are exceedingly rare, making it extremely difficult to designate them as pathogenic using classical genetic methods such as segregation within a pedigree, or by identifying multiple carriers of the variant. As such, it often remains challenging to predict clinical outcomes and make informed treatment decisions based on genetic data alone.

In an attempt to address this problem, several computational tools have been developed that estimate the functional consequences and pathogenicity of disease-associated variants (Richards et al., 2015). These tools use a variety of predictive features, such as evolutionary sequence conservation, protein structural and functional information, the prevalence of a variant in large putatively healthy control populations or a combination of annotations (Kircher et al., 2014; Lek et al., 2016; Richards et al., 2015). Despite extensive efforts, none of these tools used in isolation or combination can faithfully report on the functional effects of a large portion of disease-associated variation, and their accuracy is intrinsically limited to existing experimental training data (Grimm et al., 2015; Miosge et al., 2015; Starita et al., 2017). These limitations were clearly demonstrated in a recent study that showed that in vivo functional assays of 21 human genes in yeast identified pathogenic variants with significantly higher precision and specificity than computational methods (Sun et al., 2016). This means that, even for genes with well-characterized biological functions, there are often hundreds of variants of uncertain functional significance (Landrum et al., 2014; Starita et al., 2017). This creates a challenging situation that requires direct assessment of the functional effects of disease-associated variants in vivo (Starita et al., 2017).

Genetically tractable model organisms are crucial for discovering novel gene functions and the functional consequences of disease-associated genetic variants (Dunham and Fowler, 2013; Lehner, 2013; Manolio et al., 2017; Wangler et al., 2017). Governmental and private funding agencies are increasingly commissioning large-scale collaborative programs to use diverse genetic model organisms to decipher variants of uncertain significance (Chong et al., 2015; Gahl et al., 2016; Wangler et al., 2017). Among genetic model organisms, the nematode Caenorhabditis elegans has proven to be a particularly powerful animal model for the functional characterization of human genes in vivo (Kaletta and Hengartner, 2006). C. elegans is an ideal genetic model as it combines the throughput and tractability of a single-celled organism with the complex morphology and behavioral repertoire of a multi-cellular animal. In addition, ∼60-80% of human genes have an ortholog in the C. elegans genome (Kaletta and Hengartner, 2006; Lai et al., 2000; Shaye and Greenwald, 2011). Phenotypic analyses of transgenically expressed human genes are routinely done to confirm functional conservation and to observe the effects of disease-associated mutations. Notable examples of the utility of C. elegans to determine conserved human gene functions relevant to disease include the identification of presenilins as part of the gamma secretase complex, the mechanism of action of selective serotonin reuptake inhibitors, and the role of the insulin signaling pathway in normal and pathological aging (Kaletta and Hengartner, 2006; Levitan and Greenwald, 1995; Levitan et al., 1996; Murphy et al., 2003; Ranganathan et al., 2001). However, traditional methods for expression of human genes in C. elegans rely on mosaic and variable overexpression of transgenes harbored as extrachromosomal arrays or specialized genetic backgrounds that can confound phenotypic analysis. This presents several challenges that inhibit precise analysis of the often critical, but subtle, effects of missense variants and impede the use of these transgenic strains in large-scale drug screens.

The recent advent of CRISPR-Cas9-mediated genome editing has revolutionized structure-function analyses across model organisms (Cong et al., 2013; DiCarlo et al., 2013; Dickinson et al., 2013; Doudna and Charpentier, 2014; Friedland et al., 2013; Gratz et al., 2013; Hwang et al., 2013; Jinek et al., 2012, 2013; Li et al., 2013). This system uses a single-guide RNA (sgRNA) to precisely target a nuclease (most often Cas9) to induce a DNA double-strand break at a defined location (Doudna and Charpentier, 2014). The double-strand break can then be repaired via the error-prone non-homologous end-joining pathway (often resulting in damaging frameshift mutations) or a homology repair pathway, e.g. homology-directed repair (HDR) or microhomology-mediated end joining (Nakade et al., 2014). Following a double-strand break, exogenous DNA repair templates can be used as substrates for HDR, allowing virtually any desired sequence to be inserted anywhere in the genome. Importantly, CRISPR-Cas9 genome engineering is remarkably efficient and robust in C. elegans (Dickinson and Goldstein, 2016; Norris et al., 2015).

Here, we present a broadly applicable strategy that adapts CRISPR-Cas9 genome engineering for targeted replacement of C. elegans genes with human genes. We illustrate how the library of knockout and humanized transgenics generated with this approach can be efficiently combined with automated machine vision phenotyping to rapidly discover novel gene functions, and assess the functional conservation of human genes, and how this will allow for analysis of the effects of variants of uncertain significance with unprecedented precision. It is our hope that the human gene replacement and phenomic characterization strategy delineated in this article will serve both basic and health researchers alike, by serving as an open and shareable resource that will aid any genome engineer interested in understanding the functional conservation of human genes, and the functional consequences of their variants.

To replace the open reading frame (ORF) of an orthologous gene with a human gene, our strategy first directs an sgRNA to induce a Cas9-mediated DNA double-strand break immediately downstream of the ortholog start codon (Fig. 1A). A co-injected repair template containing ∼500 bp homology arms targeted to the regions immediately upstream and downstream of the ortholog ORF serves as a substrate for HDR. By fusing the coding DNA sequence (CDS) of a human gene of interest to the upstream homology arm, HDR integrates the human gene into the ortholog at a single copy in-frame (Fig. 1A).

To streamline genome editing we have based our method on a recently described dual-marker-selection (DMS) cassette screening protocol (Norris et al., 2015). The DMS cassette consists of an antibiotic resistance gene (Prps-27::neoR::unc-54 UTR) and a fluorescent marker (Pmyo-2::GFP::unc-54 UTR) that greatly facilitates the identification of transgenic animals (Norris et al., 2015). We chose this cassette over similar methods as it can be used in any wild-type or mutant strain amenable to transgenesis and does not require any specialized genetic backgrounds, and avoids the use of morphology and/or behavior altering selection markers that necessitate cassette excision prior to phenotypic analysis (Bend et al., 2016; Dickinson et al., 2013, 2015). In our strategy, the DMS cassette is placed between the human gene of interest and the downstream homology arm (Fig. 1A). This deletes the entire ORF of the ortholog and separates the human gene from the ortholog’s transcriptional terminator upon initial integration. In many cases, this efficiently creates a useful deletion allele of the ortholog with no human gene expression (this can be confirmed via inclusion of an epitope tag if immunological reagents are unavailable). In the unlikely event that human gene expression does occur without a transcription terminator, a second repair template using the same validated homology arms and sgRNA – and no human gene – can be integrated to create an ortholog null. A deletion allele can also be ordered from the Caenorhabditis Genetics Center where available. Importantly, the DMS cassette is flanked by two LoxP sites housed within synthetic introns that allow subsequent excision of the selection cassette via transient expression of Cre recombinase (Norris et al., 2015). DMS cassette excision connects the human gene to the endogenous C. elegans ortholog’s transcriptional termination sequence, such that a single copy of the human gene will now be expressed under the control of all of the ortholog’s 5′ and 3′ cis- and trans-regulatory machinery (Fig. 1B). Validation of the desired edit is performed using standard PCR amplification and Sanger sequencing of both the 5′ and 3′ junctions of the target region (Fig. 1A; Au et al., 2018).

For structure-function analysis of a human gene variant of uncertain significance, the variant of interest is incorporated into the HDR plasmid using standard in vitro methods such as site-directed mutagenesis, and the same genome editing process is repeated using the same validated sgRNA and homology arms. This process allows for (1) the initial generation and phenotypic analysis of a complete deletion allele in the C. elegans orthologous gene; (2) direct integration of the human gene to determine whether the human gene can compensate for loss of the orthologous gene, measuring functional conservation; and (3) structure-function analysis of the effects of variants of uncertain significance on wild-type gene function (Fig. 1B). Importantly, the vast majority of variants of uncertain significance identified in patients are heterozygous, and the integrated nature of the transgenes generated with this strategy allows for straightforward assessment of heterozygous alleles using standard genetic crosses (Fig. 1C).

As a proof of principle, we demonstrated the utility of our strategy by focusing on the critical disease-associated gene phosphatase and tensin homolog (PTEN). PTEN is a lipid phosphatase that antagonizes the phosphoinositide 3-kinase (PI3K) signaling pathway by dephosphorylating phosphatidylinositol (3,4,5)-trisphosphate (PIP3) (Li et al., 1997; Maehama and Dixon, 1998). Heterozygous germline PTEN mutations are associated with diverse clinical outcomes, including several tumor predisposition phenotypes (collectively called PTEN harmatoma tumor syndrome), intellectual disability and autism spectrum disorders (Hobert et al., 2014; Li et al., 1997; Liaw et al., 1997; McBride et al., 2010; O'Roak et al., 2012; Orrico et al., 2009; Sanders et al., 2015; Varga et al., 2009). Despite extensive study, it is currently impossible to predict the clinical outcome of a PTEN mutation carrier using sequence data alone (Matreyek et al., 2018; Mighell et al., 2018). PTEN also has several technical advantages that make it an ideal test case: (1) PTEN functions in the highly conserved insulin signaling pathway that is well characterized in C. elegans (Ozes et al., 2001; Ogg and Ruvkun, 1998; Mihaylova et al., 1999; Fig. 2B); (2) C. elegans has a single PTEN ortholog called daf-18 (Fig. 2A-D), and transgenic overexpression of human PTEN using extrachromosomal arrays has been shown to rescue reduced longevity and dauer-defective phenotypes induced by mutations in daf-18 (Liu and Chin-Sang, 2015; Solari et al., 2005); and (3) C. elegans harboring homozygous daf-18 null alleles are viable and display superficially normal morphology and spontaneous locomotor behavior (Mihaylova et al., 1999; Fig. S1).

Structure-function analyses of PTEN variants necessitate an in vivo phenotypic assay that reports on PTEN function. When wild-type worms are grown in the presence of NaCl and their food source (Escherichia coli), they display naïve attractive navigation behavior toward NaCl. This behavior is called NaCl chemotaxis and can be quantified by measuring the navigation behavior of a population of animals up a controlled NaCl concentration gradient (Ward, 1973). Previous work has shown that daf-18 is required for attractive navigation behavior up a concentration gradient of NaCl, such that animals with deletion or reduction-of-function mutations in daf-18 display innate aversion to NaCl (Tomioka et al., 2006). Interestingly, insulin/PI3K signaling normally actively promotes salt avoidance under naïve conditions and daf-18 functions in a single chemosensory neuron (ASER) to antagonize the insulin/PI3K pathway and promote salt attraction (Adachi et al., 2010; Tomioka et al., 2006). Using our machine vision system, the Multi-Worm Tracker, we developed an automated high-throughput NaCl chemotaxis paradigm (Fig. 3A,B) and replicated the finding that daf-18(e1375) reduction-of-function mutants display strong aversion to NaCl (Swierczek et al., 2011; Fig. 3C,D). We then generated a transgenic line using traditional extrachromomosal array technology that directed pan-neuronal expression of wild-type human PTEN using the aex-3 promoter (Kuroyanagi et al., 2010). Pan-neuronal expression of human PTEN is able to rescue the daf-18 reduction-of-function phenotype and restore attractive NaCl chemotaxis to wild-type levels (Fig. 3C,D). This work establishes NaCl chemotaxis as an in vivo behavioral assay of conserved nervous system PTEN functions.

Next, we used our strategy to create a daf-18 deletion allele and single-copy replacement daf-18 with the 1212 bp canonical human PTEN CDS. Complete deletion of the 4723 bp daf-18 ORF resulted in strong aversion to NaCl and chemotaxis down the salt gradient (Fig. 4A). We achieved genome editing efficiency for human gene replacement similar to what has been previously reported for plasmid-based CRISPR-Cas9-mediated DMS cassette integration alone (Au et al., 2018; Norris et al., 2015, 2017). Chemotaxis avoidance of worms harboring the daf-18 complete deletion allele generated using CRISPR-Cas9 was not significantly different from that of worms carrying the previously characterized large deletion allele daf-18(ok480), confirming effective inactivation of daf-18 (Fig. 4B). DMS cassette excision and expression of a single copy of human PTEN was unable to substitute for daf-18 and did not rescue attractive NaCl chemotaxis behavior (Fig. 4C). This result was observed in two independent single-copy human PTEN knock-in lines on several independent experimental replications (Fig. 4C). Transcription of human PTEN was confirmed using reverse transcription PCR in both knock-in lines (Fig. 4D). Sanger sequencing confirmed error-free insertion of transgenes at base-pair resolution before and after cassette excision. There are several potential reasons why expression of human PTEN using extrachromasomal arrays rescued daf-18 mutant phenotypes (Solari et al., 2005; Fig. 3C,D), whereas targeted single-copy replacement with PTEN did not (Fig. 4C). The two most prominent differences between these two technologies are the expression levels of the transgenes and the use of the endogenous 3′ UTR in the CRISPR-Cas9 knock-in versus the unc-54 myosin 3′ UTR used in most C. elegans transgenes, to ensure proper processing of transcripts in all tissues, including all constructs previously shown to rescue daf-18 phenotypes with human PTEN (Merritt and Seydoux, 2010; Solari et al., 2005; Fig. 3C,D).

In order to address these limitations and increase the speed of transgenesis, we created an alternative repair template strategy that includes a transcriptional termination sequence so that expression of the human gene begins immediately upon integration (Fig. 5A). We included the validated unc-54 3′ UTR sequence fused to the 3′ end of human PTEN (Fig. 5A). Expression of wild-type human PTEN using this genome editing strategy significantly rescued NaCl chemotaxis, indicating production of functional PTEN immediately upon genomic integration prior to cassette excision (Fig. 5B). Again, for this editing strategy, we achieved efficiency similar to what has been previously reported for DMS cassette integration alone (Au et al., 2018; Norris et al., 2015, 2017). This alternative approach also offers the added benefits of increased throughput (as a second injection step to excise the cassette was not required for human gene expression as it is in the first strategy) and the option for retained visual transgenic markers (either within the selection cassette or by adding a 2A sequence to drive reporter expression from the same promoter), which simplifies the generation and phenotypic analysis of heterozygotes and double mutants (Fig. 5A; Ahier and Jarriault, 2014; Calarco and Norris, 2018; Norris et al., 2017). Given the demonstrated versatility of CRISPR-Cas9-mediated genome editing in C. elegans, these results suggest that our strategy should be broadly applicable for in vivo analysis of diverse human disease-associated genes.

A long-standing goal has been to understand how human disease-associated variants alter normal gene function to produce sensory and learning abnormalities characteristic of diverse neurogenetic disorders. The massive number variants of uncertain significance recently implicated in the etiology neurogenetic disorders necessitates a dramatic increase in throughput of both transgenic construction and behavioral phenotyping if this goal is to be achieved (Ben-Shalom et al., 2017; Geschwind and Flint, 2015; Lim et al., 2017; Sanders et al., 2015; Starita et al., 2017; Wangler et al., 2017). By combing streamlined human gene integration with rapid machine vision phenotypic analysis of C. elegans, our strategy greatly simplifies the identification of novel conserved gene functions in complex sensory and learning phenotypes.

We explored whether daf-18 mutants displayed behavioral deficits in mechanosensory responding and/or habituation, a conserved form of non-associative learning that is altered in several neurodevelopmental and neuropsychiatric disorders (McDiarmid et al., 2017, 2018; Rankin et al., 2009; Stessman et al., 2017; Timbers et al., 2017; van der Voet et al., 2014). When a non-localized mechanosensory stimulus is delivered to the side of the Petri plate C. elegans are cultured on, they respond by eliciting a brief reversal before resuming forward locomotion. Wild-type C. elegans habituate to repeated stimuli by learning to decrease the probability of eliciting a reversal (Rankin et al., 1990). To determine whether daf-18 is important for mechanoresponding and/or non-associative learning, we examined habituation of the daf-18(e1375) and daf-18 complete deletion mutants. Compared with wild-type animals, both daf-18(e1375) reduction-of-function and daf-18 complete deletion mutants exhibited significantly reduced probability of eliciting a reversal response throughout the habituation training session, indicating mechanosensory hyporesponsivity (Fig. 6A,B). Despite this hyporesponsivity, the plasticity of responses, or the pattern of the gradual decrement in the probability of emitting of a reversal response throughout the training session, was not significantly altered in daf-18 mutants (Fig. 6A,B,E). Importantly, targeted single-copy replacement of daf-18 with human PTEN was sufficient to rescue the mechanosensory hyporesponsivity phenotype across the training session toward wild-type levels (Fig. 6D). These results identify a novel conserved role for PTEN in mechanosensory responding, a fundamental biological process disrupted in diverse genetic disorders (Badr et al., 1987; McDiarmid et al., 2017; Orefice et al., 2016). More broadly, they illustrate how the library of transgenic animals generated with our strategy can be used to rapidly characterize the role of diverse human genes in complex sensory and learning behaviors. These novel phenotypes can then be used to investigate the functional consequences of disease-associated variants in intact and freely behaving animals, and to screen for therapeutics that reverse the effects of a particular patient’s missense variant.

To demonstrate the feasibility of our human gene replacement strategy for assessing variants of uncertain significance, we set out to determine whether our in vivo functional assays could discern known pathogenic variants. Early studies characterizing the role of PTEN as a tumor suppressor suggested that impaired protein phosphatase activity was key to the etiology of PTEN disorders (Tamura et al., 1998). However, this notion was challenged by the identification of cancer patients harboring a missense mutation that changes a glycine residue in the catalytic signature motif to a glutamate, which was predicted to abolish the lipid phosphatase activity of PTEN while leaving the protein phosphatase activity intact (Liaw et al., 1997; Myers et al., 1997). Subsequent biochemical analyses supported the now widely accepted view that the lipid phosphatase activity of PTEN is critical to its tumor suppressor activity (Myers et al., 1998).

We used site-directed mutagenesis to incorporate the PTEN-G129E (PTEN, c386G>A) missense variant into our repair template and used our human gene replacement strategy to replace daf-18 with a single copy of human PTEN-G129E (Fig. 7A). Animals harboring the PTEN-G129E variant displayed strong NaCl avoidance, equivalent to that of animals carrying the complete daf-18 deletion allele, indicating loss of function (Fig. 7B). Similarly, PTEN-G129E mutants also displayed mechanosensory hyporesponsivity that was not significantly different from that of daf-18 deletion carriers (Fig. 7C). These in vivo phenotypic results accurately classify the pathogenic PTEN-G129E as a strong loss-of-function variant. In addition, by taking advantage of a pathogenic variant with well-characterized biochemical effects, these results identify a necessary role for PTEN lipid phosphatase activity in both chemotaxis and mechanosensory responding, providing novel insight into the molecular mechanisms underlying these forms of sensory processing. Taken together, these results demonstrate the potential of human gene replacement and phenomic characterization to rapidly identify the functional consequences variants of uncertain significance.

We have developed and validated a broadly applicable strategy for targeted human gene replacement and phenomic characterization in C. elegans that will facilitate assessment of the functional conservation of human genes and structure-function analysis of variants of uncertain significance with unprecedented precision. We established an automated NaCl chemotaxis paradigm and demonstrate that pan-neuronal overexpression or direct replacement of daf-18 with its human ortholog PTEN using CRISPR-Cas9 is sufficient to rescue reversed NaCl chemotaxis preference induced by complete daf-18 deletion. We further identified a novel mechanosensory hyporesponsive phenotype for daf-18 mutants that could also be rescued by targeted replacement with human PTEN. In vivo characterization of mutants harboring a single copy of the known lipid phosphatase inactive G129E variant accurately classified this variant as pathogenic, and revealed a crucial role for PTEN lipid phosphatase activity in NaCl chemotaxis and mechanosensory responding. We provide novel high-throughput in vivo functional assays for PTEN, as well as validated strains, reagents, sgRNA and repair template constructs to catalyze further analysis of this critical human disease-associated gene. More broadly, we provide a conceptual framework that illustrates how genome engineering and automated machine vision phenotyping can be combined to efficiently generate and characterize a library of knockout and humanized transgenic strains that will allow for straightforward and precise analysis of human genes and disease-associated variants in vivo (Fig. 8).

To date, the most widely used genome editing-based human disease variant assessment strategy in C. elegans uses sequence alignments to identify and engineer the corresponding amino acid change into the orthologous C. elegans gene (Bend et al., 2016; Bulger et al., 2017; Canning et al., 2018; Pierce et al., 2018; Prior et al., 2017; Sorkaç et al., 2016; Troulinaki et al., 2018). A major advantage of this approach is that, by using the C. elegans gene, intronic regulation, protein-protein interactions, subcellular localization and biochemical activity of the protein of interest are, by design, perfectly modeled by C. elegans. Even when expressed at physiologically relevant levels directly from the ortholog’s native loci, and with evidence of phenotypic rescue, it is not guaranteed that a transgenic human protein will recapitulate all functions and interactions of the orthologous C. elegans protein. This presents an important consideration when attempting to replace C. elegans proteins that must interact in extremely precise heteromeric complexes to perform their molecular functions (e.g. certain ion channel subunits) (Bend et al., 2016; Prior et al., 2017). The most obvious limitation of this approach is that it can only be used to study orthologous amino acids. Amino acids that have been conserved throughout evolution are, by definition, the least tolerant to mutation and are thus far more likely to be detrimental to protein function when mutated (Starita et al., 2017; Weile et al., 2017). Indeed, many variant effect prediction algorithms rely on sequence conservation as the main predictor that a variant will be deleterious (Richards et al., 2015; Starita et al., 2017). It is also important to note that a large portion of amino acids will not be conserved to humans (e.g. >50% of amino acids are not conserved from DAF-18 to PTEN, Fig. 2C) and alignment algorithms that identify orthologous amino acids are imperfect. Many current implementations of orthologous amino acid engineering also require constant generation of completely new sgRNAs and repair templates for subsequent edits. Human gene replacement, in contrast, allows any coding variant of uncertain significance to be studied in vivo using the same validated sgRNA and homology arms.

One potential limitation of human gene replacement is that it relies on a C. elegans ortholog to replace. Estimates suggest that there are C. elegans orthologs for 60-80% of human disease-associated genes. This means there will not be an ortholog available for a minority of human genes (Kaletta and Hengartner, 2006; Lai et al., 2000; Shaye and Greenwald, 2011). This is even less likely to be a problem for disease variant modeling, as disease-associated genes are more likely to be highly conserved (Aerts et al., 2006; López-Bigas and Ouzounis, 2004). A recent study also showed that ∼10% of human disease-associated genes are able to functionally substitute for their yeast paralogs (in addition to orthologs), further increasing the number of human genes that can be studied by replacement (Yang et al., 2017). Still, in the event that there is no suitable C. elegans ortholog or paralog available for a human gene of interest, the human gene can simply be integrated at a putatively neutral genomic location using CRISPR-Cas9 or transposon mobilization and expressed from a heterologous promoter (Frøkjær-Jensen et al., 2008, 2012, 2014). This approach can be used to screen for phenotypes induced by expression of any human gene, and to determine whether a variant exacerbates or eliminates these effects (Baruah et al., 2017).

With the rapidly expanding set of precise genome editing techniques available to C. elegans, researchers interested in variants of uncertain significance now have the freedom to choose the approach that best suits their particular needs and interests. The approaches described here provide a diverse collection of methods that can be sequentially tested in a pragmatic hierarchy of precision, beginning with direct replacement and working down until phenotypic rescue is achieved. Regardless, it will always be ideal to have corroborating evidence of variant effect from multiple techniques, indeed multiple model systems, to best inform clinical decisions.

A necessary step in the establishment of human gene functional assays is the identification of phenotypes that are rescued by or induced upon expression of the reference (wild-type) human gene, as we have done here for human PTEN and NaCl chemotaxis. Indeed, the establishment of functional assays remains a major bottleneck for variant assessment across species (Starita et al., 2017; Weile et al., 2017). Although traditional extrachromosomal array transgenes offer a quick way to establish such assays, several aspects of these transgenes can severely impede this process. These include, but are not limited to, (1) variable overexpression of transgenes, which can lead to silencing of transgenes in certain tissues, complicating phenotypic analysis (e.g. multi-copy transgenes are expressed in the soma but silenced in the germline, while low- and single-copy transgenes are expressed in both) (Kelly et al., 1997; Merritt and Seydoux, 2010); and (2) variably mosaic expression, which can make it extremely difficult to assess the rescue of partially penetrant and subtle complex phenotypes, as an animal must simultaneously carry the extrachromosomal array and be one of the members of the population that displays the partially penetrant phenotype that can often be difficult to score.

The human gene replacement approach described here allows for generation of ortholog deletion alleles directly in any wild-type or mutant strain amenable to transgenesis using the same reagents designed for replacement, thereby reducing the confounding effects of background mutations on phenotype discovery. The use of excisable selectable markers that do not severely alter morphology, baseline locomotion and several evoked sensory behaviors further simplifies phenotypic analysis and removes the need for any specialized genetic backgrounds (Norris et al., 2015; Figs 4-6; Fig. S1). Using this approach in combination with machine vision, we provide two in vivo functional assays for human PTEN, NaCl chemotaxis and mechanosensory responsivity. In particular, NaCl chemotaxis possesses several characteristics that make it an ideal functional assay: (1) a large functional range between deletion and human gene rescue to discern potentially subtle functional differences among missense variants; (2) scalability, as many plates can be run simultaneously; and (3) straightforward analysis using an automatically calculated preference index (alternatively, many laboratories score chemotaxis manually by simple blinded counting). Importantly, these reagents and functional assays can now be used in precision medicine drug screens aimed at identifying compounds that counteract the effect of a particular patient's missense variant. Further broad-scale phenomic characterization of targeted knockouts and mutant libraries, combined with new databases that curate ortholog functional annotation across model organisms, should expedite the process of in vivo functional assay development (Lee et al., 2018a; McDiarmid et al., 2018; Thompson et al., 2013; Wang et al., 2017b; Yemini et al., 2013).

One of the goals of this article is to illustrate how Cas9-triggered homologous recombination can be adapted to directly replace C. elegans genes with human genes. An exciting adaptation of this approach will be to combine targeted human gene insertions with bi-partite systems for precise spatial-temporal control of human transgene expression, as has recently been done to overexpress human genes as UAS-complementary DNA (cDNA) constructs in Drosophila (Luo et al., 2017; Lee et al., 2018b; Wangler et al., 2017). The recent and long-awaited development of the cGAL4-UAS system should allow a similar approach to be developed for C. elegans, currently the only organism for which the complete cell lineage and neuronal wiring diagram is known (Sulston and Horvitz, 1977; Sulston et al., 1983; Wang et al., 2016; White et al., 1986).

While one of the clearest uses for targeted replacement is precise structure-function analysis of variants of uncertain significance, there are several exciting applications beyond modeling disease-associated alleles. Targeted human gene replacement is also particularly well suited for investigation of the evolutionary principles that determine the replaceability of genes. By allowing for human gene expression immediately upon genomic integration, this approach should also be adaptable to essential genes. A homozygous integrant will only be obtained if the human gene can substitute for the orthologous essential gene, creating a complementation test out of the transgenesis process. This will allow for systematic and precise interrogation of the sequence characteristics and functional properties required for successful human gene replacement (Kachroo et al., 2015). A library of humanized worms would also open the door to the rich resources of tools available to visualize and manipulate human genes and proteins that are often unavailable for C. elegans researchers (e.g. high-quality antibodies, biochemically characterized or known pathogenic control variants, and experimentally determined crystal structures, Figs 2 and 7; Berman et al., 2000). Given the throughput that has been achieved for reporter gene analysis (several thousand genes) and genome editing in C. elegans, it should be possible to generate a humanized C. elegans library of similar size to those recently created in yeast (Dickinson and Goldstein, 2016; Dupuy et al., 2007; Hamza et al., 2015; Hunt-Newbury et al., 2007; Kachroo et al., 2015; Norris et al., 2017; Sun et al., 2016; Yang et al., 2017). Integrated transgenes would offer the possibility of humanizing entire cellular processes for detailed in vivo analysis in a relatively complex, yet tractable, metazoan with increasingly powerful tools for spatial-temporal control of transgene expression and protein degradation (Armenti et al., 2014; Wang et al., 2016, 2017a; Zhang et al., 2015).

Deep mutational scanning and related technologies have recently made it feasible to characterize the functional effects of virtually every possible amino acid change of a protein on a particular cellular phenotype (Fowler and Fields, 2014; Fowler et al., 2010). Several of such exhaustive sequence-function maps have been recently been generated in yeast and human cell culture systems (Findlay et al., 2014, 2018; Majithia et al., 2016; Matreyek et al., 2018; Mighell et al., 2018; Weile et al., 2017). These tools offer amazing resources that serve as ‘lookup tables’ of functional missense variation in human genes, to enable experimentally confirmed variant interpretation immediately upon first clinical presentation (Starita et al., 2017; Weile et al., 2017). An ambitious and exciting goal for the C. elegans community will be to further streamline genome engineering and high-throughput phenotyping to achieve the first comprehensive sequence-function map in a metazoan.

Worms were cultured on nematode growth medium (NGM) seeded with E. coli (OP50) as described previously (Brenner, 1974). N2 Bristol and CB1375 daf-18(e1375) strains were obtained from the Caenorhabditis Genetics Center (University of Minnesota, Minneapolis, MN, USA). daf-18(e1375) harbors a 30 bp insertion in the fourth exon and is predicted to insert six amino acids before introducing an early stop codon that truncates the C-terminal half of the protein while leaving the phosphatase domain intact (Ogg and Ruvkun, 1998).

The following strains were created for this work:

VG674, daf-18(e1375); yvEx674[paex-3::PTEN::unc-54; pmyo-2::mCherry::unc-54 UTR];

VG810-813, daf-18(e1375); yvEx810-813[paex-3::PTEN::unc-54 UTR; pmyo-2::mCherry::unc-54 UTR];

VG712, daf-18(yv3[daf-18p::PTEN+LoxP pmyo-2::GFP::unc-54 UTR prps-27::NeoR::unc-54 UTR LoxP+daf-18 UTR]);

VG713, daf-18(yv4[daf-18p::PTEN+LoxP pmyo-2::GFP::unc-54 UTR prps-27::NeoR::unc-54 UTR LoxP+daf-18 UTR]);

VG714, daf-18(yv5[daf-18p::PTEN+LoxP+daf-18 UTR]);

VG715, daf-18(yv5[daf-18p::PTEN+LoxP+daf-18 UTR]);

VG817, daf-18(yv7[daf-18p::GFP::T2A::PTEN::unc-54 UTR+LoxP pmyo-2::GFP::unc-54 UTR prps-27::NeoR::unc-54 UTR LoxP+daf-18 UTR]);

VG818, daf-18(yv8[daf-18p::GFP::T2A::PTEN::unc-54 UTR+LoxP pmyo-2::GFP::unc-54 UTR prps-27::NeoR::unc-54 UTR LoxP+daf-18 UTR]);

VG867, daf-18(yv14[daf-18p::GFP::T2A::PTEN-G129E::unc-54 UTR+LoxP pmyo-2::GFP::unc-54 UTR prps-27::NeoR::unc-54 UTR LoxP+daf-18 UTR]).

The reference PTEN CDS (UniProt consensus, identifier P60484-1) was obtained from a pCMV-PTEN plasmid (Addgene plasmid #28298) and cloned into a TOPO gateway entry clone (Invitrogen), according to the manufacturer’s instructions. The PTEN entry clone was recombined with an pDEST-aex-3p destination vector (obtained from Dr Hidehito Kuroyanagi; Kuroyanagi et al., 2010) to generate the aex-3p::PTEN::unc-54 UTR rescue construct using gateway cloning (Invitrogen), according to the manufacturer’s instructions.

The Moerman laboratory guide selection tool (http://genome.sfu.ca/crispr/) was used to identify the daf-18-targeting sgRNA. The daf-18 sgRNA sequence GGAGGAGGAGTAACCATTGG was cloned into the pU6::klp-12 sgRNA vector (obtained from the Calarco laboratory, University of Toronto, Ontario, Canada) using site-directed mutagenesis and used for all editing experiments. The daf-18p::PTEN CDS and daf-18p::GFP::T2A::PTEN::unc-54 UTR upstream homology arms were synthesized by Integrated DNA Technologies and cloned into the loxP_myo2_neoR repair construct (obtained from the Calarco laboratory) using Gibson Assembly.

C. elegans wild-type N2 strain was used for all CRISPR-Cas9 editing experiments. Genome edits were created as previously described (Norris et al., 2015). In brief, plasmids encoding sgRNA, Cas9 co-transformation markers pCFJ90 and pCFJ104 (Jorgensen laboratory, Addgene) and the selection cassette flanked by homology arms (∼500 bp) containing PTEN were injected into wild-type worms. Animals containing the desired insertions were identified by G418 resistance, loss of extrachromosomal array markers and uniform dim fluorescence of the inserted GFP.

Correct replacement of the daf-18 ORF with human PTEN was confirmed by amplifying the two regions spanning the upstream and downstream insertion borders using PCR followed by Sanger sequencing (primer binding locations depicted in Fig. 1). The genotyping strategy is essentially as described for deletion allele generation via DMS cassette insertion in Norris et al. (2015).

The forward and reverse primers used to amplify the upstream insertion region were 5′-TGCCGTTTGAATTAGCGTGC-3′ (located within the daf-18 genomic promoter region) and 5′-CCCTCAATGTCTCTACTTGT-3′ (located within the myo-2 promoter of the selection cassette), respectively.

The forward and reverse primers used to amplify the downstream insertion region were 5′-TTCCTCGTGCTTTACGGTATCG-3′ (located within the Neomycin resistance gene) and 5′-CTCAACACGTTCGGAGGGTAAA-3′ (located downstream of the daf-18 genomic coding region), respectively.

Following cassette excision via injection of Cre recombinase, the daf-18 promoter (5′-TGCCGTTTGAATTAGCGTGC-3′) and daf-18 downstream (5′-CTCAACACGTTCGGAGGGTAAA-3′) primers were used to amplify human PTEN and confirm error-free insertion at the daf-18 locus via Sanger sequencing (Fig. 1).

Total RNA was isolated from mixed-stage VG714 and VG715 PTEN knock-in animals using a GeneJET RNA Purification Kit (Thermo Fisher Scientific), according to the manufacturer’s instructions. Total RNA was treated with DNase (New England Biolabs) and purified with an RNeasy MinElute spin column (Qiagen), according to the manufacturer’s instructions. cDNA libraries were prepared from crude and purified total RNA using Superscript III (Invitrogen). All genes were amplified from cDNA libraries with Platinum Taq DNA Polymerase (Thermo Fisher Scientific) and gene-specific primer sets.

The forward and reverse primers used to amplify the PTEN CDS were 5′-ATGACAGCCATCATCAAAGA-3′ and 5′-TCAGACTTTTGTAATTTGTG-3′, respectively. The forward and reverse cmk-1 intronic control primers (Ardiel et al., 2018) were 5′-AGGGTAGGCTAGAGTCTGGGATAGAT-3′ and 5′-ACGACTCCGTTGTCGTGCATAAAC-3′, respectively.

The PTEN 1D5R reference structure (Berman et al., 2000; Lee et al., 1999) was visualized using PyMOL software (DeLano, 2002). Structural models for full-length human PTEN, DAF-18 53-506 and full-length DAF-18 were predicted using Phyre2 (Kelley et al., 2015) and visualized using PyMOL.

The chemotaxis behavioral assay was conducted on a 6-cm assay plate (2% agar), in which a salt gradient was formed overnight by inserting a 2% agar plug containing 50 mM NaCl (∼5 mm in diameter) 1 cm from the edge of the plate. A control 2% agar plug without NaCl was inserted 1 cm from the opposite edge of the plate. Strains were grown on NGM plates seeded with E. coli (OP50) for 3 or 4 days. Worms on the plates were collected and washed three times using M9 buffer, before being pipetted onto an unseeded NGM plate to remove excess buffer and select animals carrying transformation markers. Adult worms were transferred and placed at the center of the assay plates and were tracked for 40 min on the Multi-Worm Tracker (Swierczek et al., 2011). After the tracking period, the chemotaxis index was calculated as (A–B)/(A+B), where A was the number of animals that were located in a 1.5-cm-wide region on the side of the assay plate containing the 2% agar plug with 50 mM NaCl, and B was the number of animals that were located in a 1.5-cm-wide region on the side of the assay plate containing the 2% agar plug without NaCl (Fig. 3B). Animals not located in either region (i.e. in the middle section of the assay plate) were not counted toward the chemotaxis index. One hundred to two hundred animals were used per plate, and two or three plate replicates were used for each line in each experiment. Any statistical comparisons were carried out on plates assayed concurrently (i.e. on the same day).

Worms were synchronized for behavioral testing on Petri plates containing NGM seeded with 50 µl OP50 liquid culture 12-24 h before use. Five gravid adults were picked to plates and allowed to lay eggs for 3-4 h before removal. The animals were maintained in a 20°C incubator for 72 h. Plates of worms were placed into the tapping apparatus and, after a 100 s acclimatization period, 30 taps were administered at a 10 s interstimulus interval (ISI). Comparisons of ‘final response’ comprised the average of the final three stimuli. Any statistical comparisons were carried out on plates assayed concurrently (i.e. on the same day).

Multi-Worm Tracker software (version 1.2.0.2) was used for stimulus delivery and image acquisition (Swierczek et al., 2011). Behavioral quantification with Choreography software (version 1.3.0_r103552; Swierczek et al., 2011) used ‘--shadowless’, ‘--minimum-move-body 2’ and ‘--minimum-time 20’ filters to restrict the analysis to animals that moved at least two body lengths and were tracked for at least 20 s. The MeasureReversal plugin was used to identify reversals occurring within 1 s (dt=1) of the mechanosensory stimulus onset. Custom MATLAB (MathWorks) and R (https://www.r-project.org) scripts (available on request) organized and summarized Choreography output files. Final figures were generated using GraphPad Prism version 7.00 for Mac OS X. Each experiment was independently replicated at least twice. No blinding was necessary because the Multi-Worm Tracker scores behavior objectively. Morphology metrics, baseline locomotion metrics, initial and final reversal responses, habituation difference scores or chemotaxis indices from all plates were pooled, and metrics were compared across strains with ANOVA and Tukey’s honest significant difference (HSD) tests. For all statistical tests, an alpha value of 0.05 was used to determine significance.

We thank Dr Evan L. Ardiel for useful comments and discussion regarding the manuscript; Dr Kurt Haas for motivating discussions to write the manuscript; Erica Li-Leger and the Moerman laboratory for assistance with experiments; Dr John Calarco, Dr Erik Jorgensen and Dr Hidehito Kuroyanagi and their laboratories for sharing their constructs and protocols or making them publicly available; and Warren M. Meyers and Christine R. Ackerley for useful advice and discussions regarding figure design and/or protein structural modeling. We also thank the Caenorhabditis Genetic Center for strains.