Inherited peripheral neuropathies (IPNs) are a broad ranged group of genetically and clinically heterogeneous neurodegenerative disorders affecting the peripheral nervous system (PNS) with variable clinical patterns and a number of different genetic causes. Depending on the types of damaged peripheral nerves, IPNs are divided into Charcot–Marie–Tooth disease (CMT), distal hereditary motor neuropathy (dHMN), hereditary sensory autonomous neuropathy, hereditary neuropathy with pressure palsies (HNPP), and several other disorders with symptoms of muscle weakness, numbness, and pain (England and Asbury, 2004). CMT, also called hereditary motor and sensory neuropathy, is the most common peripheral neuropathy with gait disturbance, muscle weakness, and sensory defects at the extremities (Schwartz, 2019). There are demyelinating peripheral neuropathies such as CMT type 1 and axonal peripheral neuropathies such as CMT type 2. Moreover, they are further classified into many subtypes based on genetic and phenotypic classification (Pareyson et al., 2017).

While characterization of clinical features, linkage analysis, and whole exome sequencing of the study participants have uncovered a number of novel causative genes (Nam et al., 2016), researchers also developed animal models to confirm genetic causes of clinical phenotypes. First introduced 20 years ago, mouse models have been extensively used to study various causative genes such as PMP22 (CMT1A, CMT1E, and HNPP), MPZ (CMT1B), HSPB1 (CMT2F, dHMN2B), MTMR2 (CMT4B), and so on (Hong et al., 2016; Tanaka and Hirokawa, 2002). In recent years, zebrafish models were introduced to IPN studies (Bragato et al., 2016; Hong et al., 2016; Won et al., 2019). These animal models are not only useful for dissecting the role of causative genes but also for investigation of therapeutic targets. For example, transgenic mice with a mutation in the HSPB1 gene, which causes CMT2F and dHMN2B (Evgrafov et al., 2004), were assayed for the therapeutic effect of HDAC6 inhibitors, and the treatment effectively reversed axonal loss in these mice (d'Ydewalle et al., 2011).

Axons in Drosophila melanogaster or Caenorhabditis elegans are not surrounded by myelin sheaths or Schwann cells, and therefore they are not suitable models for demyelinating peripheral neuropathies, but they still can be used as a model system for peripheral neuropathies characterized by axonal impairment. The causative genes of axonal neuropathies examined using the Drosophila model so far include GARS1, GDAP1, MFN2, and YARS1 (Yamaguchi and Takashima, 2018).

C. elegans models have not been employed as much as the other animal models. Recently, however, there was a report that disruption of fzo-1, a C. elegans homologue for MFN2, induced differences in mitochondrial morphology and deficits in worm movement (Byrne et al., 2019). MFN2, a key player in mitochondrial activities (Long et al., 2019), is a causative gene for the axonal CMT2A and HMSN6A characterized by defects in mitochondrial dynamics and axonal transport, and this result therefore suggests the utility of C. elegans models for axonal peripheral neuropathies.

In an effort to find potential C. elegans peripheral neuropathy models, we made a list of C. elegans homologues of currently known causative IPN genes. We then examined the motor behaviors of the worm strains with mutations in these genes and searched for those that show the deficits, as the worm’s locomotion and body posture might be substantially matched by the IPN phenotypes. This makes monitoring their movement a relatively simple yet useful experimental approach for neurogenetics in worms (Yemini et al., 2013).

To do so, we made a simplified version of WormTracker (Yemini et al., 2013) and monitored the motor behaviors of 26 different worm strains, and compiled a database of their behavioral features. Our assay then shows that the worms with mutations in HSPB1, MFN2, DYNC1H1, and KIF1B homologues evidently exhibit phenotypes of abnormal movement, suggesting them as candidates for C. elegans peripheral neuropathy models.

Preparation of worms

C. elegans strains were cultured using standard techniques (Brenner, 1974). Worms were synchronized and grown until they reached adulthood at 25°C. The experiments were performed when they reached adulthood and started laying eggs. For the measurement, 10 to 12 individuals were moved to a new nematode growth medium (NGM) plate 10 min before recording. To control the effect of the medium on worms’ movement, we made fresh plates one day before the experiment. In case the homozygotes are lethal (rab-7, tfg-1, hars-1, and yars-1), we used only the progenies of heterozygotes for the experiment, after confirming they are heterozygous by checking their segregants.

Image analysis and statistical analysis

We recorded 1,200 images for 10 individuals for each experiment. To analyze these images automatically, we wrote custom MATLAB (MATLAB2013b; Mathworks, USA) scripts that extract the movement trajectories (Fig. 1A) and the body-skeleton curves (Fig. 1B) of individuals. From the trajectory and the body-skeleton curve, we quantified 104 behavioral and moving features.

The distribution of each feature of a mutant was compared to that of the control strain (N2) in order to determine a distribution-free overlapping index η (Pastore and Calcagnì, 2019). Taking into consideration that the overlap between two different normal distributions follows a normal distribution (Inman and Bradley, 1989), we converted the overlapping index η to the P value of a null hypothesis that the two distributions are not different. Following the same procedures in the WormTracker (Yemini et al., 2013), we checked for the significance at P ≤ 0.01 and calculated the q-value, the positive false discovery rate (Storey, 2002).

A statistical analysis was also performed using a custom written MATLAB script. All MATLAB scripts will be available upon request.

C. elegans strains with mutation of homologues of the IPN genes

To find C. elegans strains (with disruption of a gene) associated with IPNs, we first identified the homologues of causative genes, and then looked up the databases of worm strains bearing the mutation of their homologues. In Table 1, we list human genes identified as underlying causes of the axonal peripheral neuropathies from the Inherited Neuropathy Variant Browser (http://hihg.med.miami.edu/code/http/cmt/public_html/) and recent studies (Lee et al., 2020; Pareyson et al., 2017; Yamaguchi and Takashima, 2018). We also listed their homologues in C. elegans, referring to the HUGO Gene Nomenclature Committee (HGNC; https://www.genenames.org/) Comparison of Orthology Predictions (HCOP) (Wright et al., 2005) and WormBase (https://wormbase.org/), a model organism information resource (Harris et al., 2020).

Next, we searched two worm strain databases, Caenorhabditis Genetics Center (CGC, University of Minnesota, USA) and National BioResource Project (NBRP) C. elegans (Tokyo Women’s Medical University, Japan), for mutants of the C. elegans homologues of causative genes. Although not all genes had their mutant strains available, we could select 26 strains to monitor their motor behaviors (Table 2, Supplementary Fig. S1).

First, we included the mutants of HSPB1 homologues, because HSPB1 is one of the most studied causative genes for axonal peripheral neuropathies such as CMT2 and dHMN2B (Evgrafov et al., 2004; Schwartz, 2019), and relevant animal models would be valuable for developing therapeutic strategies such as HDAC6 inhibitors (d'Ydewalle et al., 2011; Kim et al., 2016; Schwartz, 2019). GARS1, which encodes glycyl-tRNA synthetase, was another gene of interest because several aminoacyl-tRNA synthetase genes had been linked to IPN (Antonellis et al., 2003; Jordanova et al., 2006; Lee et al., 2012; 2020; Vester et al., 2013). However, its mutant was not available, and therefore we included hars-1/HARS1 and yars-1/YARS1 in our assay instead. fzo-1/MFN2 is also an interesting gene related to mitochondrial dynamics, because MFN2 homologues have been one of the successful invertebrate model systems, as mentioned above (Byrne et al., 2019; Eschenbacher et al., 2012). For comparison, we also added several genes causing demyelinating peripheral neuropathies: exc-5/FGD4, lbp-5/PMP2, mtm-1, and mtm-5/MTMR2. Even though exc-5 deficient mutant was associated with CMT4H, not the axonal type of CMT, it was one of the earliest C. elegans CMT models, due to the morphological similarity between the structural defects of its excretory canals and those in tubules in CMT4H patients (Mattingly and Buechner, 2011).

We preferred a homozygote with a mutation of a single gene of interest, but such strains were not always available. In the absence of such strains, a heterozygote or a strain with multiple mutations was examined to cover various subtypes of peripheral neuropathies (Table 1). For example, disruption of aminoacyl tRNA synthetases (hars-1/HARS1, yars-1/YARS1) was lethal to worms and homozygotes demonstrated larval arrest (Pierce et al., 2011; Sönnichsen et al., 2005), and therefore we had to test heterozygotes (hars-1/+, yars-1/+) instead.

Monitoring the behaviors of worms

We adapted WormTracker and its motion analysis workflow (Baek et al., 2002; Yemini et al., 2013). However, we largely simplified both the instrumentation and the analysis. The major difference is that our setup is much simpler without a software-controlled motorized XY-stage, because we do not track a single worm. Instead, our field of view covers an entire NGM plate (35 mm petri dish), and thus we can track 10 to 12 individuals simultaneously. This reduces the spatial resolution significantly (Fig. 1A), making a more sophisticated image analysis such as distinguishing head and tails, characterization of omega-turn, and upsilon-turn events impractical. However, even at the reduced resolution, we still can obtain movement trajectories (red curves in Fig. 1A) and body-skeleton curves (red curves in Fig. 1B) of individual worms.

From the trajectories and the body-skeleton curves, we quantified 44 motion features such as speed, acceleration, reversal, turn speed, roaming/dwelling intervals, and 60 body posture features such as amplitude, looping, and bending angles, as defined in the previous studies (Baek et al., 2002). Reversal is when worms make a rapid backward motion, as shown in Fig. 1A, and looping is when an animal’s head or tail touches its body, making a loop. We will explain more of these features later in the text and in the supplementary text.

It should be noted that worms exhibit two-state moving behavior of roaming and dwelling (Ben Arous et al., 2009). As shown in Fig. 1A, worms sometimes spend their time in an inactive mode termed “dwelling” in which they do not move around much and only alternate backward and forward movement at a low speed, whereas they move fast in search of food in “roaming” mode. To disjoin the effect of the roaming/dwelling decision on worms’ capacity to move, we need to compare the motion features in roaming and dwelling intervals separately.

Comparison of locomotion and shape features between strains

For each strain, we quantified each of the 104 behavioral features for all individuals in all images collected from the measurement. For each feature, we compared the distribution of each strain to that of N2, the C. elegans wild type, to determine if the two distributions are different from each other. Furthermore, we calculated the q-value for each comparison, following the same algorithm described in the literature (Storey, 2002; Yemini et al., 2013). Briefly, the q-value (positive false discovery rate) is a measure of how likely it is a false claim that the two distributions are different.

Fig. 2 summarizes the q-values of 29 selected behavioral features for all strains assayed in this study (See Supplementary Fig. S2 for the overlapping index η). We tried to include as many features as possible to illustrate a variety of different q-value patterns, while omitting those showing similar patterns. The q-values of all 104 features are provided in Supplementary Material . In Fig. 2, it is notable that hsp-12.1, hsp-12.2, and hsp-12.3/HSPB1 mutants exhibit deficiency in their motion: they move slower in roaming intervals (speed, average speed over 10 s, maximum speed over 10 s) and they also accelerate slower (maximum lateral acceleration over 10 s). Since worms tend to keep their speed while they are roaming, the difference in the acceleration capacity is better observed in lateral acceleration as they change directions. Other strains that move slower than N2 are fzo-1/MFN2, dhc-1/DYNC1H1, and unc-104/KIF1B mutants. In Fig. 3, the boxplots of these strains illustrate their differences with N2 in detail (Figs. 3A and 3B). unc-104 worms looked physically abnormal under our experiment conditions: they did not roam and their locomotion was very limited. hsp-43/HSPB1 and tfg-1/TFG mutants were other strains that showed decreased lateral acceleration (maximum lateral acceleration over 10 s).

Body posture features of hsp-12.1, hsp-12.2, and hsp-12.3/HSPB1 deficient mutants were also significantly different from those of N2. The amplitude of the lateral undulling motion that worms make while they crawl (Fig. 3E) and the related body posture features such as the ratio of the area of convex hull of worm image (Convex-Area) to the area of worm image (Convex-Area/Area) are smaller for these worms than those of N2 (Fig. 3C). This means that these worms appear to be relatively straight. Other than the HSPB1 homologues, fzo-1/MFN2, dhc-1/DYNC1H1, ifa-4, ifb-2/NEFL, and rab-7/RAB7A also have smaller amplitude than N2.

The body of hsp-12.2/HSPB1 deficient mutant is less bent than that of N2. To quantify this, we calculated a bending angle, the angle between two tangential vectors separated by 20% of body length, for every pair over the body (Fig. 3F). Its maximum (‘maximum bending angle 20%’) indicates how much the body is bent where it is bent the most, while its standard deviation (‘standard deviation of bending angle 20%’) measures how much the bending angle changes at different parts of the body. Not only the maximum bending angle but also the standard deviation of the bending angles are smaller in the hsp-12.2/HSPB1 mutant (Fig. 3D). The body of rab-7/RAB7A and yars-1/YARS1 deficient worms are also less bent than N2.

Figs. 3A-3D present boxplots of several selected strains that exhibit differences both in locomotion and body posture features. The boxplots of corresponding behavioral features for all strains are available in Supplementary Figs. S3-S6, respectively.

Animal models are powerful tools for studying the pathogenic mechanisms of peripheral neuropathies as well as for finding therapeutic strategies (d'Ydewalle et al., 2011; Hong et al., 2016; Won et al., 2019; Yamaguchi and Takashima, 2018). Nevertheless, C. elegans models have rarely been applied until recently.

In an effort to find novel C. elegans peripheral neuropathy models, we monitored behaviors of various worms with mutation of C. elegans homologues of the causative genes and compiled a database of their behavioral features, adapting previous studies on an automated system for monitoring worms’ behavior (Baek et al., 2002; Yemini et al., 2013).

Among 11 different C. elegans homologues of HSPB1 that already had their deletion mutants ready in the CGC and NBRP, hsp-12.1, hsp-12.2, and hsp-12.3 deficient worms exhibit a significant decrease in their locomotion capacity, making these worms viable candidates for a C. elegans peripheral neuropathy model.

These worms also exhibit a different body posture compared to the C. elegans wild type. The body posture of worms is indeed a neurogenetic phenotype in worms: for example, the amplitude of lateral undulation of worms is regulated by a mechano-sensory neuron DVA (Zhen and Samuel, 2015). The amplitude has been linked to the speed of locomotion in a previous biophysical study (Shen et al., 2012). Bending angles across the body are also coordinated by excitatory ventral cord motor neurons (Wen et al., 2018).

Other reasonable candidates would be fzo-1/MFN2, dhc-1/DYNC1H1, and unc-104/KIF1B. In fact, fzo-1 mutant has already been employed to examine the role of mitochondrial dynamics on behavior and lifespan (Byrne et al., 2019). Also, it was reported that axonal transport in unc-104 worms was impaired and the worms were much slower and underwent larval arrest depending on the growth temperature (Edwards et al., 2015). Dynein (DYNC1H1) as well as kinesin (KIF1A, KIF1B) is also involved in axonal transport.

Ubiquitously expressed small heat shock protein 27 (HSP27), which is encoded by HSPB1, protects cells from a wide variety of unfavorable physiological and environmental conditions (Acunzo et al., 2012). Mutations in human HSPB1 are associated with CMT2F and dHMN2B (Evgrafov et al., 2004), and mouse HspB1 mutants showed distal axonal loss and a decrease in acetylated alpha-tubulin in peripheral nerves (d'Ydewalle et al., 2011). Mutations in MFN2, which develop dominant or recessive CMT2A, are known as the most common genetic causes of axonal CMT (Züchner et al., 2004). MFN2 mediates the dynamic balance between fusion and fission of mitochondria (Santel and Fuller, 2001). DYNC1H1, implicated to CMT2O (Weedon et al., 2011), encodes a large subunit of the cytoplasmic dynein complex, which is a type of microtubule-activated ATPase. Mutations in this gene caused neurodegeneration, anterior horn cell death, impairment of axonal retrograde transport, and perturbation of neuronal development and migration (Hafezparast et al., 2003). KIF1B, related to CMT2A1 (Zhao et al., 2001), encodes an N-terminal-type motor protein that plays an important role in the anterograde transport of mitochondria as a monomeric motor. Mouse Kif1b heterozygotes showed defects in transporting synaptic vesicle precursors and progressive muscle weakness (Zhao et al., 2001).

To conclude, we suggest C. elegans homologues of HSPB1, MFN2, DYNC1H1, and KIF1B as a model for axonal peripheral neuropathies. We hope that our behavioral phenotype database will assist the development of more C. elegans models for better understanding of pathogenesis mechanisms as well as novel therapeutic strategies.

We thank the Caenorhabditis Genetics Center (USA), the National BioResources Project C. elegans (Japan), and the Million Mutation Project (Thompson et al., 2013; https://doi.org/10.1101/gr.157651.113) for the strains used in this study. We also thank Ah Jin Lee (Kongju National University, Republic of Korea) for sharing her peripheral neuropathies gene list. This work was supported by the National Research Foundation (2018R1A4A1024506, 2019R1C1C1007124, and 2019R1A2C1087547), Republic of Korea.

K.W.C. and K.S.L. wrote the manuscript. K.S.L. designed the experiment and K.S.L. and J.S.K. performed the experiment. K.S.L. performed the image analysis and statistical analysis.

The authors have no potential conflicts of interest to disclose.

A list of causative genes of axonal peripheral neuropathies and C. elegans homologues

A list of the worm strains used in this study