Genetic influences on motor learning and performance and superperforming mutants revealed by random mutational survey of the mouse genome

Evolution depends upon genetic variations that influence physiology. As defined in a genetic screen, phenotypic performance may be enhanced or degraded by such mutations. We set out to detect mutations that influence motor function, including motor learning. Thus, we tested the motor effects of 36,444 non‐synonymous coding/splicing mutations induced in the germline of C57BL/6J mice with N‐ethyl–N‐nitrosourea by measuring changes in the performance of repetitive rotarod trials while blinded to genotype. Automated meiotic mapping was used to implicate individual mutations in causation. In total, 32,726 mice bearing all the variant alleles were screened. This was complemented with the simultaneous testing of 1408 normal mice for reference. In total, 16.3% of autosomal genes were thus rendered detectably hypomorphic or nullified by mutations in homozygosity and motor tested in at least three mice. This approach allowed us to identify superperformance mutations in Rif1, Tk1, Fan1 and Mn1. These genes are primarily related, among other less well‐characterized functions, to nucleic acid biology. We also associated distinct motor learning patterns with groups of functionally related genes. These functional sets included, preferentially, histone H3 methyltransferase activity for mice that learnt at an accelerated rate relative to the remaining mutant mice. The results allow for an estimation of the fraction of mutations that can modify a behaviour influential for evolution such as locomotion. They may also enable, once the loci are further validated and the mechanisms elucidated, the harnessing of the activity of the newly identified genes to enhance motor ability or to counterbalance disability or disease.


Introduction
No sharp boundaries can be established a priori between the types of functions of the organism that are primarily governed by one or a few gene products and those supported by larger ensembles of them that operate in concert.Rather, experimental evidence is required to establish, in a function-specific manner, when the characteristics of the organism are facilitated or degraded by individual genes or when they are subject to more numerous genetic influences.We set out to experimentally elucidate the relative contribution of these two possibilities to motor performance during a complex task such as the rotarod, additionally taking into account motor learning.The motivation for the study of movement is that it represents a trainable, complex or high-order organism function where both individual genes and groups of genes may be expected to be influential.Ample precedent exists for both kinds of influence: On the reductionist end are laboratory mice subject to spontaneous or intended single-gene mutations that degrade motor phenotype.This is also the case for some human diseases that deteriorate motor ability where the causal links between gene product, cellular and organism function are few or stereotyped.In contrast, on the opposite, more complex end, most of the behaviours considered indispensable in the natural environment require adaptation to changes in initial and intervening conditions.This adaptation is often enabled by the concerted interaction of gene products that subserve the activity of the whole cell, organ or organism rather than one of their constituent parts (Pascual et al., 2023).The disease counterpart of this concerted action is exemplified by the dependence of mitochondrial disease manifestations on the transcriptome of a functionally related gene family rather than on individual genes (Jakkamsetti et al., 2022).Of note, not all individual deleterious mutations are perceptible at the organism level.In fact, estimation of the functional impact of damaging certain individual gene products considered critical can reveal unsuspected compensation.For example, progressive suppression or abolition of the cardiac ionic current i f , long judged fundamental to initiation of the normal sinus rhythm and thus of the heartbeat, results in negligible disturbance of heart function due to extragenetic functional compensation by the current i b, Na in a process entirely circumscribed to the cell membrane (Noble et al., 1992).
Whereas single gene discovery and characterization has been widely exploited, investigation of concerted gene influences, which relies on the analysis of myriad genes, requires large throughput study of mutations and associated behaviours.In this context, one of the most impactful activities of the organism from an evolutive perspective is locomotion or, more generally, motor behaviour.As an eminently adaptive (i.e.subject to learning) behaviour (Jakkamsetti et al., 2021), motor behaviour is governed by an array of both well-known and poorly understood physiological and genetic influences.Locomotion is critically relevant from the perspective of human disease.For example, the motor consequences of stroke alone account for over half of all the neurological disability in the world (World Health Organization, 2006).
Thus, we performed a forward genomic screening of mutations that may impact motor performance under a complex motor task.Given the mutability of the laboratory mouse, we examined whether definable mouse mutations artificially induced and tested for effect in both homozygous and heterozygous states might augment or degrade motor performance as assayed on a standard (i.e.accelerating) rotarod over several consecutive testing trials performed while blinded to genotype and compared with control mice tested simultaneously.Our work was facilitated by the fact that, originally, the C57BL/6J laboratory mouse underwent a process of stringent inbreeding, fixing alleles that might impede motor performance in the homozygous state.Indeed, many wild mice and some laboratory strains exhibit motor performance inferior or superior to C57BL/6J mice.We reasoned that most randomly induced point mutations are inconsequential, but some, by causing loss or gain of function, may alter and even augment performance at the organism level.We were particularly interested in motor superperformance mutants given the far-reaching therapeutic development potential of any gain of function thus identified, since genetic gain of motor function might ultimately prove amenable to replication via the development of mechanism-related pharmacological or other interventions.

Methods
This study was approved by the Institutional Animal Care and Use Committee of UT Southwestern Medical Centre (protocol number 2016−101869), which operates under the U.S. Public Health Service and the U.S. Animal Welfare Act guidelines.All other relevant institutional regulations were followed.C57BL/6J mice were used and generated at UT Southwestern.There were no mice excluded nor unexpected events.Standard laboratory mouse nutrition was offered ad libitum.At the conclusion of this study, the intact mice proceeded to further but unrelated studies under separate institutional approval.The investigators understand the ethical principles under which the Journal of Physiology operates and this work complies with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines 2.0 animal ethics checklist.
V. Jakkamsetti and others J Physiol 602.11

Mutagenesis
Mice were generated as previously described (Wang et al., 2015).In brief, over ∼5 years we induced germline mutations in male generation 0 (G0) of C57BL/6J mice and bred them to C57BL/6J females to produce G1 (first generation after mutagenesis) males, which were subjected to whole exome sequencing to identify all non-synonymous coding and putative splicing changes induced by N-ethyl-N-nitrosourea (ENU) or resident in the background.Extensive experience with ENU indicates that, with a handful of exceptions, coding/splicing changes are the main source of the induced phenotype (Cordes, 2005).
While some of the mutations were homozygous lethal prior to weaning age and therefore were tested only in the heterozygous state, the majority were homozygous viable.To compute genome saturation, or the fraction of all genes destroyed or severely damaged and examined three or more times in the homozygous state, we used a method based on observation of a curated set of essential genes: those known to be essential for life based on the results of robust knockout mutations on the C57BL/6J background.We also used a larger set of ENU-induced mutations and checked the effect of mutations falling into genes of this type.This allowed us to correlate 'severe damage or destruction' manifested as a lethal effect with PolyPhen-2 (Adzhubei et al., 2013) scores (for missense errors), or with putative null alleles (for nonsense, start lost, stop lost, critical splicing error mutations).This analysis was performed on several thousand mutations.
Lethality prior to weaning age is a phenotype with an enormous genomic footprint, encompassing large numbers of nucleotides in approximately one-third of all genes.Assuming the types of damage that befall these genes are representative of the types of damage befalling all genes and the effect of that damage at the protein level, computation of the likelihood that each point mutation will cause damage can be generalized to measure the likelihood of severe damage or destruction to all individual genes, and to the sum of all genes in the genome (Wang et al., 2018).We concluded that, on average, only one ENU-induced mutation in six is capable of causing severe damage or outright destruction of the encoded protein.This method of estimation must be considered stringent, since certain assays are able to detect even modest damage to a protein, for example loss of 10% of enzymatic activity.
Average coverage of the exome target (62 million nucleotides in length) was ∼99.69% to a depth ≥10 nucleotides (quality score ≥40).Since we give consideration to two or more identical aberrant calls among 10 reads (P = 0.0107 by binomial probability calculation), fewer than one in 10,000 mutations within the target sequence will be missed in sequencing.False positive mutations, while exceedingly rare, are eliminated by resequencing every non-synonymous coding/splicing change identified using Ion Torrent sequencing technology in the course of genotyping G1, G2 and G3 mice (described below).
On average, 60 coding/splicing changes were identified per G1 male.Each G1 male with ≥30 mutations and adequate breeding capacity was used to generate a large pedigree.G1 males were crossed to C57BL/6J females to yield 10-15 G2 daughters.Each G1 male was then crossed to its own daughters to yield ∼40-60 G3 mice (∼600-700 animals per week derived from 12-15 pedigrees), in which most of the mutations were brought to homozygosity.G1, G2 and G3 mice were genotyped at all mutation sites identified in the G1 founder, this time using Ion Torrent sequencing.In rare instances, mutations identified by Illumina sequencing were not confirmed, and were eliminated from further consideration.All validated mutations were assessed for zygosity in G2 mothers and G3 offspring.This allowed detection of probable lethal effects, evidenced by distortion of the expected Mendelian ratio.It also allowed subsequent mapping of each observed phenotype to a causative mutation, using the statistical computation method we have described (Wang et al., 2015).

Gene selection for further validation
Candidate Explorer, a machine learning program developed by us (Xu et al., 2021), determines, across all assays of function, whether any individual mutation is likely to be validated as causative of phenotype.In brief, 60 features are used by Candidate Explorer to assess candidate mutations and score them.These features include the P value; the magnitude of the phenotypic effect; the number of homozygotes assayed; the variance of the data; the degree of overlap between phenotypic performance of homozygous mutant and homozygous reference allele populations; the existence of multiple alleles supporting causation; the predicted damage caused by each allele (assessed by another machine learning program); the predicted essentiality of the candidate gene (E score; assigned by still another machine learning program); and others.Candidate Explorer is trained on the basis of CRISPR/Cas9 validation data.

Motor testing
We used a standard rotarod (rod diameter 3 cm, fall height 20 cm, start speed 4 r.p.m., acceleration rate 1 r.p.m./8 s, Touchscreen RotaRod, Harvard Apparatus, Holliston, MA, USA) in several consecutive testing trials performed over 7 days while blinded to genotype and simultaneously testing control littermate mice.The rotarod accelerates; hence, a prolonged time to fall (or raw score) at elevated angular velocities indicates much greater performance than a similar prolongation at lower speeds.For example, in the case of the superpeforming mutant Rif1 I107T/I107T  this translates into an increase in maximum linear mouse speed of approximately from 1.0 m/min (control) to 1.4 m/min.
We conducted a rigour and effectiveness analysis by applying rotarod screening to mice mutated as above at a gradually increasing rate from 100 to 900 mice per week by a team of six testers who tested an approximately equal number of mice per week.Testing more than 900 mice per week nearly exceeded testing capacity and was thus considered prohibitive for efficient screening.Conversely, low mouse numbers also negatively impacted phenotypic mutation detection rate.Fig. 1A-E illustrates the effect of varying the number of mice per week relative to the number of genes with probably deleterious (damaging or null) mutations studied each week.Increasing weekly mouse numbers screened exerts a sharp impact on the number of mutant genes studied for each particular week.Thus, we estimated the optimal rate of screening at about 600 (but fewer than 900) mice per week, which was ultimately contingent upon weekly mouse generation rates (subject to unavoidable biologically imposed fluctuations) and can be adjusted to accommodate pedigrees that span about 100 more or fewer mice.
To render the motor behaviour results comparable across testers and time, we converted raw rotarod fall latency scores to normalized scores for gene linkage analysis and T-scores for subsequent motor behaviour analysis.Normalized scores were obtained by first deriving a scaling factor by dividing 100 by the mean wild-type mouse performance in that week of testing (Fig. 1F).The scaling factor was multiplied by the raw, unmodified score of each mouse [in wild-type, REF (reference non-mutated mice littermate to ENU mutagenized mice), HET (heterozygous mutants) and VAR (mutant homozygous) groups] tested that week to calculate their respective normalized score.The mean normalized wild-type mouse score was thus 100 for each week and the REF, HET and VAR mice scores in relation to the wild-type mean could be compared across weeks.T-scores [calculated as 50+(10×Z-score)] were derived from Z-scores (which are often used for similar forward genetics projects; Cinà et al., 2019;Kumar et al., 2011) but modified to allow mathematical operations on non-negative numbers for standardizing data across time and locations (Levin et al., 2020).T-scores also allowed for a simple identification procedure of outlier performance since the mean is set at 50 and every 10 units correspond to one standard deviation.For control, untrained wild-type mice were tested each week alongside the ENU mutagenized mice totalling 1408 mice.Rotarod T-scores for the sixth trial in these mice tested over time showed stability (Fig. 1F).As expected for wild-type mice, most of the performances were within 2 SD of the mean and a few were within 3 SD.To map phenotype to DNA loci, we used Linkage Analyzer (Xu et al., 2021), which statistically tests the association between an assay and genotype for all mice at every mutation site in every pedigree, using the dominant, additive and recessive models of inheritance.
Principal component analysis (PCA) and gene enrichment analysis.PCA was conducted on the motor behaviour parameters listed in Table S1.Overall, the parameters reflected rotarod performance scores at each trial, learning between trials and probable learning trajectories including early and late rise in performance and learning rate.To allow comparison across multiple continuous parameters of varying magnitude, the values for each parameter were normalized to range from 0 to 1 prior to conducting PCA.Gene enrichment analysis (GEA) was conducted using EnrichR (Kuleshov et al., 2016).To examine initial learning curves over just six trials, T-score values were smoothed with a median filter over two adjacent trials.Statistical analyses examining differences between groups employed original T-scores prior to smoothing.

Genome coverage (saturation) rate
A total of 36,444 such mutations were screened (in either heterozygous or homozygous state).In total, 32,075 of these mutations resided in 32,726 G3 mice, which were screened in the homozygous state at least once.By computing gene damage or destruction, we estimate that 16.3% of all autosomal genes were severely damaged or destroyed by mutations that were brought to homozygosity in five or more G3 mice, each tested six times using the rotarod.Note that the X chromosome is not mutagenized in our G3 mice: it was derived from a wild-type C57BL/6J female mated to a mutagenized G0 male, which contributed a Y chromosome to the G1 founder.However, spontaneous mutations of the X chromosome did occur and were detected.If all chromosomes are included in the saturation estimate, at least 11.4% of all genes were damaged or destroyed by mutations that were brought to homozygosity in five or more G3 mice, each tested six times using the rotarod.The average percentage of each mutation effect was: probably benign: 38.3%, possibly damaging: 15.1%, probably damaging: 38.8% and probably null: 4.5%.The remaining 3.3% can be accounted for by mutations classified as 'unknown' .

Superperformance genes
Based on the criterion of 'good or excellent' candidate ranking in Candidate Explorer (Xu et al., 2021), we noted J Physiol 602.11 15 mutations in 14 genes that either increase or decrease rotarod performance (Table 1).
Several of these loci are known causes of brain diseases with ataxia or dystonia.All are recessive except for struggles, an allele of Nfatc3 that is also associated with reduced survival.All are unambiguously assigned to a single mutation except even-steven, a phenotype that results from one of two co-segregating mutations, in  Although Lepr is expressed in the brain, obesity mutations (including Lep, which is not brain-expressed) can cause poor rotarod performance.Among the mutant genes that enhance rotarod performance, we unambiguously assigned (Candidate Explorer verification probability, CEvP = 0.974) a single recessive ENU-induced mutation, Rif1 I107T/I107T (nietzsche) that increases rotarod performance.We also discovered Tk1 V140E/V140E (CEvP = 0.689, tico), Fan1 I618N/I618N (CEvP = 0.698, hitched) and Mn1 (CEvP = 0.796, ubermus) mutations in relation to motor superperformance (Fig. 1G and H).

Functional role of four superperformance genes
We examined brain microarray data from the Allen Brain Atlas (Lein et al., 2007) for the regional expression of the superpeformance genes (Fig. 2).Fan1 displayed greater expression in the pons or medulla, Mn1 in striatum, Rif1 in cerebellum and Tk1 in medulla.Overall, these structures are generally relevant to motor performance.
There was a distinct functional role common to these four genes.Gene enrichment analysis indicated a significant association of these genes with nuclei acid biology and DNA repair and modulation.Gene Ontology Molecular Function indicated that this gene set's association with nucleoside kinase activity (Fig. 2B) was higher than that of randomly selected genes.

Distinct clusters of learning and non-learning ENU mutagenized mice
We also examined the impact of ENU mutations on motor learning.The first few trials during rotarod training involve rapid learning for the mouse, associated with learning to regulate paw position and speed to adapt to the accelerating surface (Buitrago et al., 2004;Jakkamsetti et al., 2021;Shiotsuki et al., 2010).Evidence of such learning is manifest as early as after a single rotarod trial (Jakkamsetti et al., 2021).When compared to wild-type mice, the mice with ENU mutations performed slightly worse (Fig. 3A).This is in agreement with previous studies (Eyre-Walker & Keightley, 2007) that indicate that the majority of mutations can have a neutral or deleterious impact with very few providing an advantage.Average motor learning rate was also reduced for ENU mice (Fig. 3B).However, considering the large number of ENU and wild-type mice tested, it was surprising that the difference in their respective learning rates barely reached statistical significance (P = 0.026).This suggested that there was considerable variability in the average learning rate and that a significant number of mice exhibited values overlapping with wild-type mice, thus decreasing statistical significance.Hence, to test whether behavioural clusters corresponding to different learning patterns across six trials existed, we conducted an unsupervised PCA clustering of data with multiple parameters including those that reflected different learning trajectories (Table S1).This revealed two main clusters of mouse performance, one comprising 20,672 mice and one with 1837 mice (Fig. 4A).Compared to the smaller cluster, mice from the larger cluster displayed learning trajectories similar to wild-type mice (Fig. 4B).In contrast, mice from the smaller cluster displayed flat learning curve trajectories.The smaller cluster mice did not exhibit a net positive learning rate and, in fact, were slightly below a learning rate of zero (mean = −0.11,P < 0.0001, one-sample t test against zero) and were significantly different from wild-type mice and mice from the larger cluster in their average learning rate (Fig. 4C).Thus, we henceforth refer to the larger cluster as the learning group mice and the smaller cluster as the non-learning group mice.The first nascent baseline performance for the smaller cluster was higher (Table S2), and the final performance score at the sixth trial was lower than that of mice in the larger cluster.We noted that the average performance over six trials was slightly lower than the wild-type for both ENU mutagenized learning groups, and more so for the non-learning group (Fig. 4D).We also examined the average learning rate for the four mutations to determine if their superperformance was also associated with superior learning.We found that mice with mutations in Mn1 and Fan1 did show a higher learning rate but this was not the case for mice with mutations in Rif1 and Tk1 (Fig. 4E).

Mutation rates in learning and non-learning mouse groups
Since motor learning is, arguably, critical for survival, we hypothesized that the mice that did not learn would harbour a higher number of homozygous mutations given that, in general, mutations have no impact or cause deficits in behavioural performance.Surprisingly, the non-learning mice had fewer mutation numbers compared to the learning group (Fig. 5A).This prompted us to re-examine our hypothesis that more mutations would result in deteriorating motor performance.Thus, we examined the impact of the number of mutations in a mouse on performance.However, it is possible that, with each cumulative mutation in a single mouse, the performance will eventually begin deteriorating in some animals at some mutation number, thus confounding the analysis.Therefore, we first examined the coefficient of variation of performance across mutation rates and selected 15 as the number of mutations beyond which performance variability changed drastically (Fig. 5B).We found that, as expected, total motor performance deteriorated slightly but significantly as the number of mutations increased (Fig. 5C), yet the relationship between the two was weak and the dot plot of average T-score performances against mutation number did not show any significant relationship (Fig. 5D).Interestingly, the initial performance of the mouse (i.e. the average of the first two trials) was more correlated with deterioration upon an increase in mutation number.Similarly, the average learning rate increased slightly but reliably with an increase in number of mutations in a mouse (Fig. 5E-G).There was an opposite relationship between initial motor performance and learning rate as expected.Mice that started with low initial performances demonstrated better learning (Fig. 5H).The 6096 genes with homozygous mutations that were unique to the learning group (Table S3) shared a common role in Gene Ontology Molecular Function 2023 that indicated DNA modulation with histone methyltransferase activity (Fig. 5I).

Number of mutations required for superperformance
The combination of several mutations can confer a behavioural advantage.Therefore, we estimated the mutation rate that could enable the emergence of a superior or outlier motor behaviour.Based on the wild-type mouse performance data where no mice deviated more than 3 SD from the mean, we defined outlier performance as an increase in performance greater than 3 SD.We found 39 mice with outlier performance, amounting to 0.15% of mice (Fig. 6A).In other words, for approximately every 672 ENU mice there was an outlier performing mouse.Given the total number of homozygous mutations harboured by ENU mice, one mouse became an outlier in motor performance on the rotarod approximately for every 2410 homozygous mutations.Regarding the learning rate we found 696 mice with outlier performance (Fig. 6B) amounting to 0.27% of mice.In other words, for every 38 ENU mice one was a motor learning outlier, or one mouse became an outlier in motor learning rate per approximately 696 homozygous mutations studied.

Relationship between number of mutations and survival
Elevated mutation rates can be expected to reduce survival.Thus, we examined the relationship between mutation number and post-weaning survivability.
There was a significant logarithmic relationship between mutation load in a mouse and survivability (Fig. 6C), with the number of surviving mice decreasing logarithmically as the number of mutations in a mouse increased.

Scale and effort-yield ratio relative to alternative approaches
While in principle individual knockout mutations could be generated one at a time to cover the genome and tested to identify those that enhance motor performance, this could not possibly be accomplished as efficiently as a screen of ENU-induced mutations: first, assuming there are ∼25,000 genes in a mouse (Guenet, 2005), targeted testing of each gene (using, for example, 20 mice per mutant pedigree) would require 500,000 mutant mice, which would take over 16 years if 500 mice per week are tested.Second, such an effort would not permit a comprehensive assessment of all genes: about one-third of all genes are essential for survival to weaning age, and ENU permits analysis of this fraction of the genome by inducing viable hypomorphic V. Jakkamsetti and others J Physiol 602.11 alleles, which regularly create living mice with measurable phenotypes.Moreover, in each pedigree, an average of 60 ENU-induced coding/splicing changes are examined in all zygosities.The causative mutation is usually identifiable using the automated mapping software of Mutagenetix (Linkage Analyzer), given a panel of 30-50 G3 mice.Each week 500-700 mutations can be examined in this way, within a total of 7-10 pedigrees.The alternative, i.e. the expansion of 600-900 single pedigrees per week, each bearing a mutation and tested for our purposes, would be impractical.Even if only 20 mice were generated per pedigree, we would not be able to phenotype 12,000-18,000 mice per week.The multiplex analysis permitted by the ENU approach makes it much more powerful than the single mutation approach.
The efficiency of our method (15 super-or sub-performance motor genes identified after ∼30,000 mice studied -or one gene for each ∼2000 mice) is comparable to other reports that also use ENU mutagenesis and behavioural screening for sleep (Funato et al., 2016) (two behaviour-modifying gene mutations reported for ∼8000 mice) or by comparing mice of different strains (Kumar et al., 2013) (one gene mutation reported for ∼1000 mice).
In summary, whereas several years of effort were once needed to determine the source of a phenotype via positional cloning, we have made it instantaneous.When a phenotype is detected in screening, its cause is generally known at the same time.We have also greatly increased the sensitivity with which quantitative trait mapping can be performed.Whereas formerly only phenotypes with a large effect size could reliably be ascribed to a causative mutation, we show that it is now feasible to instantly map mutations with effect sizes of a few per cent, in which there is considerable overlap between the phenotypic performance of wild-type, heterozygous and homozygous populations.These conditions have transformed forward genetics and made it possible to reliably determine which mutations cause increases as well as decreases in rotarod scores.This type of work cannot be accomplished in human populations, in outbred mice or in recombinant inbred mapping projects, where mutation density greatly exceeds the density of ENU-induced mutations.

What is to be considered 'superperformance'?
As with the term 'gain of function' , 'superperformance' depends on the type of 'performance' that one refers to, since this term applies to various biological levels of observation.For example, increased time on a rotarod may be due to greater paw movement velocity, or decreased variance of paw position at low rod speed, or even to the opposite variance at high speed.In several publications (Piochon et al., 2014;Sheppard et al., 2022;Vergouts et al., 2015), the relationship of these parameters with, for example, linear gait velocity may not be straightforward or may even appear counterintuitive.A mouse with a broad-based gait will do well on a slippery rock but poorly on a narrow liana.Thus, the quantification of motor performance upon a complex motor behaviour such as rotarod rate of learning and time to fall is likely to prove more translatable to humans than a simpler task (Jakkamsetti et al., 2021).Performance on a rotarod involves multiple motor skills including forward locomotion, balance and learning of complex paw patterns.A more mechanistic understanding of the impact of gene mutation on specific motor modalities will benefit from reductionist assays on CRISPR mice generated with the same performance enhancing mutations

Function of the genes associated with enhanced motor performance or motor learning
We identified Rif1, Tk1, Fan1 and Mn1 as genes that, when mutated, lead to motor superperformance.We have also uncovered an additional 6096 genes associated with a faster than expected (i.e.relative to other mutant mice) motor leaning rate.
The available information about the normal function of Rif1, Tk1, Fan1 or Mn1 is uneven.For example, whereas these genes are differentially expressed in brain areas potentially relevant to locomotion and whereas all have been implicated in nucleoside phosphorylation, more significant evidence is available regarding the function of Rif1.Rif1 regulates several DNA damage response processes, particularly in relation to double stranded break (DSB) repair, as it influences repair pathway selection between non-homologous end-joining (NHEJ) versus homologous recombination repair (HRR) (Scully et al., 2019).DNA repair is related to synaptic strength: loss of polymerase µ function can induce less efficient but more conservative NHEJ repair, resulting in increased mitochondrial respiration efficiency and enhanced hippocampal long-term potentiation (Lucas et al., 2013), which may underlie synaptic potentiation evoked by repeated motor activity (Whitlock et al., 2006), whereas DSB formation may be a physiological event that leads to early-response gene expression relevant for experience-driven changes to synapses (Madabhushi et al., 2015).Additionally, RIF1 influences chromatin function by directly binding to G4 DNA structures (Kanoh et al., 2015).Rif1 regulates the formation of telomeric and transcriptional RNA-DNA hybrid R-loop structures, which are sources of endogenous DSB (Tubbs & Nussenzweig, 2017) relevant to human diseases including neurological ones (Richard & Manley, 2017).Thus, the Rif1 mutation A salient association of the 6096 genes related, when mutated, to relative accelerated motor leaning is with histone H3 methyltransferase activity.Whereas numerous additional activities may be relevant given the large number of these genes, mouse histone regulation via methylation has been implicated in certain forms of synaptic plasticity (Shen et al., 2016), including hippocampal changes in relation to habituation to novel environments (Collins et al., 2019).
In summary, whereas we might have entertained a priori that ion channel and other membrane function genes would be best situated to exert the changes of neuronal function necessary for superperformance or increased motor learning, we find that the most robust associations point to DNA regulatory aspects.

Evolutionary implications
Given the rate of locomotor-impacting mutations we have identified and the estimated spontaneous mutation rate, it is possible to conjecture the probability of appearance of a modification in locomotor behaviour over time.Table 2 illustrates the estimated spontaneous mutation rates of several organisms.The table reflects wide variation and the values should be taken with further caution because the genetic background probably influences the mutation spectrum, in addition to a potential for segregating variation among inbred mouse strains (Dumont, 2019).If we take one estimate from Table 2, the probability of a homozygous mutation happening by chance through spontaneous mutations is (7.9 × 10 −9 ) × (7.9 × 10 −9 ) or 6.24 × 10 −17 .Taking into account our results indicating that 0.15% and 0.27% of all the homozygous ENU mutations studied modify locomotor performance and motor learning phenotype, about (7.9 × 10 −9 ) × (7.9 × 10 −9 ) × 0.15) mutations, or 9.36 × 10 −18 (1.19 × 10 −7 %) of mutations per replication may endow a superior phenotypic impact.Similarly, 2.13 × 10 −7 % of mutations per replication may endow superior motor learning.

Technical considerations
In our study, similarly to other forward genetics projects, the number of homozygotes available for experimentation can in practice be unpredictable.Since behavioural testing conducted prior or blindly relative to whole exome sequencing, the actual number of homozygotes during testing is unknown, with the possibility of a limited number.To provide more statistical rigour to mapping phenotype to DNA loci, we used Linkage Analyzer (Xu et al., 2021), which statistically tests the association between an assay and genotype for all mice at every mutation site in every pedigree, using the dominant, additive and recessive models of inheritance.In the absence of large normative datasets, the application of Linkage Analyzer allows for the identification of mutations with a small number of homozygotes that may be missed using traditional statistical analysis.However, a limitation of our large-scale screening study is the testing over six trials since mutations that affect late motor learning could be missed.

Conclusions
Despite the ever-increasing diversity of neurological disease mechanisms elucidated, motor impairment remains a common and poorly treatable outcome.For example, the motor consequences of stroke alone account for over one half of all the neurological disability in the world (World Health Organization., 2006).In contrast to this, some people and animals are endowed with supernormal motor ability stemming from spontaneous mutations in genes such as myostatin (Schuelke et al., 2004).Disorders as different as muscular dystrophy and spinal muscular atrophy benefit from myostatin blockade in mice (Bogdanovich et al., 2002;Long et al., 2019), suggesting that genetic or pharmacological modulation outside the primary disease locus can mitigate motor impairment.Our results suggest that superperformance can be identified at a measurable rate.After further mechanistic work, it may then be possible to induce it to augment limitations in motor performance or recovery, thus mitigating illness or disability, even when due to disparate causes or mechanisms.

Figure 1 .
Figure 1.Systematic motor evaluation of mutant mice Impact of number of mice tested per week (A-E) and rotarod scores for superperforming mice (F) normalized rotarod scores (left panel) and T-scores (right panel) of control wild-type mice over time.Most rotarod T-scores lie within the mean ±2 SD (50 ± 20) and almost all T-scores lie within 3 SD of the mean (50 ± 20).G and H, superperformance mutants.Specific mutations in the genes Rif1, Mn1, Fan1 and Tk1 induce a motor phenotype of superior performance on the rotarod test.Raw (unmodified) rotarod (G) and normalized (H) scores correspond to the fourth trial for Mn1 and Tk1 and the sixth trial for Rif1 and Fan1.For Mn1, raw scores of a single pedigree are shown (to allow statistical analysis) whereas the normalized scores include two pedigrees normalized to average wild-type mice tested at the same time.Abbreviations: WT: wild-type, REF: reference non-mutated mice littermate to ENU mutagenized mice, HET: heterozygous mutants, VAR: variants with homozygous mutations.n.s., Non-significant comparison; * P < 0.05; * * P < 0.01; * * * P < 0.005; * * * * P < 0.001.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 2 .
Figure 2. Brain expression and overall function of four superperformance genes A, Allen Brain Atlas in situ hybridization summary data for brain structures in the mouse.B, gene enrichment analysis of function.Blue shadowing indicates significance (P < 0.05) for adjusted P-values.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 4 .Figure 5 .
Figure 4. Motor behavioural cluster of ENU mutagenized mice A, PCA of mouse rotarod behaviour reveals two distinct clusters.The thick red and orange lines were determined by the density-based clustering algorithm DBSCAN.B, rotarod performance across six trials for the two clusters isolated in A and wild-type (WT) mice.C, average learning rate across six trials for the two clusters isolated in A and wild-type mice.Colours of bars represent the same groups as the colours in B. D, average performance across six trials.E, trial-by-trial data and average learning rates for the four mutations with superior performance shown in Fig. 1.Abbreviations: WT: wild-type, REF: reference non-mutated mice littermate to ENU mutagenized mice, HET: heterozygous mutants, VAR: variants with homozygous mutations.n.s., Non-significant comparison; * P < 0.05; * * P < 0.01; * * * P < 0.005; * * * * P < 0.001.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 6 .
Figure 6.Mutation rate associated with outlier motor performance A, average T-score for each mouse stacked longitudinally.B, average learning for each mouse stacked longitudinally.Red lines demarcate 3 SD above the mean.C, mutation number in a single mouse plotted against the number of surviving mice with the same number of mutations.* * * * P < 0.0001.[Colour figure can be viewed at wileyonlinelibrary.com]

Table 1 . Sub-and superperforming alleles uncovered by mutant mouse rotarod motor testing.
Ak8 or Ntng2 (Candidate Explorer favoured the former).Two separate mutations in the leptin receptor (Lepr) gene, both leading to obesity, caused poor rotarod performance.