Introduction

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

Transgenic models have proven to be a useful experimental approach to pain research, capable of revealing phenotypic differences in pain sensitivity between knockout mice with a mutated gene and wild-type control mice. In fact, transgenic studies of pain (see Mogil and Grisel, 1998, for a review) almost certainly represent the fastest growing technique in this field. The method has great power to render, with exquisite specificity, a single gene (and thus a single protein) nonfunctional. Nonetheless, knockouts have been criticized by some as poorly suited for resolving the issue for which they were intended: revealing the function of a particular protein (Routtenberg, 1995). The most common criticism is that compensatory effects of other genes may either mask the detection of the targeted gene's "phenotype", or alternatively be confused for the phenotype of the null gene. Virtually everyone agrees that the compensation problem is worrisome and points to the promise of second-generation "conditional/inducible" technology in ameliorating the impact of developmental compensations. Our focus herein is instead on challenges to the clear interpretation of transgenic experiments arising from the genetic background of the mutant mice, first highlighted by Gerlai (1996) (see also Crusio, 1996; Lathe, 1996). Although we have discussed this in passing a number of times, the increasing quantity of evidence signifying the seriousness of the problem compels us to describe these issues in more detail.

Essentially, two core problems exist. The first derives from the common practice of constructing transgenic mice as hybrids of two different mouse strains. In pain research as in most other fields, the default choices are 129, the source of virtually all embryonic stem (ES) cells, and C57BL/6, the most widespread inbred strain in biomedical research. The second problem is that one of these genomes is usually chosen as a pure background for the induced mutation. As will be discussed, confounds of transgenic experiments are most likely when a) 129 mice differ phenotypically from C57BL/6 mice, and b) C57BL/6 and/or 129 mice are phenotypically extreme responders compared with other strains. These warning signs are clearly present for transgenic studies of pain, with adverse consequences for the generalizability of knowledge obtained from transgenic experiments.

	Construction of Knockout Mice

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

Knockout mice are constructed by microinjecting transgenic ES cells, containing a gene mutated in vitro (the "transgene", usually rendered incapable of producing a functional protein product), into developing mouse blastocysts (see Zimmer, 1992). The ES cells incorporate into and develop with the host embryo cells, producing (when all goes well) a chimeric mouse whose tissues are derived partially from the injected cells and partially from the host cells. If the injected transgenic ES cells have by chance formed the germ cells of the chimeric mouse, the transgene can be passed on to future generations. The valuable germ line chimera is mated with wild-type mice to produce F₁ hybrid heterozygotes, 50% of which will contain one normal and one transgenic copy of the targeted gene. The transgenic F₁ hybrids are interbred to produce an F₂ hybrid generation including homozygous mutants (knockouts, with two copies of the mutated allele), heterozygotes (all with one transgenic allele), and wild-type mice in Mendelian proportions. Breeding strategies at this point are variable, but include a) continued interbreeding of knockout and wild-type populations as separate lines; b) rederivation of the three genotypes in each generation from heterozygotes; and c) repeated backcrossing of the F₂ hybrids with one and/or the other progenitor strain to produce congenic lines (see below).

Almost always, ES cells originally derived from one of the many substrains of the 129 mouse strain (see Simpson et al., 1997) are injected into a blastocyst of the C57BL/6 strain, which is also used as the wild-type breeding partner. Thus, most knockouts start as (129 × C57BL/6)F₂ hybrids. This standard procedure persists due to inertia; 129-derived ES cells exhibit extensive colonization of the host embryo, and C57BL/6 mice are highly successful breeders relative to 129. Furthermore, darkly pigmented C57BL/6 mice contrast well with albino or chinchilla-colored 129 mice, rendering the identification of chimeric offspring trivial. It should be noted, however, that ES cell lines can be successfully derived from other mouse strains (e.g., Lemckert et al., 1997), and albino/albino chimeras can be identified via genotyping.

	The Hitchhiking Donor Gene Confound

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

As pointed out by Gerlai (1996) and coined by Crusio (1996), this standard breeding strategy yields a linkage disequilibrium between the transgene and surrounding genes, producing a "hitchhiking donor gene confound". Contrary to the common assumption that knockout mice are isogenic (i.e., genetically identical) to their wild-type controls save for the transgene itself, in fact, the genomes of F₂ hybrid wild-type and knockout mice differ. Both genotypes consist of a random mix of strings of 129 and C57BL/6 alleles on chromosomes other than the one containing the targeted gene. However, on the targeted chromosome, knockout mice possess a significant amount of genetically linked or "hitchhiking" genetic material from the 129 donor surrounding the transgene. In contrast, wild-type mice have C57BL/6-derived DNA here. Congenic line construction aims to rectify this situation, standardizing the genetic background of these mice by removing gene alleles from one strain (usually 129) and replacing them with alleles from the other (usually C57BL/6). This is accomplished by repeated backcrossing of (129 × C57BL/6)F₁ hybrid mice to the C57BL/6 strain. After each backcross, 50% of the 129 alleles are lost. However, the complete loss of 129 alleles is limited by the probability of recombination; even after 12 backcrosses (and over 2 years of additional breeding) a 16-centimorgan long segment of 129-derived chromosome remains surrounding the transgene. A segment this size represents 1% of the mouse genome and may contain hundreds of genes. The confound is greatly compounded by the fact that, due to competition to publish in a timely fashion, the vast majority of these studies test the knockout mice well before congenic status is reachedin most cases at the F₂ hybrid stage.

The resultant problem, simply put, is that the differential phenotype of the knockout mice may be due to the mutated gene or to one or more of the hundreds of hitchhiking genes. Of course, a trait-relevant hitchhiking gene can only be problematic if it is functionally polymorphic between the strains in question, 129 and C57BL/6. The probability of encountering genes with functionally relevant alleles depends, of course, on how many polymorphic genes influence the trait. If 129 and C57BL/6 mice display large differences on a pain-related trait, the existence either of few genes with large effects on pain and/or many genes with small effects on pain in the same direction is implied. In the former case, the smaller chance of such a gene being present in the hitchhiking region would be associated with a larger confounding effect of that hitchhiking gene. In the latter case, the confounding effect would be smaller but of higher probability. Either way the overall chances of confound (i.e., false attribution of phenotypic differences between wild-type and knockout mice to the transgene) are greater if 129 and C57BL/6 mice show large phenotypic differences. It remains possible, but with lower probability, that potentially confounding hitchhiking genes may exist in situations where 129 and C57BL/6 mice do not differ phenotypically because the opposing allelic effects of such genes have cancelled each other out.

Befitting their divergent ancestries (Festing, 1996), these two strains are known to show multiple and significant neuroanatomical and behavioral differences. In at least one case, systematic analysis of these strainsconcurrently to F₂ hybrid knockouts and incipient congenics to both backgroundsled one group to conclude that the decreased locomotor activity they observed in their mutants should be attributed to a hitchhiking donor gene and not the transgene itself (Kelly et al., 1998). At the root of the problem was that 129 mice displayed 4-fold less basal locomotor activity relative to C57BL/6 mice and were unable to learn to perform the rotarod task. As will be illustrated below, we now know that the pain-related phenotypes of the 129 and C57BL/6 strains are usually significantly different from each other, and in many cases these differences are virtually identical to knockout versus wild-type differences.

	The Epistasis Confound

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

Variability in almost all traits is due to both environmental and inherited factors and their interaction. Within the genetic component lies additive genetic variance (the effect of individual genes), dominance deviation, and epistasis (Falconer and Mackay, 1996). Epistasis refers to the interaction between trait-relevant genes, whereby allelic status at one gene affects the influence on the trait of alleles at another gene. With regard to transgenic studies, alleles within one genetic background may significantly interact with the mutated gene in ways that significantly affect the observed phenotype of the mutant mice. This is not merely a theoretical concern; there are now multiple demonstrations of exactly this sort of interaction. In a small number of cases, the same knockout was deliberately placed on more than one genetic background with impressive consequences. For example, Threadgill and colleagues (1995) placed a null mutation of the epidermal growth factor receptor (Egfr) gene on three genetic backgrounds: CF-1, 129/Sv, and CD-1. Although the mutation was lethal in all cases, CF-1 mutants died soon after implantation, 129/Sv mutants died at midgestation, and CD-1 mutants lived for up to 3 weeks postnatally (see also Sibilia and Wagner, 1995). In other cases, the same mutation placed on different genetic backgrounds by different investigators yielded differential phenotypes, although variable testing procedures in each laboratory could also be responsible. Finally, in a disturbingly large number of cases, investigators have reported the alteration or loss of a previously observed phenotype as the knockouts are made increasingly congenic. Perhaps the most striking example of this phenomenon is the progressive failure to replicate the increased alcohol consumption phenotype of serotonin-1B receptor knockout mice (Crabbe et al., 1996). What is striking is that the phenotype disappeared as the mutants were made congenic to a different 129 substrain, not C57BL/6 as is more often the case. The ES cells were derived from the 129/SvPas substrain and the mutant was originally placed on a 129/Sv-ter background, but for backcrossing the 129/SvEvTac substrain was used. As more and more of the genome was replaced with 129/SvEvTac alleles, alcohol consumption decreased (Phillips et al., 1999), consistent with known 129 substrain differences in alcohol preference (Crabbe et al., 1999).

There is now considerable agreement among investigators that background strain can importantly influence the presence, size, and even direction of mutant phenotypes. Unlike the hitchhiking donor gene confound, the problem of epistatic interactions between mutant genes and those in their background will not be solved by conditional/inducible transgenic technology. The most often proposed solution, therefore, is for everyone to agree on a common genetic background so that transgenic studies may be more easily compared with each other. The recommended background seems to be C57BL/6 (Crawley et al., 1997; Gerlai, 1999) because of a) the relative abundance of existing phenotypic data in this strain, b) its relatively high fecundity, and c) its experiment-facilitating phenotypes on a number of traits important to behavior geneticists (e.g., high open field activity, good learning ability, high drug self-administration). Investigators have also been encouraged to consider placing their knockouts on both 129 (of a substrain appropriate to the original ES cell used) and C57BL/6 backgrounds specifically for purposes of comparison. Finally, a working group of geneticists proposed that a (129P3/J × C57BL/6J)F₁ hybrid background be used (Banbury Conference on Genetic Background in Mice, 1997). Although there is much to recommend this latter proposal, it has not been widely adopted. As will become clear, we believe that none of these options will suffice for pain research.

	Responses of 129 and C57BL/6 Mice on Pain-Related Phenotypes

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

The major focus of research in our laboratory is the determination of sensitivity to nociceptive and analgesic phenotypes of inbred mouse strains (see Mogil, 1999, for review). These data are collected so that genetic correlations can be investigated and so appropriate progenitor strains can be identified to facilitate gene identification via quantitative trait locus (QTL) mapping (e.g., Mogil and Adhikari, 1999; Mogil et al., 1999a,b). Figure 1 illustrates the relative sensitivity of 129 (specifically the 129P3/J substrain, formerly known as 129/J) and C57BL/6 (specifically, C57BL/6J) mice compared with other inbred strains (including A/J, AKR/J, BALB/cJ, C3H/HeJ, C57BL/10J, C58/J, CBA/J, DBA/2J, LP/J, RIIIS/J, SJL/J, SM/J, SWR/J strains; all obtained from The Jackson Laboratory, Bar Harbor, ME; the position of DBA/2 mice is shown explicitly for reasons described below) in a variety of assays of nociception, hypersensitivity (i.e., hyperalgesia or allodynia), and analgesia. One should bear in mind the caveat that although the 129 substrain tested in our laboratory was 129P3/J, the genetic background of most- knockout mice includes alleles from other 129 substrains (e.g., 129S1/Sv-+^p+^Tyr, 129S6/SvEvTac, 129X1/SvJ, 129P2/OlaHsd) and substrain differences have been documented (Simpson et al., 1997). Most of these differences are found in noncoding microsatellite markers and not genes, however, and we were unable to demonstrate significant phenotypic differences among two 129 substrains (129P3/J and 129S6/SvEvTac) in either thermal nociception or morphine analgesia (Mogil and Wilson, 1997).

View larger version (57K): [in this window] [in a new window]   Fig. 1.   Sensitivity of 129P3/J, C57BL/6J, and DBA/2J mice compared with other inbred strains in 30 assays of nociception, hypersensitivity, and analgesic manipulations (see text for citations). See key for symbols. Strains are placed according to their standardized score (i.e., z-score) on that assay relative to the mean of all strains tested. n.s., not significant. *Significant (P < 0.05) difference between 129P3/J and C56BL/6J strains. aAutotomy was found to be genetically correlated with assays of thermal nociception (Mogil et al., 1999b). bMore subjects have been tested since publication of the original findings (Mogil et al., 1999a); the difference is now highly significant. cBased on half-maximal analgesic dose (AD50) values for drugs tested at multiple doses and percentage analgesia scores for electroacupuncture and drugs tested at single doses. dMean absolute value z-scores (±S.E.M.) on the 26 to 30 assays in which these strains were tested. This measure, described in Mogil et al. (1999a), is an index of average deviation from the grand mean of all strains, or phenotypic distinctiveness. Higher values indicate greater deviation from the grand mean and thus increased distinctiveness. The strain ranking 1 on this measure (C57BL/6J) is thus the least representative of the 12 strains examined; the strain ranking 12 (DBA/2J) is the most representative.

In all assays of thermal nociception examined, including tail withdrawal from hot or cold water or from a focused 12.5-W light bulb, and hindpaw withdrawal from a focused light bulb in Hargreaves' test, C57BL/6 mice were more sensitive than 129 mice, showing shorter latencies to make a withdrawal response (Mogil and Adhikari, 1999; Mogil et al., 1999a; Wan et al., 2000). C57BL/6 mice were within the two most sensitive strains tested in four of the five assays, and 129 mice were moderately sensitive at best. In assays of chemical nociception, including the writhing, formalin, bee venom, and capsaicin tests (Mogil et al., 1999a; Lariviere et al., 2000), C57BL/6 mice were variably sensitive, but 129 mice were among the two least sensitive strains in each. Finally, in two assays of mechanical sensitivity, the tail-clip test and the von Frey test, these two strains showed significant differences (Mogil et al., 1999a; Lariviere et al., 2000).

Figure 1B shows the relative performance of the two strains in models of pain hypersensitivity, including peripheral nerve injury (Chung model)-induced thermal hyperalgesia and mechanical allodynia and dynorphin-, carrageenan-, and bee venom-induced hypersensitivity (Mogil et al., 1999a; Lariviere et al., 2000). Whereas C57BL/6 mice are fairly resistant to developing thermal hyperalgesia, 129 mice were found to be exquisitely sensitive to the development of mechanical allodynia.

Strain surveys of sensitivity to analgesic manipulations also show that 129 and C57BL/6 mice exhibit differential phenotypes (Mogil and Wilson, 1997; Flores et al., 1999; Kest et al., 1999; Wan et al., 2000; Wilson et al., 2000). Figure 1C illustrates the fact that C57BL/6 is relatively insensitive to a number of analgesic compounds. By contrast, the 129 strain is rather sensitive to analgesics. These strain differences involve differential potency (by up to 100-fold), efficacy, and duration of analgesic action (see also Crain and Shen, 2000).

Furthermore, we (Kest et al., 2000) have supported and extended the recent findings of others (Kolesnikov et al., 1998; Crain and Shen, 2000; E. Bilsky, E. Castro, M. Ibrahim, T. Malan, and F. Porreca, unpublished data) that 129 mice do not appear to exhibit analgesic tolerance to or dependence on morphine. In these previous studies, the 129/SvEv substrain was tested; we show that 129P3/J mice similarly lack morphine tolerance and dependence in stark contrast to C57BL/6 and most (but not all) other strains. Other documented 129 versus C57BL/6 phenotypic differences that may manifest themselves in transgenic studies of pain include locomotor activity, anxiety, learned helplessness, and conditioned avoidance (see Crawley et al., 1997).

The relative differences between 129, C57BL/6, and other strains with respect to the phenotypes described in Fig. 1 are illustrated via a principal components analysis (see Fig. 2). The large distance between 129 and C57BL/6 strains indicates their overall distinctiveness relative to each other with respect to pain-related traits.

View larger version (19K): [in this window] [in a new window]   Fig. 2.   Principal components analysis of nociceptive and analgesic phenotypes listed in Fig. 1. Individual strains are plotted along the two derived dimensions that explain the most variance in the data. Only strains used in the vast majority of strain surveys were considered. In this analysis, phenotypically similar strains are represented as geometrically closer. The 129P3/J and C57BL/6J strains are similar to a small number of adjacent strains but dissimilar to most strains and highly dissimilar to each other. On the other hand, the central location of the DBA/2J strain indicates that it is most similar to (i.e., representative of) all the strains shown.

	Implications and Recommendations

Top Abstract Introduction Construction of Knockout Mice The Hitchhiking Donor Gene... The Epistasis Confound Responses of 129 and... Implications and... References

The implications of these sometimes extreme differences in pain-related phenotypes for transgenic studies are obvious. As previously discussed, the possibility of the hitchhiking donor gene confound is higher when knockout mice resemble 129 mice and wild-type mice resemble C57BL/6 mice. Table 1 summarizes the direction of significant knockout versus wild-type differences in all studies known to us where transgenic mice were tested for nociceptive sensitivity in common assays featuring a significant 129 versus C57BL/6 strain difference. As shown, "129-like" knockout mice are seen far more often than "C57BL/6-like" knockout mice. In many cases, the knockout versus wild-type differences described in these studies are essentially phenocopies of 129 versus C57BL/6 data collected in our laboratory, although this in and of itself is not proof of confound. It should also be noted that since most transgenic studies of pain have investigated genes thought to play a role in pain production, the absence of such genes would tend to reduce pain sensitivity. It is unfortunate, therefore, that 129 mice are already less sensitive to pain in many assays than C57BL/6.

Several at least partial solutions to the hitchhiking donor gene confound have been proposed (e.g., Zimmer, 1996; Mogil and Wilson, 1997; Crusio et al., 1998; Gerlai, 1999); a full discussion of these is beyond the scope of this review. These solutions include 1) the use of conditional/inducible knockouts; 2) rescue experiments, either using gene "knock-ins" or the missing protein itself; 3) construction of transgenic mice on pure backgrounds or on F₁ hybrids of two pure backgrounds; 4) comparing the phenotypes of knockout mice derived from alternate breeding schemes; and 5) simply comparing the phenotype of the mutant mice to that of the progenitors extemporaneously. The most elegant of these solutions is likely the use of inducible mutations, in which the gene of interest is switched on or off by the experimenter at will. Within-subject comparisons can then be madethe ideal control. Although long promised, such technology is still not widely used. Once it is generally available many existing transgenic studies of pain may need to be performed again using second-generation mutants. We have argued previously that testing the progenitor strains along with wild-type and knockout mice is a simple way of evaluating the likelihood of interpretational confounds (Mogil and Wilson, 1997). However, we are aware of only a single transgenic study of pain that has presented such data (Ugarte et al., 2000).

Hitchhiking donor genes aside, the data presented herein suggest that neither 129 nor C57BL/6 is a representative genetic background on which to place a mutation of relevance to paineither targeted or induced by chemical mutagenesis. It has long been recognized that 129 is an outlier on a wide variety of morphological and behavioral traits, with many substrains showing a deficient or absent corpus callosum, for example (see Festing, 1996). By contrast, many behavior geneticists have been far more sanguine about the use of the C57BL/6 strain as a genetic background despite the long-known genotypic uniqueness of the C57 and C58 strain families (e.g., Taylor, 1972). Although the C57BL/6 strain is certainly useful for many behavior genetic studiesfor example, it is the only major strain that will perform well in the Morris water maze (see Crawley et al., 1997)many of its phenotypic characteristics contrast sharply with other inbred and outbred mouse strains. We argue that a trade-off exists between choosing a background strain to maximize one's likelihood of demonstrating a mutant phenotype (good) and assessing that phenotype on a background where rare alleles interact epistatically with the mutation (bad).

Thus, since one must ultimately choose some genetic background, we believe that the ability to generalize one's results would be best served by choosing an average-responding strainone with common alleles at trait-relevant genesas has been recommended (Crawley et al., 1997). Another advantage of using a strain with an average phenotype is that effects of the mutation in either direction may be detected, uncomplicated by floor or ceiling effects. The strain survey data collected in our laboratory can be used for selecting such a strain. We have suggested that the DBA/2 strain is a good genetic background for studies of nociceptive sensitivity since its phenotype is consistently moderate (Mogil et al., 1999a), a finding supported by the reanalysis shown in Figs. 1 and 2. Of course, it cannot be assumed that the DBA/2 strain is an average responder for any phenotype under investigation; this should be tested empirically in each case. Producing multiple knockout lines on different genetic backgrounds concomitantly is a superior, if less practical, strategy. Other options include placing null mutations onto F₁ hybrid or outbred (i.e., randomly bred) backgrounds; such mice are far healthier and more representative of "mousedom" than inbred strains lacking genetic heterozygosity. Trade-offs exist here too, however. Just as the genetic homozygosity of inbred mice is an unnatural situation, so is the heterozygosity at every locus characterizing F₁ hybrids. Outbred mice from commercial suppliers are partially inbred (due to breeding bottlenecks) and genetically undefined. In theory, wild-derived inbred mice (e.g., CAST/Ei, SPRET/Ei) are an excellent choice but in practice will present large technical hurdles for pain researchers because of their hyperactivity.

It has been argued by Gerlai (1996, 1999) that the most interesting results are to be obtained from studies that specifically examine interactions of the gene of interest with the genetic background. If a more representative genetic background is used, we will learn about the function of the gene in the typical mouse. If one cannot generalize even to the murine species as a whole, implied generalizations to other species (i.e., humans) are increasingly suspect. In some cases, however, it will be desirable for findings to be obtained in strains of mice more prone to chronic pain states. For instance, the DBA/1 strain of mice is highly susceptible to collagen-induced arthritis (Paska et al., 1986). By comparing the effects of mutations on DBA/1 versus other pure backgrounds, one could ask the important question of how the results from a "normal" strain generalize to a susceptible strain. We hope that such complex issues will be eventually addressed.

Transgenic technology has exhibited great power to study the role of proteins for which no other methods of investigation exist. Undoubtedly, much has been learned from these studies, and this will continue to be the case. We do hope, however, that the data presented herein will convince those who perform and read transgenic studies of pain of the limitations of the method as currently practiced. The transgenic technique has considerable potential, but this potential must be approached with just as much caution. Possible confounds aside, knockouts are not a magic bullet; findings from transgenic studies must be interpreted within the context of converging evidence derived from alternate (i.e., conventional) biological strategies.