Chronic pain is a major healthcare problem affecting the daily lives of millions with enormous financial costs. The notorious variability and lack of efficient pain relief pharmaceuticals provide both genetic and therapeutic challenge. There are several genetic approaches that aim to uncover the molecular nature of pain phenotypes into their genetic components. Gene mapping using model organisms for various pain phenotypes has led to the identification of novel genes affecting susceptibility and response to pain stimuli. Translational studies have succeeded to tie those genes to human pain syndromes, thus suggesting new targets for drug discovery. In this short review, a perspective on pain genetics and the trajectory from pain phenotype to pain gene involving fine-mapping strategies, bioinformatic analysis and microarray profiling alongside human association analysis will be introduced. This integrated approach has led to identification of CACNG2 as a novel neuropathic pain gene affecting pain susceptibility both in mice and humans. It also serves as a prototype for efficient and economic discovery of pain genes. Comparisons to other methods as well as future directions of pain genetics will be discussed as well.
Pain is medically defined as ‘an unpleasant sensory and emotional experience associated with actual or potential tissue damage, or described in terms of such damage’ . The perception of pain is an adaptive response that signals the presence of damaging and life-threatening events (e.g. traumatic injury, a potentially injurious environment and infection) [2, 3]. The inability to detect noxious stimuli confers a serious risk to our survival and well-being. Individuals suffering from congenital abnormalities that render them incapable of detecting painful stimuli often face conditions that can be life threatening . Pain can be divided into three classes: (i) nociceptive, (ii) inflammatory and (iii) neuropathic. Nociceptive pain such as a stubbed toe results in a felt sensation (pain) matching the stimulus (noxious). This is a classic example of ‘normal’ pain. Pain arising, for instance, from tissue injury or infections involves increased responsiveness because of the release of inflammatory mediators in the injured tissue. Thus, the felt sensation (pain) may no longer match the stimulus (non-noxious). In inflammatory pain, a reduction of the threshold of the nociceptors is observed. Both normal and inflammatory pain represent a key feature of the pain system – its operation as an alarm bell. The pain felt provides warning and protection . Neuropathic pain, however, reflects an abnormal functioning of the pain system. Pain no longer acts as an alarm factor; rather, it becomes a problem by producing a ‘false alarm’. While conventional medicine copes with normal and inflammatory pain quite efficiently, neuropathic pain remains inadequately treated.
The International Association for the Study of Pain defined neuropathic pain as ‘pain initiated or caused by a primary lesion in the nervous system’ . Pain following nerve injury is paradoxical. The paradox is that in many cases the outcome of nerve damage (trauma, pathology, etc.) is not to eliminate painful signals. Rather, nerve trauma or disease resulting in cut axons is frequently associated with the initiation of symptoms that include: (i) tenderness to stimuli in the denervated body part (hyperalgesia and allodynia), (ii) pain evoked by normal movement or weight bearing and (iii) spontaneous pain .
The fact that pain is present (and even increases) after nerve injury indicates the complex nature of the changes and events taking place in both peripheral nervous system (PNS) and central nervous system (CNS), following radical damage to the nerves. Several pathological processes affecting peripheral nerves, sensory ganglia and CNS structures can induce neuropathic pain.
Pain usually results from activation of nociceptive afferents by actually or potentially tissue-damaging stimuli. Pain may also arise by activity generated within the nervous system without adequate stimulation of peripheral sensory endings. This may be due to disease, trauma or anything, that affects the peripheral nerves. Frequently, the results of such injuries are the development of chronic, often intractable pain.
This category includes various chronic ailments, for example, sciatica pain, painful diabetic neuropathy, and phantom limb pain.
Phantom limb pain
The ‘phantom pain’ phenomenon is a fascinating case of neuropathic pain. The amputation of a limb is commonly followed by the sensation that the absent body part is still present. Non-painful phantom sensations may take a specific position, shape, or movement of the phantom, feelings of warmth or cold, itching, tingling, or electric sensations, and other paraesthesias. Pain in the body part that is no longer present occurs in 50–80% of all amputees . Phantom pain is common after limb amputation, breast removal and also in body parts that have been denervated but are still present (anesthesia dolorosa) [8-11]. Phantom pain has many features: at one end of the spectrum it is limited to simple, short-lasting and rarely occurring painful shocks in a missing body part; at the other end of the spectrum it can be a constant, excruciatingly painful experience during which the individual has a vivid and intense perception of the missing body part. It seems to be more severe in the distal portions of the phantom and can have a number of characteristics such as stabbing, throbbing, burning or cramping. Its onset can be immediate, but it may also appear for the first time many years after the amputation. From a behavioral perspective, although many amputees exhibit psychological symptoms such as anxiety, depression and isolation, there is no evidence that the phantom limb per se represents a psychological disturbance. The variability of pain experience in general poses a significant obstacle when it comes to research. In order to address this issue, several animal pain models have been developed bearing the advantage of uniformity and controlled conditions.
Neuropathic pain models in rodents
Rodent models have proven valuable in identifying pathophysiological mechanisms for behavioral and emotional disorders . In the pain research field, extensive research has characterized the behavioral phenotypes for an array of rodent models of pain including neuropathic pain [13-15]. Rodent models can be divided into two primary groups according to (i) localization [dorsal or ventral root, dorsal root ganglion (DRG) or peripheral nerves] and (ii) the type of lesion (transection, tight or loose ligature, crush, etc.) . It is important to note that because of the lack of communication, a direct evaluation of neuropathic pain in rodents is not possible. Instead, an alternative approach is used: observation of changes in response threshold to mechanical and thermal stimuli, self-attack of the denervated limb (autotomy) and other measurements are used as indications for the presence of neuropathic pain. Wall et al.  have presented the neuroma model of neuropathic pain as an efficient model assessing the appearance of spontaneous pain following radical nerve injury. The neuroma model mimics the clinical symptoms of phantom limb pain. A complete denervation of the paw is achieved by cutting both the sciatic and saphenous nerves. As a result, some animals present a dramatic change in their behavior, developing ‘autotomy’, a behavior that comprises scratching and biting of the numb paw. Pain level is monitored by scoring autotomy. The neuroma model has successfully been applied to both rats and mice for the purpose of physiological and genetic analyses [18-20]. Moreover, in a comprehensive study of various pain models in animals, Mogil et al. [21, 22] have supplied persuasive description of pain behavior and its variability among mouse strains suggesting their usefulness in both behavioral and genetic studies. The variability of the neuroma model in various mice strains is presented in Fig. 1. The assessment of pain behavior in animals also enabled the pre-selection of the sex (i.e. females vs males) based on the observed variation . The results of the above studies gave clear indication of the heritable nature of pain phenotype and indeed triggered several studies aimed to map genetic components affecting pain behavior [24, 25].
In general, genetic studies of pain comprise experiments where the focus is either on genes' role in pain processing and signaling or on the effect genetic factors may have on the predisposition to pain measured as severity, onset, latency and response to pain treatments and so on. In this review, the focus will be on the identification of genes associated with the susceptibility to neuropathic pain or in other words, how individualized differences (reflected in their genetic makeup) affect pain behavior.
Neuropathic pain, including phantom pain, is affected by both the nature of the neural injury and psychosocial factors. Its notorious variability among individuals, even when the underlying nerve damage is identical, gave the indication of a significant genetic contribution to the amount of pain felt . This variability, which is mainly due to genetic differences, is the key feature in pain gene identification.
Established rodent models for chronic pain can be combined and applied in a preferred mapping population, allowing the mapping of genetic factors associated with pain under verified pain exposure. Using this approach enables to bypass the pain variance obstacle among human cohorts. Such an approach has yielded the identification of loci affecting pain that will be discussed later in this review.
A major and a known challenge of pain geneticists is the nature of pain as a complex trait. In contrast to Mendelian traits in which a genotype–phenotype correlation is rather clear and defined, complex traits exhibit continuous phenotypes that obscure the association of genotype with the investigated phenotype . The complex inheritance of such traits is attributed to a number of contributing genes and genetic variants (even hundreds), environmental factors and gene–environment interactions .
A major advantage of mapping genetic factors associated with any trait (positional cloning) lies in the fact that genome-wide scan technology represents unbiased hypothesis testing. It requires no a priori assumption about the underlying physiology to be made. Despite significant successes in quantitative trait locus (QTL) detection, the actual identification of the gene behind the QTL remains a daunting task. There are several ways to detect and map QTL, but all are multistage processes. The use of model organisms such as rodents can help to maximize the chances of revealing QTLs and their underlying gene(s) .
In the section below, the arsenal of mapping strategies that were considered in the quest for pain genes identification will be briefly described. For further information and for those who considered the implementation of animal models in their research, a comprehensive description of mapping strategies and mapping statistical considerations appears in detail in Refs [29-32].
Traditionally, the starting point of QTL detection required the creation of a mapping population, usually either an F2 or backcross (BC) population. Subsequently, the estimation of the QTL location in the genome and the definition of its confidence interval (CI) are determined. This interval mapping procedure is performed using the same population used for the previous stage. Applying this procedure each chromosome is scanned, looking at pairs of markers at a time. The assigned probability value logarithm of odds (LOD) score indicated on the most significant regions to contain a QTL (the highest LOD score value). The next significant step toward QTL identification lies in the application of the appropriate fine-mapping strategy. The choice needs to take into consideration the trait mode of inheritance, resource availability and more [30, 32]. There are several strategies aimed at providing QTL fine mapping at high resolution. Starting with the preliminary steps described above, fine mapping is needed in order to reduce the mapping interval and the number of relevant candidate genes.
Chromosome substitution strains (CSS)
CSS has been suggested as a valuable resource for QTL mapping. Those lines are based on a donor strain (D) and a host strain (H). Each CSS carries both copies of chromosome 1…19 or sex chromosome from the donor strain, whereas all other chromosomes from the host strain are intact and homozygous. By this way, any effect can be attributed to a defined chromosome . Fine mapping using CSS can be carried out in the form of congenic line production. This step involves further backcrossing of the CSS in which the QTL was mapped. The results would be a set of CSS – recombinant lines each carrying a relatively small single segment from the donor strain on the background of the host strain (i.e. congenic line). It needs to be noted though that the production of the above lines is time consuming and subsequent genotyping needs to be applied .
Recombinant inbred lines (RILs)
RILs considered to be an excellent genetic resource. RILs are formed by intercrossing a pair of inbred strains followed by recurrent inbreeding of the progeny producing a panel of inbred animals, each with a unique combination of the progenitor genomes. Once a series of RIL is genotyped, this information will be useful for future experiments. In case where the trait of interest showed contrasting phenotype in RIL progenitors, using RIL for mapping an effect can be highly productive as a QTL can be mapped to a relatively small interval. The high mapping resolution of the RIL was the basis for more advanced mapping strategies.
Collaborative Cross (CC)
In order to exploit the high mapping resolution of the RIL and to overcome the inherent limitation of RIL as it is based on two strains only, the CC has been developed. The construction of the CC was based on genetically diverse set of eight founder inbred strains. Systematic outcrosses of the founder strains followed by inbreeding resulted in a population that provide a unique opportunity to observe phenotypic variation via massive novel allelic combinations . The usefulness of the CC as a reference population tailored for the dissection of complex trait was exemplified for various quantitative traits such as behavior, body weight and blood chemistry .
Recombinant inbred segregation test (RIST)
The RIST approach applies the advantage of RILs. Assume that a set of RIL presenting contrasting phenotype for the trait of interest is selected resulting in QTL detection to a certain interval on a certain chromosome. The RILs obtaining recombination points in the interval harboring the QTL are crossed with both parental lines to generate two segregating populations. The QTL will segregate in one population but not in the other population with relation to the recombination point, thus localized to small interval. The resolution of the fine mapping is dictated entirely by the number of the recombinations found in the mapped interval. The main advantages of the RIST approach are: feasible number of animals required even for small effects and the time frame that requires only two generations .
Recombinant progeny testing (RPT)
RPT forms another useful fine-mapped strategy. Assume that a BC population is generated from two distinct inbred strains, and QTL is mapped. Individuals with various recombination points along the mapped intervals are isolated. The selected individuals are backcrossed to one of the original progenitors and a new population is produced and selectively phenotyped. The comparison between the new RPT families (each carrying a unique genetic pattern along the area of interest) reduced the CI to a smaller interval with a corresponding reduction in relevant candidate gene number. RPT strategy requires only three generations for fine mapping. Furthermore, it can be very useful where there is an interest in two particular inbred strains for a certain trait but with no available panel of RIL. The application of the RPT strategy has enabled the fine mapping of loci for various traits .
In silico analysis
A complementary approach to gene mapping is in silico analysis. Application of in silico tools can significantly contribute to the detection of genes and rare variants. Data mining and automated tracking of new knowledge facilitate locus mapping. In addition, in silico prioritization of candidate genes plays an indispensable role in dealing with linked or associated loci . The basic in silico mapping approach in the mouse relies on known differences between mouse inbred strains and uses a high density of markers to derive a strain distribution pattern that can be used for mapping. For example, assume that we look at a trait of interest for four inbred strains. Group 1 (A–B) shows certain phenotype and group 2 (C–D) presents contrasting phenotype. By scanning all known sequences of the four strains one may focus only on sequences (haplotypes or single SNP, for example) that show complete cosegregation of alleles and the phenotype in one group but not in the other. Doing so can speed up candidate gene prioritization [38, 39]. A highly informative reservoir is the Mouse Phenome Database . ‘In silico’ not only describes the search for particular information (data mining) that is stored in computer-based resources but also includes task such as organizing, analyzing, and predicting increasingly complex data arising from molecular and clinical studies with the aid of a computer. In recent years, the implications of in silico analyses have shown its efficacy in terms of detecting novel loci for various traits [41-43].
Genome-wide association studies (GWAS) were developed in order to address some of the shortcomings of traditional linkage tests, especially the low power of standard linkage analyses [44, 45]. Association means a relationship that is defined by the non-random occurrence of a genetic marker with a trait, which suggests an association between the genetic marker (or a marker close to it) and the investigated phenotype such as disease pathogenesis. This approach is hypothesis free, i.e. there is no pre-existing hypothesis about a particular gene or locus, and a null hypothesis that no detectable association exists. As in any classical association design, GWAS is based on employing two groups of unrelated affected and healthy individuals selected from the population. An allele is considered to be associated with the trait if it occurs at a significantly different frequency among cases when compared with controls.
A true association between a particular allele and a trait can occur in two situations. Association will obviously occur if the studied allele directly influences the trait. Such causative allele will have functional properties that increase the risk of developing the disease. Second, the studied allele might be indirectly associated with the disease by being in correlation with the causative allele. This correlation is termed linkage disequilibrium, i.e. the non-random association between alleles at different adjacent loci.
mRNA expression profiling
An independent approach applied for causal gene identification is whole transcriptome microarray analysis. The importance of microarray analysis lies in the actual discovery of expression alterations that relate to the trait of interest. As a complementary tool for gene mapping, expression profiling offers the opportunity to explore the functional consequences of a defined (but not completely characterized) genetic difference at the molecular level even before the identity of the causal locus itself is known . For example, gene expression profiling of the DRG in chronic pain models has revealed the alteration of various genes , some of which were subsequently shown to be key modulators of pain . Nonetheless, gene expression analysis using microarray technology suffers from well-known limitations including poor sensitivity and dynamic range, a requirement for substantial amounts of RNA, and a limited capacity to identify new transcripts or RNA splice sites.
An integrated approach for candidate gene selection
In principle, each of the methods specified above can stand on its own. An accelerating number of independently studies using either gene mapping, in silico mapping, mRNA profiling or GWAS are being published on a regular basis. Nevertheless, in this research, we aimed to show how the integration of various methods can lead to the identification of the gene(s) underlying a QTL, in a case such as neuropathic pain. The rationale, of course, is that the combination of different approaches can compensate for the caveats and disadvantages of any single method.
As a platform, this approach aims to facilitate gene discovery for many traits so long as the basic components of animal models and distinguishable phenotype are applicable and exist. Figure 2 presents a simple-to-follow scheme of the integrated approach. The actual ‘integration’ starts once a QTL that is/are associated with the trait has been mapped. Then, it is divided into two major levels of screening: (i) QTL analysis and (ii) gene expression analysis (Venn diagram – Fig. 2). For the QTL analysis, initial mapping, fine mapping and bioinformatic analysis are preformed in a sequential manner. Once the QTL was mapped to a manageable size interval (and number of candidate genes), the use of sequence databases enables further reduction of the number of relevant genes (, ).
In parallel to the QTL analysis, a second methodology is already in action. Gene expression analysis provides an independent source of information that can be of great value for the identification of the actual genes affecting a trait. The causative gene must be expressed in a trait-relevant tissue and may also exhibit varying expression levels for different states related to the trait. This analysis may highlight potentially relevant genes. This combination has the potential to reduce the list of candidate genes down to a single gene or a mere few [41, 51].
The results of the combined methods allow the selection of a leading candidate gene. It is advised, however, that this selection will be followed by examination of its relevance using a mutant mouse strain as a useful complementary tool , that is, the third level in this approach (Fig. 2, graph diagram). When one succeeded to tie the selected candidate gene to the trait of interest, the transnational stage is coming into effect in the form of human association study for the selected candidate only (Fig. 2, human cohort diagram).
Beyond the purpose of actual gene identification in a field that is eager for new genetic targets, this approach enables the moving in a timely fashion from QTL to quantitative trait gene (QTG). The ultimate proof is not complete until one can specify which polymorphic sites in the identified gene actually cause the difference in the trait phenotype – the quantitative trait nucleotides (QTN). For that purpose, the use of additional molecular methods, perhaps augmented by pharmacological approaches, needs to be considered. We argue that the integrated approach has the potential to move one rapidly also to this final stage.
It is important to add that even with the decreasing costs of genome sequencing, executing large-scale GWAS is still very expensive. Therefore, identifying genes in animal models and then translate the data to humans can still be more cost effective.
In the section thereinafter, how the integration of the methods specified above led to the identification of CACNG2 as a single gene behind a mapped QTL will be described step-by-step . In order to avoid unwarranted biases, we adopted a hypothesis-free experiment avoiding a priori knowledge of the expected genetic component(s).
The starting point of gene identification was QTL detection. Seltzer et al.  reported a QTL for neuroma model pain in the mid-part of Chr.15 in a study based on the RILs panel of the C57BL/6J and A/J mice. They called it Pain1. We confirmed this in chosen progenitors inbred strains (C3H/Hen and C58/J) . The selection of those strains was based on their extreme contrasting pain phenotype and the low variability in those strains (Fig. 1). C3H/Hen and C58/J BC population were generated for initial QTL detection (Fig. 2, gray mouse image). We conducted a whole genome screening and linkage tests, and the link between a region on chromosome 15 and the neuropathic pain phenotype was verified. The mapping results to the same region of Chr.15 based on two independent mapping populations enhanced the generality of the findings.
The next significant step toward QTL identification lies in the application of the appropriate fine-mapping strategy. The choice needed to take into consideration the trait mode of inheritance, resource availability and more [30, 32]. As it was mentioned above, fine mapping is needed in order to reduce the mapping interval and the number of relevant candidate genes. Two efficient fine-mapping techniques were applied: RPT and RIST. Individuals carrying different recombination along the interval containing the QTL Pain 1 were selected to create the RPT population, and the pain behavior between the RPT families was compared.
In order to accelerate fine mapping we applied the RIST strategy in parallel. The progenitors of the RIST were C57BL/6J and A/J RILs carrying recombination points within the mapped interval. Minimizing the number of possible causal genes was the first stage of the project. This reduction was obtained by combining the RPT and RIST fine-mapping strategies (Fig. 2, Venn diagram).
The pain behavior of the progenitors of each population was well characterized [21, 23], and the combination of the results achieved its goal to a dramatic interval size reduction, down to 4.2 Mbp from an initial mapping interval of ∼50 Mbp (25 cM) . The usage of a dense SNP map of the mouse has in the past shown to be useful for reduction in the number of causal genes in a given region . The in silico analysis – a combination of comprehensive pain behavior profiles for >10 inbred mouse strains we had plus SNP information, enabled the exclusion of a significant number of genes. This in silico analysis included polymorphic SNP from all genomic function classes (e.g. Coding non-synonymous, 3′/5′ UTR, splice sites, etc.). In this analysis, Cacng2 was highlighted as a possible gene for further investigation, because of the presence of several polymorphic SNP, distinguishing between several pain-susceptible and pain-tolerant inbred strains. Those SNPs were located mainly in introns and in the 3′ UTR.
Coding non-synonymous SNPs are accepted as an important function class as they directly affect the translated product of the gene. Nevertheless, SNPs in other regions such as the UTR regions and introns are considered as important sites for altering gene expression and affecting traits [55-57]. Therefore, inclusion of those SNP in this analysis was justified.
In parallel to the QTL mapping and the bioinformatic analysis, mRNA expression profiling was applied. As it was explained above, microarray analysis is a powerful tool when gene contribution to the phenotype (as it is reflected in the gene's expression) needs to be assessed. The expression profiling was based on the comparison between pain-susceptible vs pain-tolerant mouse strains and between two conditions: ‘sham’ (i.e. no nerve injury) and ‘nerve injured’ . We found Cacng2 expression to be altered both in the response to nerve injury (sham vs nerve injury) and when the expression of Cacng2 in a neuropathic pain-susceptible strain (C3H) was compared to four different pain-tolerant strains. The analysis of data from microarrays is non-trivial. Problems may arise in developing a means for ranking genes, which is typically based on the degree of difference between two experimental conditions. The threshold of fold change >1.5 and p value <0.001 were taken in order to minimize false discoveries (Fig. 2, Venn diagram).
The results of the combined methods allowed us to select a leading candidate gene. The prioritization process following the major part of this analysis (Fig. 2) highlighted Cacng2 as the most reasonable candidate gene, as it was the only gene within the fine-mapped interval to meet all the four screening criteria: (i) complete cosegregation of SNPs between genotype and phenotype, (ii) functional relevance to pain by PubMed search, (iii) significant regulation following nerve injury, and (iv) differential expression in high vs low autotomy strains. No other gene met any three of the criteria, and only nine met two criteria. Furthermore, additional strengthening support came from immunolabeling on tissue sections. This assay confirmed that Cacng2 is indeed prominently expressed in the PNS (L5-DRG) and that it is localized to neurons, rather than glia or other cells.
Mapping results are statistical by nature and expression profiling has its own limitations. Expression analysis can reveal the overall number of transcripts that are regulated by axotomy. However, additional methods are needed to identify which individual transcripts play an important role in pain sensation itself. In this case, testing the possible effect of Cacng2 in a single gene mutation (e.g. knockout mice) was necessary. For this purpose, we exploited the naturally occurring hypomorphic mutant ‘stargazer’ . Using stargazer mice we applied two tests: (i) pain behavior observations based on the autotomy phenotype and (ii) electrophysiological assay. Stargazer mice (loss of the Cacng2 transcripts) showed significant pain behavior, whereas both mice heterozygous for the mutation and w.t. mice showed virtually no pain phenotype at all. The second test applied to Cacng2 KO stargazer mice was electrophysiological analysis of spontaneous discharge generated in afferent neurons axotomized by nerve injury. A noticeable number of studies have supplied evidence that the appearance of spontaneous afferent discharge is an important substrate of spontaneous dysesthesias and pain in a variety of neuropathic pain conditions [5, 59-61]. The incidence of spontaneous activity was almost double in the mutated mice compared with heterozygous and w.t. mice. Our comprehensive experiments using stargazer mice further pointed out the probable role of Cacng2 in neuropathic pain.
Human pain gene
The advantage of rodents as model animals for neuropathic pain research is well known. A key end point, however, is to use these data for human benefit. Our successful identification of Cacng2 as a gene that contributes to pain susceptibility in the mouse encouraged us to see whether this finding has relevance in the human context.
The major goal of pain research is to bring relief to those who suffer. Progress can take in the form of promoting awareness of the chronic syndrome and, of course, pushing forward toward better pain therapy. It was natural that following the identification of Cacng2 as a pain gene in the mouse we would look for a potential role in human cohorts as well. Phantom breast pain is phantom pain affecting women who undergone breast removal (mastectomy) [8, 62, 63]. We analyzed the human homologous CACNG2 in a group of 549 breast cancer patients. Applying standard association analysis, a three SNP haplotype was identified as associated with the tendency to develop neuropathic pain following mastectomy (Fig. 2, human cohort).
The fact that a particular woman had the bases A-C-C at three specific adjacent SNPs provided considerable predictive power that postoperative pain might develop (odds ratio = 1.7). That is, the presence of this A-C-C haplotype, which can be determined in advance in a simple blood test, could be used to inform the surgeon that special care is needed to minimize nerve injury. It will be stressed, however, that it is essential to test the reproducibility of this genetic finding in additional mastectomy pain cohorts before the medical usefulness of the information can be assured.
The relevance of animal models to human disease phenotypes is a source of long-standing debate in the scientific community [14, 64, 65]. Nevertheless, at least in the case of CACNG2 the relevance of animal models to address a human phenotype has been proven.
A discussion regarding the physiological function of CACNG2 in pain is somewhat beyond the context of this review. Further investigation is of prime interest. CACNG2 has been studied mainly in the context of epilepsy [66-68]. The relationship between chronic pain and epilepsy is still elusive. However, some of the best neuropathic pain killers have been developed from anticonvulsant/antiepileptic drugs, thus suggesting a connection between the two neurological phenomena, at least from a pharmacological perspective.
Parallel approaches for pain genes identification
Our research approach contrasts with other alternatives that are common in the pain field. So far, the large majority of studies in pain genetics have relied on candidate gene approaches. The pre-selected candidate strategy has the advantage that the ability to tie genotype to phenotype is within reach. If there are data regarding the gene product and its pattern of expression, etc., a connection to the trait can be established. Several genes associated to different chronic pain phenotypes have been found. Examples are as follows: COMT , the μ-opioid receptor , DRD4  and other genes. The weakness of the candidate gene approach is that it is subjected to unwarranted biases because a number of a priori assumptions have to be made in the selection of the candidates to be tested. Alternatively, GWAS is a well-based strategy, with a major advantage of high resolution that enables one to uncover even loci with a small effect on the phenotype. Taking into account that most of the statistical obstacles concerning the huge sample sizes and multiple testing have been resolved, GWAS became a standard methodology for the discovery of novel causal genes [45, 72]. However, when it comes to the pain community, GWAS has been used only rarely. The main reason is that reliable GWAS needs to be based on a large number of accurately phenotyped individuals, with corresponding matched controls. The numbers needed for GWAS are unusually difficult to obtain in the pain field because studies require subjects with identical pathology. Pain GWAS can be executed only by collaborative multisite efforts as it has been performed for other traits [73, 74]. In addition, such collaboration will demand standardized pain assays that are still uncommon among many pain laboratory and clinics. Recently, such collaboration has yielded significant results for migraine . But, this still stands as an exception in the pain genetics arena.
Neely et al.  adopted a slightly different approach by adding another model organism to the equation. A genome-wide Drosophila screen was conducted in order to identify novel thermal nociception genes. A single candidate gene was later selected out of the primary screening output of more than 500 potential candidates, based on its known connection to widely used analgesics for neuropathic pain in humans. The selected gene (α2δ3) was subsequently tested for association to pain phenotype on mice and humans. The selection of α2δ3 from among many candidate genes based on a priori knowledge of physiological characteristics can be viewed as a drawback, narrowing down the range of gene discovery. The first investigation stage consisting in an unbiased wide-genome screening was practically superfluous. In contrast, in the integrated approach described above, no physiological information was considered prior to the selection of the candidate gene Cacng2. Rather, selection was based on convergence of different genetic approaches. Another approach is illustrated in a recent study by Costigan et al.  who used a mouse expression microarray profiling as a single screening tool for pain gene identification. Following their analysis, several genes (>100) were marked as possible candidate genes. The relevance of the selected gene (KCNS1) was tested in several human cohorts of various chronic pain syndromes.
Probably, the best ‘proof of concept’ of the integrated approach can be viewed through recent findings achieved in the same manner. Recently, Mogil et al.  have used similar approach to uncover the role of vasopressin-1A receptor in pain sensitivity. The implementation of this approach by the same group led to the identification of the P2X7 receptor (P2X7R) as yet another novel pain gene affecting chronic pain sensitivity in both mice and humans . Our successful identification of CACNG2 as a neuropathic pain gene as well as more recently published studies recommend the integrated approach as a platform for the unbiased dissection of complex traits.
Future directions of pain genetics
Integrating genetics, bioinformatics and expression data can be highly useful for understanding more comprehensively the complex features of pain genetics and physiology and how genetic polymorphisms affect pain phenotype, susceptibility to pain after nerve injury, for example. Nevertheless, the addition of more analytical tools such as cognitive methodologies and advanced imaging techniques brings considerable optimism for deciphering the elusive ‘pain matrix’.
From a different angle, the emerging world of stem cell research and regenerative medicine bears great promise for both basic science and pain treatments. For example, the ability to reprogram cells makes it possible to focus on neurological features of the individual patient, characterize them and perform screening for different agents [80-82]. In a recent study, Franchi et al.  showed the treatment of neuropathy by using neuronal stem cells in an animal model.
Pain genetics has just matured from its juvenile stage. It can be benefited from the methods and techniques pioneered by others. As networking between different disciplines (such as the different ‘omics’) becomes routine in science, it is hoped that future discoveries will illuminate understanding of not only pain physiology and genetics per se but also other traits such as human cognitive development, behavior and perception.
I thank The Hebrew University Center for Research on Pain. Ariel Darvasi and Marshall Devor provided constructive comments on the manuscript.