Twin studies of pain

Authors


  • None of the authors have conflicts of interest relating to the contents of this manuscript.

Corresponding author: Christopher Sivert Nielsen, Norwegian Institute of Public Health, Division of Mental Health, Box 4404 Nydalen, 0403 Oslo, Norway.

Tel.: +47 21078277

fax: +47 22353605

e-mail: chsn@fhi.no [PO BOX 4404]

Abstract

Twin studies provide a method for estimating the heritability of phenotypes and for examining genetic and environmental relationships between phenotypes. We conducted a systematic review of twin studies of pain, including both clinical and experimental pain phenotypes. Fifty-six papers were included, whereof 52 addressed clinical phenotypes. Of the most comprehensively studied phenotypes, available data indicates heritability around 50% for migraine, tension-type headache and chronic widespread pain, around 35% for back and neck pain, and around 25% for irritable bowel syndrome. However, differences in phenotype definitions make these results somewhat uncertain. All clinical studies relied on dichotomous outcomes and none used pain intensity as continuous phenotype. This is a major weakness of the reviewed studies and gives reason to question their validity with respect to pain mechanisms. Experimental pain studies indicate large differences in heritability across pain modalities. Whereas there is evidence for substantial common genetic risk across many clinical pain conditions, different experimental pain phenotypes appear to be associated with different genetic factors. Recommendations for future research include inclusion of pain intensity scaling and number of pain sites in phenotyping. Furthermore, studies examining the genetic relationships between pain phenotypes, in particular between clinical and experimental phenotypes, should be prioritized.

The utility of twin studies in the post-genomic era

Using twins to examine the contribution of heredity and environments to variation in human traits is among our oldest technique for the systematic study of human genetics [1]. Twin studies have seen widespread application, and are still a very useful approach in the post-genomic era, perhaps more so today, than earlier. By exploiting the difference in genetic similarity between monozygotic (MZ) and dizygotic (DZ) twin pairs, twin studies provide a cost-effective and robust method of estimating heritability and for examining the genetic relationships between phenotypes. In this section, we briefly describe these applications, with particular emphasis on their relevance for genetic association studies.

Heritability

Heritability is most commonly estimated in classical twin studies including MZ and DZ twins that have been raised together. Typically these studies estimate the contribution of additive genetic effects (A) to the variance of a phenotype (narrow sense heritability). In addition, classical twin studies also partition environmental influences into those that are common to both co-twins (C) and those that are unique to each twin (E). Importantly, the latter also includes measurement error and random fluctuation over time. Non-additive genetic effects of dominance (D) or epistasis (I) are confounded with A and C in the classical twin study, but can be estimated in extended twin designs including other family members, or by dropping the C parameter from the model. Thus both ACE and ADE models are determinate in a classical twin design. It should be noted that heritability is a characteristic of the phenotype in a given population and may vary as function of sex, age and differences in exposure to environmental influences.

Most phenotypes are heritable to some degree, and showing that pain is heritable is perhaps not in itself of huge importance. However, these estimates are useful, among other reasons, because they can provide basis for selecting appropriate pain phenotypes for genetic association studies. If two phenotypes are thought to be expressions of the same underlying genotype, then the phenotype with the higher heritability is likely to be the better choice for genetic studies. In many cases lower heritability reflects greater measurement error or poorer stability over time, both of which are undesirable phenotype characteristics for genetic association studies. A second advantage of knowing the heritability of a phenotype is that it serves as a benchmark for gaging progress in identifying the variants associated with polygenetic phenotypes.

Genetic relationships between phenotypes

Where two correlated phenotypes are examined together, twin studies can estimate the degree to which the phenotypic correlation is mediated by common genetic factors. If several phenotypes are included, this can be extended to genetic factor analysis. In other words, such studies tell us to what extent phenotypes are distinct or overlapping from a genetic standpoint. For instance, a twin study including five anxiety disorders revealed that most of the genetic variance in these phenotypes was explained by a single underlying genetic factor [2]. Thus from a genetic standpoint, the distinction between different types of anxiety appears to be unwarranted. This is of great importance for selecting phenotypes for case–control studies. If, for instance, pain from rheumatoid arthritis (RA) was highly correlated genetically with other and more common types of muscular-skeletal pain, then a genetic association study comparing RA patients with the remaining population would have little chance of success, since pain-associated variants would occur with high frequency in the control group. Ideally, case definition would therefore not be based on manifest characteristics, but on the underlying genetic architecture, as determined by this type of twin analysis. Extended application of multivariate twin studies can also be used to examine causal relationships between phenotypes in what is known as co-twin control designs. This is of particular relevance for defining endophenotypes, where a crucial assumption is that the endophenotype lies on the causal pathway between the genotype and exophenotype, but this topic lies beyond the scope of this review.

The assumptions of classical twin studies

The validity of results from classical twin studies depends on certain assumptions. First, it is assumed that environmental influences are shared to an equal extent by MZ and DZ co-twins (equal environments assumption). If MZ co-twins experience more similar environments than DZ co-twins, this can conceivably inflate estimates of heritability by creating greater similarity among MZ co-twins. Though it seems likely that this assumption is breached to some degree, there is evidence that the effect on heritability estimates is small [3, 4]. Second, it is assumed that the genetic similarity between parents is the same as between random individuals in the population (random-mating assumption). However, even for phenotypes such as intelligence where considerable assortative mating occurs, the underestimation of broad sense heritability (A+D) is modest, and it is mainly the relative contribution of A vs D that is affected [5].

What is pain phenotype?

Under normal circumstances, pain is a sensory and emotional response to ongoing or impending tissue damage. The tissue pathology may arise from any number of causes and should not be considered part of the pain phenotype. In fact, in cases of idiopathic (functional) pain, objective pathology is not known, though this does not necessarily mean that it does not exist. Though objective tissue pathology is not part of the pain phenotype, the severity of pathology may influence degree of pain and it is therefore a confounding factor in quantifying pain phenotypes. This confounding can be avoided to a greater or lesser extent, in order of degree as follows:

  1. Pain ratings of experimental stimuli, where stimulus intensities can be standardized across subjects, bypass the issue of confounding. This is a major strength of these models and a compelling reason for including them in genetic studies of pain. However, though experimental pain assays are known to correlate with, and to some extent predict, clinical pain it remains to be shown that experimental pain measures are genetically related to clinical pain.
  2. Pain ratings among patients suffering from a condition where objective measures of disease severity are available and can be controlled for statistically, also represent a viable option. Though controlling for objective pathology will not eliminate the issue of confounding, it will reduce the problem.
  3. Pain ratings of clinical conditions without controlling for objective measures of disease severity may also be relevant, as the relationship between disease severity and degree of pain is typically weak and it can be argued that objective disease characteristics are of less importance for determining pain levels than the pain sensitivity of the individual [6].
  4. Finally, it can be argued that diagnoses where pain is the only criterion or a necessary criterion also have some relevance to pain genetics, on the assumption that pain must reach a certain level to result in referral and diagnosis of the condition. This is primarily a relevant approach for functional pain conditions, but has serious disadvantages compared to option (3) above.

In our view phenotypes that are based of objective measures of pathology alone are not relevant to pain genetics. Objective pathology may or may not be accompanied by pain [7], and measures such as magnetic resonance (MR) imaging of lumbar disk degeneration are uninformative about nociceptive processing. Though increasing the specificity of a diagnosis by including blood samples, imaging, etc. is desirable for studying the etiology of the disease, it may reduce relevance with respect to understanding mechanisms underlying pain.

Systematic review

Twin studies of pain have been addressed in two general reviews of pain genetics previously [8, 9]. However, due to the broader scope of these reviews, they have been selective with respect to the studies that were included and discussed. Consequently, we conducted a systematic review of the literature in order to gain an overview over the available findings.

Methods

We searched MEDLINE (1946 to 24 June 2012), EMBASE (1974 to 2 June 2012), and PSYCHINFO (1967 to week 2 June 2012) databases for English language publications. The search terms ‘twin study’ or ‘twin studies’ or ‘twin pair*’ were used to identify the study type. This search was crossed with a search for pain phenotypes using the terms ‘pain*’, ‘fibromyalg*’, ‘headache*’, ‘migraine*’, ‘angina*’, ‘neuralgia*’, ‘zoster’, ‘irritable bowel*’, ‘ulcer*’, ‘osteoarthr*’, ‘arthr*’, ‘temporomandibular joint*’, or ‘musc* skel*’. The selection of specific diagnoses supplementing the general term ‘pain’ was made from the following: (i) conditions where pain is a necessary criterion for diagnosis (i.e. headache), (ii) diagnoses that represent a large fraction of the chronic pain cases [e.g. osteoarthritis (OA)] and finally (iii) diagnoses that appeared in a preliminary search for ‘twin study’ and ‘pain’. This search yielded a total of 452 hits. The titles and abstracts were reviewed by the first and last author independently for relevance. In cases of disagreement or uncertainty, decisions were made after in depth assessment of the full papers. Only original research reports in peer-reviewed journals were included. Reports were considered relevant if the phenotype definition included some assessment of the degree of pain (categories (1–3) above) or presence of pain (category (4) above). This also included studies of conditions where pain was not explicitly assessed, but where pain was a necessary criterion for diagnosis [e.g. headache, fibromyalgia (FM)] or where diagnosis in the absence of pain was considered rare or unlikely (e.g. IBS, Migraine). Studies reporting exclusively on objective pathology were excluded (e.g. disk herniation determined by MR), as it is well established that the correlation between pain and objective disease characteristics is weak for most conditions [10]. Studies that did not provide heritability estimates or genetic factor estimates were excluded. Therefore studies reporting concordance rates without formal estimation of heritability were omitted from this review, but are for completeness listed as Supplementary References. This resulted in a total of 58 papers. These were reviewed in detail for information on phenotype definition, sample size, etc., as seen in Tables S1–S6. Heritability estimates (h2) are reported as broad sense heritability for ADE models (i.e. A+D), else as narrow sense heritability (A).

For phenotypes where sufficient data was available from three or more studies of independent population based samples, meta-analysis was carried out for an overall heritability estimate across studies. In cases where two studies reported data from the same or overlapping sample the largest study was chosen. Studies that examined children or elderly subjects exclusively were not included in the meta-analyses, as heritability may vary over age and there was insufficient data to test whether such age differences in heritability were present. For studies reporting separate heritability estimates for men and women, these estimates were entered independently. The same approach was used where age-stratified estimates were reported. The estimates from each study were weighted by the number of complete twin pairs in the study. Meta-analysis was carried out in stata/ic version 10.1 (StataCorp LP 2009) using the Metan package 24.3 (http://www.stata-journal.com/sj9-2.html).

Results

Headaches

Studies of headaches are listed in Table S1. Migraine was examined in 15 studies [11-25], most of these using International Headache Association (IHS) criteria. Heritability estimates range from 33% to 77%, the highest obtained by Svensson et al. in a pediatric sample [20]. The largest study conducted by Mulder et al. includes data from nearly 30,000 pairs in eight countries [16]. This study finds an overall heritability of 46%, and evidence for non-additive effects (D). Though two smaller studies seem to indicate a somewhat higher heritability among women than men [12, 13], this is not confirmed in the cross-national study. Meta-analysis is presented in Fig. 1, showing an overall heritability [95% confidence interval (CI)] of 45% (CI = 41–49%). Overall, the data for migraine are consistent and indicate that close to half the risk of developing migraine is attributable to genetics in adult populations.

Figure 1.

Forest plot of heritability estimates with 95% confidence intervals from studies of migraine.

Tension-type headache has been examined in three studies [20, 26, 27]. Ulrich et al. applied IHS criteria finding non-significant heritability of 19% [27], Svensson et al. used clinical assessment of questionnaire responses and found 68% heritability among children [20], whereas Russel et al. relied on self-diagnosis (migraine excluded), reporting heritability of 48% for men and 44% for women [26]. The latter study is the largest, including more than 7000 pairs and is likely to be the most accurate estimate for adults. Three additional studies examined recurrent headaches without further classification [12, 13, 28]. However, as tension-type headache is the most common type of headache it is likely to be the issue for the majority of cases. Honkasalo et al. reported heritability of 43% for men and 48% for women [12], whereas Larsson et al. in a sample of nearly 13,000 pairs reported 49% for men and 31% for women [13]. Finally, Kato et al. reported heritability of 41% for men and women combined [28]. Taken together the data indicate heritability around 40–45% for tension-type headache, but heterogeneity in phenotype definition makes this conclusion somewhat uncertain. Meta-analysis could not be conducted due to missing CIs and other issues with the reporting.

Temporomandibular joint disorder (TMD) was only examined in one study, reporting heritability of 27% [18]. However, as this study was based on questions about persistent pain in the general area in question, it is not clear to what extent the case definition was selective for TMD or whether other conditions, such as chronic ear-aches may have been included.

Back and neck pain

Studies of back and neck pain are listed in Table S2. Low back pain (LBP) is addressed by eight studies [29-36] and two additional studies report on back pain in general [37, 38]. Phenotype definitions are quite heterogeneous, making direct comparison somewhat problematic. Heritability estimates range from zero to 68%. The lowest figures were obtained by El-Metwally et al. in a pediatric study [30] and by Hartvigsen et al. in a study of elderly subjects [37]. This may possibly reflect methodological problems with pain assessment in these age groups. This interpretation gains some support from the low prevalence estimates in these studies compared to other studies examining back pain without additional criteria. The remaining five studies of adults are somewhat more consistent, with estimates ranging 30–68%. For the largest studies, Hestbaek et al., Hartvigsen et al., and Nyman et al., this range is reduced to 30–44% [31, 32, 35]. Meta-analysis of studies of LBP and back pain in general was performed and indicates an overall heritability of 34% (CI = 30–39%) (Fig. 2). However, as there is great heterogeneity in phenotype definitions and prevalence estimates this result is somewhat uncertain. MacGregor et al. compared back pain in general with severe back pain with radiation, finding that the latter had higher heritability, albeit not significantly so [34]. A similar trend is reported by Battie et al. where LBP with hospitalization has the highest heritability (29).

Figure 2.

Forest plot of heritability estimates with 95% confidence intervals from studies of low-back and back pain.

Neck pain (NP) was examined in six studies [31, 34, 35, 38-40]. Hartvigsen et al. report no heritability in a sample of elderly subjects [40], but find heritability of 39% in a later study (2009) of adults [31]. For the remaining four studies, estimates range from 24% to 58%. The largest study, conducted by Fejer et al., included nearly 11,000 twin pairs and found significantly higher heritability among males (52%) than females (34%) [39]. None of the other studies approach this sample size. Meta-analysis was attempted for three studies [31, 35, 38], but failed due to significant heterogeneity (I-squared = 88.5%; p < 0.001).

Visceral pain

Studies of visceral pain are listed in Table S3. The most studied visceral pain phenotype is irritable bowel syndrome (IBS) where heritability is reported by five studies [28, 41-44]. ROME-II diagnostic criteria were applied in two, others used approximations of these criteria and one used medical records, where available. In addition one small study examined functional bowel disorder (FBD) [45] – a broader classification which includes IBS. Though heritability estimates ranged from zero to 48% (58% for FBD), and appear quite inconsistent, estimates from three studies ranged from 22% to 27% [28, 42, 44]. Moreover, two of these (Svedberg et al. [44]; Kato et al. [28]) employed very large samples giving increased confidence in their results. There was insufficient data to conduct meta-analysis of IBS studies.

A number of other visceral pain conditions have been examined [17, 42, 46-50], but none by more than one study, and sample sizes are too small to yield conclusive results (see Table S3 for details).

Arthritic and inflammatory conditions

Studies of arthritic and inflammatory conditions are listed in Table S4. Only two studies of RA met the inclusion criteria. Van der Woude et al. found 66% heritability [51]. MacGregor et al. performed reanalysis of data that previously reported as concordance estimates, and found 65% and 53% heritability among Finnish and UK twins, respectively [52]. Both these studies included only twin pairs where at least one twin was diagnosed with RA.

OA was examined in two studies. Kirk et al. used self-report data that were validated against a sample of clinically diagnosed patients [53]. Heritability estimates varied considerably across sites (0–53%), most likely due to lack of power. When OA was considered independent of body site heritability was found to be 25% for men and 36% for women, but CIs were large and included zero. Kujala et al. analyzed self-report data and estimated heritability of 0% in men and 44% in women [54]. Small samples size and other weaknesses make it doubtful whether the large sex difference in heritability is valid. In addition, Charles et al. examined joint pain in a small sample and found 12% heritability [55]. In sum, these studies are too small to reach definite conclusions about the contribution of genetic factors to OA. Addition phenotypes examined include carpal tunnel syndrome, ankylosing spondylitis, psoriatic arthritis and gout [56-60].

Widespread pain

Studies of widespread pain are listed in Table S5. Chronic widespread pain (CWP) and FM have similar diagnostic criteria and will be treated together. Two studies have examined CWP/FM in adults, both in large samples Kato et al. found that CWP has a heritability of 48% for men and 54% for women using American College of Rheumatology criteria [61]. In a later publication, most likely conducted on the same sample, they found an overall heritability of 53% as part of a multivariate analysis [28]. Markkula et al. (2009) employed cluster analysis of questionnaire items, defining a ‘likely FM group’ which was then validated against a clinical FM sample [62]. Heritability was estimated at 51%. The size of these studies and the consistency of their results give grounds for confidence in these estimates. At odds with this are results from a smaller study of CWP in 11 year olds, conducted by Mikkelsson et al., where no evidence of heritability was found [63]. This may possibly be due to issues with phenotype definition or problems with obtaining valid results from this age group, as noted above for LBP. A nearly 10-fold difference in the prevalence of CWP/FM between the three studies, which were all conducted in Scandinavian populations, underscores that phenotypes may be quite different from study-to-study. A different approach was taken by Williams et al. [38]. They investigated pain reporting at different anatomical sites, and performed a multivariate analysis of genetic associations between pain sites. They found that a single common factor explained 95% of variance in the data and that this factor had a heritability of 46%. Similarly, Nyman et al. found that combined neck and back pain had a higher heritability (60%) than either NP (30%) or back pain (24%) alone (see Table S2) [35]. Finally, Roysamb et al. reported heritability of number of body sites with pain, 24% for women and 31% for men [64], but as only four sites were included and three of these where located in the midline, the relevance, if any, for CWP/FM is unclear. Reported data from the studies of CWP/FM was insufficient for meta-analysis. However, the available adult studies are consistent and indicate heritability around 50% for muscular-skeletal pain at multiple sites.

Experimental pain phenotypes

Studies of experimental pain phenotypes are listed in Table S6. We identified four studies that reported heritability estimates for experimental pain. MacGregor et al. examined pressure pain threshold in a sample of 609 twin pairs, finding insignificant heritability and considerable shared environmental (C) effects [65]. However, this study has major weaknesses that make the validity of the results uncertain. Co-twins were examined together in the same room, which may have inflated estimates of C, and pressure pain threshold was only examined once, whereas common practice is to repeat the measurement three or more times, discarding the first assessment which tends to be unreliable.

The remaining three studies appear to use solid methodology, but were all performed on rather small samples. Norbury et al. used a procedure involving burn injury and reported on a number of measures associated with this procedure, as well as independent measures of chemically induced pain [66]. Heritability estimates ranged from zero to 55%. Notably, heritability for heat pain threshold was found to be 53% in contrast to findings by other authors below. Genetic correlations between phenotypes were not reported.

Nielsen et al. examined visual analog scale pain ratings of the cold-pressor test and of heat stimuli, with heritability estimates of 54% and 25%, respectively [67]. This study reported a genetic correlation of 0.35 between phenotypes, and only a minor amount of the phenotypic variance in cold-pressor pain and heat pain could be explained by genetic influences that are associated with both phenotypes (6% and 3%, respectively). The study concluded that the genetic causes of variance in these pain modalities are mainly independent.

Finally, Angst et al. examined heat-pain threshold, and time to pain and tolerance time during the cold-pressor test [68]. This was done in a pharmacogenetic design, where pain tests were conducted during baseline, placebo and alfentanil (a strong short acting opioid) conditions. Heritability at baseline was high for cold-pressor tolerance (49%), and considerably lower for heat pain threshold (20%) and cold-pressor time to pain (17%). Heritability for baseline vs alfentanil difference scores was highest for cold-pressor time to pain (60%), lower for cold-pressor tolerance (30%) and non-significant for heat-pain threshold (12%). It is not clear why the cold-pressor measure that has the lowest heritability at baseline has the highest heritability as an index of analgesia, though ceiling effects in cold-pressor tolerance during the alfentanil condition may be a contributing factor.

Co-twin control studies and studies of genetic correlations

A large number of studies have examined genetic relationships between phenotypes, some of them reporting heritability estimates and meeting our inclusion criteria, others not. Results from these studies are complex, and a full review is not possible within the scope of this article. However, the striking finding is that among these studies, only one experimental and one clinical study have reported on genetic relationships between pain phenotypes. Nielsen et al. found almost no common genetic influence for heat and cold-pressor pain [67], whereas Williams et al. found that nearly all variance in pain reporting at different muscular-skeletal body sites can be explained by one common factor [38].

Discussion

The validity of the clinical phenotypes

Our review identified 52 studies of clinical pain phenotypes. A common characteristic of all these studies was lack of pain scaling. Though presence of pain, and in some cases intensity or frequency, was a defining characteristic of the diagnoses, none of the studies used pain intensity as a continuous phenotype in its own right. Rather, analysis was restricted to dichotomous diagnoses even where information on degree of pain was available. Aside from loss of statistical power, this omission makes the relevance of these studies to the field of pain highly uncertain and it is not clear to what extent the estimated genetic effects reflect pain processing or the tissue pathology causing pain. As diagnosis of the clinical conditions reviewed depends on the presence of pain, one might expect that patients that experience much pain are more likely to be diagnosed than those who experience little pain. However, the extent to which this effect contributes to the case–control definition is unclear. Presumably, conditions where pain is the only symptom have greater relevance to pain genetics than conditions where additional objective criteria must be met. It is therefore worth noting that both tension-type headache and CWP appear to have heritability in the upper range among the phenotypes examined.

Of special interest is Williams et al.'s finding that nearly all the risk of developing pain at different (muscular-skeletal) body sites was explained by a common factor, and that this factor has a heritability of 46% [38]. Assuming that the cause of pathology may be quite diverse across body sites and across subjects reporting pain, this finding supports the view that the specific tissue pathology is of lesser importance in determining pain reporting. This may indicate that the confounding between pathology and pain is not pronounced, and that the heritability pain at specific body sites, such as back or neck, mainly reflects genetic effects on pain processing. It also indicates that examining pain across body sites may be a sensible approach in future studies. However, as the study did not include headaches or sites where visceral pain predominates, it is not clear whether the conclusions are valid beyond the muscular-skeletal domain.

Why do heritability estimates vary?

As noted above, heritability is a characteristic of a phenotype in a given population, and may vary as a function of sex, age and environmental exposures. Several studies report separate heritability estimates for men and women. However, few of these studies actually tested qualitative and quantitative sex differences, and examination of CIs indicates that separate reporting was only warranted in a minority of the studies. In these cases, lack of replication in independent samples makes results uncertain. This does not mean that sex differences in the genetic architecture of pain do not exist. On the contrary, there is evidence from animal research of sex specific molecular mechanisms in pain and it is likely that functional variants affecting these mechanism exist [69, 70]. Rather, limited statistical power and other issues, such as of poorly defined pain phenotypes, may have masked sex effects in many of the included studies.

A few studies reported heritability estimates by age group within the same sample. Fejer et al. found significantly decreasing heritability for NP with increasing age – a finding that most likely reflects accumulating influence of random environmental influences over the life-span [39]. A limiting factor in the reported age comparisons is that they did not include children. Three studies examined painful conditions in pediatric samples. A Finnish study found no evidence of heritability for CWP [60], whereas adult studies indicate that about half the risk is explained by genetic factors. Another study of the same Finnish sample, found no evidence of heritability for LBP in children [30], whereas our meta-analysis estimates 34% heritability for back pain. The third study found considerable higher heritability for migraine and tension-type headache among Swedish children [20] than data from comparable studies of adults. Taken at face value, these results indicate that heritability increases with age for muscular-skeletal conditions, but decreases with age for headaches, an interpretation we find unlikely. Alternatively, results may be explained by the way phenotypes were defined in the Finnish sample, i.e. defining pain in terms of frequency rather than intensity.

Differences in environmental exposures, such as work load, may also affect heritability. As data on relevant environmental exposures, such as work load were not available, this issue was not treated in our review. However, since all studies were performed on samples from industrialized countries, and the majority from Scandinavian countries, it is likely that variation in environmental exposures is greater within samples than between samples.

Keeping these caveats in mind, sample size appears to be the single most important factor affecting consistency of findings across reviewed studies. Where two or more large studies have examined the same phenotype, results tend to be quite consistent.

The heritability of specific clinical conditions

Though the number of the twin studies examining painful clinical conditions is fairly large, relatively few studies addressed each phenotype and several phenotypes were only examined in one study. Compounding this problem, many studies were underpowered, in particular those addressing phenotypes with low prevalence. Nevertheless, there is substantial data on some phenotypes that warrant tentative conclusions. Results indicate heritability around 45–50% for migraine, tension-type headache and CWP, around 35% for back pain, and around 25% for IBS in adult samples. Data for children and elderly subjects are too sparse to reach conclusions. With the exception of migraine, there is considerable heterogeneity across studies in phenotype definition, which along with differences in sample composition contribute to uncertainty in the estimates. Nevertheless, it is notable that CWP is in the upper range of the heritabilities reported and IBS is in the lower. Though both conditions are classified as functional disorders, the importance of genetic contributions appears to be quite different. In a study by Kato et al. it was shown that IBS shares genetic etiology with psychiatric disorders to a greater extent than does CWP, indicating that differences in the genetic contribution to these conditions is qualitative as well as quantitative [28]. The classification of conditions as functional or not may therefore not be a useful distinction in genetic pain studies.

Experimental pain phenotypes

Experimental pain studies represent an approach to examining pain processing independent of pathology. These methods are also the only feasible approach to studying pharmacological efficacy in a twin design, as was done by Angst et al. [68]. Experimental pain was assessed in four studies, three of which we consider to be methodologically sound. Findings indicate large differences in heritability across pain assays. It is possible that much of this variation is due to differences in the reliability of these measures over time. For instance, cold-pressor pain, which was found to have heritability around 50% in two studies [67, 68], has been reported to have good test-retest reliability [67]. Heat pain measures, found to have low heritability in two of three studies [63, 64], have been reported to show poor test-retest stability despite high within-session consistency [71], though other studies have reported higher stability over time [72]. This may indicate that the exact experimental procedures are important, for instance the number of stimulus repetitions, scaling methods, timing, etc., and further research will be needed to establish which of these parameters are crucial. Be that as it may, the experimental pain studies were all quite small with large CIs in their estimates and larger studies will be needed to determine which experimental pain assays show the closest association with genetic disposition.

Relating experimental to clinical pain

A striking finding in our review of the literature is the complete lack of studies examining genetic relationships between experimental and clinical pain phenotypes. For many clinical phenotypes there is a high degree of co-morbidity – as reflected in Williams et al.'s finding of a common heritable risk factor for pain at different body sites [38]. In contrast, experimental pain modalities are typically poorly correlated [73] and Nielsen et al. found that almost none of the variance in heat pain was explained by genetic factors associated with cold-pressor pain and vice versa [67]. Animal studies have also shown minimal genetic correlation across pain modalities [74]. This begs the question: If there is a common risk factor underlying many or most clinical pain conditions, but there is great genetic heterogeneity between experimental pain assays, which experimental model is best suited as an endophenotype for clinical pain? In our view, this is among the most important pain research questions that need to be addressed in future studies.

Limitations

Our ambition to provide a complete review of twin studies of pain has met with several challenges. The first of these is that the definition of what constitutes a pain study is not very clear. We have chosen to exclude studies of conditions where diagnoses are exclusively based on other criteria than pain, though pain may very well be the most salient symptom to the patient. Using a more liberal definition would include more studies, but with lower thematic relevance, whereas a stricter criterion would reduce the number considerably. In particular, clinical conditions where objective criteria for diagnosis are available may be underrepresented. However, our focus in this review is not on the heritability of diagnoses, but rather on the pain associated with these diagnoses. From this perspective, our inclusion criteria may be too liberal, and strict definition might actually lead to no clinical studies being included, due to lack of pain scaling. Undoubtedly, some of our decisions are open to debate.

A second challenge is that most twin studies are conducted on samples drawn from National twin registries. Consequently, several studies use the same or overlapping sample. As far as possible, such overlap has been taken into account in reporting the data and in conducting meta-analyses, by eliminating the smallest study reporting on the same phenotype in the same sample. However, information on sample composition is somewhat insufficient in some studies and we cannot rule out that some overlap may have occurred.

Finally, all the reviewed studies were conducted in industrialized countries. As discussed above, differences in environmental exposures may have considerable impact on heritability. Generalizing findings to populations in non-industrialized societies is therefore not warranted.

Recommendations

Despite the large number of twin studies addressing pain in some form or other, we have identified several weaknesses. Future studies would be greatly improved by taking the following into account:

  1. Clinical studies need to measure degree of pain using standardized scales. In addition to pain intensity, other dimensions such as pain duration, and impact on daily functioning should be assessed where possible.
  2. Assessing the number of pain sites through body maps or check lists is of importance, as widespread pain appears to be a promising phenotype for genetic studies.
  3. Clinical studies should attempt to compare more severe with less severe pain phenotypes. So far the data seems to indicate that the latter is less heritable.
  4. Larger experimental pain studies need to be conducted, using multiple pain modalities of proven long-term reliability. It is worrisome that very few studies have actually examined the stability of experimental pain assays over time, an obvious prerequisite for using these assays in genetic analysis.
  5. Relationships between pain phenotypes need to be examined. Results for clinical phenotypes reported by Williams et al. [27] need to be replicated and extended to encompass pain outside the muscular-skeletal category. For experimental pain, analysis needs to encompass a comprehensive set of pain assays tested in larger samples. Most importantly genetic associations between clinical and experimental pain phenotypes need to be examined. Such evidence would be of great value for planning and executing of the next generation of experimental pain genetics studies.

Acknowledgement

This work did not receive external funding.

Ancillary