These data illustrate an approach to investigating causal relationships among pain, mood, and genetic polymorphisms patients who undergo a surgical procedure that produces variable degrees of pain relief. The degree of surgical relief of pain at the first two postoperative time points explains 20% of the variance in mood at 1 year (p <0.0001), but there is also additional inter-individual variation in mood (Fig. 4), some of which may result from inherited genetic variation. Because surgical relief of pain is a large abrupt change occurring at a fixed time point, this experimental design is well-suited for studying physiological events over time.
Although the galanin-2 receptor and mu opioid receptor are plausible candidates to mediate effects of pain on mood, and showed graphical patterns consistent with a pain-gene interaction, we cannot prove these specific effects because of our modest sample size. Although this was the largest prospective study of pain from a uniform lesion that we could identify at the outset of the project, a several hundred patient cohort does not provide sufficient power to correct for the tests of dozens of genes and multiple analytical models unless the relative risk conferred by the polymorphism is more than 2.5, larger than most common polymorphisms for which a link to medical diseases have been established . However, major medical centers perform thousands of many types of pain-relieving operations each year, making possible more powerful searches of this type. The number of simultaneous statistical tests supported by candidate studies increases steeply with sample size. For example, just an eight-fold increase in N permits a million-fold increase in independent tests, sufficient to examine the genome in detail .
Another limitation of this study is the lack of clinical diagnosis of anxiety or depressive disorder. Across cohorts of patients, the SF-36 MH subscale correlates strongly with research psychiatric diagnoses and changes with successful treatment, but cannot provide individual diagnoses . The SF-36 pain measure does not have optimal precision either, for estimating the actual pain level over many months. Bellamy et al.  found that a 0–10 point numerical scale and 100 mm VAS were more sensitive than a 5-category pain intensity scale, while Jensen and McFarland  reported that the average of 7 pain measurements at different times gives a better estimate of actual pain than 1–2 measurements. Quality of life researchers are currently seeking to improve diagnostic precision with computerized adaptive algorithms that choose items to hone in on each subject's response range.
It is possible that some of patients with the most severe mood disorder may have refused to return questionnaires or contribute DNA, lessening the study's power to examine this link. Another possible gap is that in the 16 years since this study was begun, pain psychologists learned that styles of coping with pain and environmental stressors are as important determinants of many pain outcomes as mood. This study did not include detailed measurements of pain catastrophizing , pain self-efficacy, and stressors in the work and personal environment , which one might consider for a new prospective study of pain and mood.
While we cannot exclude a false positive in this study, the possible mood mediating effects of the galanin-2 and mu opioid receptor polymorphisms should be studied in additional cohorts of patients with pain. The neuropeptide galanin is widely expressed in the central nervous system, including areas regulating emotionality . It has been implicated in a wide range of physiological functions including pain control and cognition and in behaviors such as anxiety and depression . We recently reported that haplotypes in the galanin gene were associated with anxiety-associated alcoholic phenotypes in humans . The three galanin receptor subtypes, GALR1, GALR2, and GALR3 , are widely distributed in mood-related brain areas such as hypothalamus, central amygdaloid nucleus, and thalamus  and may mediate anxiety-associated behavior . No common functional polymorphism has yet been identified in the human GALR2 gene. There is strong linkage disequilibrium (LD) between all SNP pairs we genotyped, and on the related chromosome region (HapMap; http://www.hapmap.org). Although the SNPs used in our study are not located within GALR2, they are within 5 kb of the start and end sites of the gene (Fig 4), and are within a haplotype block encompassing GALR2, its regulatory elements, and neighboring genes. The nominally significant pain-gene interaction on mood of three out of four SNPs in this region and the haplotype may reflect the contribution of a functional allele in GALR2 or the genes located nearby.
The mu opioid receptor is also a plausible candidate to mediate between pain and mood. The endogenous opioid system and μ-opioid receptors modulate affective behaviors  as well as affective components of acute pain . Opioid treatment of chronic pain is often accompanied by striking improvements in mood . Although the human mu opioid receptor gene (OPRM1) has polymorphisms that affect receptor function and are associated with some behavioral phenotypes , no associations with depression and anxiety disorders have yet been reported . The three OPRM1 SNPs most closely associated with pain-related mood scores in our study are in high linkage disequilibrium and in the same haploblock, which also includes the previously reported functional non-synonymous SNP Asn40Asp (rs1799971). However, the latter SNP was not associated with mood scores. Therefore we assume that the association signal, if replicated, could be attributable to another functional allele that is still unknown.
In this study, we have tested mood effects of genes that we had already genotyped based on their involvement in chronic pain processing. For future studies on pain-mood interaction, one might select additional candidate genes reportedly associated with anxiety and depression; e.g., the genes for corticotropin-releasing hormone (CRH) , the adenosine A(2A) receptor , the dopamine D4 receptor , and tryptophan hydroxylase .
It is interesting that among patients in the quartile with the least surgical pain relief, mood did not deteriorate, and actually improved from 6 months on. Within this quartile of patients (Fig. 4), for example, pain levels at 24 months after surgery were equivalent to the presurgery levels, but mood at 24 months was considerably improved from baseline. It would be interesting to examine individual differences in resilience factors, including genetic resilience factors, that are associated with improvements in mood in the face of the surgical failure to relieve pain.
The preceding analysis of gene effects mediating mood responses to unrelieved pain in surgical patients is a gene × environment (G × E) interaction study. Thus far, the most widely cited examples of G × E interactions on behavioral endpoints [17, 18] involve an environmental stress. Although our most suggestive results for the genes we examined pertained to unique pain-gene interactions on mood, the same analytic method might help to identify other genes that have a general effect on mood, demonstrated by a main effect in the regression. A potential advantage of this study design is that patients are all subject to a similar stress – several months of a painful condition followed by major surgery – and then experience variable levels of residual pain, a stressor that can be well-quantified and corrected for in the analysis. Without detailed measurements of environmental stress, it may be difficult to identify genes for depression or anxiety. For example, although twin studies suggest that the heritability of unipolar major depression disorder is 40–50%  it has been difficult to replicate candidate gene associations, possibly because the environmental circumstances triggering depression are so varied . A major surgical procedure may provide a somewhat uniform stressful situation. Within this setting, the two factor analysis shown above may remove a large amount of variance in late mood due to pain, increasing sensitivity to a main effect of a genetic polymorphism involved in general mood-regulating mechanisms not specific to patients with pain.
It is possible that alternative statistical methods may better detect "mood" or "pain-mood" genes in the types of surgical cohort studies illustrated above. For example, patients without any pain might be excluded from the pain-mood analyses, or different time-points might be chosen for the pain and mood variables. The relative merits of alternative methods can be compared more persuasively once a robust gene effect is identified to use as a gold standard.