Polymorphisms and Circulating Levels in the Insulin-Like Growth Factor System and Risk of Breast Cancer: A Systematic Review

The insulin-like growth factors (IGF) I and II are peptide growth hormones that promote cellular proliferation of normal breast epithelial cells (1, 2). In the circulation, IGF-I and IGF-II form complexes with one of six different binding proteins (IGFBP), the vast majority (>90%) being with IGFBP-3 and an additional acid-labile protein subunit (3). The biological actions of the IGFs are transduced by a series of cell surface receptors. Binding of the IGFs to their binding proteins increases their half-lives from 12 minutes to >12 hours and is thought to sequester them, preventing interaction with their cell surface receptors (2). In the circulation, these tertiary complexes remain intact due to the presence of binding protein–specific protease inhibitors. In extravascular tissues, where such inhibitors are absent, cleavage of the binding proteins by a family of specific proteases reduces the stability of these complexes and liberates IGF-I and IGF-II, allowing them to interact with their cell surface receptors (4).

The IGF axis has been shown to play a role in cellular transformation and mammary carcinogenesis (5). Transgenic mice, which overexpress IGF-I (6) and IGF-II (7) specifically in the mammary gland, have an increased incidence of breast adenocarcinoma (6). Regulation of expression of IGF-I and IGF-II is complex. IGF-I is synthesized primarily in the liver, in response to growth hormone secretion, but both IGF-I and IGF-II are produced at other sites, including the breast (1, 2). Increased local production of IGF-I and IGF-II by tumors during the progression of several cancers, including breast cancer (8), suggests that activation of the IGF-I receptor is important for neoplastic growth in vivo.

The first epidemiologic studies to investigate a relationship between circulating levels of IGF-I, IGF-II, and/or IGFBP-3 and breast cancer in human subjects were published in the early 1990s (9-11). Many subsequent studies have sought to replicate—or refute—the findings of these studies. A series of twin studies (12-16) have shown that serum levels of IGF-I, IGF-II, and IGFBP-3 are determined by a combination of genetic and environmental effects. For IGF-I, estimates for the proportion of variance that is explained by genetic effects range from >80% (12, 13) to 38% (14). Several more recent epidemiologic studies have examined the role of genetic polymorphisms within or around the structural genes for IGF-I, IGF-II, and IGFBP-3 in determining serum levels of these growth factors and/or risk of breast cancer.

The purpose of this study is to review all published studies that have sought (a) associations between circulating levels of IGF-I, IGF-II, and their binding proteins and risk of breast cancer; (b) associations between polymorphisms in the IGF system and circulating levels of the protein products; and (c) associations between polymorphisms in the IGF system and breast cancer risk.

Relevant English-language articles published between January 1966 and October 2003 (inclusive) were sought from PubMed and EMBASE using a combination of thesaurus and free-text terms. The databases were searched using the keywords “insulin-like growth factor,” “IGF*,” “serum level/s,” “plasma concentration/s,” “plasma level/s,” “circulating IGF*,” “IGF*, serum,” “IGF* concentration/s,” “IGF* level/s,” “plasma IGF*,” “polymorphism,” “genetic,” and “microsatellite” alone or in combination with “breast cancer risk” or “case control or cohort.” Searches using the MESH terms “breast neoplasms,” “insulin-like growth factor I,” “insulin-like growth factor II,” “polymorphism (genetics),” and “plasma” were also carried out. Reference lists within all relevant articles and reviews were searched to identify publications (including published abstracts from conferences) not captured by the computerized searches. (Details of the search strategy are available on request.)

English-language articles were reviewed and were included if they were based on human subjects and investigated (a) associations between circulating levels of IGF-I, IGF-II, and/or their binding proteins and the risk of breast cancer; (b) associations between polymorphisms in or near the transcribed sequences of the genes coding for the IGFs/IGFBPs and circulating levels of their protein products; (c) associations between these polymorphisms and breast cancer. Articles that only investigated the role of IGFs/IGFBPs as prognostic markers rather than predictors of breast cancer risk or measured expression of IGFs/IGFBPs at the mRNA rather than the protein level were not considered in this review.

Each publication was independently reviewed by two of the authors using a standardized data extraction sheet. For studies of IGFs/IGFPs in relation to breast cancer risk, odds ratio (OR) estimates were extracted for the highest versus lowest concentration categories, with 95% confidence intervals (95% CI) and P_{linear trend}. The change per unit was recorded if IGF/IGFBP levels were analyzed as a continuous variable. For studies that reported an OR for the highest category versus the lowest category and a 95% CI or equivalent measure of variance, a combined estimate was calculated using inverse variance weighting (17). Because of differences in study design, exposure categorization, and adjustment for confounders, the combined estimate cannot be interpreted as a quantitative estimate of any specific comparison and should be used simply as evidence for or against a positive association between circulating levels and breast cancer risk. The percentage of total variation across studies that was due to heterogeneity rather than chance was quantified as I² (18). For studies examining genotype in relation to circulating IGF/IGFBP levels, the frequency of the common allele in the study populations was computed, and for studies of genotype in relation to breast cancer risk, the ORs for common homozygotes versus heterozygotes and for common homozygote versus rare homozygotes were extracted (or calculated) wherever possible. Disagreement between two investigators was resolved by discussion with all authors.

The first study to examine the association between circulating IGF levels and breast cancer risk was published by Peyrat et al. (9). By the end of October 2003, 21 studies had reported on the association between circulating levels of IGFs/IGFBPs and breast cancer risk. Eight case-control studies (9, 11, 19-24) were excluded from this review because their sample sizes were small, study subjects were recruited opportunistically, and comparisons of means (or median) levels of IGFs/IGFBPs between cases and controls were not adjusted for possible confounders, and no OR estimate was provided (or could be calculated). In some (19-21, 23), women with benign breast disorders were taken as controls, but later evidence indicates that IGFs/IGFBPs may be implicated in the etiology of these diseases (25). (An appendix with further details is available on request).

Main Findings. The characteristics of the 13 studies included in this review are summarized in Table 1. The study by Jernström and Barrett-Connor (26) was nested within the Rancho Bernardo cohort, but as blood samples were collected only after the diagnosis of breast cancer this study was regarded as a case-control study.

Menopausal status at the time of blood draw may modify the effect of IGFs and IGFBPs on breast cancer risk (27), so the main results from the studies summarized in Table 1 are presented for premenopausal and postmenopausal women separately wherever possible. IGF-I levels decline with age (28) and all studies that stratified on menopausal status at the time of blood draw reported higher levels at premenopausal ages.

For premenopausal women, one of three case-control studies and three of five prospective studies reported a significant linear trend and an OR of at least 2 for women who were in the highest category of circulating IGF-I level relative to those in the lowest (refs. 27, 29-31; Table 2; Fig. 1). The weighted average of the IGF-I effect estimates across the studies [after exclusion of the study by Petridou et al. (32) where IGF-I was analyzed as a continuous variable] was 1.6. There was, however, moderate evidence of heterogeneity between the studies (I² = 51%), with two reporting ORs of <1.0 (33, 34). Restricting the analysis to women who were both premenopausal and aged <50 years at the time of blood draw (27) or to those who were premenopausal at the time of blood draw and aged <50 years at the time of diagnosis of breast cancer (35) increased the magnitude of the IGF-I effect (Table 2). In one study (31), however, the effect of premenopausal levels of IGF-I was stronger when the analysis was restricted to women who had breast cancer after age 48 years, but the point estimates were based on rather small numbers of cases (Table 2). Five studies were adjusted for circulating levels of IGFBP-3; the effect of IGF-I was strengthened after this adjustment in one study (27) but not in the others (refs. 29, 30, 33, 35; Table 2).

None of the six prospective studies examining IGF-I and breast cancer risk in postmenopausal women found an OR that was statistically significantly different from one (Table 2; Fig. 1). Two of the five case-control studies found a positive trend in the OR of breast cancer with increasing levels of IGF-I (29, 36). The weighted average of the study-specific IGF-I effect estimates (after exclusion of the studies where data were analyzed as a continuous variable; refs. 26, 32, 36) was 1.09 and there was no evidence of heterogeneity between studies (I² = 1.5%). Five studies adjusted for circulating levels of IGFBP-3, but this adjustment made little difference to the results (refs. 27, 29, 30, 33, 37; Table 2).

Two case-control studies provided data for women with no stratification by menopausal status or age (Table 2). One showed a strong positive association between the IGF-I/IGFBP-3 ratio, a marker of IGF-I bioavailability, and breast cancer risk, with women in the top quintile having a 7-fold increase in risk relative to those in the bottom one (10). The other reported nonsignificant positive associations with levels of IGF-I and the IGF-I/IGFBP-3 ratio but based on a small sample (only 40 cases; ref. 38).

Only two studies, both with a case-control design, have examined the role of circulating levels of IGF-II on breast cancer risk and neither found statistically significant evidence of an association, although the point estimate from the largest (29) was consistent with a positive effect (Table 2).

One of the two case-control studies that examined the effect of premenopausal IGFBP-3 levels on breast cancer risk and three of the four prospective studies reported OR estimates of at least 2 (refs. 29-31, 35; Table 3; Fig. 1), but some had wide 95% CI, and in one study (35), a borderline statistically significant effect was seen only when the analysis was restricted to women who were ages <51 years at the time of breast cancer diagnosis (Table 3). The weighted average of the study-specific IGFBP-3 effect estimates was 1.62, consistent with a positive association, but with evidence of moderate heterogeneity between studies (I² = 54%). In addition to the data presented in Table 3 or Fig. 1, Bruning et al. (10) reported statistically significantly lower circulating levels of IGFBP-3 in premenopausal cases than in controls (P = 0.028) but gave no estimate of the effect size.

Among postmenopausal women, most studies found no association of IGFBP-3 levels with breast cancer risk (Table 3; Fig. 1), although one case-control study (29) reported a statistically significant positive linear trend. The weighted average of the estimates for IGFBP-3 was consistent with an OR of 1.0 and evidence of moderate heterogeneity between studies (I² = 35%). In addition, Bruning et al. (10) reported similar IGFBP-3 levels in postmenopausal cases and controls (P = 0.86). No overall association with IGFBP-3 levels was observed in the only study that did not stratify by menopausal status (38). Hankinson et al. (27) reported a “nonsignificant inverse association between IGFBP-3 and breast cancer risk,” but no effect estimates were provided.

Few studies have examined the association of circulating levels of other IGFBPs with breast cancer risk (Table 3). Four prospective studies analyzed IGFBP-1, but none found a statistically significant effect; however, one study (37) collected nonfasting blood samples. Four prospective studies investigated the role of IGFBP-2, and of these, one study (30) reported a statistically significant protective effect, which was restricted to postmenopausal women.

Study Design and Study Populations. Studies where controls were recruited “opportunistically,” using convenience samples, such as other hospital patients without breast cancer or employees, may be subject to selection bias (38). Studies where controls were drawn from approximately the same population as cases are less likely to be biased (10, 29). Case-control comparisons nested within large prospective studies are the most informative, as they reduce the potential for selection bias as well as measure IGFs/IGFBPs levels before disease onset.

In studies with a non-nested case-control design, the possibility of reverse causality cannot be excluded, as IGF/IGFBP measurements were done only after breast cancer had been diagnosed and sometimes after treatment. Although breast tumor cells express and secrete IGFs, particularly IGF-II and IGFBPs, it is unlikely that tumor production significantly affects postdiagnostic measurements of IGFs and IGFBP-3, as the basal circulating levels of these proteins are very high and their clearance from circulation is very slow. IGFBP-2 is an exception, as it normally circulates at much lower levels. Of much greater concern in case-control studies is the fact that circulating concentrations of IGF-I and IGFBP-3 progressively decline in patients with cancer, consistent with a catabolic host response in which metabolism can be substantially affected even if the tumor is extremely small in relation to body mass. Treatment effects and postdiagnostic changes in lifestyle, particularly diet, further affect circulating levels of the IGFs and their binding proteins (39).

All prospective studies in this review were based on incident cases, but for many case-control studies, it is unclear whether incident or prevalent cases (or both) were recruited. Studies based on prevalent cases (26) could have been affected by survival bias, as circulating levels of IGFs/IGFBPs may predict prognosis of breast cancer (24).

Case and control definitions varied between studies. Some were restricted to histologically confirmed cases (32, 34, 36, 38) or to early disease (10) or operable disease (33). Two studies (27, 34) included both invasive and in situ tumors, but only one presented data separately for each one of these type of tumors (34).

Potential Confounding Variables and/or Effect Modifiers. Mean circulating levels of IGF-I and IGFBP-3 increase from birth to puberty and progressively decline throughout the remainder of life in both sexes (28). The majority of studies took account of the potential confounding effect of age either by matching cases and controls on this variable or by adjusting for it (Tables 1-3).

Ethnic differences in circulating levels of IGF-I and IGFBP-3 have been reported (40, 41). Only five studies provided explicit information on the ethnic origin of their study subjects. Of these, four studies were restricted to a single ethnic group [African American (36), Chinese (29), Caucasian American (26), or White women (30)] and one (38) was matched on ethnicity.

Some studies (26, 29-33, 35-37) adjusted for nutritional intake and/or anthropometric measurements, but adjusted and unadjusted estimates were similar in the few studies that provided both. In principle, however, such adjustment would not be appropriate if these factors lie in the causal pathway between IGFs/IGFBPs and breast cancer. Nutritional intake is a strong determinant of IGF-I plasma concentration in humans, with high-energy diets increasing and energy restriction decreasing circulating levels (42, 43). Adult height and weight are associated with breast cancer risk (44), and some studies (45-47), but not all (46-49), have reported associations of circulating levels of IGF-I and/or IGFBP-3 in adults with height in childhood and adulthood and with adult body mass index. Thus, circulating levels of IGF-I and IGFBP-3 could reflect the relationship among nutrition, growth, and breast cancer risk (42, 50).

Exclusion criteria also varied (Table 1). Some studies excluded women with conditions that are thought to affect circulating levels of IGFs/IGFBPs, such as diabetes mellitus, hepatic disorders, endocrine dysfunction, or nutritional-related problems. Oral contraceptives (51) and hormone replacement therapy (HRT; ref. 47) decrease circulating levels of IGF-I; therefore, the use of these hormones should be taken into account when investigating the role of IGF-I on breast cancer risk. However, the IGF-I effect estimates from the few studies that excluded users (10, 26) or matched/adjusted for hormone use (27, 31, 34, 37) were not consistent (Table 1).

Sample Size. The total number ranged from 30 (36) to 513 (34) for cases and from 30 (36) to 987 (34) for controls, with four case-control studies (26, 32, 36, 38) but no prospective study, having <100 breast cancer cases. Few studies were large enough to satisfy conventional criteria of adequate statistical power, however. For example, ∼300 or 800 cases and equal numbers of controls would be required in each menopausal stratum to ensure that the study would have 90% power to detect an OR (comparing the top quartile with the bottom quartile) of 2.0 or 1.5, respectively, at the 5% significance level. Few studies have these many cases and controls in each menopausal stratum.

Blood Sample Collection and Laboratory Assays. Some studies collected samples after a period of fasting or at particular times of the day. A fasting sample is essential for measurement of IGFBP-1, as circulating levels change acutely, under insulin regulation, throughout the day. There is much less variation for IGF-I, IGFBP-3, and IGFBP-2. Small nighttime variations in IGF-I and IGFBP-3 levels have been described, but these are probably due to posture-related fluid redistribution (52).

Samples were stored at temperatures ranging from −20°C to −80°C. Only a few studies provide information on storage time, but all prospective studies and one case-control study (29) matched cases and controls on time of sample collection. The case-control study by Petridou et al. (32) was not matched on storage time but adjust for it in the analyses. In some studies (e.g., ref. 30), samples had been thawed previously and then refrozen, and this may have affected measurements, particularly of IGFBP-3, which is susceptible to degradation.

IGF/IGFBP concentrations were determined by a variety of assays, including RIA, immunoradiometric assay, and ELISA, either using commercial kits or in-house assays. Commercial kits for measuring IGF-I and IGFBP-3 were developed for the diagnosis of growth hormone disorders, such as acromegaly, growth hormone deficiency, or resistance rather than the investigation of relatively small interindividual variations within normal populations. It is now apparent that the performance of many assays is far from optimal when used to rank individuals within the normal concentration range. The large variation seen between assays, and between batches for some assays, is likely to attenuate the ability of epidemiologic studies to detect statistically significant exposure-disease associations and to provide precise point estimates of the effect.

Conclusions. Overall, the findings are consistent with a positive association between premenopausal levels of IGF-I and IGFBP-3 and subsequent risk of breast cancer. For IGF-I, this is consistent with findings from laboratory-based research and with studies showing similar effects on breast mammographic density (25). For IGFBP-3, the picture is more complex, as the original hypothesis was that high circulating levels of IGFBP-3 would protect against breast cancer by sequestering IGF-I and preventing it from interacting with cell surface receptors. The epidemiologic evidence, however, suggests that in premenopausal women at least high levels of IGFBP-3 may be associated, independently or as a marker of other biological processes, with an increased risk of breast cancer. IGFBP-3 has been found to exert dual regulatory effects on IGF-I action. By binding IGF-I, IGFBP-3 also increases the half-life of IGF-I, protecting it from degradation and hence increasing the amount that can reach local tissues. Thus, although IGFBP-3 can inhibit the action of IGF-I on cell proliferation and apoptosis, it may also enhance its effects by increasing pericellular concentrations of IGF-I (53).

Although the relative risks associated with high levels of IGF-I and IGFBP-3 are likely to be relatively modest, this exposure could still account for a considerable proportion of breast cancer cases, as the percentage of premenopausal women exposed to high circulating levels of these proteins is high. Assuming a linear association between premenopausal levels of these proteins and risk of subsequent breast cancer, with OR estimates in the second, third, and fourth quartiles relative to the bottom quartile of 1.2, 1.4, and 1.6, respectively (the latter being consistent with the weighted average of study-specific effect estimates found here), the population attributable risk fraction for high levels of each one of these proteins would be ∼20%. These calculations are, however, rather simplistic, as the effects of IGF-I and IGFBP-3 on breast cancer risk may not be independent.

Main Findings. Twelve of the 13 articles that investigated circulating levels of IGF-I in relation to polymorphisms focused on a simple tandem repeat (STR) that lies 1 kb 5′ to the IGF-I gene transcriptional start site. This 5′ STR was first identified by Rotwein et al. (54). In normal Caucasian populations, the common allele is the 19 CA repeat allele, which occurs at a frequency estimated to be between 59% (55) and 70% (56), except in one study based on a sample of 56 individuals (57) in which it was estimated to be 30%. Overall, the repeat lengths vary from a minimum of 10 repeats to a maximum of 23 repeats with two alleles, the 19 repeat allele and the 20 repeat allele predominating in Caucasians (70% and 17%, respectively; ref. 58). In other ethnic groups, the distribution of alleles was less extreme. In Black women, for instance, the 18, 19, 20, and 21 repeat alleles occurred at frequencies of 16%, 38%, 19%, and 14%, respectively (56).

Rosen et al. (55) first reported that the 19/19 genotype was associated with decreased levels of serum IGF-I in men and women. Table 4 (and Fig. 2 for Caucasians) summarize the characteristics of this and the 11 subsequent publications that have sought to replicate this association. These 12 publications represent 10 independent comparisons; Jernström et al. (56, 58) report analyses on overlapping samples, as do Vaessen et al. (59) and Rietveld et al. (60). Each study has carried out a slightly different main comparison. Some compared homozygotes for the 19 repeat allele (19/19) with all other genotypes combined (19/− and −/−), some compared all 19 allele carriers (19/19 and 19/−) with those who had no copies of the 19 repeat allele (−/−), and some compared homozygotes for the 19 repeat allele (19/19) with heterozygotes (19/−) and non–19 allele carriers (−/−) separately. Most studies adjusted for age and gender, and several studies (56-58, 60-63) adjusted for additional covariates. Of these 10 comparisons, 3 (55, 58, 64) reported a significant (P < 0.05) association between the 19 allele and lower circulating levels of IGF-I, 5 (57, 62, 63, 65, 66) found no genotype effect, and 2 (59-61) reported a significant association between the 19 allele and higher circulating IGF-I levels (Table 4).

Study Design and Study Populations. Two studies (63, 65) only report results for cases and controls combined, although Giovannucci et al. (65) report that the results did not differ when cases and controls were analyzed separately. If, however, there were a relationship between serum IGF-I levels and disease status and some but not all of this variation were due to the genotype being studied, then the estimate of the genotype effect would be biased if it were estimated from data combining cases and controls.

Seven of the studies in Table 4 (55, 58-62, 64, 65) were based on Caucasian or predominantly (>99%) Caucasian populations. Kim et al. (63) studied Korean women and showed a dramatically different allele frequency for the 19 repeat allele (14% for Koreans versus 59-70% for Caucasians). Their study population, however, was taken from women attending a menopause clinic for bone density assessment and includes 158 (53%) women with osteopoenia and 75 (25%) women with osteoporosis.

Three of the studies (56, 57, 66) include men and women from more than one population of origin, and all three reported a significant difference in the frequency of the 19 repeat allele between one or more of their ethnic groups. In addition, one study (56) found a significant difference in serum IGF-I levels between Black women and White women, whereas another (66) found a significant difference between Latino White women and the other racial/ethnic groups in their multiethnic study. Thus, any analysis of genotype in relation to serum levels that does not take into account ethnic group, such as the analysis by Jernström et al. (56) may be confounded. Simply adjusting for ethnic group, however, may not be appropriate if the allele does not have the same effect in different ethnic groups and this cannot be tested unless the study is powered to test for interactions. This may be particularly important for polymorphisms of no functional significance where any effect is likely to be due to linkage disequilibrium with some other (functional) variant. Kato et al. (57) and DeLellis et al. (66) presented their results stratified by ethnic group, but the number of subjects in each stratum was small. The largest single group in the Kato et al. study was White men (n = 33) and the largest in the DeLellis et al. study was Latino White women (n = 68).

Method of Genotyping. In all of the studies, the region of IGF-I containing the 5′ STR was amplified by PCR using the same primers as those initially used by Rosen et al. (55). These primers generate a product of ∼180 to 200 bp depending on the number of CA repeats (19 CA repeats = 192 bp). Differently sized alleles were separated by PAGE, except in the article by Kato et al. (57) in which the size of the CA repeat was determined by DNA sequencing of the amplified product. The frequency of the 19 repeat allele was unusually low in this study (30% versus 59-70%) and there were no 19 repeat allele homozygotes at all. Sequencing of regions of highly repetitive DNA may cause difficulties, although this is most likely to be the case for heterozygotes where the electrophenogram may be difficult to interpret. Departures from Hardy-Weinberg equilibrium (HWE) in the control population may be indicative of genotyping problems (67). Some studies tested for departures from HWE in controls (59, 62, 64), but many did not.

Sample Size. In a recent letter examining the replication validity of genetic association studies, Ioannidis et al. (68) found that the first study published tended to report more extreme estimates of disease association than subsequent studies, particularly when the first study published had a relatively small sample size. The first association study examining the relationship between serum IGF-I levels and the STR 1 kb upstream of the IGF-I gene was based on measurements in 116 controls and reported a 19% (crude) difference between 19/19 homozygotes and 19/− and −/− genotypes combined. Only two of the subsequent studies (59, 61) reported differences of similar magnitude (>15%) between genotypes, but in both of these studies, the comparison was for 19/19 versus −/− and the difference was in the opposite direction.

Based on the data in Table 2, and assuming a multiplicative, codominant model, with a frequency of 0.65 for the 19 allele, between 550 (assuming mean serum levels of ∼200 ng/mL) and 1,000 controls (assuming mean serum levels of 150 ng/mL) are required for 90% power to detect a difference of 7.5% between each genotype at 5% significance. Two of the studies from Table 4 genotyped and measured circulating levels of IGF-I in >500 controls from a single population of origin.

Other IGF-I Polymorphisms Data on other polymorphisms in or around the IGF-I gene are sparse. Rasmussen et al. (69) found no nonsense, frameshift, or missense mutations in the coding sequences of the IGF-I gene in 82 probands of type II diabetics. Arends et al. (70) investigated three STRs: the 5′ STR (see above), a STR in the second intron of the IGF-I gene, and a STR (D12S318) that lies 3′ to IGF-I. They showed greater than expected (P = 0.02) transmission of the 191-bp allele of the intronic STR in their family based study of 124 children born small for gestational age. They also showed lower mean serum levels of IGF-I (expressed as SD scores) in children carrying the 191-bp allele.

The National Center for Biotechnology Information public dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/) has mapped two additional single nucleotide polymorphisms upstream of the human IGF-I gene. One is an A/T polymorphism at −421, and the other is a C/T polymorphism at −1229 with respect to the major transcription start site in promoter 1. Promoter 1 (upstream of exon 1) is the major promoter used in mammals; it is active in all tissues in which IGF-I is expressed and produces ∼75% of transcripts in the liver (71). We did not identify any studies examining these polymorphisms in relation to serum levels.

Polymorphisms and Serum Levels of IGF-II and IGFBP-3 Four studies have investigated polymorphisms in and around either the IGF-II gene or the IGFBP-3 gene (Table 4). For IGF-II, several polymorphisms 5′ to the gene, 3′ to the gene, and within intronic sequences have been identified (72). Serum levels, however, have only been investigated in relation to one of these, an A/G single nucleotide polymorphism in the 3′ untranslated region (73). For this single nucleotide polymorphism, AA homozygotes had higher mean serum IGF-II levels than GG homozygotes (683.3 ± 146.9 versus 614.0 ± 124.0 ng/mL; P = 0.01).

For the IGFBP-3 gene, we identified three studies, all of which examined an A/C polymorphism at −202 bp relative to the transcriptional start site. This polymorphism occurs close to the basal promoter within a region that in the rat and bovine genes is rich in binding sites for hormone receptors, including growth hormone, estrogen, thyroid hormone, and glucocorticoids, many of which are conserved in the human IGFBP-3 gene (74). In all three studies, circulating levels of IGFBP-3 decreased as the number of copies of the A allele decreased (AA > AC > CC). Consistent with these in vivo findings, Deal et al. (75) showed that in an in vitro transient transfection assay the C allele had 50% lower activity compared with the A allele.

Conclusions. For the IGF-I 5′ STR, there is no convincing evidence of an effect of genotype on serum levels of IGF-I. The IGFBP-3 −202 polymorphism occurs in a location that could plausibly affect expression levels of the gene and in the three studies investigating this polymorphism in relation to circulating levels of IGFBP-3 the evidence of a modest effect is consistent. The single study investigating the ApaI polymorphism in the IGF-II 3′ untranslated region is suggestive but further data will be needed to confirm or refute this effect.

Main Findings. Four studies have sought a relationship between IGF-I 5' STR and cancer risk (Table 5). Of these, only the first published study (76) found a statistically significant OR (95% CI) of 0.47 (0.21-1.00) for a comparison based on the 19 repeat allele. The only comparison made was of women with one copy of the 19-bp allele versus women with no copies of the 19 allele as, in addition to the unusually low frequency of the common 19 allele in this study (30% versus 59-70% in other studies), there were no common homozygotes (19/19).

Method of Genotyping. As with studies of circulating levels of IGF-I in relation to genotype, all four studies used the same PCR primers as those initially used by Rosen et al. (55) and only one (76) used DNA sequencing rather than size fractionation by PAGE. Case genotypes would not generally be expected to conform to HWE, except in a codominant model where risk alleles are assumed to act multiplicatively (77, 78). Only one study (66) reported testing for HWE in the control population.

Ethnicity. Combining data from more than one population of origin has the potential for confounding by population stratification in association studies. The majority of cases and controls were matched on ethnic group in two studies (57, 76), and the analysis was stratified by ethnic group in another (66). All women in the study by Figer et al. (79) were of Israeli Jewish origin, but the proportion who were of Ashkenazi origin differed among sporadic cases (58%), BRCA1/BRCA2 carriers (98%), and controls (39%).

Sample Size. Under a multiplicative codominant model with a frequency of 0.65 for the 19 allele, ∼400 cases and 400 controls are required for 90% power to detect an OR of 2.0 between women homozygous for the 19 repeat allele and women with two non–19 alleles at a significance of 5% (based on a test for trend of all three genotypes). Substantially larger numbers would be needed to detect more modest effects. Approximately 1,200 cases and 1,200 controls would be needed to detect an OR of 1.5. The study by Missmer et al. (61) genotyped 463 cases and 622 controls from a single population of origin; therefore, if a true association between breast cancer risk and the 19 repeat allele in Caucasians does exist, it is likely to be more modest than the estimate of a 2-fold risk originally suggested by Yu et al. (76).

IGF-II and IGFBP-3 Polymorphisms and Breast Cancer We were unable to find any studies investigating breast cancer risk in relation to polymorphisms in the IGF-II gene. Only one study has sought an association with the −202 IGFBP-3 polymorphism (80). The ORs (95% CI) reported by this study [0.99 (0.77-1.26) for CC versus AC and 0.97 (0.72-1.30) for CC versus AA in which 677 cases and 834 controls were genotyped] were entirely compatible with random variation.

Conclusions. Despite the extreme OR found for the IGF-I 5′ STR in the relatively small study by Yu et al. (76), none of the three subsequent studies found a statistically significant difference in breast cancer risk for any comparison of 19 repeat allele with any other allele/genotype. The evidence available for the A/C polymorphism 5′ of the IGFBP-3 gene is also consistent with chance. Modest relative risks in mozygotes with two copies of these alleles versus noncarriers cannot be ruled out.

The evidence from the studies included in this review suggests a positive association between premenopausal levels of IGF-I and IGFBP-3 and subsequent breast cancer risk. For postmenopausal levels of IGF-I and IGFBP-3, the available evidence is consistent with no effect.

A limitation of any systematic review is the possibility of publication bias. There is, however, no reason to suppose that studies that suggest a protective effect would be more or less likely to be published than those showing an increased risk or that publication bias is more likely in studies of premenopausal women than in those conducted on postmenopausal women. The consensus of a modest risk conferred by higher IGF-I levels and higher IGFBP-3 levels in premenopausal women does, therefore, seem plausible, although the IGFBP-3 effect is opposite to that originally proposed.

Almost all of the studies in this review defined menopausal status by the time of 30), which analyzed menopausal status at the time of diagnosis of breast cancer rather than at the time of blood collection. The prospective studies indicate that premenopausal levels of IGF-I and IGFBP-3 may affect the risk of both premenopausal and postmenopausal breast cancer, raising the possibility that IGF-I/IGFBP-3 levels at younger ages are more biologically relevant possibly because of a synergistic relationship with endogenous sex hormones (81). The association of breast cancer risk with adult height, which in turn is a marker of circulating IGF-I levels in childhood, would be consistent with a hypothesis that IGF-I/IGFBP-3 measurements at younger ages are more relevant. Alternatively, because IGF-I and IGFBP-3 levels decline with age (28), the failure to observe an effect for postmenopausal levels may simply reflect larger assay error when circulating levels are low. Future studies should analyze the effect of age at blood collection as well as age at breast cancer diagnosis to resolve this issue. An improvement in the validity of current laboratory assays would also contribute to more accurate estimation of the IGF/IGFBP effects on breast cancer risk in these subgroups of women.

Circulating levels of IGFs are determined by clearance as well as production and hence are influenced by the six high-affinity IGFBPs and their complex interactions. Tissue concentrations in the breast are affected by local production and clearance rates as well as by circulating levels, and there is evidence of local expression of both IGF-I (82) and IGF-II (83, 84) in breast cancer patients. For logistical and ethical reasons, epidemiologic studies rely on measurements of levels of IGFs/IGFBPs in the serum or plasma, but local concentrations in breast tissue itself are likely to be the more relevant.

Serum levels of IGF-I, IGF-II, and IGFBP-3 are determined by a combination of genetic and environmental effects (12-16), and polymorphisms that influence the level of expression of the structural genes are likely to affect lifetime exposure to IGFs/IGFBPs by both endocrine and autocrine mechanisms. Unfortunately, polymorphisms in sequences that directly affect gene expression are difficult to find (85). This may account for the fact that all of the studies examining genotype, and either serum IGF-I levels or breast cancer risk directly examine a single STR upstream of the IGF-I gene. It is not clear why the length of this STR should influence serum levels. An association between rare alleles of a STR upstream of the HRAS gene and common cancers, including breast cancer, has been reported (86), but for breast cancer at least, this observation remains contentious (87, 88). An in vitro study (89) reports that the length of a STR proximal to an enhancer element in the first intron of the EGFR gene influences transcription levels. There are no known regulatory elements near the IGF-I STR, although it does lie at the 3′ border of a region of high sequence identity between man and mouse. The paucity of data on other polymorphisms that might influence expression levels may be due to a genuine lack of such variants. Despite the evidence of a genetic effect on serum IGF-I levels, there is no reason to believe that this effect operates in cis. The production of IGF-I by the liver is regulated by a variety of factors, including growth hormone and insulin (2), and clearance is influenced by IGF-II, IGFBPs, and acid-labile protein subunit (3, 53).

A functional effect of the IGFBP-3 −202 A/C polymorphism is biologically plausible, and both in vitro and in vivo evidence is consistent with reduced expression from the C allele and it is surprising that only one study has examined this polymorphism in relation to breast cancer risk. Investigators may have been deterred by the large sample sizes required to estimate small relative risks. Even if the relative risk in AA homozygotes was 1.3 (the upper bound of the 95% CI for the point estimate), ∼2,500 cases and 2,500 controls would be required for 90% power at 5% significance. As the frequency of the “risk” allele is so high (∼0.5), a relative risk of 1.3 in AA homozygotes would correspond to a population attributable fraction of 12%, and even a relative risk of 1.10 would correspond to a population attributable fraction of ∼5%. Risk alleles with a lower population frequency may confer a higher risk to the individual but will account for a much lower population attributable fraction. For instance, a germ line mutation (1100delC) in the cell cycle checkpoint kinase CHEK2 has been shown to confer a relative risk of ∼2-fold, but the frequency of this variant is probably 1% (90) or less (91) in the general population, resulting in a population attributable fraction for this variant of only ∼1%.

Epidemiologic studies into the role of circulating levels of IGFs and/or polymorphisms in/around IGF-I, IGF-II, and IGFBP-3 genes on breast cancer risk must be adequately designed. Large studies (or combined analyses of data from consortia of smaller groups) are essential. Only prospective studies can exclude reverse causality in relation to circulating IGF/IGFBP levels and breast cancer risk, but cross-sectional studies on healthy individuals are appropriate for investigating associations between polymorphisms and circulating levels of IGFs/IGFBPs, and case-control studies are the most cost-effective choice for examining the effect of polymorphisms on breast cancer risk.