Circulating Insulin-Like Growth Factors and IGF-Binding Proteins in PSA-Detected Prostate Cancer: The Large Case–Control Study ProtecT

Prostate cancer is the commonest male cancer in industrialized countries, but options for primary prevention, particularly by lifestyle interventions, are currently limited (1). Observations that mortality is several-fold greater in “western” than Asian countries and among immigrants to high-mortality countries (2) point to the metabolic consequences of “westernization” (high meat, milk, and alcohol intake; low exercise; and obesity) as being potentially important in prostate cancer. A common end point of westernization is obesity, insulin resistance, and perturbations of the insulin-like growth factor (IGF) system (3). Nutritionally regulated IGFs and their binding proteins (IGFBP) not only play a key role in somatic growth but also activate potentially carcinogenic intracellular signaling networks (3). The IGF system, therefore, may link westernized diet, obesity, and metabolic factors with prostate cancer (4).

Epidemiologic studies have observed positive associations of circulating IGF-I with prostate cancer (5). Our recent systematic review, however, found that most studies were small (mean = 180 cancers), contained a heterogenous admixture of cases (screen detected, clinically detected, metastatic and fatal), and the results were highly inconsistent (6), with effect-sizes being larger for retrospective (pooled OR per SD increase in IGF-I = 1.26; 95% CI: 1.05–1.52) compared with prospective (1.07; 0.97–1.18) studies. The retrospective studies were largely based on clinically detected disease and may be more likely to contain advanced cases, which could lead to an overestimation of associations as a result of reverse causality. Furthermore, detection bias is a concern in studies lacking standardized methods to determine case–control status (7). For example, if prostate cancer in men with benign prostatic hyperplasia (BPH) is more likely to present symptomatically and hence be diagnosed than in men without BPH, studies based on comparing clinically identified prostate cancer with unscreened controls could show positive associations of IGF-I with prostate cancer if IGF-I causes BPH, not cancer (7). Few epidemiologic studies investigate components of the IGF system other than IGF-I or IGFBP-3. However, a genome-wide association study has identified a susceptibility locus for prostate cancer in the region of the IGF-2 gene (8), and IGFBP-2 is upregulated in prostate cancer cell lines (9), regulates the tumor suppressor gene PTEN (10), increases in progressing prostate cancer (11, 12), and decreases after prostatectomy (13).

We investigated associations of circulating IGF-I, IGF-II, IGFBP-2, and IGFBP-3 with prostate cancer prevalence, stage, and grade in men who had PSA-detected disease, allowing inference to focus on the development of early-stage cancers as opposed to progression of cancer. Our study involved nearly 3,000 case–control pairs identified from among more than 110,000 men who attended a PSA testing clinic in primary care, attenuating detection bias, and making it the largest project to date that we are aware of to investigate the IGF system in prostate cancer. On the basis of previous research, albeit with predominantly clinically detected cases (5, 6, 14), our primary hypothesis was that IGF-I and IGF-II would be positively associated and IGFBP-3 would be inversely associated with risk of PSA-detected prostate cancer and that the magnitude of associations would be stronger for advanced versus localized disease, suggesting a role for IGFs in the progression rather than initiation, of prostate cancer. Our secondary hypothesis was that IGFBP-2 would be positively associated with prostate cancer, given previous suggestions that this peptide may be a tumor marker (9–13).

We carried out a cross-sectional case–control analysis in which cases and controls were identified from (nested within) the Prostate Testing for Cancer and Treatment (ProtecT) study (15). In the ProtecT study, all (approximately 227,000) men aged 50 to 69 years registered at 337 general practices from 9 U.K. cities (centers) were invited, between 2002 and 2009, to have a PSA test at a prostate check clinic. More than 110,000 of these men attended the clinic. Participants with a raised PSA level (≥3.0 ng/mL, approximately 11% of men tested) underwent digital rectal examination (DRE) and 10-core prostate biopsy, and those with confirmed localized prostate cancer were invited to take part in a randomized trial comparing radical prostatectomy, radical radiotherapy, and active monitoring. Histologic material obtained at biopsy was assigned a Gleason score by specialist uropathologists following a standard proforma and then categorized as low (score < 7), mid (score = 7), and high (score > 7) grade. There was no central review of histology, but the uropathologists participated in a blinded audit scheme. Cancers were staged clinically (physical exam, DRE, PSA, biopsy, isotope bone scan where indicated), using the TNM staging system, as either localized (T1/T2, NX, M0) or advanced (T3/T4, N0-3, M0-1). Trent Multicentre Research Ethics Committee approved the ProtecT study and allied prostate cancer research under the auspices of the Prostate Mechanisms of Prostate cancer and Treatment (ProMPT) study. There is no overlap in study period with a previous study based on the feasibility phase of ProtecT, conducted between 1999 and early 2001 (16).

The sample size (3,000 cases, 3,000 controls) was determined a priori to detect ORs of 1.22 comparing the highest versus lowest 3 quartiles of IGFs or IGFBPs at 5% significance, 90% power. Cases were selected from among all men diagnosed with localized or advanced cancer who had provided a blood sample for research. Men who had no evidence of prostate cancer (PSA below the 3 ng/mL threshold or above the threshold but with one or more sets of negative prostate biopsies) and who had provided a blood sample for research were eligible for random selection as controls (approximately 80,000 potentially eligible controls). We stratum matched cases to controls (1:1) by 5-year age group and recruiting general practice.

PSA was measured using standard methods in hospital laboratories at each of the study centers. Blood samples for IGF measurement were drawn at the time of the PSA test, frozen at −80°C within 36 hours, then transferred for assay on dry ice. Concentrations of IGF-I, IGF-II, and IGFBP-3 were measured in the laboratory of Professor Holly by in-house radioimmunoassay (RIA) which measures total IGF-I (17), IGF-II (18), or IGFBP-3 (19) levels, including all forms of IGFBP-3 that have undergone minor fragmentation. Serum IGFBP-2 was measured using a one-step sandwich ELISA (DSL-10-7100; Diagnostic Systems Laboratories). The molar ratio of IGF-I:IGFBP-3, which may reflect bioavailability of IGF-I, was calculated as (0.13 × IGF-I concentration in ng/mL)/(0.025 × IGFBP-3 concentration in ng/mL).

All assays were carried out blind to knowledge of case or control status. Each stratum-matched case–control pair were thawed and assayed together wherever possible (this occurred 79% of the time; occasionally case–control pairs were separated at the assay stage, for example, if not all samples in the individual boxes used for transport were assayed together, thus separating the pairs without the knowledge of the laboratory staff who were strictly blinded to case–control status). Serum pools were collected with high and low levels of each analyte; these were aliquoted and assayed in every assay to assess between-assay variability, and replicates of the same pool were analyzed within each assay to assess within-assay variability. Assays were undertaken between 2007 and 2010 and each assay typically contained 40 samples for IGF-I and IGF-II, 32 samples for IGFBP-2, and 80 samples for IGFBP-3. The intraclass correlations (ICC) for within-assay variability for IGF-I, IGF-II, IGFBP-2, and IGFBP-3 were 0.86, 0.91, 0.95, and 0.88, respectively; the ICCs for between-assay variability were 0.66. 0.84, 0.81, and 0.71.

Self-reported data on ethnicity, smoking, alcohol, exercise, occupation, urinary symptoms, history of BPH, family history of prostate cancer (father and brother), height and weight [to calculate body mass index (BMI) kg/m²] were collected from questionnaires completed before receipt of the initial PSA test result.

IGFBP-2 was log transformed before analysis as its distribution was positively skewed. IGF-I, IGF-II, and IGFBP-3 distributions were approximately normally distributed. To investigate associations of IGFs and IGFBPs with prostate cancer risk, serum IGFs and IGFBPs were categorized into quintiles, based on the control distribution. ORs and 95% CIs for associations of each quintile of IGF or IGFBP with risk of prostate cancer, using the lowest quintile as the reference group, were estimated using conditional logistic regression to account for the matching variables (5-year age group and recruiting general practice), further adjusted for exact age at recruitment as a continuous variable to prevent residual age confounding. In models assuming a linear dose–response relationship, we computed ORs for associations of prostate cancer per SD increase in IGF or IGFBP using conditional logistic regression adjusted for exact age.

We compared our basic conditional logistic regression model, with models also controlling for the following potential confounding factors (each entered separately): BMI (continuous), height (continuous), smoking (current smoker, ex-smoker, never smoked, or unknown), genetic susceptibility (indexed by family history of prostate cancer in father or brother, categorized as yes or no), occupational social class (managerial/professional class, intermediate class, working class, or unknown, based on the social categorization of Rose and O'Reilly (20), physical exercise (low, moderate, high, or unknown intensity; ref. 21), and current alcohol consumption (never, special occasions, lowest third, middle third, highest third). Where these variables were not observed to confound the associations of IGFs and IGFBPs with prostate cancer (by altering the regression coefficient by ≥10%), they were not subsequently included in fully adjusted models. Duration of serum sample storage and assay kit number were observed to cause some variation in IGFs or IGFBPs and were controlled for in all models; although not associated with case–control status, inclusion of these variables in the models reduced the degree of unexplained variation in effect estimates (narrowing the CIs).

In several previous studies (6) models including IGF-I and IGF-II were additionally adjusted for IGFBP-3 because levels of IGFBP-3 regulate the bioactivity of IGFs (22). We therefore also present such models. Likewise, we present models of the association of IGFBP-3 with prostate cancer that are additionally adjusted for IGF-I or IGF-I plus IGF-II because IGFBP-3 is partly regulated by these factors (23). All components of the IGF system can affect the concentrations of others, either through changes in equilibria, and consequently clearance, or via feedback at the pituitary. We therefore also developed a model that was mutually adjusted for all measured IGFs and IGFBPs, for completeness. However, we made an a priori decision to interpret these mutually adjusted models with caution because of the complexity of the interrelationships between IGF-I, IGF-II, IGFBP-2, and IGFBP-3; for example, because IGFBP-3 is itself partly regulated by IGF-I and IGF-II (23), controlling IGF-I or IGF-II associations for IGFBP-3 may represent statistical overadjustment, attenuating, or biasing real causal associations (24). We have previously reported on associations of IGFs and IGFBPs with BMI in ProtecT (25) and shown associations of these peptides with BMI, particularly for IGFBP-2. However, further controlling for BMI in the models made little difference to the effect estimates and we present parsimonious models not controlling for BMI.

In a case-only analysis, we assessed associations of IGFs and IGFBPs with cancer stage (advanced vs. localized) and grade (high vs. mid vs. low grade). Stage analysis was done using unconditional logistic regression, and grade analysis was done using ordinal logistic regression, both models adjusting for exact age and the study center where the recruiting general practice was based (9-level variable). We investigated whether there was any evidence that associations of IGFs and IGFBPs with prostate cancer differed by age (≤60 years, >60 years) using likelihood ratio tests for interaction.

We conducted a sensitivity analysis in which we compared cases with those controls who had also undergone a biopsy because of a raised PSA, but who had no histologic evidence of prostate cancer (“biopsy-negative” controls). We also investigated by linear regression whether IGF or IGFBP concentrations were associated with PSA level among controls, to determine whether IGFs could cause an increase in PSA levels that may be unrelated to prostate cancer.

Finally, we undertook a meta-analysis of the published literature, stratified by cancer detection method (PSA or routinely detected), to obtain pooled estimates of the world-wide literature on associations of IGF-I and IGFBP-3 with prostate cancer. We classified “PSA-detected studies” as those that were nested within PSA-based screening trials for prostate cancer, to capture studies with case–control pairs that would be very similar to our own. All other studies were classified as “routinely detected,” although we accept that these will likely contain an admixture of screen-detected and clinically detected cases. We identified relevant studies from a published systematic review (6) and searched for newly published articles up to September 2011. To compare across studies, we calculated the log OR or risk ratio per SD increase in IGF or IGFBP level using the method of Chene and Thompson (26) or Greenland and Longnecker (ref. 27; detailed methods presented in Rowlands (2009; ref. 6), comparing cases with controls. We then used these estimates to carry out random and fixed effects meta-analysis using the metan command in Stata (28). Our inference is based on random effects models because we identified substantial heterogeneity; however, we present fixed effect models for completeness. The I² statistic was calculated as a measure of the degree of heterogeneity across studies. The meta-analysis was further stratified by study design (retrospective vs. prospective) because of our previous finding that study design is associated with effect size (6), and by whether or not effect estimates were mutually adjusted for other components of the IGF system, because this may have been a source of between-study heterogeneity. All statistical analyses were carried out using Stata 11.0 (Stata Corp., 2009).

We identified 2,890 men with prostate cancer, of whom 2,699 (93.4%) provided a blood sample for research and were eligible for our study. We obtained serum samples for 2,686 (99.5%) eligible cases and 2,766 stratum-matched controls, randomly selected from the sample of all potential controls (approximately 80,000) who provided sufficient blood sample for research purposes (87.8% of all potential controls provided a blood sample). The 13 eligible cases that we could not include in our analysis had an insufficient volume of blood to carry out the assays. Most men self-classified themselves as white ethnicity (99%). Cases were more likely to report a family history of prostate cancer, but BMI, height, and other lifestyle characteristics did not differ between cases and controls (Table 1). IGF-I concentrations were on average lower, and IGF-II, IGFBP-2, and IGFBP-3 higher, in cases compared with controls (Table 1).

Of the 2,686 cases, 2,355 (88%) were localized, 311 (11%) advanced, and stage was not available for 20 (1%). Of the advanced cases, most were T3a (67%), only 7 (2%) had metastases. Most men had low-grade (Gleason < 7) cancer (1,808; 67%), with 720 (27%) mid-grade (Gleason = 7) and 152 (6%) high-grade (Gleason > 7) cancers (not available for 6 men). A total of 176 controls had PSA levels ≥ 3 ng/mL and subsequent negative prostate biopsies. IGF-I and IGF-II concentrations were correlated with each other (r = 0.21) and with IGFBP-3 (r = 0.37 and r = 0.58, respectively); IGFBP-2 was inversely correlated with IGF-I (r = −0.16), IGF-II (r = −0.11), and IGFBP-3 (r = −0.19).

In analyses controlling for age, assay kit, and storage duration (model 2), there was no evidence of an association of IGF-I with prostate cancer (OR: 0.99 per SD increase in IGF-I; 95% CI: 0.93–1.04; Table 2). IGF-II was positively associated with prostate cancer (OR: 1.16; 1.08–1.24), although there was evidence against the assumption of linearity (P < 0.01), likely due to the second quintile having a similar OR (1.45) to the fourth (1.43). The odds of prostate cancer increased with increasing IGFBP-2 (OR: 1.18; 1.06–1.31) and IGFBP-3 (OR: 1.27; 1.19–1.36).

When we additionally controlled for IGFBP-3 (model 3), the previously observed null association between IGF-I and prostate cancer became inverse (OR: 0.85; 0.79–0.91); further adjustment for all other measured components of the IGF system (IGF-II, IGFBP-2, and IGFBP-3) made little difference to the IGFBP-3–adjusted model (OR: 0.81; 0.75–0.87; P < 0.001).

The positive association of IGF-II disappeared after controlling for IGFBP-3 (OR: 0.99; 0.91–1.08) and remained null after controlling for all other measured components of the IGF system, IGF-I, IGFBP-2, and IGFBP-3 (OR: 1.04; 0.95–1.13; P = 0.4). IGFBP-3 remained positively associated with prostate cancer after controlling for IGF-I (OR: 1.42; 1.32–1.53) and additionally controlling for IGF-II (OR: 1.40; 1.28–1.52) or IGF-II and IGFBP-2 (OR: 1.42; 1.30–1.55). The observed positive association of IGFBP-2 with prostate cancer remained after controlling for all other measured components of the IGF system, IGF-I, IGF-II, and IGFBP-3 (OR: 1.10; 1.028–1.17; P = 0.005).

There was no evidence of an interaction of age with IGFs or IGFBPs on prostate cancer risk (all P_interaction > 0.39). There was no strong evidence that associations of any of the IGFs or IGFBPs with prostate cancer differed according to stage (Table 3) or grade.

IGF-I was weakly positively related to PSA level in controls, but IGF-II, IGFBP-2, and IGFBP-3 were not related to PSA level (Supplementary Appendix Table S1). In sensitivity analyses, associations of IGF-I with prostate cancer did not change appreciably when biopsy-negative controls were used, although CIs were wide, reflecting imprecision around point estimates with a small sample size (Supplementary Appendix Table S2). The magnitude of the positive associations of IGF-II and IGFBP-3 became stronger (OR = 1.43 per SD increase in IGF-II; 95% CI: 1.15–1.77; OR = 1.36 per SD increase in IGFBP-3; 95% CI: 1.13–1.63), but the direction of the previously observed positive association with IGFBP-2 was reversed (OR 0.88 per SD increase in IGFBP-2; 95% CI: 0.75–1.03).

We identified 4 publications based on PSA-screened populations (2 retrospective/2 prospective) and 51 studies of routinely detected cancers (34 retrospective/17 prospective; see Supplementary Appendix for reference list). The forest plots and pooled OR for these studies combined with the current study are shown in Figs. 1 (IGF-I) and 2 (IGFBP-3). The pooled IGF-I ORs per SD increase in IGF-I for PSA-detected studies, from random effects models, were 1.02 (95% CI: 0.79–1.33; I²: 81%) for retrospective studies and 1.04 (95% CI: 0.92–1.17; I²:0%) for prospective studies. For routinely detected cancers, the pooled IGF-I ORs were 1.19 (95% CI: 1.00–1.41; I²:92%) for retrospective studies and 1.08 (95% CI: 0.99–1.17; I²: 57%) for prospective studies. For IGFBP-3, the pooled ORs were 1.06 (95% CI: 0.85–1.34; I²:84%) for retrospective, PSA-detected cancers; 0.97 (95% CI: 0.86–1.09; I²:1%) for prospective, PSA-detected cancers. For routinely detected cancers, the pooled IGFBP-3 ORs were 0.73 (95% CI: 0.57–0.93; I²:94%) for retrospective studies and 1.03 (95% CI: 0.95–1.12; I²:40%) for prospective studies.

This is the largest study to date investigating associations of the circulating IGF system with prostate cancer. In models that were not mutually adjusted for other components of the IGF system, we found positive associations of IGF-II, IGFBP-2, and IGFBP-3 levels, although no association of IGF-I, with PSA-detected prostate cancer. There was no strong evidence that any of the IGFs or IGFBPs were related to prostate cancer stage or grade, although we were underpowered to detect modest associations. The magnitude of the increased risk associated with the highest versus lowest quintile of IGF-II, IGFBP-2, and IGFBP-3 is similar to that seen for men with a family history of prostate cancer in a first-degree relative (approximately 50%–100% increased risk).

Our finding of a positive association of IGF-II with prostate cancer is supported by a genome-wide association study that identified a susceptibility locus for prostate cancer in the TH-INS-IGF2AS-IGF2 region on chromosome 11p15.5 (rs7127900; 8). IGF-2 gene overexpression, which can occur via loss of imprinting or LOH (29), has been observed in prostate (30), as well as colorectal (31) and more aggressive ovarian (32) cancers. Few epidemiologic studies have investigated circulating IGF-II in prostate cancer, but two studies report that raised IGF-II was associated with increased prostate cancer risk, the association being stronger in advanced disease (16, 33). That the positive association of IGF-II with prostate cancer disappeared after controlling for IGFBP-3, however, indicates that the observed effect may be secondary to the positive association of IGFBP-3 with prostate cancer. A possible secondary effect of the tumor on levels (reverse causality) should be considered, as positive associations of IGF-II with prostate cancer have mostly been observed in small retrospective studies (6) and IGF-II levels were found to decrease after removal of colorectal adenomas, precursors of colorectal cancer (34). Tumor load in early screen-detected prostate cancers (such as were present in our study), however, would be expected to be negligible, making reverse causality an unlikely explanation of our findings.

There have been very few epidemiologic studies investigating IGFBP-2 in prostate cancer, but some have observed increased IGFBP-2 levels in prostate cancer (12, 35, 36), particularly advanced (12) compared with early disease (37). IGFBP-2 inversely regulates the tumor suppressor gene PTEN (10) that when lost is associated with IGFBP-2 overexpression (38) and prostate cancer progression (39, 40). Thus, serum IGFBP-2 may be an accessible marker of PTEN status and cancer progression. As with IGF-II, reverse causality should be considered as an explanation for our findings because there is some evidence that IGFBP-2 may be originating from the prostate (13), and small studies indicate that serial measures of IGFBP-2 increase over time in prostate cancer that is progressing (12) but decrease after radical prostatectomy (13). Further research, based on serial measures of IGF-II and IGFBP-2 after diagnosis of prostate cancer, is required to determine whether these peptides are tumor markers of potential clinical importance.

We observed a strong, linear positive association of IGFBP-3 with screen-detected prostate cancer that gets stronger with adjustment for IGF-I and IGF-II, in line with a recent large study (41) and a previous meta-analysis of prospective studies (5). IGFBP-3 has antiproliferative and proapoptotic effects, as well as stimulatory effects on cell growth, for example, by increasing the concentration of IGFs near the IGF-IR (42) and, independent of IGF-binding, activating mitogen-activated protein kinase signaling (43, 44). Our meta-analysis (Fig. 2) reveals highly heterogeneous results, however (I² = 92%). A possible reason for inconsistent IGFBP-3–prostate cancer associations could be assay differences: IGFBP-3 exists as both intact and fragmented forms in the blood, each of which may be present with multiple additional posttranslational modifications. Which of these forms is measured in any assay is poorly defined, and it is possible that different IGFBP-3 assays used in different studies are measuring different forms of the peptide, leading to heterogeneity in effect estimates.

Our findings for IGF-I do not seem to be in line with previously published meta-analyses (5, 6, 14). Previous meta-analyses, however, have combined PSA detected and clinically detected prostate cancers and, therefore, consist of a mixed group of cancer types. Cases detected by PSA testing, such as those in ProtecT, represent a very different phenotype [low volume, well-differentiated, localized cancers (45), a substantial proportion of which will never become clinically identifiable (46)], compared with routinely detected cases, which have progressed and have either become large enough, or have metastasized, to be diagnosed routinely. As shown by others (47), it is likely that at least some risk factors for the initiation of prostate cancer differ from those that cause a subgroup of such cancers to eventually progress into clinically manifest disease. The meta-analysis presented here indicates no or weak associations of IGF-I with PSA-detected prostate cancer. Although the pooled estimate from the stratified meta-analysis of cross-sectional studies was largely driven by our ProtecT data, importantly we also observed no association of IGF-I with PSA-detected prostate cancer in prospective studies, an analysis which did not include ProtecT. In contrast, there was a positive association of IGF-I with routinely detected cancer, which was particularly marked in retrospective studies. Positive associations seen in previous reports involving clinically detected disease could, therefore, indicate that IGF-I is not associated with the development of early-stage cancers but with progression to clinically identifiable disease, in line with some studies (but not all; ref. 5) that observed strong positive associations of IGF-I with clinically advanced or aggressive, but not localized, prostate cancer (6). An effect of IGF-I on progression, but not initiation, would be consistent with IGF-I–mediating effects of a western diet on the prevalence of clinically detected and fatal prostate cancers (which varies considerably across the world; ref. 48) but not on the prevalence of autopsy-detected subclinical cancers (which does not vary to the same extent; ref. 49). We observed inverse associations of IGF-I with prostate cancer after controlling for IGFBP-3 or all other measured components of the IGF system. Inverse associations of IGF-I with localized prostate cancer have previously been observed (50–52) but have not previously elicited much attention because of imprecision in effect estimates. As previously stated, we made an a priori decision to interpret associations observed in mutually adjusted models with caution due to the complexity of the IGF–IGFBP-3 relationship and the risk of statistical overadjustment and bias (24). Nevertheless, the findings are of interest given that the correlation of IGFPB-3 with either IGF-I or IGF-II is only moderate (probably because IGFBP-3 levels are primarily regulated by growth hormone) and IGFBP-3 has important IGF-independent effects (43, 44).

The limitations of the study include the fact that very few men had metastatic or high-grade disease, limiting our ability to identify differences in associations by stage or grade; residual confounding due to unmeasured variables may be present; and it comprised mainly white British men, so the results may not be generalizable to other ethnic groups. The IGFs and IGFBPs were measured around the time of diagnosis, so levels may be influenced by the presence of disease; however, it is difficult to understand why the presence of cancer should lower levels of IGF-I and thus explain the null/inverse association we observed for IGF-I with prostate cancer. Our cancers were screen detected, so were very early in their pathogenesis, and a large proportion of them are likely to remain indolent (46); however, we feel that this is a strength of the study as it allows focused inference about associations with the development of early-stage cancers rather than the progression or spread of cancer to those that are clinically apparent. Measurement of circulating IGF and IGFBP levels may not represent biologically active levels in the tissues; therefore, we cannot exclude that IGF-I at the tissue level may play a role in the initiation of early prostate cancer. This is an important consideration given our findings for IGFBP-3, which has been shown to increase the concentration of IGFs near the IGF-IR of cells (42). Assays measuring free IGF-I are available but were not used in this study, as there is some debate that they only measure a subset of the more bioavailable IGF-I in the circulation (53, 54) and free IGF-I in the circulation may be unrelated to free IGF-I in the tissues where the proportion of IGFBPs is very different. An investigation of free IGF-I levels in 545 prostate cancer cases from the Physicians' Health Study found no association between free IGF-I and prostate cancer risk (54). Figures 1 and 2 show wide variation between studies in whether or not associations were mutually adjusted for other components of the IGF system. There were too few studies to undertake a formal meta-regression analysis of the impact of mutual adjustment. However, stratifying the results by whether or not mutual adjustment was done did not reduce the observed heterogeneity and visual inspection does not indicate that effect estimates were markedly different when unadjusted versus mutually adjusted estimates were computed.

This study provides the largest assessment of the role of the IGF system (IGF-I, IGF-II, IGFBP-2, and IGFBP-3) in the development of PSA-detected prostate cancer. The evidence suggests potentially important roles for IGFBP-2 and IGFBP-3 (and possibly IGF-II) but no association of IGF-I. We speculate that circulating IGF-I has a limited role in the development of early prostate cancer but may remain an important risk factor for disease progression. Thus reduction of circulating IGF-I is unlikely to be a useful strategy for the primary prevention of prostate cancer. Future research should focus on large studies that can confirm or refute a role of IGF-II, IGFBP-2, and IGFBP-3 in PSA-detected prostate cancer and investigate the association of IGF-I with progression of PSA-detected prostate cancer; such analyses would determine whether IGFs have a potential clinical translational role in predicting which men with localized disease might progress and hence require immediate radical treatment.

M-A. Rowlands had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. No potential conflicts of interest were disclosed by the other authors. The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the Department of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

The authors thank for the tremendous contribution of all members of the ProtecT study research group, and especially the following who were involved in this research: Prasad Bollina, Sue Bonnington, Lynne Bradshaw, James Catto, Debbie Cooper, Michael Davis, Liz Down, Andrew Doble, Alan Doherty, Garrett Durkan, Emma Elliott, David Gillatt, Pippa Herbert, Peter Holding, Joanne Howson, Mandy Jones, Roger Kockelbergh, Howard Kynaston, Athene Lane, Teresa Lennon, Norma Lyons, Hing Leung, Malcolm Mason, Hilary Moody, Philip Powell, Alan Paul, Stephen Prescott, Derek Rosario, Patricia O'Sullivan, Pauline Thompson, Lynne Bradshaw, Sarah Tidball. They also thank the men who participated in this study and the NIHR Cambridge Biomedical Research Centre. Ross Harris carried out the dose–response calculations for the meta-analysis. Val Laundy, Semih Dogan, Li Zeng, Ola Wojtowicz, and Kalina Zdunek conducted the IGF assays.

The authors also thank for the support of the National Cancer Research Institute (NCRI) formed by the Department of Health, the Medical Research Council (MRC), and Cancer Research UK.

This work was supported by Cancer Research U.K. project grant C18281/A7062. The ProtecT study is funded by the U.K. Health Technology Assessment (HTA) Programme of the NIH Research, HTA 96/20/99; ISRCTN20141297. The NCRI provided funding through ProMPT (Prostate Mechanisms of Progression and Treatment) and this support is gratefully acknowledged. G. Davey Smith and R.M. Martin work within the CAITE centre, which is supported by the MRC (G0600705) and the University of Bristol. D. Gunnell, F. Hamdy, J. Donovan, and D.E. Neal are NIHR Senior Investigators.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.