The breast cancer specimens were obtained from primary breast cancer patients in an ongoing population-based cohort, BC Blood Study, consisting of 1,116 patients. The patients were included between October 2002 and June 2012 at the Skåne University Hospital, Lund, Sweden as previously described (29). The included patients were between 24 and 99 years old and had no previous history of cancer within the last ten years. A research nurse measured the patients’ anthropometric factors: weight, height, waist and hip circumference, and breast volume (30, 31) prior to surgery. In addition, patients were asked to fill in questionnaires regarding lifestyle factors as well as reproductive history and use of exogenous hormones or other medications during the past week. Patients were classified as diabetic if they self-reported use of any kind of antidiabetic treatment.

Clinicopathological data and information about breast cancer recurrences and deaths were obtained from medical records, pathology reports, the Regional Tumor Registry and the Population Registry (32–34). Adjuvant breast cancer treatment was administered according to standard of care and was recorded until the first breast cancer event. In patients without any breast cancer events, treatments were recorded until last follow-up or death prior to July 1, 2014. Patients may have received more than one type of adjuvant treatment during follow-up. As of November 2005, human epidermal growth factor receptor 2 (HER2) evaluation was introduced and trastuzumab was included in the adjuvant treatment setting. These were entered as a covariate in subgroup analyses of patients. Aromatase inhibitors (AIs) are mainly offered to postmenopausal patients and the impact of nuclear InsR on AI-treatment response was therefore only analyzed among patients ≥50 years.

This study was carried out in accordance with the recommendations of the local ethics committee at Lund University with written informed consent from all patients. All patients gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the local ethics committee at Lund University (Dnr 75-02, Dnr 37-08, Dnr 658-09, Dnr 58-12, Dnr 379-12, Dnr 227-13, Dnr 277-15, and Dnr 458-15).

The final study cohort consisted of 900 patients after excluding patients who received preoperative treatment (n = 51), patients with only ductal carcinoma in situ (n = 39), patients with distant metastasis ≤0.3 years from baseline (n = 8), or patients with no evaluable invasive tumor tissue on tissue microarray (TMA) (n = 118). The report followed the Reporting recommendations for tumor MARKer prognostic studies (REMARK) criteria (35).

Representative tumor regions of formalin-fixed paraffin-embedded tissue blocks were collected from surgical specimens and assembled in a TMA containing duplicate 1.0 mm cores, using a semi-automated tissue array device (Beeches instruments, Sun Prairie, WI, USA). Freshly cut sections were automatically deparaffinized and pretreated using the PT Link system (DAKO, Glostrup, Denmark). All slides were stained simultaneously in one batch to obtain identical conditions in order to minimize staining variation due to technical factors. Immunohistochemistry was performed using the Autostainer Plus from DAKO with the EnVision FLEX high-pH kit, according to the manufacturer’s instructions (DAKO, Glostrup, Denmark) with the InsR (β-subunit) (GR36, Calbiochem; dilution 1:50) antibody. The InsR antibody detects the β subunit, which is identical for the isoforms InsR-A and InsR-B. InsR expression in the present study thus reflects the total amount of InsR. ERα was routinely stained for clinical purposes at the Department of pathology in Lund and ERα status was obtained from pathology reports (32–34). In Sweden, an ERα cutoff of >10% of stained nuclei is still used in the clinic compared to the 1% cutoff that is used in other countries.

The immunohistochemical staining was evaluated by two independent observers (SBj, AR) blinded to tumor characteristics and patient information. Reexamination was performed in case of discrepancy (1.5%) until consensus was reached. The evaluated tumors were divided into negative or any positive nuclear staining. In case of bilateral tumors, all scores were based on the same tumor when evaluating the combined nuclear, cytoplasmic and membrane InsR expression. The prognostic impact of the cytoplasmic and membrane InsR staining intensities have been reported previously (9).

All statistical analyses were calculated using he SPSS software versions 22 (IBM). χ² tests were used for analyses between dichotomized expression levels in different cellular compartments. Logistic regression tests were used for analyses between expression levels and categorical variables, presenting both crude odds ratios (OR) and adjusted OR (OR_adj) for time between surgery and staining (TBSAS, years). Kaplan–Meier curves and LogRank tests were used for univariable survival analyses. Cox regression was used for multivariable analyses providing hazard ratios (HRs) with 95% confidence intervals (CIs) adjusted for age (continuous), invasive tumor size (≥21 mm or skin or muscular involvement), any axillary lymph node involvement, histological grade III, ER status, BMI (≥25 kg/m²), TBSAS (years), and treatments. Two-way interaction terms between nuclear InsR and BMI (≥25 kg/m²), and between nuclear InsR and ER status, and three-way interaction terms between nuclear InsR, BMI (≥25 kg/m²), and ER status were calculated and used in adjusted Cox regression analyses to investigate potential effect modifications. In case of bilateral tumors (n = 15), the highest score was applied. The tumor characteristics from the same tumor were used in all analyses where tumor characteristics were included. Sensitivity analyses were performed using the scores for the contralateral invasive tumor. Restriction analyses were performed where patients who reported pre-operative treatment with anti-diabetic medications were excluded.

Breast cancer events were defined as local or regional recurrences, contralateral cancer or distant metastasis, and the time to event is referred to as event-free survival (EFS). For distant metastasis-free survival (DMFS), only distant metastases were considered events, and death due to any cause was the only considered event for overall survival (OS). Patients without events were censored at the last follow-up or death prior to July 1, 2014. Patients were followed from inclusion to the first breast cancer event, distant metastasis, or death, respectively.

Power calculations assuming 900 patients with an accrual interval of 10 years and additional follow-up time of two years and a frequency of 25% of positive nuclear InsR tumors showed that the study was able to detect true HRs between 0.762 and 1.342 with 80% power and an α of 5% (36). All P-values presented are two-tailed and nominal and were not adjusted for multiple testing since this is an exploratory study. A P-value of less than 0.05 was considered significant.

Figure 1A illustrates the distribution of nuclear InsR staining. There were 900 tumors (88.4%) with available nuclear InsR scores of which 214 (23.8%) tumors were positive for nuclear InsR. There was a significant association between longer TBSAS and positive nuclear staining: nuclear InsR⁺: 9.0 years (IQR 7.0–10.0, P < 0.001), nuclear InsR⁻: 5.0 years (IQR 3.0–8.0), indicating that the proportion of patients with nuclear InsR⁺ tumors decreased over time with the highest proportion observed among patients included before 2006 (Figure 1B). There were significant inverse correlations in all time intervals between nuclear expression of InsR and membrane expression (negative, positive; all r_s ≤ −0.15, all P-values ≤ 0.047). For cytoplasmic intensity (negative, weak, moderate, strong) and dichotomized cytoplasmic InsR expression (negative/weak versus moderate/strong) there were inverse correlations that were significant in all time intervals with the exception of the years 2008–2009 (all r_s ≤ −0.12, all P-values ≤ 0.048). InsR was rarely present in all three compartments simultaneously.

Tables 1 and 2 present both crude and TBSAS adjusted associations between patient and tumor characteristics and nuclear InsR expression. In the crude analyses in Table 1, nuclear InsR⁺ was significantly associated with lower weight, BMI, and waist-to-hip ratio, as well as younger age. These factors were no longer significant after adjustment for TBSAS. In both the crude and the adjusted models in Table 2, nuclear InsR⁺ was significantly positively associated with HER2 amplification and postoperative adjuvant trastuzumab treatment, but not with ER⁺ status. Nuclear InsR⁺ was also significantly positively associated with postoperative adjuvant chemotherapy in the adjusted model.

There was an inverse correlation between nuclear InsR expression and TBSAS in all three patients groups stratified by BMI status as normal-weight, overweight, and obese (<25.00, 25.00–25.99, 30.00+; all r_s ≤ −0.31, all P-values < 0.001). In contrast, there were no associations between nuclear InsR and BMI in the different time intervals (<2006, 2006–2007, 2008–2009, 2010+), indicating that the association between nuclear InsR and BMI did not change over time.

Patients were followed for up to 11 years with a median follow-up of 5.0 years for patients still at risk. Of the 900 patients, 107 had had any breast cancer event and 67 of these patients had a distant metastasis. Eighty-six patients died due to any cause and 52 of these patients had had a reported breast cancer event prior to death.

Nuclear InsR was not a prognostic marker for EFS, DMFS, or OS among all patients. However, there were significant effect modifications depending on BMI on the associations between nuclear InsR and EFS (P_interaction = 0.007), DMFS (P_interaction = 0.040), and OS (P_interaction = 0.027). After adjustment for prognostic factors and TBSAS, normal-weight patients with nuclear InsR⁺ tumors had a non-significant increased risk for any breast cancer event and non-significant shorter DMFS and OS. In contrast, overweight or obese patients with nuclear InsR⁺ tumors had a significantly decreased risk of any breast cancer event [adjusted HR (HR_adj) 0.48, 95% CI: 0.25–0.92], and of death due to any cause (HR_adj 0.35, 95% CI: 0.17–0.72), and a non-significant longer DMFS.

There were also significant effect modifications of ER status on the association between nuclear InsR and EFS where nuclear InsR⁺ expression in ER⁺ tumors was associated with a non-significant higher risk of any breast cancer event and with lower risk in patients with ER⁻ tumors (P_interaction = 0.003). A similar effect modification between nuclear InsR and ER status was observed regarding DMFS (P_interaction = 0.043), but not with respect to OS. A combined nuclear InsR/ER expression score for each tumor was therefore calculated resulting in four different groups: InsR⁻/ER⁺ (n = 609), InsR⁺/ER⁺ (n = 182), InsR⁺/ER⁻ (n = 32), and InsR⁻/ER⁻ (n = 76). One patient had missing ER status.

The combined expression of nuclear InsR and ER was investigated in relation to prognosis in different breast cancer treatment groups. For endocrine treatment, only patients with ER⁺ tumors were included. Any breast cancer event was used as a marker for poor adjuvant treatment response.

Sensitivity analyses were performed due to the 15 patients with available bilateral tumors. Nine patients had evaluable invasive tumor tissue on the contralateral side and five tumors changed dichotomized status for the patients with two evaluable tumors. The effect estimates remained essentially the same in all sensitivity analyses using the dichotomized status for the contralateral tumor. In total there were six analyses that went from significant to borderline significant and three analyses that went from borderline to significant.

There were 29/898 patients (3.2%) who self-reported any kind of antidiabetic treatment based on the preoperative questionnaire (two missing) who were excluded in the restriction analyses. Among these 29 patients, there were two breast cancer events, no distant metastasis, and nine deaths. The frequency of nuclear InsR was non-significantly lower in patients who reported anti-diabetic treatment (13.8%) compared with other patients (24.2%; P = 0.20). There were ten analyses that changed from significant to borderline significant and four analyses that changed from borderline to significant, but the effect estimates remained similar.