Transforming growth factor beta 1 levels predict echocardiographic changes at three years after adjuvant radiotherapy for breast cancer

Altogether, 63 patients with available serum samples, receiving postoperative RT for breast cancer or ductal carcinoma in situ (DCIS) were included in this prospective, observational, single-center study. The included patients had breast-conserving surgery (n = 62) or a mastectomy (n = 1) before RT, but none received chemotherapy. Earlier publication describes the key inclusion and exclusion criteria in detail [11]. The Tampere University hospital ethics committee approved the study (R10160) and informed consent was obtained from all participants.

A detailed description of the RT protocol is found in an earlier publication [12]. Briefly, the planning target volume (PTV) received either 50 Gy in 2 Gy fractions or 42.56 Gy in 2.66 Gy fractions. The remaining breast after breast-conserving surgery or the chest wall after mastectomy with margins constituted the PTV. For the one patient with axillary node-positive disease, PTV included the axillary and supraclavicular areas.

TGF-β1, PDGF and N-terminal pro-brain natriuretic peptide (proBNP) were analyzed from serum samples drawn at the start of RT, at the end of RT and 3 years after RT. Concentrations of TGF-β1 and PDGF-AB were measured by enzyme-linked immunosorbent assay with reagents from R&D Systems Europe Ltd. (Abingdon, UK). Samples were stored at − 80 °C and all samples were analyzed simultaneously with reagents from the same batch. Samples from before and after RT were also reanalyzed [10]. The detection limit and the interassay coefficient of variation were 7.8 ng/ml and 5.1% for TGF-β1 and 3.9 ng/ml and 3.5% for PDGF-AB, respectively.

Echocardiographic examination was performed at the same time points as when the serum samples were drawn. A single cardiologist (ST) performed all the examinations by a commercially available ultrasound machine (Philips iE33 ultrasound system; Philips, Bothell, WA, USA) and a 1–5 MHz matrix-array X5–1 transducer, as described previously [13, 14].

The median and interquartile range (IQR) were calculated for variables with skewed distributions. The Wilcoxon signed-rank test was used to test for change in a variable between measurements before RT, after RT and at 3 years. The linear relationship between continuous variables was determined using Spearman’s correlation. The Mann-Whitney U-test for continuous variables was used to test for differences in the biomarkers, echocardiographic measurements or radiation doses between two groups. Group based trajectory modeling was used to determine the two trajectory groups [15]. The trajectory groups were created according to the three measurements of TGF-β1 or PDGF in each patient as a continuous outcome measure and the groups represent clusters of individuals with similar trajectories and outcomes over time [16]. Models were fitted by using the flexmix package [17] of the statistical program R, version 3.3.0, from the R Foundation for Statistical Computing [18]. The relative goodness of fit was assessed using Bayesian information criteria (BIC). Fischer’s exact test was used to test for differences in categorical variables in the two trajectory groups. Linear regression was used to test multivariable associations with the change in GLS over 3 years. IBM SPSS Statistics software, version 25 for Windows (Armonk, NY, USA) was used for statistical testing. P-values under 0.05 were considered statistically significant.

Including all 63 patients, the median TGF-β1 decreased significantly during RT, p = 0.003 (Table 1). At the three-year follow-up, the median TGF-β1 level remained lower than before RT, p = 0.001. The median PDGF also decreased during RT, p = 0.001. The difference in median PDGF levels between before RT and at 3 years remained significant, p = 0.046. The median proBNP (n = 62) was stable during RT, p = 0.325, but increased by the three-year follow-up, p < 0.001.

The correlations of TGF-β1 and PDGF at corresponding time points and the changes between these time points are shown in Table 2. There were significant correlations between the TGF-β1 and PDGF as well as between the TGF-β1 and proBNP levels (Table 2), but PDGF and proBNP did not correlate together.

Age inversely correlated with the TGF-β1 level at 3 years (rho = − 0.310, p = 0.013) and the change in TGF-β1 from before RT to 3 years (rho = − 0.280, p = 0.026). There was no significant correlation between TGF-β1 and body mass index (BMI) or time from surgery to RT. Furthermore, there was no statistically significant difference in the TGF-β1 levels before RT in groups with different comorbidities or the use of medication, e.g. hypertension, hypothyrosis, coronary artery disease (CAD), smoking, diabetes and use of aromatase inhibitors (AI), tamoxifen, angiotensin convertase inhibitors (ACE), acetylsalicylic acid (ASA), or statins.

PDGF at 3 years also inversely correlated with age (rho = − 0.348, p = 0.005), but not with BMI or time from surgery to RT. Statin-users had lower median PDGF levels before RT than did nonusers, 12.0 (10.2–16.5) ng/ml and 16.5 (13.4–20.4) ng/ml, respectively (p = 0.024). Furthermore, tamoxifen-users had lower median PDGF levels after RT than nonusers did, 12.1 (11.6–17.3) ng/ml and 15.6 (12.8–20.1) ng/ml, respectively (p = 0.041). There were no differences in the median PDGF levels according to the other baseline characteristics.

The TGF-β1 and PDGF levels before RT, after RT, at 3 years, and the changes between these time points correlated with different structural and functional parameters in echocardiography. Echocardiographic measurements are presented in Additional file 1: Table S1. The significant correlations for TGF-β1 and PDGF after RT and at 3 years and echocardiography are presented in Additional file 2: Table S2.

Trajectory analysis was performed to group the patients by behavior of TGF-β1. Group 1 (n = 19) had significantly higher levels of TGF-β1 before RT, after RT and at 3 years than group 2 (n = 44), p < 0.001 for all time points (Table 3). In group 1, there was a tendency for TGF-β1 levels to decrease from before to after RT and to the three-year follow-up, p = 0.066 and p = 0.080 respectively. In group 2, there was a significant decrease in the TGF-β1 level from baseline to after RT, p = 0.023, and to the three-year follow-up, p = 0.006. The groups were similar in baseline characteristics (Table 3), and there were no significant differences in the proBNP levels at the three time points between the groups.

Echocardiographic parameters by the two trajectory groups are shown in Table 4. Baseline measurements were similar between the two groups. The IVS at 3 years, the PW after RT and the PW at 3 years were significantly different between the groups, p = 0.016, p = 0.039 and p = 0.010, respectively. There was a tendency for a difference in GLS at 3 years, p = 0.081. During RT, there was a significant change for group 1 in IVS (p = 0.036), PW (p = 0.030), TAPSE (p = 0.021) and scIBS (p = 0.030). For group 2 there was a significant change during RT in TAPSE (p = 0.030) and scIBS (p = 0.007). During the three-year follow-up, GLS and scIBS significantly worsened from baseline in group 1, p = 0.013 and p < 0.001, respectively. In group 2, PW decreased and scIBS increased from baseline to 3 years, p = 0.028 and p = 0.002, respectively. Radiation doses to the heart, left ventricle (LV), right ventricle (RV) and the left anterior descending artery (LAD) were similar between the two trajectory groups (Additional file 3: Table S3).

To further explore the association between TGF-β1 and GLS suggested by correlation and the significant worsening in trajectory group 1, multivariable linear regression analysis was performed. In the model, TGF-β1 trajectory group 1 (β = 0.27, p = 0.013), left-sided breast cancer (β = 0.39, p = 0.001) and the use of AI (β = 0.29, p = 0.011) were significantly associated with a reduction in GLS from before RT to 3 years. Additionally, there was tendency for age to be associated (β = 0.18, p = 0.071) with worsening GLS during the three-year follow-up. These factors explained 33% of the change in GLS.

A trajectory analysis was also performed for PDGF. PDGF levels were significantly higher at all time-points in group 1 (n = 8) than in group 2 (n = 55), p < 0.001 (Additional file 4: Table S4) for all time-points. The groups did not differ in baseline characteristics (Additional file 4: Table S4). The change in PDGF was only significant in group 2 from before to after RT, p = 0.001.

Only scIBS at 3 years was significantly higher in group 1 than group 2, p = 0.044. The elevated PDGF levels in group 1 were not associated with more changes in echocardiographic parameters, but the group 1 was too small for a meaningful comparison (Additional file 5: Table S5). Furthermore, radiation doses to the heart, LV, RV or LAD were similar in the groups (Additional file 4: Table S4).

The most important finding in our study was the association of elevated TGF-β1 before RT with a decline in LV systolic function, namely, impairment in GLS during the three-year follow-up. This association was apparent in the correlation between TGF-β1 and GLS at 3 years and further with the trajectory analysis in which patients were grouped into two groups according to TGF-β1 behavior. Group 1 had significantly higher baseline TGF-β1 levels than group 2. At baseline, the echocardiographic parameters were similar, but RT induced a thickening of the IVS and PW during RT in group 1, but not in group 2. These changes most likely depict RT-induced inflammatory changes. During the three-year follow-up, group 1 had a significant worsening of GLS and group 2 did not. In multivariable analysis, the trajectory group 1 remained a significant factor in predicting GLS worsening during the 3 years in addition to AI use and left-sided breast cancer, which we have reported in earlier to affect echocardiographic parameters [12, 13]. Left-sided breast cancer associates with higher radiation doses to the heart [19], which probably explains the significance of the side. Radiation dose is a significant factor determining cardiovascular risk of breast cancer patients and dose-volume constraints are in clinical use to decrease this risk [20]. Radiation doses were similar between the groups, as the grouping mostly reflects a difference in baseline TGF-β1. The worsening in GLS is probably the most clinically significant echocardiographic change, as GLS is an excellent early predictor of major adverse cardiac events [21]. Additionally, the correlation between the levels of TGF-β1 and proBNP, an accepted marker of heart failure [22], at all three time-points further supports the association of elevated TGF-β1 levels and cardiovascular changes.

In previous literature, two other studies with adjuvant external beam RT [8] and intracavitary partial breast brachytherapy [9] reported elevated circulating baseline TGF-β1 levels in patients who developed radiation-induced fibrosis of the breast. Studies with genetic mutations also support the idea that baseline TGF-β1 levels may predispose individuals to normal tissue toxicity from RT. Circulating levels of TGF-β1 are, at least partially, controlled genetically [3]. The TGF-β1 C-509 T variant allele (rs1800469) is associated with elevated levels of circulating TGF-β1 [3]. Prospective studies present contradictory evidence of the association between TGF-β1 C-509 T variant allele and radiation-induced fibrosis of the breast [4, 5]. There is some evidence that mutations in TGF-β1 genes are linked to cardiovascular risk. In patients receiving postoperative RT for breast cancer, patients with the TGF-β1 29C > T variant allele, associated with low TGF-β1 levels, had an increased cardiovascular risk with HR 1.79. However, in this study, there was no association between CV mortality and radiation dose [7]. Furthermore, the role of TGF-β1 is not clear in cardiovascular diseases. The evidence on the role of TGF-β1 in atherosclerosis is contradictory, but most studies suggest TGF-β1 inhibits atherosclerosis [23]. In contrast, elevated levels of TGF-β1 are associated with, for example, hypertrophic cardiomyopathy [23].

Earlier, we reported that TGF-β1 and PDGF decrease during RT [10], but this study shows that TGF-β1 remains at a lower level at 3 years compared to before RT. The decreases were small but statistically significant. In a suggested model, RT is thought to induce a cytokine release [2], but since we only have measurements for before and after RT, a cytokine release could have taken place earlier during the RT course. One previous study with intraoperative RT (IORT) for breast cancer reported that IORT had no effect on the TGF-β1 levels measured from wound fluid [24]. We found that a decrease in TGF-β1 during RT and the three-year follow-up correlated with worsening LV systolic function e.g. GLS. Additionally, the three-year change was correlated with an increased echodensity of the myocardium, scIBS, and LV measurements. The inverse correlations between the change in TGF-β1 levels and proBNP levels, further suggest that the decrease in TGF-β1 may be associated with an increased cardiovascular risk.

We also found associations between elevated baseline PDGF levels and LV systolic function, e.g. worsening GLS, and LV diastolic function, e.g. decreased mitral E. The change in PDGF during RT also predicted impairment in GLS over 3 years. The trajectory analysis did not further support the usability of PDGF in predicting RT-induced echocardiographic changes. Additionally, there were no correlations between PDGF and proBNP. Therefore, despite the strong correlation between the two biomarkers, TGF-β1 seems to be more usable than PDGF in predicting the cardiovascular effects of RT. To our knowledge, no previous studies exist about the PDGF and RT-induced toxicity in humans.

Although we now present results with 3 years of follow-up, the follow-up time is still short, considering that the increased risk of cardiovascular effects of RT takes years to manifest. Longer follow-up will show whether the echocardiographic changes lead to clinical cardiovascular morbidity. This may require larger studies than ours. Additionally, the underlying cause of elevation of TGF-β1 is not known nor do we have information on how well other risk factors of cardiovascular diseases, such as elevated blood pressure, diabetes, and hyperlipidemia, are controlled in the patients. These factors could influence the results.