This study was approved by Baylor College of Medicine Institutional Animal Care Use Committee. This was a longitudinal study of 90 wild-type C57BL/6 mice (Jackson Laboratories, Bar Harbor, ME) aged 6-8 weeks randomized to doxorubicin (DOX group, n=45), doxorubicin + dexrazoxane (DOX/DEX group, n=30) or placebo with saline injection (control group, n=15). The DOX group received doxorubicin 3 mg/kg once per week by intraperitoneal injection for a total of 6 weeks. The DOX/DEX group received doxorubicin 3 mg/kg plus dexrazoxane 30 mg/kg intraperitoneal injection once per week for a total of 6 weeks. These doses were chosen to simulate the development of anthracycline-induced cardiotoxicity. The control group received intraperitoneal saline (same volume as the DOX group) once per week for 6 weeks.
The key measures assessed by the CMR were LV and right ventricle (RV) size, LVEF, RVEF, myocardial edema evaluated by average T2-relaxation time, myocardial fibrosis assessed by ECV and myocardial strain. The CMR variables were assessed at 4 weeks (during chemotherapy administration), 8 weeks (2 weeks after conclusion of chemotherapy) and 12 weeks (6 weeks after conclusion of chemotherapy). Images were obtained 48-72 hours after dexrazoxane and doxorubicin administration on week 4. The primary outcome of the study was the presence of ventricular dysfunction, which was defined as decrease in LVEF or RVEF below 50%. The CMR parameters were analyzed by individuals blinded to the treatment groups.
Cardiac Magnetic Resonance Imaging
Serial CMR examinations were performed on a Bruker 9.4-T CMR imaging machine as previously described (15, 16) For the CMR examinations, the mice were anesthetized by isoflurane (induction by 4-5%; maintenance by 1-2% in oxygen from a precision vaporizer). Once anesthetized, the mice were placed in the cradle with electrocardiographic electrodes attached via tape to a front and back paw of the mice. To quantify the RV and LV volumes, the EF, and the LV mass, a fast gradient-echo, low-angle shot sequence was performed in the ventricular short-axis with the following parameters: flip angle 20o; repetition time, 8.85 ms; echo time, 2.36 ms; matrix, spatial resolution, 0.13 x 0.15 mm. The conventional CMR parameters of RV and LV end-diastolic volume, the LV mass, the RV and LV EF were analyzed on the fast gradient-echo sequence using commercially available software, CVI 42 (Calgary, CA). To assess the myocardial edema, T2-relaxation time was quantified. A respiratory and electrocardiogram-gated, multiecho, ventricular short-axis, spin-echo sequence was performed with 5 spin-echo times (7.8, 15.6, 23.4, 31.2, 39 and 46.8 ms). The sequence was performed prior to contrast administration with the following parameters: single slice at mid-ventricular level, slice thickness of 1 mm, matrix of 128 x 128, spatial resolution of 0.23 x 0.23 mm, and 2 averages. Triggering was done at usually every second to third cardiac cycle with a repetition time of 200 to 400 ms. To quantify T2-average, the endocardial and epicardial border on the mid-ventricular slice, as it had the best image quality, was traced for all echo times and analyzed using commercially available software, CVI 42 (Calgary, CA). A region of interest was traced in the LV free wall and interventricular septum and averaged together. Myocardial fibrosis was assessed by measuring the extracellular volume fraction (ECV). The ECV was calculated by pre-contrast and post-contrast T1 measurements. For measurement of the T1 time, a gadolinium-based contrast agent (0.2 mmol/kg) was diluted in saline at a 1:10 ratio and given via intraperitoneal injection. Similarly to T2 quantification, the T1 was measured at the LV mid-ventricular level in a short-axis slice, both pre-contrast and at several intervals post-contrast using a Look-Locker technique. The Look-Locker sequence was electrocardiogram-gated with a non-slice selective inversion pulse with the following parameters: gradient-echo readout, flip angle 10o, repetition time 2.2 ms, echo time 1.6 ms, in-plane resolution of 0.13 x 0.15 mm, slice thickness of 1 mm, repetition time per segment of 22 ms, and number of averages 6 (pre-contrast) or 4 (post-contrast). The post-contrast sequence was performed every 6-8 minutes following intraperitoneal contrast injection. A region of interest was drawn in the LV free wall and interventricular septum at each time interval, plotted over time and then averaged together for myocardial T1 time. A similar process was performed for a region of interest in the blood pool for ECV calculation. The ECV was calculated using known hematocrit for the mice strain. The global circumferential and longitudinal strain (GCS and GLS, respectively) were acquired at a mid-ventricular short-axis slice for GCS and a single 4-chamber slice for GLS by using a Complementary Spatial Modulation of Magnetization (CSPAMM) technique. The following parameters were used for each sequence: flip angle 20o, repetition time7 ms, echo time 3 ms, in-plane resolution of 0.14 x 0.14 mm, slice thickness of 1 mm. The gridlines were analyzed for global peak strain values using commercially available software, Myocardial Solutions (Morrisville, NC).
The R statistical environment was used to perform a one-way ANOVA followed by a post-hoc analysis using the Tukey HSD test on the CMR parameters in relationship to the treatments. Pearson and Spearman correlations between the CMR parameters at the measured time points were calculated using the R statistical environment. Correlations were considered statistically significant for p-values less than 0.05.
Ventricular Dysfunction Predictive Modeling
A logistic generalized linear model was developed in the R statistical environment to predict ventricular dysfunction. Ventricular dysfunction in our model occurs when the LVEF or RVEF was less than or equal to 50%. The parameters measured at 4 weeks were used to predict dysfunction at either 8 or 12 weeks and the parameters measured at 8 weeks were used to predict dysfunction at only 12 weeks. For a one-parameter model a Chi-squared test was used to determine if the model performed better than a constant intercept only model. When we added a second parameter to the model a Chi-squared test was again used, but we compared the two-parameter model to the following one-parameter models: LVEF, RVEF, GLS or T2 time. Receiver operator curves were generated using the ROCR package in the R statistical environment (17).