Principal Strain Analysis for Early Detection of Radiation-Induced Cardiotoxicity in a Mouse Model

Abstract

Purpose:

Radiation-induced cardiotoxicity (RIC) is common in patients receiving thoracic radiation and a major risk factor for morbidity and mortality. The development of novel approaches for early detection and mitigation of RIC remains an acute unmet need. The objective of this study is to develop a mouse model of RIC that recapitulates the progression of cardiac dysfunction seen in patients receiving thoracic radiation and to develop novel cardiac strain markers that exhibit higher sensitivity in detecting subclinical RIC over existing approaches.

Methods:

We developed a mouse model of RIC through image-guided whole heart irradiation of male C57BL/6J mice using two radiation regimens (8Gy × 5 and 24Gy × 1). We developed a pipeline for analyzing anatomical and principal strains derived from cardiac magnetic resonance (CMR) imaging obtained at baseline and at 3-months and 6-months following radiation.

Results:

Both radiation regimens used for whole heart irradiation caused a progressive decline in both anatomical and principal cardiac strains over time. The minimum principal cardiac strain detected subclinical decline in cardiac contractility at an earlier time point than the traditional anatomical cardiac strains. We also observed asymmetric changes in contractility at the epicardium and endocardium relative to averaged cardiac strain across the full thickness of the left ventricle following cardiac irradiation, further reinforcing the limitations of existing methods that do not capture the heterogeneity in cardiac strain changes along the transmural axis.

Conclusion:

We have developed a mouse model of RIC that recapitulates time-dependent deterioration in myocardial contractility noted in patients receiving thoracic radiation. We also developed CMR imaging-derived novel principal strain cardiac markers that detect subclinical deterioration in cardiac contractile function earlier than traditional anatomic cardiac strain markers. If successfully translated into patients, our novel approach of measuring CMR imaging-derived cardiac principal strain analysis may enhance detection of subclinical RIC in patients receiving thoracic radiation.

1. Introduction

Cancer continues to be a leading cause of global mortality, with an estimated 20 million new annual cases worldwide [1]. Around 50% of these patients receive radiation therapy (RT) as a component of their treatment, including patients with some of the most common malignancies such as lung cancer and breast cancer [2]. Although thoracic RT improves survival in many patients, radiation-induced cardiotoxicity (RIC) is a major risk factor for morbidity and mortality [3, 4]. Standard evaluation of RIC involves the assessment of global cardiac function through changes in left ventricular ejection fraction (LVEF). Despite its utility in evaluating advanced stages of RIC, LVEF is inadequate for characterizing early subclinical ventricular dysfunction [5–7].

The emergence of advanced imaging technologies has enabled “high-fidelity” characterization of left ventricular (LV) contractile function. Cardiac magnetic resonance (CMR) imaging offers excellent soft tissue contrast and temporal resolution [8, 9], thereby enhancing the fidelity of evaluating the contractility of the cardiac muscle. Cardiac strains represented in the form of American Heart Association (AHA) segmentation maps have shown sensitivity for detection of pathologies such as hypertrophic cardiomyopathy [10], myocardial infarction [11], and cardiac fibrosis [12]. Recent preclinical studies using murine models have demonstrated superior sensitivity of anatomical cardiac strains, i.e., strains reported along the anatomical axes, namely, the radial, circumferential, and longitudinal (RCL) axes over organ level measures of cardiac function (such as LVEF) to detect sub-clinical cardiac dysfunction [13, 14]. Despite their utility, anatomical cardiac strains are indirect measures of myofiber contractility, and are confounded by the orientation of muscle fibers in the LV. Indeed, myofiber arrangement is subject-specific, with fiber orientation not always aligning along with the RCL axes. For instance, a patient with impaired myofiber contractility in the presence of longitudinally aligned myofibers may present with a similar global longitudinal strain (GLS) to a patient with normal myofiber contractility and circumferentially aligned fibers. Therefore, anatomical cardiac strain measurements performed without considering fiber orientation may not accurately capture cardiac function in all cases [15, 16].

Since myofibers serve as the mechanical driver of cardiac contraction, “fiber strain,” i.e., the metric of fiber shortening [17], serves as the most sensitive marker of contractility impairment in the myocardium. However, quantifying fiber strain in vivo requires establishing the three-dimensional fiber architecture, which remains technically challenging. Despite recent advances in diffusion tensor imaging (DTI), which hold the promise of providing fiber strain [18], their clinical translation and applications to small animals remain limited. In the absence of reliable measurement of myofiber strains, strains along the principal axes, i.e., the strains corresponding to the directions of maximum deformation of the LV, are expected to offer greater correlation with myofiber shortening relative to anatomical strains. In their analysis of three-dimensional (3D) cardiac strains, Pedrizetti et al, [16] observed striking correlations between the principal axes and the fiber orientation. Despite the perceived advantages, the utility of principal cardiac strains in detecting subclinical cardiac dysfunction remains largely unexplored. Moreover, the common practice of reporting cardiac strain markers as single global representative measures of organ-level cardiac contractility limits the detection of transmural (i.e. epicardial to endocardial) variation in cardiac injury and adaptations, thereby limiting the sensitivity of detecting asymmetric regional injuries along the myocardial thickness. RIC-related cardiac remodeling has been observed to impact specific transmural regions at varying depths of the LV [18]. These regional injuries are expected to affect the spatial heterogeneity of cardiac strain patterns, thus necessitating a comprehensive characterization of transmural cardiac strains to complement global cardiac strain markers for effective risk-stratification.

In summary, current commonly used methods of cardiac strain analyses are limited to (a) anatomical strain analyses which are confounded by the lack of information regarding fiber orientation, thereby undermining their ability to accurately capture myocardial impairment in all cases, and (b) single global value representation of the cardiac strains, which yield a gross measure of myocardial contractility, without considering the changes along the transmural axis. To overcome these limitations, we propose a pipeline for analysis of principal cardiac strain markers that enables characterization of both global and transmural cardiac strain changes in response to radiation. We hypothesize that our novel approach will enhance detection of radiation-induced subclinical ventricular dysfunction over existing approaches. Employing a mouse model of RIC, we present for the first time a longitudinal study to investigate changes in global anatomical and principal cardiac strains using our comprehensive cardiac strain analysis pipeline and further characterize heterogeneity in strain changes along the transmural axis.

2. Methods

2.1. Murine RIC models

Whole heart irradiation was delivered under image guidance to 8-week old male C57BL/6J mice on a small animal irradiator (X-RAD 225Cx- Precision X-ray Inc) using the following parameters: tube potential of 225 kV with a 13-mA current, 5 mm focal spot, and 0.3-mm copper filter at 2.69Gy/min. Radiation was delivered as a single beam along the anterior-posterior axis under 1–5% isoflurane anesthesia. Two radiation dosage regimens were used: 8Gy delivered on five consecutive days (8Gy × 5) (BED 146.67Gy, α/β=3 [19]) and 24Gy delivered in a single fraction (24Gy × 1) (BED 216.00Gy, α/β=3 [19]) (Fig. 1A). All animal procedures were approved by the Institutional Animal Care and Use Committee. Mice were purchased from Jackson Laboratories and housed in institutional animal facilities with 12hr day/night light cycles in group housing with ad libitum food and water.

Figure 1.

(A) Schematic representation of the longitudinal study to investigate radiation-induced cardiotoxicity (RIC) in a murine model using cardiac magnetic resonance (CMR) imaging-derived comprehensive strain analysis. Whole heart irradiation regimens of 8Gy × 5 or 24Gy ×1 were delivered under image guidance to induce RIC in 8-week old male C57BL/6J mice (n = 5 mice per treatment arm). CMR imaging was performed in these mice at baseline (0M), and at 3-months (3M) and six-months (6M) post-RT. Mock irradiated mice were used as controls. (B) Multiscale description of the comprehensive strain analysis, including the calculation of anatomical and principal cardiac strains in the left ventricle (LV). An arbitrary section of the LV tissue is highlighted to represent the orientation of the anatomical (R,C,L) and principal axes. R: Radial, C: Circumferential, L: Longitudinal, Epi: epicardium, Endo: endocardium, Min. and Max: Minimum and maximum principal axes.

2.2. Longitudinal CMR image acquisition

CMR imaging and subsequent strain analyses were performed for both RT groups (n = 5 each) at three time points: baseline or pre-RT (0M), three-months (3M) and six-months (6M) post-RT. Similar studies were carried out in a control group (n = 5) of sham irradiated mice at two time points (0M and 6M post-RT. Short-axis (SA) CMR images were acquired for all mice using a 7T Bruker Avance III HD scanner with a 72 mm quadrature proton volume coil. All mice were anesthetized prior to the procedure using 1–5% isoflurane. Fast low angle shot images were acquired at six slice positions along the SA using the following parameters: TE =1.6 ms, TR = 9 ms, flip angle = 20°, readout bandwidth = 65789 Hz, resolution = 200 μm, slice thickness = 1 mm, field of view = 32 × 32 mm, averages = 4. The SA stack of images (Supplementary Fig. S1) was used to calculate the LVEF and mass (LV mass) by isolating and segmenting the endocardium at end diastole (ED) and end systole (ES) using Segment [20].

2.3. Cardiac strain analysis

A non-rigid image registration (NRIR) framework employing the diffeomorphic demons algorithm was used to derive cardiac deformation from the stack of SA slices [21, 22]. The deformation was initially calculated in terms of cartesian displacements at the time-frame corresponding to ES with respect to ED.

2.3.1. Principal and anatomic strain analysis

The Cartesian displacement vectors calculated through NRIR were used to derive cardiac strains in the anatomical and principal axes. The anatomical axes in each representative cubic element of the myocardium were defined along the radial and, circumferential directions by identifying spatially varying tangential and normal vectors in the LV geometry (Fig. 1B). The strain tensor in anatomical axes (including both normal and shear components) was calculated as the orthonormal transformation of the Cauchy-Green strains (Cartesian coordinates) to the anatomical coordinates, as shown below

$${\mathit{E}}_{RCL}=\left[\begin{array}{lll}{E}_{RR}\hfill & E_{RC}\hfill & E_{RL}\hfill \\ E_{CR}\hfill & E_{CC}\hfill & E_{CL}\hfill \\ E_{LR}\hfill & E_{LC}\hfill & E_{LL}\hfill \end{array}\right]$$

where E is the Cauchy-Green strain tensor, and E_RCL is the tensor of Cauchy-Green strains in the anatomical axes, wherein the normal (diagonal) strain components, denoted by E_RR, E_CC, E_LL, correspond to the shortening of the representative cube (Fig. 1B) in radial and circumferential directions, respectively. In contrast to the anatomical strains that do not consider shearing components (namely, E_RC and E_RL) and offer confounded measures in describing fiber strain, principal strains, by definition, incorporate both normal and shear strain terms. Consequently, the principal axes (Fig. 1B), representing the directions of maximum and minimum deformation experienced by the cubic element during the cardiac cycle, are considered to be closely correlated with the myofiber strains. Namely, principal strains were derived from the eigenvalue decomposition of E as:

$$\left|\mathit{E}-\lambda \mathit{I}\right|=0\Rightarrow \left\{\begin{array}{l}\text{max}(\lambda )=E_1:\text{maximum principal strain}\hfill \\ \text{min}(\lambda )=E_3:\text{minimum principal strain}\hfill \end{array}\right.$$

More specifically, E₁ calculated at a specific spatial location (pixel) within the ventricular wall, represents the principal direction of the maximum expansion of the LV at that point. In this case, E₁ would correspond to the maximum thickening of the LV wall. Conversely, E₃ signifies the principal direction with the maximum contraction experienced by the ventricular wall. In this study, since we acquired CMR images in the SA plane, we focused our strain analysis in the R-C plane (Fig. 2A). Thus, the involved anatomical strains included the normal strains E_RR, E_CC, and the shear strain E_RC. Accordingly, the maximum and minimum principal strains were calculated in the R-C plane and evaluated against E_CC and E_RR to maintain consistency in the comparisons.

Figure 2.

(A) American Heart Association (AHA) bullseye segmentation plots of the left ventricle comprising of the base, mid, and apical slices. (B) Representative heat maps of end-systolic strain calculations in AHA segmentation plots along the circumferential (E_CC) and radial (E_RR) axes, and minimum (E₃) and maximum (E₁) principal axes in mice at 0M (baseline) and at three-months (3M) and six-months (6M) post-RT after receiving 8Gy × 5 whole heart irradiation.

2.3.2. Transmural cardiac strain calculations

Cardiac strains were calculated at every pixel in the CMR stack between individual time frames, and values pertaining to the LV were isolated through segmentation [20]. Cardiac strains were calculated at ES with respect to ED, through temporal discretization of cardiac cycle and augmentation of the strain through time [21]. Strain transmurality was determined by identifying epicardial and endocardial regions from the segmented images. To this end, a Laplace boundary-value problem on the isolated “full thickness” LV wall (with Dirichlet boundary conditions applied on the endo- and epicardial regions) was used to divide the LV wall into three layers ranging from endocardium to epicardium (Fig. 1B) and averaging respective strains over epicardial and endocardial regions [23]. In contrast, a single global metric averaging the regional distribution of normal or principal strains was calculated as the “full thickness average,” analogous to the standard calculations of global radial and circumferential strains (GRS and GCS).

2.4. Statistical analysis

Cardiac strains for all mice were analyzed using GraphPad Prism 9, and results were presented as mean ± standard deviation (SD). One-way ANOVA analysis with repeated measures and Tukey’s correction was performed to compare the means of the 8Gy × 5 and 24Gy × 1 groups at 3M and 6M post-RT time points with their corresponding baselines (0M). Paired Student’s t-test was used to determine the statistical differences between the global strain measures in representing cardiac motion at the endocardium and epicardium at all time points for the 24Gy × 1 group. Student’s t-test was also performed to compare both radiation groups with the control mice at the 6M post-RT time point. All results are presented for the entire LV and p < 0.05 indicated significance for all analyses.

3. Results

3.1. Detection of subclinical RIC via principal cardiac strain analysis

Traditional cardiac imaging studies evaluate strain along pre-defined anatomic axes (i.e., radial, circumferential, and longitudinal or ‘RCL’ axes). Since, the orientation of LV fibers do not always align along RCL axes, the principal strain - which reflect the axis along which maximum contraction of the LV is observed - may offer an alternative for highly sensitive assessment of subclinical cardiac dysfunction without a prior knowledge of fiber orientation. To investigate the utility of principal strain analysis in improving the sensitivity of detecting RIC, we established a mouse model of RIC by delivering whole heart irradiation (8Gy × 5 or 24Gy × 1) or sham irradiation to male C57BL/6J mice using a small animal irradiator. Since most previously published murine cardiac irradiation models employed a single fraction regimen, we employed this fractionation scheme to enable comparison with existing models [13, 14, 24, 25]. However, this regimen lacks clinical significance as a single fraction is almost never employed for treatment of patients with lung cancer. As a more clinically relevant alternative, we employed a 5-fraction regimen, which is more common for treatment of early stage lung cancer patients. CMR imaging was obtained at baseline before cardiac irradiation, and at 3 months (3M) and 6 months (6M) post-RT (Supplementary Fig S1). We applied our custom pipeline to evaluate changes in the anatomic and principal cardiac strains (Figs. 1 and 2). Strain analyses revealed significant time-dependent deterioration in LV contraction in both radiation groups along both anatomic axes and the principal axes (Figs. 2 and 3 and Supplementary Fig S2 and S3). Both the global anatomic strains and principal strains were significantly worse at 6 months relative to pre-RT baseline (Fig. 3). At 3 months post-RT, anatomic strains showed numerical decline relative to pre-RT baseline (with the exception of global radial strain in the 24Gy × 1 group) although this decline did not reach statistical significance (Figs. 3A and B). In contrast, the decline in minimum principal strain for both the 8Gy × 5 and 24Gy × 1 arms was significant at both 3M and 6M post-RT [p: 0.022 for min. principal strain at 0M vs 6M; p <0.001 for min. principal strain at 0M vs 6M for 8 Gy × 5; p: 0.051 for min. principal strain at 0M vs 3M; p: 0.002 for min. principal strain at 0M vs 6M for 24 Gy × 1] (Fig. 3C). Similarly, mice that received both 8 Gy × 5 and 24 Gy × 1 showed statistically significant decline in minimum principal strain at 6 months when compared to control mice at the same time point [p: 0.010 for control vs 8 Gy × 5 at 6M; p<0.001 for control vs 24 Gy × 1 at 6M] (Supplementary Fig. S4). No significant change in LVEF was noted at 6M post-RT in any of the irradiated groups or the control group (Supplementary Fig. S5A). Although there was a trend towards increase in LV mass at 6 months relative to pre-RT baseline in the 24 Gy 1 arm, it did not reach statistical significance [p: 0.082] (Supplementary Fig. S5B). Taken together, these findings suggest that the minimum principal strain enables detection of subclinical RIC at an earlier time point (i.e. 3M post-RT) than the traditional anatomical strains, although given the exploratory nature, our findings need to be validated in future studies using diverse preclinical models.

Figure 3.

Global estimation of cardiac strains along the circumferential (A), and radial axes (B), as well as minimum principal (C), and maximum principal axes (D) at baseline (0M), three- (3M) and six-months (6M) post-RT for mice in the control and 8Gy × 5 and 24Gy × 1 regimens of whole heart radiation. Results are presented as the full-thickness mean ± SD, i.e., the mean and SD of all the strain values calculated in every region of the left ventricle (LV). GCS: global circumferential strain, GRS: global radial strain

3.2. Characterization of radiation-induced transmural heterogeneity in strain changes

Cardiac strains are commonly reported as averaged global value across the whole thickness of the ventricular wall. Such an approach limits evaluation of potential differences in the injury and/or adaptation patterns along the transmural axis. To overcome this limitation, we estimated strains at both the epicardium and endocardium. For this analysis, the ventricular wall was divided into three layers: Endo, Mid, and Epi, making up of 25%, 50%, and 25% of the wall thicknesses, respectively. We observed time-dependent deterioration in contraction at both the epicardium and the endocardium as measured by both the circumferential strain and the minimum principal strain (Figs. 4A–D). Next, we compared changes in averaged global strain across the full thickness of the LV against changes in strain at the epicardium and endocardium. The averaged global strain was calculated using the formula (Epi+Mid+Endo)/3. Our analysis demonstrated significant differences between averaged global straincompared to strain at the epicardium and endocardium, especially at 6 months (Fig. 5A–D), further reinforcing the heterogeneity in changes in strain along the transmural axis (p: 0.007 GCS vs Epi; p: 0.001 GCS vs Endo; p: 0.032 Global min vs Epi; p: 0.003 Global min vs Endo; all at 6 months post-RT). Our pipeline, which effectively captures such heterogeneity, thus, provides superior resolution in characterizing RIC along the transmural axis.

Figure 4.

Time-dependent changes in transmural cardiac strain for mice that received 24Gy × 1 in whole heart irradiation along the circumferential (A, B) and minimum principal axes (C, D). Cardiac strain changes are depicted at both endocardium (A, C) and epicardium (B, D) at baseline, three- and six-months post-RT. Endo and Epi CS: circumferential strains at the endocardium and epicardium, GCS: global circumferential strain, Endo and Epi E₃: minimum principal strains at the endocardium and epicardium.

Figure 5.

One-way ANOVA analysis highlighting the statistical differences between the global and regional (endo- and epicardial) cardiac strains at all three time points along the circumferential (A, C) and minimum principal axes (C, D). Results are presented at baseline, three- and six-months post-RT for mice that received 24Gy × 1 in whole heart irradiation. Endo and Epi CS: circumferential strains at the endocardium and epicardium, Endo and Epi E₃: minimum principal strains at the endocardium and epicardium

4. Discussion

Despite significant technological improvements in the delivery of RT, RIC remains a major contributor to morbidity and mortality in patients receiving thoracic RT. The molecular mechanisms underlying RIC remain poorly understood, and there are currently no clinically-validated approach for the early detection and/or mitigation of RIC in patients receiving thoracic radiation. Since tissue analysis of the heart following thoracic RT is not practically feasible, pre-clinical models of RIC that faithfully recapitulate the progression of cardiac dysfunction following exposure to radiation are urgently needed to elucidate mechanisms of RIC and to enhance early detection strategies.

Recent preclinical studies in murine models [13] [14] have demonstrated the utility of cardiac MRI in defining deterioration in cardiac strain at an early time points (8–10 weeks after cardiac irradiation) and before changes in LVEF manifest (typically around 12 months in murine models) [14]. These studies represent important advances for studying mechanisms of RIC. However, the limitations of the existing models include reliance on anatomic cardiac strains for detection of subclinical RIC and lack of resolution along the transmural axis to capture heterogeneity on cardiac strain changes. To overcome these limitations, we established a mouse model of RIC in a wild-type C57BL/6J mouse background and developed a pipeline for analyzing both anatomic strains and principal cardiac strains with high resolution that enables detection of heterogeneity in strain changes along the transmural axis. While exploratory in nature, our study suggests that minimum principal strain may enable detection of RIC at an earlier time point than anatomical strains.

A possible explanation for potential improved performance of the principal strain markers may be related to the torsional mechanics of the LV contraction and consequent transverse shear [11, 26]. LV torsion is widely attributed to the helical arrangement of the myofibers, which results in a counter-rotational twist between the base and apex of the LV during ejection. Since principal strain analyses accommodate normal and shear strains in their calculations, they have shown greater association with the resulting myofiber shortening in health and disease [15, 16]. The analyses in this study were restricted to in-plane strain measurements limited to a stack of SA images. Thus, the principal strains and axes are indicative of the in-plane myofiber orientation. As such, the inclusion of the longitudinal axis will provide additional depth to the investigation of principal cardiac strains. The pipeline presented herein does not have any limitations for such analysis and can readily applied to images acquired in both LA and SA and the subsequent changes in the magnitudes and alignment of the principal cardiac strains will be investigated in future studies. Nevertheless, principal cardiac strain changes promise to provide additional insight into radiation-driven cardiac remodeling. Indeed, the association of principal axes with the LV microstructure may enable enhanced detection of subclinical RIC in cancer patients than existing anatomical strains and will be investigated in future studies.

We have further elaborated our pipeline for measuring both global (averaged) cardiac strains across the full thickness of the ventricular wall as well as cardiac strains at the epicardium and endocardium. We have previously compared the methodology employed in this study with a commonly used commercial software package, cvi42, and demonstrated good concordance in global anatomical strain measurements [22]. The regional strain analysis at the epicardium and endocardium affords greater resolution in studying transmural heterogeneity in cardiac pathology. This further emphasizes the notion that assessing cardiac strains at endocardium and epicardium in conjunction with global anatomical and principal strain markers offers comprehensive insights into anatomical and functional complexity of the heart and provides insights into asymmetric effects of radiation on the heart based on anatomical differences.

Limitations of our study include small sample size, which limits ability to distinguish smaller differences in cardiac strain (especially at earlier time points), lack of tissue correlates (including correlation with fiber architecture and tissue contractile function) and lack of correlation between strain changes and cardiac substructure dosimetry (due to whole heart irradiation). Nevertheless, our study represents an important advance in identifying novel cardiac strain markers that enhance detection of sub-clinical RIC (especially at earlier time points) as well for studying transmural changes in cardiac strain induced by radiation at a higher resolution. Future studies using partial heart irradiation protocols in combination with ex vivo tissue analyses (both histological and biomechanical) will address some of the limitations of our study. Similarly, replicating our studies in multiple mouse strains using both male and female sexes, and employing both hypofractionated and conventionally fractionated radiation regimens will further strengthen the generalizability of our platform and findings. Efforts are also underway to determine if principal cardiac strain analysis will enable early identification of subclinical RIC in patients receiving thoracic RT, and have the potential to improve patient care.