Incremental value of multiparametric cardiac magnetic resonance imaging for non-invasive identification of significant acute cardiac allograft rejection: a prospective and biopsy-proven study

Abstract

Aims

This study aimed to evaluate the association between cardiac magnetic resonance imaging (CMR) multiparameters and significant acute cardiac allograft rejection (SR), and assess the incremental value of CMR multiparameters over conventional serum examinations for identifying SR.

Methods and results

Heart transplantation (HTx) recipients with endomyocardial biopsy and healthy controls were prospectively recruited for CMR assessment. CMR feature tracking was performed to evaluate the left ventricular (LV) global strain in all three directions. The last serum examinations including N-terminal pro-brain natriuretic peptide (NT-proBNP) before anti-rejection therapy were recorded. Participants were divided into three groups: control, SR [acute cellular rejection grade ≥ 2R and/or antibody-mediated rejection (AMR) grade ≥ pAMR1], and NSR (non-SR). Finally, 30 controls (43.3 ± 13.6 years, 26 males) and 51 HTx recipients comprising 23 SRs (48.6 ± 12.6 years, 24 males) and 28 NSRs (42.7 ± 14.9 years, 16 males) were enrolled for analysis. Compared with NSRs, SRs showed elevated NT-proBNP (7797.0 ± 7527.6 pg/mL vs. 3334.6 ± 5935.3 pg/mL, P < 0.001), worse LV global longitudinal strain (GLS) (−9.7 ± 3.1% vs. −13.1 ± 2.9%, P < 0.001), and increased native T₁ (1384 ± 80.1 ms vs. 1321 ± 69.9 ms, P < 0.001) and T₂ values (50.9 ± 2.7 ms vs. 45.7 ± 4.3 ms, P < 0.001). In multivariable analysis, LVGLS (OR = 0.76, 95% CI, 0.59–0.98, P = 0.03) and T₂ value (OR = 1.35, 95% CI, 1.10–1.65, P = 0.01) were independently associated with SR after NT-proBNP adjustment. Furthermore, the likelihood ratio test showed LVGLS (P = 0.002) and T₂ value (P < 0.001) had incremental value over NT-proBNP for identifying SR.

Conclusion

LVGLS and T₂ value were independently associated with SR, providing incremental value for non-invasive identification of significant rejection in HTx recipients.

Graphical Abstract

Open in new tabDownload slide

Introduction

Heart transplantation (HTx) has proven to be a successful treatment for end-stage heart failure, demonstrating outstanding short- and long-term survival rates.¹ Although with the advances in immunosuppression, acute cardiac allograft rejection (ACAR) is still the major complication in the first year and the primary concern for 5 years after HTx.^2–4 Currently, endomyocardial biopsy (EMB) remains the gold standard for the identification of ACAR.⁴ However, EMB is invasive with an overall complication rate of 4.1% and has a limitation of sampling error because of localized sampling.^5,6 Due to the patchy distribution of rejection, a non-invasive tool with full coverage assessment of the left ventricle (LV) may help overcome these limitations.⁷

Cardiac magnetic resonance imaging (CMR) serves as the gold standard for evaluating cardiac structure and function in vivo and allows assessing myocardial tissue characteristics non-invasively of the entire LV. For functional evaluation, besides obtaining the conventional functional parameter LV ejection fraction (LVEF), the emerging CMR feature tracking (CMR-FT) technology provides the ability to evaluate LV myocardial deformation during the whole cardiac cycle quantitatively and has been validated for the early diagnosis of ischaemic and non-ischaemic cardiomyopathy.⁸ However, the correlation between CMR-FT-derived strain parameters and ACAR has not been fully recognized. As for myocardial tissue characterization, T₂ value is elevated in tissue with high water content related to myocardial oedema.⁹ Native T₁ value increases in the presence of diffuse fibrosis, myocardial oedema, and inflammation. Additionally, the extracellular volume (ECV) fraction serves as a quantifiable measure of myocardial interstitial volume.⁹ These CMR tissue parameters may be useful in assessing ACAR. Assessment incorporating the different strengths of CMR parameters may complement each other and offer a way to optimize the use of CMR for transplant rejection identification.⁴ Whereas, there are few studies on a combined diagnosis of CMR multiparameter indicators. Serum high-sensitivity cardiac troponin I (hs-cTnI) and N-terminal pro-brain natriuretic peptide (NT-proBNP) are routine surveillances of HTx recipients and have been shown to be related to ACAR.^4,10 The incremental value of multiparametric CMR over conventional serum examinations for the identification of ACAR has not been characterized so far.

Considering only significant ACAR (SR), which is defined as acute cellular rejection (ACR) grade ≥ 2R and/or antibody-mediated rejection (AMR) grade ≥ pAMR1, needs further immunosuppression treatment in present clinical practice,⁴ using EMB as the reference standard, the purposes of this study were to (i) evaluate the association between multiparametric CMR indicators and SR and (ii) to explore the incremental value of the CMR parameters over conventional serum examinations for the identification of SR in HTx recipients.

Methods

Study participants

In this prospective cross-section observational study, we consecutively recruited HTx recipients with abnormal clinical symptoms and signs suspected of rejection, regardless of the time since transplantation, for EMB and CMR examinations at a single centre (Fuwai Hospital, Beijing, China) from May 2019 to November 2023. All participants provided written informed consent. The institutional review board of Fuwai Hospital approved this study, and all procedures were performed by the Declaration of Helsinki.

All participant’s EMB reports and electronic clinic records were collected and recorded. Severe coronary artery vasculopathy (CAV) was diagnosed according to the consensus of the International Society for Heart and Lung Transplantation (ISHLT).¹¹ The exclusion criteria were: (i) age < 18 years; (ii) contraindications to CMR, including claustrophobia and glomerular filtration rates ≤ 30 mL/min; (iii) poor CMR image quality; and (iv) presence of concomitant severe CAV, systemic infiltrative disease, or any other disease suspected to affect CMR results, which were excluded to improve the accuracy of the study findings. The last serum examinations before any further anti-rejection treatment were recorded. The interval between serum examinations and EMB, as well as between CMR and EMB, is <3 days. A total of 30 gender- and age-matched healthy controls were also recruited. All the healthy controls had no previous history of cardiovascular diseases, risk factors of cardiovascular, current cardiovascular treatment, or any abnormal imaging or laboratory findings.

CMR protocol

All CMR examinations were conducted on a 3.0 T MRI scanner (Magnetom Skyra, Siemens Healthcare, Erlangen, Germany) equipped with a phased-array cardiovascular coil, using electrocardiographic respiratory gating. Cine images were acquired using a standard balanced steady-state free precession (bSSFP) sequence with retrospective electrocardiogram gating in short-axis views from base to the apex of LV and long-axis two-, three-, and four-chamber views. To cover the entire LV, nine short-axis slices were performed. All cine images were acquired with 25 phases per cardiac cycle. The scan parameters were as follows: repetition time (TR) 39.60 ms, echo time (TE) 1.45 ms, flip angle (FA) 45°, field of view (FOV) 330 × 380 mm, slice thickness 8 mm, and SENSE factor 2. Late gadolinium enhancement (LGE) images were acquired 10–15 min after intravenous contrast administration of gadolinium-DTPA contrast (0.2 mmol/kg Magnevist/Gadavist, Bayer, Leverkusen, Germany) utilizing a breath-held phase-sensitive segmented inversion recovery sequence in the same views as the cine images.

Native T₁ and T₂ maps were acquired during breath-holding at three identical short-axis locations: the base, middle, and apex of the LV. A bSSFP-based modified Look-Locker inversion recovery [MOLLI 5(3)3] images were obtained before and 10–15 min after contrast administration to acquire native and post-contrast T₁ maps, respectively. T₁ imaging scan parameters included TR 277.32 ms, TE 1.08 ms, FA 35°, FOV 360 × 380 mm, slice thickness 8 mm, and a SENSE factor of 2. To calculate T₂ maps, a true fast imaging with steady-state free precession sequence was used at matched slices. Related imaging parameters were as follows: TR 194.74 ms, TE 1.29 ms, FA 12°, FOV 360 × 380 mm, slice thickness 8 mm, and a SENSE factor of 2.

CMR post-processing

All CMR parameters were measured using the commercially available software (CVI42 version 5.3.6, Circle, Calgary, Alberta, Canada). The endocardial and epicardial contours of the LV myocardium were manually delineated (excluding the papillary muscles) in both end-diastolic and end-systolic phases on short-axis cine images to generate conventional CMR parameters including LV end-diastolic volume (LV EDV), LV EDV index (LV EDVi), end-systolic volume (LV ESV), ejection fraction (LVEF), cardiac output (CO), and so on. Additionally, endocardial and epicardial contours were traced throughout the whole cardiac circle on the two-, three-, four-chamber, and short-axis cine images to obtain the CMR-FT parameters. The LV global longitudinal strain (LV GLS) was obtained by tracking the long horizontal-axis cine images whereas the LV global circumferential (LV GCS) and the LV global radial strains (LV GRS) were derived from the short-axis cine images.

LGE was assessed as presence or absence by visualization. Native T₁ and T₂ values were generated from the corresponding maps respectively after delineating the epicardial and endocardial contours on basal, middle, and apical slices. ECV fraction map was generated using the patient haematocrit level obtained on the day of CMR as the formula: ECV = (ΔR1 myocardium/ΔR1 blood) × (1 − haematocrit), where R1 = 1/T1 and ΔR1 is the change in relaxation time rate between native and post-contrast T₁ maps. The global values of native T₁, T₂, and ECV were calculated from all three slices.

All post-processing was performed by a single reviewer (P.Z., 3 years of review experience). Interobserver agreement was assessed in a randomly selected 15 HTx recipients by another independent reviewer (Z.D., 4 years of review experience) who was blinded to the analyses of the previous reviewer. Intraobserver agreement was also assessed by P.Z. in these 15 HTx recipients. All above CMR readers were blinded to the EMB results and rejection therapy.

EMB and the definition of SR

EMB conducted via a standardized right internal jugular approach was stained with haematoxylin and eosin, and the results were reported by a pathologist blinded to CMR findings. The EMB results, according to the ISHLT cardiac allograft biopsy grading system, encompassed ACR grades (0R, 1R, 2R, and 3R) and AMR grades (pAMR0, pAMR1, pAMR2, and pAMR3).⁴ The definition of SR was ACR grade ≥ 2R and/or AMR grade ≥ pAMR1. Non-significant ACAR (NSR) was defined as ACR grade < 2R with AMR grade pAMR0.

Statistical analysis

Continuous variables or categorical variables are reported as means ± SDs or numbers with percentages, respectively. Shapiro–Wilk test was used to assure normality of continuous variables. χ² test or Fisher’s exact test was performed for the comparison of categorical variables as appropriate. Student’s t-test or Mann–Whitney U test was performed to compare two continuous variables as appropriate. One-way analysis of variance with post hoc Bonferroni tests or Kruskal–Wallis test was applied for the comparison of three continuous variables as appropriate. Uni- and multivariable logistic analyses were performed to identify the independent predictors of SR. Parameters from serum examinations and CMR with a P < 0.05 in univariable logistic analysis were included into the multivariable logistic models. Variables selected from the univariable analysis also calculated the variance inflation factor results to avoid potential overfitting. Likelihood ratio test was used to evaluate the incremental value of CMR parameters over the cardiac serum examinations. Receiver operating characteristic (ROC) analysis was applied to evaluate the performance of the models. The corresponding area under the ROC curve (AUC) and the 95% confidence interval were reported as well. Youden’s index was performed to determine the related sensitivity and specificity of the models. Intraobserver and interobserver reliabilities were described by intraclass correlation coefficients (ICCs) or kappa coefficient as appropriate. Statistical significance was assumed at P < 0.05. Statistical analyses were performed using the R Studio (version 4.1.2) and SPSS (version 25).

Results

Participant’s characteristics

A total of 57 HTx recipients were recruited. A total of six HTx recipients were excluded, including five who had concomitant severe CAV and one due to poor CMR image quality. The flow chart of this study is presented in Figure 1. Finally, 51 HTx recipients (46.0 ± 13.8 years, 78.4% male) and 30 healthy controls (43.3 ± 13.6 years, 86.7% male) were included for analysis. The average time from HTx to study was 3.9 ± 3.3 years in all recipients. The interval between CMRs and EMBs was 2.2 ± 1.6 days. There were 28 (54.9%) NSRs (12 were 0R and 16 were 1R of ACR grade; all were pAMR0 of AMR grade) and 23 (45.1%) SRs (10 were 2R, 3 were 3R of ACR, and 10 were ≥pAMR1 regardless of ACR grade) according to EMB results. We found no evidence of a difference in HTx indications between NSR and SR groups, including ischaemic heart disease, hypertrophic cardiomyopathy, and dilated cardiomyopathy (all P > 0.05). The participant’s characteristics are shown in Table 1.

Serum examinations and CMR measurements of controls, NSRs, and SRs

Compared with controls, heart rate (68.0 ± 11.5 bpm vs. 87.0 ± 10.0 bpm, P < 0.001) was increased; LVEF (61.9 ± 4.7% vs. 55.4 ± 11.7%, P = 0.047) and LV GLS (−13.1 ± 2.9% vs. −16.2 ± 2.1%, P < 0.001) were impaired; and native T₁ (1321 ± 69.9 ms vs. 1222.0 ± 40.9 ms, P < 0.001), ECV (34.5 ± 5.3% vs. 27.6 ± 2.0%, P < 0.001), and T₂ (45.7 ± 4.3 ms vs. 41.8 ± 2.0 ms, P < 0.001) were elevated in NSRs. LGE was more prevalent in NSRs compared with controls (57.1% vs. 0, P < 0.001). We found no evidence of differences in LV ESV, LV EDV, LV EDVi, CO, LV GRS, and LV GCS between NSRs and controls (Table 2 and Figure 2).

For the comparisons between NSRs and SRs, NT-proBNP (7797.0 ± 7527.6 pg/mL vs. 3334.6 ± 5935.3 pg/mL, P < 0.001) were elevated; LV GLS (−9.7 ± 3.1% vs. −13.1 ± 2.9%, P < 0.001) was worse; and native T₁ (1384 ± 80.1 ms vs. 1321 ± 69.9 ms, P < 0.001) and T₂ (50.9 ± 2.7 ms vs. 45.7 ± 4.3 ms, P < 0.001) were elevated in SRs. However, we found no evidence of differences in creatinine, hs-cTnI, heart rate, LV ESV, LV EDV, LV EDVi, LVEF, CO, LV GRS, LV GCS, ECV, and the presence of LGE between SRs and NSRs (Table 2 and Figure 2).

Uni- and multivariable analyses for the identification of SR

Univariable logistic analysis showed that NT-proBNP (P = 0.04), LV GLS (P < 0.01), native T₁ (P = 0.01), and T₂ (P < 0.01) were significantly associated with SR (Table 3). The AUC values for the identification of SR in HTx recipients were as follows: T₂ (0.85, 95% CI, 0.73–0.96), LV GLS (0.79, 95% CI, 0.67–0.92), NT-proBNP (0.77, 95% CI, 0.64–0.90), and T₁ (0.71, 95% CI, 0.57–0.85) as shown in Figure 3. Due to the limitation of the SR sample size (n = 23) and statistical power, we limited each model to three variables (respecting the 1:10 rule). Therefore, we paired LV GLS, native T₁, and T₂ in three combinations, each coupled with NT-proBNP, constructing three models. In multivariable logistic regression analysis, LV GLS (OR = 0.76, 95% CI, 0.59–0.98, P = 0.03) and T₂ (OR = 1.35, 95% CI, 1.10–1.65, P = 0.01) were independently associated with SR after NT-proBNP adjustment as shown in Table 3.

Incremental identification value of CMR parameters

In the likelihood ratio test (Figure 4A), the sequential addition of LV GLS and T₂ to the NT-proBNP model (−2 log likelihood = 64.5, χ² = 5.7) improved the model identification power progressively for SR (−2 log likelihood = 55.8, χ² = 14.4, P = 0.002 and −2 log likelihood = 50.3, χ² = 20.0, P < 0.001, respectively). When added the combination of LV GLS and T₂ to the NT-proBNP model, the final model got further improvement for the identification of SR (−2 log likelihood = 44.9, χ² = 25.3, P = 0.02). In ROC analysis, the combined model of NT-proBNP, LV GLS, and T₂ could accurately identify SR in HTx recipients with AUC of 0.88 (95% CI, 0.77–0.98), sensitivity of 96%, specificity of 79%, negative predictive value of 97%, and positive predictive value of 77% (Figure 4B). When added LV GLS to a baseline model incorporating native T₁ and T₂ values, the likelihood ratio test demonstrated that LV GLS provided incremental value for the non-invasive identification of SR (−2 log likelihood = 50.3, χ² = 20.0 and −2 log likelihood = 44.9, χ² = 25.3, P = 0.02) (see Supplementary data online, Figure S1). Representative examples of multiparametric CMR in two HTx recipients with EMBs are shown in Figure 5.

Interclass and ICC analysis

In order to avoid the influence of the CVI 42 auto-save function, the re-measurements of strain parameters, native T₁, T₂, and ECV values were performed after all the traces of the previous measurements were completely deleted. The inter- and intraobserver agreements for various parameters were excellent, as shown by ICC: 0.90 (95% CI, 0.72–0.96) and 0.93 (95% CI, 0.83–0.97) for LV GRS; 0.91 (95% CI, 0.75–0.97) and 0.93 (95% CI, 0.83–0.98) for LV GCS; 0.92 (95% CI, 0.77–0.97) and 0.94 (95% CI, 0.82–0.98) for LV GLS; 0.95 (95% CI, 0.85–0.99) and 0.97 (95% CI, 0.94–0.99) for native T₁; 0.92 (95% CI, 0.84–0.97) and 0.95 (95% CI, 0.93–0.97) for T₂; and 0.93 (95% CI, 0.86–0.99) and 0.96 (95% CI, 0.94–0.99) for ECV. Additionally, the inter- and intraobserver agreements for LGE were good, shown as kappa coefficients, which were 0.70 ± 1.9 and 0.84 ± 1.5.

Discussion

This prospective and biopsy-proven study demonstrated three main findings: (i) LV GLS was worse, and native T₁, T₂, and NT-proBNP were elevated in SRs compared with NSRs; (ii) LV GLS and T₂ were independent predictors of SR; and (iii) LV GLS and T₂ both provided incremental value for the identification of SR. The combined model of NT-proBNP, LV GLS, and T₂ had excellent sensitivity and negative predictive value.

CMR tissue characterizations

Myocardial inflammation with oedema is the pathological basis of ACAR.⁴ On CMR, the increased tissue water content (oedema) leads to the elevation of both T₁ and, especially, T₂ relaxation times in the myocardium. T₁ and T₂ values exhibited a strong consistency with pathologically defined rejection^12–16 and were supposed to be highly sensitive approaches for detecting ACAR.¹⁷ However, elevated T₁ is less specific for active inflammation, as it may also occur in fibrotic areas where free water accumulates.¹⁸ Previous CMR studies have found native T₁ and T₂ to be useful tools for the identification of ACAR.¹⁷ In our study, we found that T₂ demonstrated an independent association with SR, whereas native T₁ did not maintain such independence after T₂ adjustment. Since T₂ more accurately reflects the severity of oedema and has been shown to outperform native T₁ and ECV in multiple studies over the years,^17,19,20 a possible explanation for our results is that myocardial oedema is more remarkable in the SR group in our study, thus making T₂ a more prominent diagnostic marker. Another possible explanation is that native T₁ values can be influenced by both oedematous and fibrotic changes. Therefore, when partially affected by fibrosis, the diagnostic efficacy of T₁ may be limited. Overall, our results indicated that T₂ might be a more effective parameter in the identification of SR than native T₁ and ECV, which requires further validation in future larger, multicentre imaging-pathology researches on HTx recipients.

CMR-FT-derived LV GLS

The conventional LVEF encompasses both longitudinal and circumferential fibre functions, without the ability to distinguish between these components. However, the CMR-FT-derived myocardial strain parameters are sensitive to LV subclinical systolic dysfunction and provide a more accurate assessment of both overall and local myocardial functional status than LVEF.²¹ We found that LVEF and LV GLS were both impaired in NSRs compared with healthy controls, which may be responses to abnormal cardiac function. However, despite LV GLS being worse in SRs compared with NSRs, there was no difference in LVEF. Comprehensively considering all findings, LV GLS may be a more sensitive biomarker of impaired subendocardial function and it was worse earlier than LVEF in SR. One possible explanation is that longitudinal myocardial fibres are located in the subendocardium and appear to be very sensitive to disturbances from various pathologies (such as perfusion abnormalities, myocardial fibrosis, and myocardial oedema).²² Lv et al.²³ and Sade et al.¹⁴ both used ultrasound speckle tracking techniques (EST) and found that EST-derived LV GLS was impaired in HTx recipients and independently associated with ACAR (ACR ≥ 2R). Consistent with these studies using EST, we found that the CMR-FT-derived LV GLS was worse in SRs compared with NSRs and also independently associated with SR.

Incremental value of LV GLS and T₂

We found that the ability for the identification of SR over the NT-proBNP was ever-increasing when including LV GLS, T₂, and the combination of LV GLS and T₂ step by step. It is shown that LV GLS and T₂ complemented each other in strain and tissue aspects for the identification of SR in HTx recipients. In this study, the efficacy of T₂ in the identification of SR was better than LV GLS, which suggests that changes in myocardial tissue may precede or be more severe than strain. This may help future prospective studies and trials select CMR parameters as surrogate endpoints to assess immunosuppression treatment effects. We constructed a combined model of NT-proBNP, LV GLS, and T₂ with an excellent negative predictive value of 97%, which might be useful in avoiding unnecessary EMB in clinic practice.

Limitations

There were some limitations in this study as follows: (i) Although a relatively small participant sample, this is a prospective and biopsy-proven study with comprehensive cardiac examinations, including T₁ and T₂ mapping of CMR. (ii) AMR is less well characterized, complex to characterize, and has a worse prognosis than ACR.²⁴ However, we did not compare the CMR parameters between ACRs and AMRs in this study because of the limited number of HTx recipients. Further differentiation of CMR parameter characteristics between these two is of great clinical significance. A prospective clinical trial with a large sample size to validate our findings and to compare CMR parameters between ACR and AMR is needed in the future.

Conclusion

In conclusion, LV GLS and T₂ value were independently associated with SR in HTx recipients, providing incremental value for the non-invasive rejection identification after HTx.

Supplementary data

Supplementary data are available at European Heart Journal - Cardiovascular Imaging online.

Acknowledgements

We are grateful to all of the participants who made this work possible.

Funding

This study is supported by the National Key R&D Program of China (numbers 2021YFF0501400 and 2021YFF0501404) and the Key Project of National Natural Science Foundation of China (number 81930044).

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

References

Author notes