Simultaneous [18F]fluoride and gadobutrol enhanced coronary positron emission tomography/magnetic resonance imaging for in vivo plaque characterization 

Abstract

Aims

¹⁸F-sodium fluoride ([18F]fluoride) and gadobutrol are promising probes for positron emission tomography (PET) and magnetic resonance imaging (MRI) characterizing coronary artery disease (CAD) activity. Unlike [18F]fluoride-PET/computed tomography (CT), the potential of PET/MR using [18F]fluoride and gadobutrol simultaneously, has so far not been evaluated. This study assessed feasibility and diagnostic potential of [18F]fluoride and gadobutrol enhanced dual-probe PET/MR in patients with CAD.

Methods and results

Twenty-one patients (age, 66.7 ± 6.7 years) with CAD scheduled for invasive coronary angiography (XCA) underwent simultaneous [18F]fluoride (mean activity/effective dose: 157.2 ± 29.7 MBq/3.77 ± 0.72 mSv) and gadobutrol enhanced PET/MR on an integrated PET/MRI (3 T) scanner. Optical coherence tomography (OCT) was used as reference. Target-to-background ratio (TBR, [18F]fluoride-PET) and contrast-to-noise ratio (CNR) values (MRI, gadobutrol) were calculated for each coronary segment. Previously suggested PET/CT-TBR thresholds for adverse coronary events were evaluated. High-risk plaques, i.e. calcified and non-calcified thin-cap fibroatheromas (TCFAs) were predominantly located in segments with a TBR >1.28 (P = 0.012). Plaques containing a lipid core on OCT, were more frequently detected in segments with a TBR >1.25 (P < 0.001). TBR values significantly correlated with maximum calcification thickness (P = 0.009), while fibrous cap thickness was significantly less in segments with a TBR >1.28 (P = 0.044). Above a TBR threshold of >1.28, CNR values significantly correlated with the presence of calcified TCFAs (P = 0.032).

Conclusion

Simultaneous [18F]fluoride and gadobutrol dual-probe PET/MRI is feasible in clinical practice and may facilitate the identification of high-risk patients. The combination of coronary MR-derived CNR values post gadobutrol and [18F]fluoride based TBR values may improve identification of high-risk plaque features.

Graphical Abstract

Open in new tabDownload slide

Introduction

Risk stratification and treatment selection in patients with subclinical and advanced coronary artery disease (CAD) remains a major challenge for cardiologists and clinicians around the globe. CAD evaluation remains focused on the detection of coronary artery obstruction; however a considerable proportion of patients suffering from acute myocardial infarction (AMI) may not present with prior symptoms or exhibit clinically apparent obstructive CAD.^1,2 Therefore, plaque formation and progression may remain subclinical and undetected until the first cardiac event.³ Even in patients with known CAD, assessment of disease activity remains elusive.⁴ Recently, [18F]fluoride (¹⁸F-sodium fluoride) emerged as a powerful molecular probe to detect active microcalcification in response to injury.^4–6 In a recent study, Kwiecinski et al.⁴ demonstrated the potential of coronary [18F]fluoride positron emission tomography/computed tomography (PET/CT) to predict future cardiac events including AMI in patients with CAD.

Non-contrast and contrast-enhanced (CE) magnetic resonance coronary vessel wall imaging (cMRI) has been shown to enable detection and characterization of atherosclerosis and coronary plaque burden, and thus may have potential to improve current risk assessment and therapeutic strategies.^7–11 Combining CE-cMRI derived plaque characterization with additional information derived from [18F]fluoride-PET using hybrid PET/MR imaging may not only yield the potential of advanced characterization and identification of coronary high-risk patients, but also reduce patient exposure to ionizing radiation, which would be favourable in younger patients or serial examinations. Previous studies using [18F]fluoride coronary PET/MR demonstrated feasibility of coronary plaque assessment.^12,13

In this study, we investigated the potential of simultaneous [18F]fluoride and gadobutrol enhanced PET/MR for detection and characterization of coronary atherosclerosis in patients who underwent invasive X-ray coronary angiography (XCA) and optical coherence tomography (OCT) as a reference.

Methods

Study design and participants

This prospective study was approved by the German federal office for radiation protection, the local ethics committee for clinical investigations and was performed in accordance with the Declaration of Helsinki. All patients provided written informed consent. Between September 2018 and January 2020, 21 patients scheduled for XCA due to clinical necessity were enrolled for an additional [18F]fluoride-PET/MR scan and OCT. All patients were in cardio-respiratory stable condition.

Exclusion criteria included: age <50 years, renal insufficiency (creatinine clearance <35 mL/min), pregnancy, inability to provide self-dependent informed consent, contraindications to cardiac MR imaging (i.e. allergies to gadolinium-based contrast agents, claustrophobia, metallic implants, e.g. pacemakers or defibrillators, central nervous system aneurysm clips, known epilepsy, asthma, and severe liver disease). PET/MR was performed with days after XCA/OCT (median 1 ± 5.7 days following XCA/OCT).

X-ray coronary angiography and optical coherence tomography

XCA was performed in all patients in an interventional catheterization laboratory (Allura Xper FD20 laboratory, Phillips, Baltimore, MD, USA). At least two orthogonal views were acquired for each coronary artery. Following the XCA procedure, OCT was performed in 19 patients (preferentially in coronary arteries with high plaque burden according to study purpose and or adapted to clinical considerations). In case of complete absence of atherosclerosis on XCA in non-stent regions, no OCT was performed (n = 2). Iodinated contrast agent (Imeron) was injected for OCT image acquisition by an automatic pump system (Angiomat Illumena, Liebel-Flarsheim, Austin, TX, USA). For offline analysis, OCT datasets were transferred in raw data format. XCA/OCT operators were blinded to PET/MR data.

Positron emission tomography/magnetic resonance

[18F]fluoride was administered intravenously 60 min prior to the PET/MR scan. All scans were performed using a 3 T hybrid PET/MR scanner (Siemens Biograph, Siemens Healthcare, Erlangen, Germany). PET and MR Image acquisitions were performed simultaneous and PET data were recorded in list-mode format. The initial PET image reconstruction used an iterative ordinary Poisson ordered-subsets expectation maximization algorithm with 21 subsets and 3 iterations incorporating point-spread function resolution modelling,¹⁴ a 344 × 344 × 127 matrix, and a 2-mm full-width-at-half-maximum Gaussian post-reconstruction filter (Siemens OP-OSEM algorithm). Attenuation correction for the body in the PET reconstruction was estimated using MR imaging. PET data were reconstructed for 25 min without ECG or respiratory gating to decrease the noise level. MRI and PET scans were performed simultaneously on a single bed centred over the heart. A standard Siemens segmentation approach based on a Dixon-VIBE MR sequence was used for attenuation correction. The UMAP was created from a Dixon MRI sequence acquired at the beginning of the PET recording.

All scans were performed in supine position using an 18-channel matrix coil. Scout scans were acquired to identify heart structures, while trigger delay and acquisition window were determined on a cine 4-chamber view. Subsequently, a T1 scout determined the patient-specific inversion time to null signal from blood. 3D whole-heart coronary MR angiography (CMRA) was performed using a T2-prepared fast low-angle shot (FLASH) sequence while 3D whole-heart coronary vessel wall imaging (CE-CMR) was performed with an inversion recovery (IR) prepared T1-weighted turbo FLASH sequence with fat suppression [fat saturation (FatSat)].⁸ The time delay between contrast injection and CE-CMR was ∼35 ± 2 min.

PET/MR image analysis

PET/MR image analysis was performed after the inclusion of the last patient. For qualitative and quantitative image analysis, dedicated PET/MR image analysis software (Visage 7.1, San Diego, CA, USA) was used. PET and MRI datasets were automatically fused for analysis based on the spatial information. Coronary arteries were divided into eight coronary segments (proximal, mid, distal RCA; left main; proximal, mid/distal LAD; proximal and distal LCX) as described before.^10,15 For each segment, SUVmax values were determined in fused MR and PET images by placing a 3D circular region of interest (ROI) with a diameter of 10 mm in the vessel centre longitudinally scanning the vessel. SUVmax values were entered in a pseudonymized database. Measurements were performed by two independent readers blinded to further patient data. For target-to-background ratio (TBR) determination, SUVmax values were divided by the average SUVmean obtained from a spherical ROI with 10 mm diameter placed in the right atrium at the level of the RCA ostium. Areas with visually higher-than-average accumulation of [18F]fluoride in the blood pool or from extra-cardiovascular structures have been voided. A segment TBR value <1.25 was considered as ‘[18F]fluoride-uptake negative’, a TBR value >1.25 as ‘[18F]fluoride-uptake positive’.¹⁶ A TBR value >1.28 was recently identified in PET/CT studies as a possible threshold for increased risk of future adverse coronary events^4,14 and was defined as threshold for segments potentially harbouring coronary ‘high-risk’ plaques in this study. A segment was considered as TBR >1.25 or TBR >1.28, if segment the TBR value exceeded the predefined TBR threshold of >1.25 or >1.28. Contrast-to-noise ratio (CNR) values were analysed independently of PET data. For that purpose, the mean signal intensity (SI) of each coronary segment (SI segment) was obtained by a 2D delineation of the vessel including the vessel wall.⁹ Afterwards, SI segment was subtracted with the mean SI of blood in the ascending aorta [SI blood—area with signal standard deviation (SD) <10] and finally divided by background noise. Background noise was defined as the SD of the SI in a ROI placed ventrally to the patient’s chest wall outside the body [(SI lesion−SI blood)/noise]. Segments were analysed in three different planes according to their main extent resulting in three CNR values for each segment. The obtained TBR and CNR values were independently analysed and registered in the pseudonymized database assigned to their segments. PET/MR readers were also blinded to OCT and XCA data.

XCA and OCT image analysis

XCA information was taken from official clinical records and verified independently by two readers. Coronary lesions were localized according to the 8-segment model mentioned above. Lesions were then classified by the degree of their angiographically determined luminal stenosis (0= no evidence of atherosclerosis, 1 = <25%, 2 = 25–50%, 3 = 51–75%, 4 = >75%, 5 = 100%, and 6 = previous stent).¹⁷ OCT frames were analysed using appropriate software (ILUMIEN OPTIS PCI Optimization System, St. Jude Medical) by two experienced readers blinded to further patient information. Plaques were characterized according to established OCT-criteria^18,19 and were assigned to six plaque types as previously described^8,20,21: (0) normal coronary segment with no plaque, (I) fibrotic plaque (mainly fibrous tissue), (II) fibrocalcific plaque (>10% confluent calcium, no necrotic core), (III) fibroatheroma (>10% lipid core covered by fibrous tissue), (IV) calcified fibroatheroma (fibroatheroma with >10% dense calcium). (V) non-calcified thin-cap fibroatheromas (TCFA, cap thickness <65µm, >10% necrotic core), and (VI) calcified TCFA (TCFA with >10% confluent calcium). Plaque Types V and VI were considered as ‘high-risk’ plaques. The plaques were allocated to the predefined coronary segments according to their main extension. XCA and OCT data were added to a pseudonymized database.

Statistical analysis

We applied random-intercept multi-level regression techniques to account for the hierarchical data structure²², as we measured several segments (usually eight) in 21 patients. For continuous variables, standardized regression was used to obtain correlations, while we applied logistic regression for dichotomous variables.

All analysis was done in R 3.6.1²³ with the packages lme4²⁴ and lmrTest for obtaining P values²⁵.

Results

Patient baseline characteristics

A total of 21 patients (81% men, mean age: 67 ± 7 years) scheduled for XCA also underwent [18F]fluoride-PET/MR (mean injected activity: 157.2 ± 29.7 MBq; PET mean effective radiation exposure: 3.77 ± 0.72 mSV). Patients had multiple cardiovascular risk factors (hypertension 86%, hyperlipidaemia 81%, and diabetes mellitus 43%) (additional patient demographics and baseline characteristics are shown in Supplementary data online, Table S1). OCT was performed in 19 of 21 patients. The patient majority presented with advanced CAD: 1-vessel: n = 3 (14.3%), 2-vessel: n = 2 (9.5%), 3-vessel: n = 14 (66.7%), and no CAD: n = 2 (9.5%). Most patients received immediate intervention: 15 patients (71.4%) underwent PCI, 2 patients were scheduled for CABG (9.5%), 1 was scheduled for aortic valve repair (4.8%), and only 2 patients with CAD did not require acute intervention (9.5%). One patient (4.8%) had no signs of atherosclerosis on angiography and MRI. In this patient, average TBR across all segments was 1.02 ± 0.08 and no segments with TBR > 1.25 or TBR > 1.28 were detected.

[18F]fluoride-PET/MR

Out of 168 coronary segments, 129 (76.8%) segments were identified and available for PET analyses with high correlation between both readers (ICC = 0.97). The analysed segments showed an overall averaged TBR of 1.62 ± 0.65. Across the entire study population, segment assigned TBR values significantly correlated with the degree of coronary artery stenosis in corresponding segments determined by angiography (P < 0.001; Figure 1). Notably, this correlation was significantly higher when segment TBR values were <1.28 (P = 0.003) and almost disappeared when TBR values were >1.28 (P = 0.193).

CMR analysis

Out of 168 coronary segments, 79 (47%) were presentable in good image quality and available for analysis. Correlation between both readers was good (ICC = 0.9). Average patient CNR values were 21 ± 15.5 and were significantly lower in one patient with no CAD (6.7 ± 5.4; P < 0.001). Across the entire study population, segment assigned CNR values significantly correlated with the degree of coronary stenosis determined by angiography (P < 0.005; Figure 1). Moreover, segment CNR values significantly correlated with corresponding segment TBR values (P = 0.003; R² = 0.129) across all segments (Supplementary data online, Figure S3).

OCT findings

OCT identified 56 segments (n = 51 with plaque Types I–VI; n = 5 no plaque) of which 48 (n = 44 plaque Types I–VI; n = 4 no plaque) could be assigned to corresponding PET/MR-analysed coronary segments. Out of all OCT-detected calcified TCFAs (Type VI; total n = 14), which were assignable to PET/MR-segments (n = 10), all were matched to segments with a TBR value >1.28 (100%). Two non-calcified TCFAs (Type V; n = 2) were detected by OCT, one could not be assigned to a PET/MR defined segment and one was matched to a segment with TBR <1.25. Therefore, the pre-defined ‘high-risk’ plaques (Types V and VI) were more frequently located in segments with a TBR >1.28 (P = 0.012), mainly driven by the location of calcified TCFAs. In total, 16 calcified fibroatheromas (plaque Type IV) were detected by OCT and 16 could be matched with corresponding PET/MR segments, of which n = 13 were matched to segments with TBR >1.28, n = 2 to segments with TBR between >1.25 and TBR <1.28 and n = 1 to a segment with TBR <1.25. Non-calcified fibroatheroma detected by OCT (Type III; total n = 5), which could be matched to corresponding PET/MR-segments (n = 4) were observed in segments with TBR >1.28 (n = 3) and TBR >1.25 (n = 1). OCT-detected fibrocalcific plaques (Type II; total n = 14), that could be matched to PET/MR segments (n = 13), were observed in segments with a TBR >1.28 (n = 10), a segment with a TBR between >1.25 and TBR <1.28 (n = 1) and segments with a TBR <1.25 (n = 2). Purely fibrotic plaques (Type I) were not detected by OCT. Out of five coronary segments detected by OCT, n = 1 segment was matched to TBR > 1.28 and three were matched to TBR <1.25 (mean TBR 1.2 ± 0.18; see also Supplementary data online, Table S2). Plaque Types III–VI, therefore containing a lipid core, were more frequently detected in segments with TBR >1.25 (P < 0.001). TBR values significantly correlated with the maximum calcium thickness (P = 0.009). Notably, fibrous cap thickness was significantly thinner in segments with TBR >1.28 (P = 0.044). Furthermore, when applying a segment TBR threshold >1.28, CNR values significantly correlated with calcified TCFAs (plaque Type VI) detected on OCT (P = 0.032).

Discussion

This study demonstrates for the first time the clinical feasibility of identifying coronary high-risk plaques non-invasively with simultaneous [18F]fluoride and gadobutrol enhanced PET/MRI (Figure 2). Such an approach may have high clinical impact, as it combines the potential of PET and contrast-enhanced MRI for coronary plaque detection and characterization, while also reducing patient exposure to additional, usually CT-associated ionizing radiation and potential complications of invasive XCA/OCT, which would be advantageous in younger patients and for serial examinations.

Delayed contrast-enhanced coronary MR imaging allows direct assessment of contrast agent uptake in the coronary wall. Non-specific extracellular contrast agents, such as gadobutrol, used in clinical practice rapidly extravasate into the vessel wall and are thought to enhance areas with increased distribution volume, delayed clearance, or increased neovascularization.²⁶ Areas of delayed contrast enhancement were shown to correlate with the severity of atherosclerosis in comparison to CT and quantitative coronary angiography.¹¹ In patients with giant cell/Takayasu arteritis, transient contrast agent uptake was measured, potentially linking this pattern to acute inflammation and oedema formation.²⁷ Moreover, in patients with systemic lupus erythematosus and subclinical coronary vasculitis, increased coronary contrast enhancement was observed and compared with CAD patients displayed a diffuse pattern of enhancement.²⁸ Several studies demonstrated that both plaque composition (e.g. fibrous vs. atheromatous)²⁹ and inflammation^10,27,28 may be associated with the retention of extracellular contrast agents.

Regarding the [18F]fluoride uptake, Adamson et al.¹² recently demonstrated good correlation between [18F]fluoride uptake in native coronary arteries between PET/MR and PET/CT (current ‘reference standard’). In this study, we also observed a significant correlation between coronary plaque enhancement (elevated CNR values) on MRI and the degree of stenosis by XCA (P < 0.005). Moreover, segment assigned CNR values showed a significant correlation with corresponding TBR values (P = 0.003). Therefore, coronary segments affected by atherosclerosis showed both increased [18F]fluoride and gadobutrol uptake.

In this study, we observed a significant correlation between TBR values and the degree of coronary stenosis detected by XCA (P < 0.001). However, this correlation was most significant in segments with a TBR <1.28 (P = 0.003), more likely reflecting atherosclerotic plaque burden than disease activity. In segments categorized as TBR >1.28, no significant association between TBR values with XCA determined degree of stenosis was observed (P = 0.193). In these segments, TBR values may predominantly reflect disease activity that has not been attended by relevant stenosis yet. This interpretation is in line with previous findings. In a study conducted by Kwiecinski et al.,⁴ increased coronary [18F]fluoride uptake was associated with higher SMART risk scores (P = 0.03) and higher calcium scores (P = 0.003) but not with stenosis presence or severity (P > 0.10). Further studies demonstrated no significant association between increased [18F]fluoride plaque uptake and the severity of coronary obstruction but with more high-risk plaque features detected by intravascular imaging.^6,30

An increase in necrotic core volume and attenuation of the thin fibrous cap has been identified as strong risk factors of imminent AMI.³ Ongoing inflammation resembles a key factor that drives atherosclerosis and plaque progression^31–33 and may result in microcalcification (disease activity) as a means to stabilize plaque and which can be detected and quantified by [18F]fluoride-PET.^5,6 The OCT-analysis in this study revealed coronary segments from native vessel wall to TCFAs with advanced calcified necrotic cores. We detected 4 segments on OCT with no evidence for plaque (Type 0), which were predominantly matched to segments considered as ‘[18F]fluoride-uptake negative’ (TBR <1.25; n = 3). Plaques containing a lipid core detected by OCT (Types III–VI) were significantly more often matched to segments considered as ‘[18F]fluoride-uptake positive’ (TBR >1.25; P < 0.001). Moreover, OCT identified 10 advanced calcified TCFAs (Type VI) that were exclusively matched to segments with ‘high-risk’ TBR >1.28 (100%). For that reason, plaques from the predefined ‘high-risk’ group (plaque Types V and VI; n = 11) were detected significantly more frequent in segments with a TBR >1.28 (P = 0.012). Consequently, the obtained fibrous cap thickness was significantly thinner in segments with a TBR >1.28 (P = 0.044). Moreover, a significant correlation between TBR values and maximum calcium thickness (P = 0.009) measured on the OCT-frame with the largest extend was identified.

In a study by Lee et al.,³⁰ no difference was found in the prevalence of TCFAs and minimum cap thickness between [18F]fluoride-positive and [18F]fluoride-negative plaques. Moreover, no differences in calcium area and maximum calcium arc were observed. However, other plaque high-risk features such as higher maximum lipid arc and more frequent prominent microvessels were associated with [18F]fluoride-positive plaques. The majority of invasive or non-invasive studies including ours identified one or more high-risk plaque features in lesions with increased [18F]fluoride uptake,^6,14,30 most likely representing different stages or dynamics of disease activity.

Regarding CE-cMRI, previous studies suggest increased gadolinium uptake correlating with plaque burden in stable disease and inflammation in unstable patients. Gadolinium uptake seems pronounced in mixed calcified/non-calcified⁷ or calcified¹¹ plaques. In patients following AMI, a significantly increased atherosclerotic plaque gadolinium enhancement was observed.¹⁰ Furthermore, in patients with sub-acute MI increased gadolinium enhancement was detected in coronary culprit lesions.⁹ In this study, CNR values showed a significant correlation with calcified TCFAs (Type VI; P = 0.032) above the ‘high-risk’ TBR threshold of >1.28. The combination of gadobutrol and [18F]fluoride uptake together may allow a more detailed non-invasive characterization of coronary plaque features. [18F]fluoride-PET is currently being investigated in several clinical trials and showed already encouraging results in identifying coronary high-risk patients. [18F]fluoride-PET may have the potential to close the diagnostic gap between non-obstructive or subclinical lesions and acute coronary events, identifying high-risk patients early and guide them towards more dedicated therapies^4,31 requiring serial and quantitative monitoring of therapy response. As the risk of radiation-dependent adverse effects increases with radiation exposure, radiation should be minimized to the lowest necessary dose. Hybrid PET/MR scanners and dedicated MR-contrast agents may not only contribute to a significant reduction of radiation exposure but also add further prognostic and diagnostic information, especially when combined with coronary [18F]fluoride-PET.

Limitations

Our study has several limitations. First, we used TBR values derived from PET/CT studies, it remains to be seen if TBR thresholds derived from PET/CT studies can be completely applied for hybrid PET/MR imaging. In general, [18F]fluoride plaque uptake may be influenced by blood pool clearance¹² and administered [18F]fluoride activation. TBR determination can be significantly influenced by determination of mean blood pool activity, which is currently not standardized for hybrid PET/MR. A substantial number of patients had known CAD and were already pre-treated with previous PCI/stent-implantation, statin intake or had a recent myocardial infarction (<6 months). It remains unclear, to what extend this influenced coronary tracer uptake. However, discrimination may be more accurate in younger patients with less advanced CAD and without previous treatment.

As described before,¹² coronary stents often cause severe artefacts in MR. However, in most patients, adjacent coronary segments were available for MR analysis. We simplified the AHA coronary arteriogram¹⁷ and narrowed it to an 8-segment model, LAD was divided in a proximal and distal segment (mid and segments of distal LAD that were together considered for analyses). Furthermore, OCT-detected plaques and angiographic results were assigned to the segment of their main extension, which may create a lack of definition or blurring in segment assignment. However, most plaques did not maintain segment borders either and coronary disease activity may more likely resemble rather a pan-coronary than a strictly localized process.

Conclusion

Our study demonstrates the clinical feasibility and potential of dual-probe coronary [18F]fluoride and gadobutrol enhanced PET/MR imaging in patients with established or suspected CAD in a small real-world patient cohort. Increased uptake of Gadobutrol and [18F]fluoride was detected in coronary high-risk plaques. In this study, we applied TBR thresholds derived from PET/CT studies, however combining the strengths of [18F]fluoride PET and gadobutrol enhanced MR may enable reduced radiation and add further information for coronary plaque characterization and disease management.

PET/MRI of coronary arteries remains challenging. Artefacts caused by previous coronary interventions (i.e. coronary stents) and patient movement may impair PET/MR analysis considerably. According to our experience, PET/MR seems to be most appropriate in proximal coronary arteries, younger and untreated (i.e. no previous stent) patients, that may represent the patient group with the greatest benefit from PET/MR. Further studies are now warranted to fully assess the potential of simultaneous [18F]fluoride and gadobutrol enhanced PET/MR in CAD and evaluate or define PET/MR-specific TBR and CNR thresholds.

Supplementary data

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

Acknowledgements

T.H.W. is a participant in the BIH-Charité Clinician Scientist Program funded by the Charité–Universitätsmedizin Berlin and the Berlin Institute of Health.

Funding

This work was supported by the Sonderforschungsbereich (SFB), Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—SFB 1340/1 2018, B01. PET/MRI use was supported by DFG fund INST 335/543-1 FUGG.

Data Availability

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

References

Author notes