Atrial Fibrosis by cardiac MRI is a correlate for atrial stiffness in patients with atrial fibrillation

Aims: A relationship between left atrial strain and pressure has been demonstrated in many studies, but not in an atrial fibrillation (AF) cohort. In this work, we hypothesized that elevated left atrial (LA) tissue fibrosis might mediate and confound the LA strain vs. pressure relationship, resulting instead in a relationship between LA fibrosis and stiffness index (mean pressure/LA reservoir strain). Methods and Results: Sixty-seven patients with AF underwent a standard cardiac MR exam including long-axis cine views (2 and 4-ch) and a free-breathing high resolution three-dimensional late gadolinium enhancement (LGE) of the atrium (N=41), within 30 days prior to AF ablation, at which procedure invasive mean left atrial pressure (LAP) was measured. LV and LA Volumes, EF, and comprehensive analysis of LA strains (strain and strain rates and strain timings during the atrial reservoir, conduit and active phases) were measured and LA fibrosis content (LGE (ml)) was assessed from 3D LGE volumes. LA LGE was well correlated to atrial stiffness index (LA mean pressure/LA reservoir strain) overall (R=0.59, p<0.001), and among patient subgroups. Pressure was only correlated to maximal LA volume (R=0.32) and the time to peak reservoir strain rate (R=0.32), among all functional measurements. LA reservoir strain was strongly correlated with LAEF (R=0.95, p<0.001) and LA minimum volume (r=0.82, p<0.001). Conclusion: In our AF cohort, pressure is correlated to maximum LA volume and time to peak reservoir strain. LA LGE is a strong marker of stiffness.


Introduction
Atrial brillation (AF) is the most common arrhythmic disease at 1 to 2% of the U.S population, with the forecast of increasing prevalence [17]. AF originates from a trigger and substrate, with substrate being a large brotic atrium. As atrial brosis stiffens the LA [5], this may cause additional increased atrial pressure, or reduced LA strain. The role of elevated cardiac pressure in AF and left atrial (LA) remodeling is important [32,39]. Firstly, elevated pressure is thought to increase LA volumes, ultimately leading to stretch and to development of atrial brosis [33], and potentially lead to new onset AF [22]. Secondly, many categories of patients with elevated cardiac pressure progress to AF, including patients with heartfailure [1], hypertrophic cardiomyopathy [26], and obstructive sleep apnea [21] among others.
The triad of strain, pressure and stiffness (e.g. brosis) are crucial in understanding atrial remodeling. A simple to understand model of strain (e, equals stress / stiffness) on a spherical shell of radius r, and thickness b, yields [34,37] : Eq. 1.
In non-AF subjects, atrial strain changes are often advocated as an early sign of pressure increase [3,4,14,35,36]. LA strain as a correlate of pressure is an attractive possibility. To date, there are no imaging biomarkers to estimate cardiac pressure accurately; many echo correlates (E/e', etc) are not fully robust.
Others have studied LA strain as a direct biomarker of LV lling pressure [3,4] phase of the LA, and the time indices of these phases (both time of peak strain and strain rate for each phase, all normalized by RR) (Fig. 1A). The conduit strain is measured by the difference of those two peaks. The conduit strain time is the duration of time measured between these two peaks.
In all subjects, when possible, three LA volumes (LAVi) (minimum, maximum and pre-atrial kick volumes) were measured using the bi-plane area method [40], and two EFs (active and total). Active LAEF was de ned here as (LAV act -LAV min )/LAV min . LAV act is the volume at begin atrial-systole. LA stroke volume (LASVi) was considered as the difference between the maximum and minimum LA volumes. Standard LV volumetric parameters of EF, mass, end-diastolic volume (EDVi), end-systolic volume (ESVi), stroke volume (SVi) were measured. Volumes and mass were indexed by patients' body surface area (BSA). The total number of functional correlates analyzed was 25. Additional File 1 shows the analysis of LA strains in one patient. All data (LA LGE, LA strain) was analyzed blinded to other metrics and was performed for research purposes.
LGE post-processing Quantitative analysis of the LA LGE was performed by manual segmentation using 3D Slicer, used to indicate the pixels containing LA myocardium on each slice of the 3D volume ( Fig. 1B). Pulmonary vein ostia were included, up to 5mm into each PV ostia; enhanced valvular pixels were excluded. Using the manual wall segmentation, the enhanced pixels were determined using a patient-speci c threshold. The patient-speci c threshold was determined using an image intensity ratio (vs. blood signal) of about 1.4 of mean blood pool signal, as previously described [30]. The threshold was adjusted to a level matching the signal intensity of the enhanced mitral valve (which is known to be partly composed of collagen), but also to exclude blood pool pixels. The inter and intra-observer variability of this approach has been reported [30] (intra-observer variability: bias ± 2SDs = − 0.3% ± 2.9%, ICC = 0.94; inter-observer variability: bias ± 2SDs = 0.6% ± 7.3%, ICC = 0.71). All LGE analysis were performed by a single expert (DCP), with more than 10 years of experience in LA LGE analysis. The volume of segmented enhanced LA myocardium was then calculated (LGE in mls) and also normalized by the total LA myocardium volume (LGE%) [30].
Pressure measurement.
Invasive left atrial pressure was measured prior to the catheter ablation, during which patients were under general anesthesia. Average atrial pressure was measured from the LA using an 8.5 F sheath following transseptal puncture. The electrophysiologists were blinded to pre-procedural LA LGE ndings.

Stiffness Index
We calculated atrial stiffness index as used in echocardiography, i.e. LA pressure/LA reservoir strain [4] [16], using invasive mean atrial pressure.

Patient categorizations
Motivated by Eq. 1, and prior studies [14], patients were further characterized by presence or absence of enlarged minimum LAVi (> 30 ml/m 2 ) (representing mean LAVI + 3SDs [28]) to further examine the relationships in more homogeneous cohorts. We also investigated categorization by elevated or normal pressure, using the median value from our cohort (LAP = 12 mmHg).  Overview of relationships Figure 3 shows a heat map of all correlations for the total cohort, with any non-signi cant correlations plotted as R = 0. Some general trends can be observed. Pressure is not correlated to any functional metrics, with exceptions of maximum LA volumes, and time to peak strain rate in the reservoir phase.
Atrial LGE is mostly correlated with LA volumes, strains and most strongly, stiffness index.
Most LA strain and LA volume metrics are strongly correlated (R > 0.6) with each other, except for timings. For example, LAEF correlates with LA reservoir strain, with r = 0.95, p < 0.001. Conventional LV function parameters do not correlate with LA functional parameters strongly.

Correlations with atrial LGE
Atrial brosis (LGE in mls) was highly correlated to LGE % (R = 0.78) ( p < 0.001). Neither LGE (mls) or LGE% correlated with age, but LGE(%) did correlate with BMI (R = 0.54, p < 0.001). Table 2 shows the correlations found with LGE. Atrial brosis (LGE mls) correlated most strongly with stiffness index (R = 0.57, p < 0.0001) (Fig. 4, Table 2). When analyzing patients with normal and elevated minimum LAVi separately (Table 2), we found that LA LGE more strongly correlated with metrics of strain in patients with normal volume, e.g. reservoir strain showed a stronger relationship with LGE (R = 0.63 p = 0.001), but not LA volume. Among patients with elevated LA volume, LA LGE became well correlated with LA volumes (R > 0.67, p < 0.003), but not LA strains. This shows that strain indices do re ect brosis but more strongly among a group with normal LA volume. Stiffness index was most consistently (i.e. for both elevated and normal volumes) correlated with atrial brosis. LGE correlations were mainly stronger than LGE (%) (see Table 2 vs. Table 3), but showed the same trends. For those with observable active function, LGE (mls) was signi cantly correlated only with active peak strain rate (R = 0.36, p = 0.03) ( Table 2), of all active strain parameters.     Table 1) pressure was signi cantly correlated (among the total cohort) only with increased BMI (R = 0.27, p = 0.026), increasing maximum LAVi (R = 0.32, p = 0.0076) and increasing time to peak reservoir strain rate (R = 0.32, p = 0.008) (Fig. 4). No other correlate was found.
Functional Correlates with LA volumes LA reservoir strain was very strongly correlated to many volume indices (LV SVi, LVEF, LVESVi, and all three LA volumes) (Fig. 3), but most strongly with minimum LAVi (R = 0.82, p < 0.001).

Discussion
There are several new ndings from this study. Importantly, this study is the rst to identify LA LGE as an index of atrial stiffness, as observed by its moderately strong relationship to "stiffness index" (Pressure over reservoir strain) in all patients (R = 0.59 for LGE), con rmed in subgroups of with normal and elevated volumes. The relationship of LA LGE (a surrogate biomarker of brosis) to stiffness index and to strain (especially for patients with normal volumes) is expected based on Eq. 1, and because collagen increases stiffness [5].
While pressure did not correlate to LA strains, there were signi cant relationships between pressure and maximum LA volumes, and time to peak reservoir strain. The increases in LA volume with pressure of the thin-walled LA is expected; indeed an echocardiographic study found a remarkably similar effect, if slightly weaker (r = 0.24) [9] in 100 patients with hypertrophic cardiomyopathy. The relationship of cardiac timings with pressure is known in other contexts [25]. For example, increasing isovolumic relaxation time (IVRT) [7] has been associated with elevated pressure, as has deceleration timings of early diastolic ow [10] and pulmonary vein ow [15]. In this study, pressure was weakly to moderately associated with timing to peak reservoir strain rate, associated with LA lling speed. An overstretched atrium might be slower to ll,. The nding that strain and pressure are not correlated in the general AF cohort agrees with a recent report comparing atrial strain and pressure [14], where a subset of AF subjects was analyzed separately, with the nding that strain was generally low and the relationship not signi cant. LA LGE was found to be modestly correlated to pressure in a study of AF subjects [31], which we did not nd, possibly because they had a greater range of elevated pressures and did not use mean LAP, as in this study.
Finally, LA strain is strongly correlated to LAEF (R = 0.95) and LA minimum volume (R = 0.82), which questions the uniqueness of the LA reservoir strain metric. A recent study showed that every point on the LA strain vs. time curve is strongly linearly related to LV strain vs. time curve with a slope equal to the ratio of maximal LV and LA volumes [23].
LA LGE volume (mls) showed consistently stronger correlations with strains, stiffness index, and other metrics than LGE (%). The concept of normalizing LGE in mls by volume of atrial myocardium was introduced by Oakes et al. [24] and is analogous to the enhancement percentage used for ventricular LGE. However, in AF where the hallmark of disease is both an enlarged and brotic myocardium, LGE (%) may be less useful as a biomarker than LGE (mls), since normalizing elevated LA LGE by an elevated LA volume certainly will obscure abnormal ndings.

Limitations
Although effort was made to exclude patients who were imaged during AF, some patients might have been incorrectly included or excluded in this study. Pressure was measured prior to the ablation procedure, and its reliability is likely to be less than if it were measured for clinical decision-making; furthermore, average pressure was obtained instead of the complete pressure curve and some data may have been acquired during non-sinus rhythm. It is intuitive that pressure changes (i.e. from begin to end LV systole), might be a more relevant correlate of reservoir LA strain magnitude. Additionally, the cohort included fewer patients with very high pressures, compared to other studies [38]. We were not able to use wall-thickness of the LA as another factor in the strain-stiffness-pressure relationship [2]. Increasing the cohort size would have been useful.

Conclusion
In conclusion, our study of AF patients demonstrates that LA LGE is a marker of stiffness (Pressure/LA reservoir strain). Atrial reservoir strain is linked to LA volume, and thus the relationship of brosis to strain is stronger among patients with normal volumes. LA pressure did modestly correlate to maximum LA volume, and with cardiac phasic timing of peak reservoir strain. In contrast to that reported in other cohorts, in AF patients LA strain is not a good correlate of left atrial pressure.

Declarations
Funding: The authors acknowledge funding from NIH: NIH 1R01HL144706, Development of MR-derived parameters of LV diastolic function: Validation and Comparison to LV and LA brosis.
Data Availability Statement: The strain analysis package is available upon reasonable request from the authors. The data consists of pressure, strains, functional metrics, and atrial brosis, and will be available from the last author upon reasonable request.  Figure 1 Left atrial strain and brosis measurement. A) Comprehensive left atrial strain measurements evaluated nine metrics of strain. These consisted of strain at three phases (reservoir, conduit and, active phase), and timings to peak strain. Additionally, the time derivative of the strain curve yielded three strain rates and times to peak strain rate in the three phases. Timings were normalized by RR. B) In two subjects, atrial LGE slices from a 3D volume are shown with segmentations (in green), using a subject speci c threshold, determined by including signal associated with the valvular enhancement, but excluding blood pool signals.

Figure 2
Representative examples of strain and atrial LGE. A) Strain curves for four patients, with normal LA volume, pressure and strain (blue), elevated atrial LGE but normal volume (grey), elevated atrial LGE and high volume (orange), and a patient with high pressure and volume but not signi cant LGE (yellow). The higher LA volume patients have reduced strains. In the high LA LGE subject with normal volume, strain is also reduced. In the high pressure subject, the time to peak reservoir strain rate increased. B-C) show representative color-coded slices from the atrial LGE images for the normal patient (B), and the patient with elevated LA LGE (C).  Left atrial brosis (LGE (mls)) correlates.
LGE was correlated with strains, volumes and strain rates.
Correlations were stronger among subjects with normal minimum LA volume. Correlations with Stiffness index was strongest. All shown correlations were signi cant (p<0.01).

Figure 5
Correlates of mean atrial pressure in the whole cohort (the other correlate, BMI, is not shown).

Figure 6
When comparing groups with normal vs. elevated pressure (LAP≥12mmHg, n=33), reservoir and conduit strain times and maximum LAVi were found to be different.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.