Regional Changes in Myocardial Strain Predict Ventricular Remodelling after Myocardial Infarction in a Large Animal Model

65 To identify early predictors of late left ventricular remodelling (LVR) post-myocardial infarction (MI) and related 66 molecular signatures, a porcine model of closed-chest balloon MI was used. LVR was assessed by cardiac 67 magnetic resonance imaging (CMRI) at baseline, 12-48 hours (acute), and 5-6 weeks (chronic) post-MI. Changes 68 in myocardial strain and strain rates were derived from CMRI data. Tissue proteomics was compared between 69 infarcted and non-infarcted territories. Peak values of left ventricular (LV) apical circumferential strain (ACS) 70 changed over time together with peak global circumferential strain (GCS) while peak GLS epicardial strains or 71 strain rates did not change over time. LVR post-MI enhanced abundance of 39 proteins in infarcted LV 72 territories,21 of which correlated with LV equatorial circumferential strain rate (ECSR). The strongest 73 associations were observed for D-3-phosphoglycerate dehydrogenase (D-3PGDH), cysteine and glycine-rich 74 protein-2 (CG-RP-2), and secreted frizzled-related protein 1 (sFRP1). Results indicate that early changes in 75 regional peak ACS predict late LV remodelling and LVR post-MI is associated with augmented levels of D- 76 3PGDH and sFRP1, which show the strongest association with peak ECSR. These findings might help to prevent 77 LVR post-MI by influencing/directing LV unloading strategies or by pharmacological control of tissue levels of 78 D-3PGDH and sFRP1. 79 early strain-based predictors of late post-MI in a preclinical

The strain-based metric used in this study was formulated to be robust and reproducible across 129 sites/users/software, use a transparent, non-proprietary algorithm, and be sufficiently sensitive to characterise 130 local ventricular function in a layer-wise manner, as recently reported by our group 18 . The methodology is 131 summarised on the LV schematic in The reference length, L0, was the endocardial perimeter length at end-diastole, and this was then compared to 135 the endocardial perimeter length, L, throughout the cardiac cycle for a given CMRI slice. Between seven and 136 nine short-axis slices were contoured depending on the specimen and time point; for example, the degree of 137 eccentric hypertrophy and specimen growth could increase the number of slices for a given specimen over time. 138 Strain rate, , as defined by Equation 2, was also calculated for the circumferential direction, where ( , ) and 139 ( , +1 ) are endocardial lengths on one cine MRI stack ( ) at consecutive points in time ( ), ∆ is the time 140 between successive images, and 0 ( ) is the end diastolic length for that cine slice. 141 The same short-axis slices were also divided into three 'vertical regions': apex, equator/mid, and base with two 143 or three slices per region, and regional circumferential strains and strain rates were calculated. Similar analyses 144 were conducted for the epicardium. To assess statistically significant changes through time, peak strain and 145 strain rate values were measured. The strain values were averaged over 2-3 slices per region rather than from 146 only one slice as reported by others 19 2 . Accordingly, the global CS defined here is the average of 7-9 regional 147 slices, rather than of only 3 slices reported by others 19 . 148 Four-chamber long-axis data were used to investigate LV longitudinal strains and strain rates in both the 149 endocardium and epicardium. Volumes were calculated for LV end-systole and end-diastole by multiplying the 150 area within the endocardial contour for a given slice by the slice thickness, and then all slice volumes were 151 summed to give the ventricle blood pool volume. LV end-systolic volume indexed to weight (LVESVi) was 152 calculated by dividing the end-systolic volume by the body surface area (BSA) of the animal. BSA was found 153 through the relation suggested by Kelley et al. shown in Equation 3 20 . 154 The endocardial and epicardial contours were manually traced on all short and long axis images in the 156 commercially available software package OsiriX (Pixmeo, Geneva, Switzerland), by one experienced user, and 157 checked by another. Contours were analysed and strain & strain rate values were calculated using an in-house 158 7 Proteomics analysis of infarcted vs remote viable myocardium was performed in five animals in keeping with 161 established methods 21 . Tissue homogenization was obtained with ceramic beads with Ripa Buffer and a 162 protease and phosphatase inhibitors cocktail. The BCA method was used to quantify the protein concentration 163 and samples were prepared at 2 mg/ml for the mass spectroscopic analysis (MSA). Additional details are 164 available in the Online Supplement. 165 166

Statistical Analysis 167
Non-parametric analysis was performed. Variables are presented as medians and confidence intervals. LVEF, 168 peak values of global LS, global and regional CS, and peak values of corresponding strain rates measured at 169 acute and chronic time-points post-MI were compared with the baseline data using a Kruskall-Wallis test. 170 Observed significant differences were analysed further by using Mann-Whitney U tests. A p-value of <0.05 was 171 considered statistically significant, but due to the high number of hypotheses tested, Bonferroni corrections were 172 performed which suggested a p<0.0024 as statistical significance. One-way ANOVA was used for initial 173 assessment with Gabriel's test to find differences between pairs of means. Linear regression and correlation 174 analyses were performed to assess relationships between the scar weight and other mechanical properties, and 175 between biomarker expression and strain. Statistical analyses were performed in IBM SPSS (IBM Corp. For each protein, an abundance ratio between infarcted and non-infarcted samples was calculated. Proteins 180 found to be at least two-fold more expressed in the infarcted myocardium were correlated with the endocardial 181 strain data of the acute phase, with those showing the strongest correlation (R 2 ≥ 0.95) being more closely 182 evaluated; a series of univariate linear regression models were performed to correlate each identified protein to 183 each mechanical variable. This statistical analysis was conducted using R version 3.4.4. For western blotting 184 analysis quantification of band intensity was performed using AlphaEase v5.5 software followed by background 185 subtraction and correction for protein loading. For evaluation of the differences between the protein expression 186 in the non-ischaemic and ischaemic myocardium, the two-tailed unpaired Student's t-test was used. 187

Characterisation of MI by CMRI and serial troponin I release 191
CMRI outcome is shown in Table S1, Online Supplement. Overall mean LVEF dropped from 56.6% ± 2.5% at 192 baseline to 45.3% ± 7.6% at 4 to 72 hours (acute) and to 49% ± 4.6% at 5-6 weeks (chronic). Mean LV scar size 193 was 16.9g ± 9.1g at the acute time point and 9.38g ± 5. global metrics, such as LVEF, GCS and GLS, myocardial strains from all 10 experiments were included as these 205 indices should be able to characterise the severity of an infarct & the subsequent LV remodelling regardless of 206 the affected coronary territory. LV remodelling, occurring due to myocardial tissue's response to the imposed 207 occlusion, encompasses changes in ventricular shape, volume, and function throughout the cardiac cycle. 208 Similarly, MRI data from all baseline, pre-MI scans were retained in the analysis. Nevertheless, for, the 209 statistical analysis of myocardial strain changes in ACS at acute and chronic timepoints only data from the 8 210 LAD experiments were included, as only these cases were expected to determine an MI affecting the apical 211 region i.e. that covered by the ACS metric. The same approach was taken for the other regional strains ECS and 212 BCS. An evaluation of the regional strains for the CX territory was not performed, because with only n=2 213 experiments in this sub-group such an evaluation would not have been meaningful. 214

Long-axis global and transmural LV strains
9 Changes in myocardial strains over time for all animals are shown in Table 1 and Table S3 (Table S4, Online Supplement). No significant changes were seen in epicardial CS, with all p>0.03 (Table S4). 227 Endocardial and epicardial strain rates did not differ at the acute or chronic time-points vs. baseline (p>0.008; 228 Table S4). No significant correlations were observed between scar weight and 229 circumferential strains. The intra-observer variability for LVEF and LVEDV were 3% and 2% respectively, and 230 the inter-observer variation were 8% and 7% respectively. 231

LV Global Function and Associations with Strain 232
LVEF and left ventricular end systolic volume index (LVESVi) were measured by established CMRI methods 233 (Table S1, Online Supplement) with relevant strain contours derived by hand (Table 1) (Figure 4). LV volumes indexed to weight were assessed given the substantial weight gain 237 observed from acute to chronic time-points. This showed that there were no significant changes in LVESVi 238 from acute to chronic time-points (Table 1). No difference was seen in LVEF values between those measured 239 by CMRI and the relevant strain contours derived by hand. 240

Myocardial proteomics, LV strains, and validation of D-3PGDH and sFRP1 by western blotting 241
Proteomics data and correlation with LV strains are reported in Figure 5 and Table 2. 5981 proteins were 242 identified, and 39 proteins were increased in infarcted territories ( Figure 5). For the analysis correlating 243 proteomics with strains, proteomics data from 4 hearts was used as strains were not available for the 5 244 experiments. Significant linear correlations were found between endocardial circumferential strain rate (ECSR) 245 and 21 of the proteins increased in the infarcted territories ( Table 2). The proteins showing the strongest 246 correlation (R 2 ≥ 0.95) with the ECSR were: D-3-phosphoglycerate dehydrogenase (D-3PGDH, R 2 = 0.96, p = 247 0.01), cysteine and glycine-rich protein-2 (CG-RP, R 2 = 0.95, p = 0.02), and secreted frizzled-related protein 1 248 (sFRP1, R 2 0.96, p = 0.01). Western blotting for D-3PGDH and sFRP1 confirmed that the level of D-3PGDH 249 and sFRP1 protein in the infarcted myocardium was significantly increased compared to the non-infarcted 250 myocardium (both P<0.05, Figure 6, Figure S2 in Supplemental file). Western blotting for CG-RP showed no 251 difference. 252

253
This study identifies an association between early change in regional strain, late LVR post-MI and enhanced 254 abundance of D-3PGDH and sFRP1 proteins. An early change in regional ACS was observed (but not in global 255 measures such as GCS, LVEF or GLS), which was predictive of late LVR. Additionally, LVR post-MI is 256 associated with an hyperexpression of 39 myocardial proteins, of which 21 correlated specifically with ECSR, 257 with D-3PGDH and sFRP1 exhibiting the strongest correlation with ECSR. 258 Our findings show early changes in regional ACS and GCS strains reflecting accurately the affected myocardial 259 territories. This was associated with early changes in strain that predicted late LVR: ACS: baseline: -37.5%, 260 chronic: -19.1%, p=0.002; GCS: baseline: -34.9%, acute: -23.8%, p=0.002, chronic: -27.7%, p=0.006. 261 GLS has been suggested as a predictor of late LVR in STEMI patients 10 and in an open-chest coronary ligation 262 porcine MI model 11 . It is argued that GLS may predict LVR as the LV apical region affected by ischemia 263 contains more longitudinal fibres, which contributes more to the local contractile performance and are less 264 affected by ischemia 12 . However, the distribution of the circumferential fibres across the LV might reflect 265 changes to longitudinal and circumferential deformations 12 , therefore suggesting that CS metric might add 266 significantly to gauge predictive information on myocardial deformation 22 . 267 In our study GLS did not change over time. The results seem to suggest that regional CS might be more 268 sensitive than GLS in predicting late LVR and quantifying LV function. Averaging strains over few slices 269 within a specific LV region, as opposed to inspecting individual slices or global metrics, might ensure that small 270 differences in image location (from patient movement or from scans at different times or different patients), 271 become less critical when comparing data, boosting the reproducibility and robustness of the method. In 272 addition, performing strain over smaller volume/regions is associated with less variation, hence with higher 273 potential of identifying smaller changes. Accordingly, CS has been shown to be an effective indicator of MI, 274 marker of LV function, and infarct transmurality (25). These findings, if confirmed, might affect the type and 275 timing of pharmacological and/or mechanical LV unloading approaches post-MI to prevent heart failure (27). 276 Correlations were also found between GCS and LVEF as well as between GLS and LVEF. Both GCS and 277 LVEF were calculated using the same short-axis data, and so a strong correlation was expected based on 278 geometrical considerations. Long-axis data was used for GLS, so the correlation found between these 279 parameters suggests that GLS might be able to detect MI and changes to LV function in keeping with findings 280 by others 26 . 281 The occurrence of MI and related ischemia/reperfusion injury trigger a storm of molecular signalling, cellular 282 remodelling, inflammatory reaction and fibrosis leading to scar formation and LVR 27,28 . Farah and colleagues 29 283 defined LV remodelling as an increase of 10% in ventricular end-systolic or end-diastolic diameter, and found a 284

58% incidence of LV remodelling after an anterior MI compared with other studies. In the Acute Myocardial 285
Infarction Contrast Imaging (AMICI) trial, the term 'reverse REM' was employed to denote a >10 % reduction 286 in LVESV found at 6 months in 39% of patients following PPCI 30 , being the only independent predictor of 2-287 year event-free survival. Based on this definition we found that 75% of our experiments had an LV remodeling 288 at both LVEDV and LVESV at 5-6 weeks at serial CMR. 289 Binek et al. found that ischaemia triggers changes in the levels of many myocardial proteins, some of which are 290 linked to contractile function or systolic wall thickness 31 . Proteomics analysis in this study showed 39 hyper 291 expressed proteins, 21 of these being strongly correlated with early changes in regional ECSR. The western 292 blotting analysis showed that D-3PGDH and sFRP1 are significantly expressed within the infarcted 293 myocardium. This finding might indicate their involvement in the early changes in ECSR as well as in 294 determining LVR post-MI. D-3PGDH is the key enzyme for the L-Serine biosynthesis pathway that branches 295 from glycolysis. It participates in a metabolic network interlinking folate and methionine cycles to support cell 296 proliferation and an amplification of function has been associated with a pro-oncogenic role 32 . sFRP1 acts as an 297 inhibitor of the Wnt signalling pathway by binding to Wnt proteins and preventing their association with 298 Frizzled receptors (34). Interestingly, sFRP1 protein has been associated with reduced scar size, improved 299 cardiac function and decreased neutrophil infiltration in a mice model of coronary ligation, indicating a 300 protective role of this protein via reduction of post-MI inflammation 34 . This anti-inflammatory role has been 12 suggested by others in rodents but not in pigs. sFRP1 to suppress the Wnt pathway has potential clinical 302 translation for novel therapies aiming to reduce scar size post-MI and warrants further investigation. 303 There are limitations to this study. The animals did not have atherosclerotic disease, which might have 304 determined a different proteomic profile. However, the MI size and other CMRI measures were in keeping with 305 what observed in humans. In addition, the animals gained a substantial amount of weight over the study period 306 with a possible confounding effect on scar size and proteomics. However, it has been suggested that the use of 307 CMRI parameters indexed to the weight of the animal can minimise this effect. Finally, a relatively small 308 number of animals (n=10) was used, with strain analyses and proteomics undertaken on sub-groups: non-309 parametric statistical tests were used to compensate. Another limitation is related to the lack of information on 310 the dynamic changes that occur after MI: due to the nature of our study design we were unable to characterise 311 the dynamic proteomic processes as previously described by other authors 31 . 312 In conclusion, this study reveals novel associations between MI, early changes in regional ACS, prediction of      were not consulted to develop patient relevant outcomes or interpret the results. Patients were not invited to contribute 445 to the writing or editing of this document for readability or accuracy.

Figure 1: Depiction of workflow from imaging to derived final strains. 449
Short-axis CMRIs with endocardial (green) and epicardial (blue) borders were traced, stacked and grouped into three regions: base, 450 equator/mid, and apex. Strain was calculated for all slices, and the mean was then found by averaging strains in their regions. Finally, 451 the resultant regional and global circumferential strains were found for each time point.    Figure 1 Depiction of work ow from imaging to derived nal strains. Short-axis CMRIs with endocardial (green) and epicardial (blue) borders were traced, stacked and grouped into three regions: base, equator/mid, and apex. Strain was calculated for all slices, and the mean was then found by averaging strains in their regions. Finally, the resultant regional and global circumferential strains were found for each time point.   Volcano plot representation of proteomics Abundance ratios for changes in each protein are shown as log10 of p-value of infarcted/health segments within the same hearts (n=5) Figure 6 Quanti cation of D-3PGDH and sFRP1 proteins by western blotting All data presented as Mean ± SEM; n=5 in each group. Quanti cation of D-3-phosphoglycerate dehydrogenase (D-3PGDH) and secreted frizzled-related protein 1 (sFRP1) in lysates of the infarcted myocardium (I) and non-infarcted myocardium (N). A=Representative western blot of D-3PGDH; B=Densitometric quanti cation of D-3PGDH; C=Representative western blot for sFRP1; D=Densitometric quanti cation of sFRP1 expression; E=Representative western blot for anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH, used for control of protein loading); * P<0.05 vs. non-ischaemic myocardium. MW: molecular weight. Statistical Test used: Mann-Whitney. ( Figure S1 shows the full blots for all the proteins shown in this gure)

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