Cardiovascular Response to Beta-Adrenergic Blockade or Activation in 23 Inbred Mouse Strains

We report the characterisation of 27 cardiovascular-related traits in 23 inbred mouse strains. Mice were phenotyped either in response to chronic administration of a single dose of the β-adrenergic receptor blocker atenolol or under a low and a high dose of the β-agonist isoproterenol and compared to baseline condition. The robustness of our data is supported by high trait heritabilities (typically H2>0.7) and significant correlations of trait values measured in baseline condition with independent multistrain datasets of the Mouse Phenome Database. We then focused on the drug-, dose-, and strain-specific responses to β-stimulation and β-blockade of a selection of traits including heart rate, systolic blood pressure, cardiac weight indices, ECG parameters and body weight. Because of the wealth of data accumulated, we applied integrative analyses such as comprehensive bi-clustering to investigate the structure of the response across the different phenotypes, strains and experimental conditions. Information extracted from these analyses is discussed in terms of novelty and biological implications. For example, we observe that traits related to ventricular weight in most strains respond only to the high dose of isoproterenol, while heart rate and atrial weight are already affected by the low dose. Finally, we observe little concordance between strain similarity based on the phenotypes and genotypic relatedness computed from genomic SNP profiles. This indicates that cardiovascular phenotypes are unlikely to segregate according to global phylogeny, but rather be governed by smaller, local differences in the genetic architecture of the various strains.

Jaxpheno2 study correlates poorly with CV-PGX data. Yet, since this study was also discordant with other MPD records, including those of Deschepper, we believe that HW and BW, and in particular its ratio (i.e. VWI) are nevertheless likely to be well reproducible phenotypes.
Results were more divergent for cardiac atria, as average AWI of most strains was lower in the CV-PGX study than in Deschepper's (Pearson r 2 =0.21, Spearman ρ 2 =0.36, p=0.07; Figure 3). This is most likely due to the small size of the atria and may be a result of subtle differences in the dissection process rather than intrinsic strain variability.

Systolic blood pressure and heart rate
In conscious ctr mice, average SBP varied from 86±8 mmHg in strain BTBRT + tf/J to 119±6 mmHg in strain SWR/J, whereas average HR extended from 442±35 beats/min in strain Balb/cByJ to 713±44 beats/min in strain SJL/J (Figure 2). Both variables were approximately uniformly distributed across strains but the patterns of distribution were not correlated (i.e. correlation r=0.25, p=0.26; Suppl. Figure 6A). HR means of closely related strains (i.e. strains C3H/HeJ vs C3H/HeOuJ and Balb/cJ vs Balb/cByJ) were very similar, while SBP values were slightly more divergent. Uni-modal distribution of baseline HR suggests that strains with high pulse have higher baseline sympathetic and lower parasympathetic activity than strains with low pulse. This is consistent with the idea that HR is an indicator of the strain-specific homeostatic state of autonomous nervous system activity [3].
The MPD contains four, respectively five multistrain datasets of mean SBP and HR measured by tailcuff, and two sets of mean HR determined by ECG monitoring of conscious animals. The sensitivity of both traits to non-genetic perturbations is reflected by the relatively low correlations of these records, even though the number of strains included in the comparisons is relatively small (Suppl. Figure 3).
Thus, only one of the nine possible pair-wise comparisons between distinct SBP datasets (i.e. Deschepper1 vs Svenson1, 7 strains in common, Pearson's r 2 =0.76, p<0.05; Suppl. Figure 3) and two of the thirteen possible combinations between HR records are significantly correlated (i.e. Jaxwest1 pulse and Hampton1, 7 strains in common, Pearson's r 2 =0.81, p<0.01; Sugiyama1 and Hampton1, 13 strains in common, r 2 =0.37, p<0.05). CV-PGX pressure and pulse values were significantly correlated with the data of Sugiyama1 for SBP (14 strains in common with the CV-PGX dataset, r 2 =0.63, p<0.01 [4]) and with those of Jaxwest1 and Hampton1 projects for HR (Jaxwest1: 7 strains in common with the CV-PGX dataset, r 2 =0.74, p<0.05; Hampton1: 16 strains in common with the CV-PGX dataset, r 2 =0.40, p<0.01 [5]), while the other (CV-PGX:MPD) pairs were more discordant (Suppl. Figure 3). The numbers of positively correlated (CV-PGX:MPD) datasets corroborate those of the (MPD:MPD) comparisons. Altogether, they are higher than expected by chance (i.e. 2/14 have a p-value<0.05 for SBP, and 4/19 have p<0.05 for HR), suggesting that despite marked susceptibility to the environment, both blood pressure and heart rate are also controlled in part by genetic determinants.

Details on phenotypes measured under atenolol or isoproterenol treatment
Heart rate under anaesthesia The opposite action of the drugs on pulse rate was clearly detected under anaesthesia (see HR (ECG) panel in Figures 2 and 4), but means, amplitudes and significance of responses were generally lower than in conscious mice. Thus, maximal and minimal values were attained at 618±25 beats/min in strain C57BL/6J treated with iso10 and at 297±27 beats/min in strain C57BLKS/J infused with ate. The chronotropic effect of iso10 reached significance (p<0.05) in nine strains only, the positive dose-effect of iso was observed in a single strain (SM/J), and the blocking effect of ate was close to negligible in sixteen strains (Figure 4). In agreement with results obtained in baseline condition, correlations between pulse measured in conscious animals or in anaesthetised mice were weak, irrespective of the drug treatment (Suppl. Figures 3 and 6). This latter information further emphasises the strain-specific confounding action of anaesthesia on pacing responses.

ECG intervals under anaesthesia
The overall modifications induced by both drugs on PR, QRS, and QT intervals of ECG waves followed expected trends in the majority of the strains, even though these measurements suffered from the confounding effects of anaesthesia. Treatment with iso10 induced a significant decrease of PR in ten strains, of QRS in four strains, and of QT in four strains (Figures 2 and 4). The same trends applied to the majority of the other strains without reaching nominal significance (p<0.05), whereas the opposite effect of ate was milder. Interestingly, the chronotropic action of iso was often not distributed equivalently across all ECG intervals. In strain KK/HlJ for instance, QRS interval was dosedependently reduced under β-stimulation. This decrease was matched by a slight increase of ST (data not shown), whereas HR, PR, and QT were barely affected. In contrast, QRS shortening was paralleled by increased HR and decreased PR and/or QT intervals in strains Balb/cByJ, C3H/HeJ and to a lesser extent FVB/NJ. Similarly, the reduction of QT intervals under β-stimulation was matched by a higher pulse rate in seven strains, whereas HR was hardly affected in SJL/J mice. Also, the increase of HR under iso was not systematically paralleled by significantly shorter QT. In A/J mice, QT was even slightly increased, whereas the most stringent positive chronotropy affected PR intervals.
Altogether, these interval-specific patterns argue in favour of differential and strain-specific genetic modulators of cardiac electrical activity.

Body weight
As mentioned in the main text body, stimulation with iso10 was paralleled by an elevated gain of body mass in strain C58/J when compared to untreated individuals. Thus, ctr mice gained 0.5±0.27 g over the two weeks of the experiment (i.e. a relative increase of 2.2% compared to BWS), whereas animals exposed to iso10 increased their weight by 2.9±0.42 g (+13.6%). The same trend was significant in nine additional strains placed under iso10 stimulation (p<0.05), whereas little change was seen in the others (Figures 2 and 4). The maximum relative difference of BWG between ctr and iso10-treated individuals reached 14.5% of BWS in strain 129S1/SvImJ (i.e. ctr: -2.4%; iso10: +12.1%). These data are consistent with observations made previously in CD-1 mice, in which chronic administration of 30 mg/kg iso per day for 13 days not only led to increased HW, but also to higher BWG (i.e. relative increases of 22% in iso-treated vs 8% in ctr mice, when compared to BWS [6]. Further experimental evidence relating elevated BWG with augmented food consumption, muscle mass, and brown fat, as observed in CD-1 mice, is currently lacking. The range of extra body weight gained under iso10 was strain-specific (Figures 2 and 4). Concomitantly, we observed slightly weaker correlations of (CV-PGX iso10 :MPD) than (CV-PGX ctr :MPD) pair-wise comparisons (Suppl. Figure 3).
The impact of β-blockade on body mass was negligible in most strains. Strain A/J however behaved in a peculiar manner as these mice tended to loose weight under all treatments, while BW of ctr animals (i.e. 6.9% of BWS). It is unlikely that these losses were caused by an undetected diseased-state as both treated and untreated animals were housed together. Moreover a similar trend, although statistically not significant, was recently reported in 12-15 wk-old A/J males challenged with five consecutive daily doses of 100 mg/kg iso [7,8].

Correcting for multiple testing
The p-values reported in Figure 4 and Suppl. Figure 4 are obtained from the Wilcoxon ranksum test.
They reflect the probability of the observed data for a particular phenotype p and strain s, given that the distributions for two different treatments (ctr vs ate, iso1, iso10, or iso1 vs iso10) have the same median value (the null hypothesis). Small p-values indicate strong evidence for rejecting the null hypothesis, i.e. suggesting that those phenotypes have different medians. We reshuffled 10'000 times the data amongst each two treatment groups and computed p-values from these permutations (corresponding to the null hypothesis) as controls. The fraction of controls with a ranksum statistic larger than the one observed for non-permuted data gives an estimate of the p-value that is independent of the phenotypic distribution. We find that the ranksum p-values (as reported in Figure   4) are well-calibrated with respect to the permutation p-values (in contrast to those from t-tests, see

Suppl. Figure 11).
When judging the significance of a statistical test within the context of many tests, the significance threshold has to be adjusted. In the well known Bonferroni correction the nominal significance threshold α=0.05 is lowered to α'=α/N (N being the number of tests). For example, focusing on one phenotype but looking simultaneously at N=23 strains gives α'≈0.002. Similarly, focusing on one strain but considering all N=27 phenotypic response variables also gives α'≈0.002.
The Bonferroni correction procedure is very conservative because it controls the family-wise error rate, keeping low the number of false positives (FP), but also true positives (TP). A more appropriate alternative is to control the false discovery rate (FDR), i.e. FP/(FP+TP). We used the Benjamini-Hochberg (BH) step-up procedure to control the FDR [9], but -to account for the dependencies of the tests -obtained the null distribution of p-values from 10'000 random permutations of the phenotypes as was done in [10]. BH-corrected p-values for each phenotype (considered over all strains) are shown in Suppl. Figure 12. Conversely, BH-corrected p-values for each strain (considered over all phenotypes) are shown in Suppl. Figure 13. Moreover, in Suppl. Figure 14 we indicate globally adjusted significance values, correcting for both multiple strains and phenotypes. In Suppl. Figure 15 we plotted BH-corrected p-values against the uncorrected p-values. This figure highlights the increase in power if one is willing to control the FDR, rather than the family-wise error rate. Indeed most BHcorrections are significantly milder than those obtained by the Bonferroni correction (which would result in all points being on the black lines). This is also evidenced in Suppl. Figure 16 and Suppl. Figure 17 which compare the number of significant observations with BH-and Bonferroni-corrections, respectively, against those without any corrections. Indeed, while more than half of the nominally significant observations (without any corrections) involving an iso1 and/or iso10-treated group of animals remain significant with BH-correction, less than 5% would survive the Bonferroni-correction.
For ate vs ctr (which gave only 107 nominally significant observations to start with), a single observation (HR of C57BLKS) passes the Bonferroni-correction, while almost a third remains significant after BH-correction.

z-scores and correlations within drug conditions
For each phenotype p, strain s and treatment, a z-score was calculated as: where x ps is the mean of phenotype p under a given treatment (t=ctr, ate, iso1, iso10) in strain s, and μ p and σ p are the mean and the standard deviation, respectively, of the average measurement for phenotype p across all strains. To address the structure of the z-scores, we built four matrices (one per drug condition), each containing one row for each strain s and one column for each phenotype p.
Hierarchical bi-clustering was applied to determine the patterns of strain and phenotype similarities in the matrices. Phenotypes were clustered according to absolute similarity while strains were clustered according to signed similarity. The branches of the dendrograms illustrating the clusters were plotted with the same colour as long as the average linkage distance was less than 20% of the maximal distance.

Supplementary Figure legends
Supplementary Figure 1 Phenotypic strain means. Bar graphs representing the full dataset of phenotypic strain means and standard deviations across the four drug conditions. White bars: ctr; blue bars: ate; orange bars: iso1; red bars: iso10. Strains are ranked by increasing HR (TC) means of ctr mice. See Table 1 for abbreviations. HR std: standard deviation of HR (TC) strain means; SBP std: standard deviation of SBP strain means.

Supplementary Figure 2
Heritabilities. Heritabilities (H 2 ) were calculated separately for each phenotype and each treatment over all strains and are presented in a colour-code using the colour scale shown on the right. Every column represents a separate treatment, while every row denotes a phenotype. Phenotypes are highly heritable within the set of CV-PGX mouse strains, with H 2 ranging from 0.53 to 0.98.

Supplementary Figure 3
Comparison of CV-PGX and MPD strain means across fourteen phenotypes. Each matrix summarises the data of all pair-wise Pearson's comparisons between and across CV-PGX and MPD projects for a single phenotype. Pearson r 2 values are indicated in the top triangle of the matrix, the significance of the correlations is shown by a colour scale, and the number of strains considered for each pair-wise comparison is indicated in the bottom triangle of the matrix. Underlined r 2 values denote negative correlations. Abbreviations are given in Table 1 for phenotypes and in the boxed legend for multistrain MPD datasets. Detailed experimental protocols and links to original publications of the MPD projects are available at http://www.jax.org/phenome [1]. Of note, Hampton1 and Jaxwest1 HR means were recorded by ECG in conscious mice, whereas all other HR means, including Jaxwest1 pulse, were obtained by a tail-cuff system. SBP was measured by tail-cuff in all projects. In complement to Figure 3, this figure also shows the comparisons of CV-PGX means recorded in ateand iso-treated mice with CV-PGX and MPD data measured in untreated animals.

Supplementary Figure 4
Intra-strain significance of drug treatments. Data are presented for the phenotypes shown in Suppl. Figure 1 as in Figure 4 (main text). Intra-strain p-values of phenotypic data recorded in treated vs untreated groups (Wilcoxon ranksum statistics) are presented as bar graphs on a -log 10 scale. The threshold of significance is indicated by red lines (p=0.05) and -log p-values are signed according to the directionality of the effect induced by the drugs. When significant, p-values for testing responses under iso10 vs iso1 are indicated by coloured stars (i.e. *: p<0.05; **: p<0.01; ***: p<0.001; red star: phenotypic mean under iso10 > phenotypic mean under iso1; blue star: phenotypic mean under iso10 < phenotypic mean under iso1). Blue bars: ate vs ctr; orange bars: iso1 vs ctr; red bars: iso10 vs ctr.
P-values smaller than 0.001 (bars extending beyond the green lines or ***) hold up against Bonferroni correction for multiple testing of either all strains for a given phenotype or all phenotypes for a given strain. Strains are ranked as in Figure 2 and Suppl. Figure 1. See Table 1 for abbreviations. Correcting for testing multiple strains. C. Correcting for testing multiple phenotypes. The ratio between the number of significant observations with Bonferroni-corrected p-values over those with uncorrected p-values was used to set the background colour of each field (see colour bar).                         Suppl. Figure     Suppl. Figure 6D        Suppl. Figure   Suppl. Figure 7D VW     Suppl. Figure 9B  Suppl. Figure 15 33 Suppl. Figure 17