*Article* **Plasma Interleukin-10 and Cholesterol Levels May Inform about Interdependences between Fitness and Fatness in Healthy Individuals**

**Francesco Sartor 1,2, \*, Jonathan P. Moore <sup>2</sup> and Hans-Peter Kubis 2**

<sup>1</sup> Department of Patient Care and Monitoring, Philips Research, 5656 AE Eindhoven, The Netherlands

<sup>2</sup> College of Human Sciences, Bangor University, Bangor LL57 2EF, UK; j.p.moore@bangor.ac.uk (J.P.M.); h.kubis@bangor.ac.uk (H.-P.K.)

**\*** Correspondence: francesco.sartor@philips.com; Tel.: +31-(0)-615-509-627

**Abstract:** Relationships between demographic, anthropometric, inflammatory, lipid and glucose tolerance markers in connection with the fat but fit paradigm were investigated by supervised and unsupervised learning. Data from 81 apparently healthy participants (87% females) were used to generate four classes of fatness and fitness. Principal Component Analysis (PCA) revealed that the principal component was preponderantly composed of glucose tolerance parameters. IL-10 and high-density lipoprotein, low-density lipoprotein (LDL), and total cholesterol, along with body mass index (BMI), were the most important features according to Random Forest based recursive feature elimination. Decision Tree classification showed that these play a key role into assigning each individual in one of the four classes, with 70% accuracy, and acceptable classification agreement, κ = 0.54. However, the best classifier with 88% accuracy and κ = 0.79 was the Naïve Bayes. LDL and BMI partially mediated the relationship between fitness and fatness. Although unsupervised learning showed that the glucose tolerance cluster explains the highest quote of the variance, supervised learning revealed that the importance of IL-10, cholesterol levels and BMI was greater than the glucose tolerance PCA cluster. These results suggest that fitness and fatness may be interconnected by anti-inflammatory responses and cholesterol levels. Randomized controlled trials are needed to confirm these preliminary outcomes.

**Keywords:** VO2max; anti-inflammatory; machine learning; PCA

#### **1. Introduction**

In the 1950s, first observational evidence emerged showing that physically active individuals had a lower risk of cardiovascular disease (CVD) [1]. This evidence was later corroborated by the protective effect found for cardiorespiratory fitness (CRF), as shown in the Aerobics Center Longitudinal Study in 1989 [2,3]. Since then, several reviews, systematic reviews, and meta-analysis have confirmed and highlighted the protective role of CRF regardless the level of fatness [4–7]. According to the "fat but fit paradox", people who have a high level of CRF may be better protected from the risk of CVD than leaner people who have low CRF [8]. However, only a small proportion of US citizens can be considered "fat and fit", and obesity is independently associated with low CRF, simply because obese people are generally less active [9].

Lahoz-Garcia et al. [10] showed an interesting partial mediation of CRF between diet and obesity in schoolchildren, meaning that higher CRF contributes, for the same diet, to a lower fat mass (FM). Consistently, others have found that moderate to vigorous physical activity levels, thus higher CRF, were independently associated with a lower atherogenic index of plasma, namely blood fat strongly related with CVD, regardless of diet; and that central adiposity mediated, in other words explains, the relationship between moderate to vigorous physical activity levels and atherogenic index of plasma [11]. This would rule

**Citation:** Sartor, F.; Moore, J.P.; Kubis, H.-P. Plasma Interleukin-10 and Cholesterol Levels May Inform about Interdependences between Fitness and Fatness in Healthy Individuals. *Int. J. Environ. Res. Public Health* **2021**, *18*, 1800. https://doi.org/10.3390/ ijerph18041800

Academic Editor: Paul B. Tchounwou

Received: 21 January 2021 Accepted: 7 February 2021 Published: 12 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

in favor of the protective role of higher CRF against CVD risk. Moreover, poor CRF has been associated with glucose intolerance [12] and a higher risk of insulin resistance in apparently healthy individuals [13]. Furthermore, it has been hypothesized that low CRF could provide an early sign of insulin resistance [14].

Obesity has been shown to be associated with low level systemic inflammation in connection with increased adipose tissue mass [15,16]. In turn there is evidence, in animal studies, of the possible role of inflammation on over-nutrition [17]. However, physical activity may counteract over-nutrition behavior at the hypothalamic level by means of anti-inflammatory signaling mediated interleukin-10 (IL-10) [17]. An anti-inflammatory role of IL-10 has been found also in rat skeletal muscle tissue [18]. In humans it was found consistently that intensive cycling is able to increase, 1 hour after the exercise, gene expression of several interleukins including IL-10, but not IL-6 [19]. High intensity exercise showed an acute, 30 minutes, IL-10 and IL-6 increase in overweight-obese inactive individuals, but this increase was not elicited by moderate intensity exercise [20]. Nevertheless, two weeks of high intensity exercise in overweight-obese unfit individuals did not show a chronic increase in IL-10 nor in IL-6 [21,22]. Rather, a chronic elevation of IL-10 found in obese women was reduced by 12 weeks of lifestyle intervention, including 30 minutes of exercise a day, only in those obese women who did not have metabolic syndrome [23]. Furthermore, higher serum concentration of IL-10 was found in older adults with a higher volume of physical activity [24]. Additionally, animal models show a possible protective role of anti-inflammatory signaling on cardiac function (i.e., left ventricular end-diastolic pressure) [25], a finding supported in human studies involving coronary heart disease patients, obese and diabetic individuals [26,27].

To further investigate the relationship between cardiovascular fitness and body composition characteristics i.e., fatness, we used a database, which combined demographic, blood lipids, insulin resistance, and inflammatory variables in association with CRF and FM% values. Our approach was to create a categorical variable composed of four classes, based on CRF and FM% levels. The four classes or categories are termed High Fatness with High Fitness (HFHF), High Fatness with Low Fitness (HFLF), Low Fatness with High Fitness (LFHF) and finally Low Fatness with Low Fitness (LFLF). The cutoff levels between categories were identified according to the literature [28,29]. We have applied a data driven approach consisting of four steps. First is an unsupervised learning phase, where the variables are clustered using Principal Component Analysis (PCA) [30]. PCA allows clustering of the variables into principal components. Second, a supervised learning phase was deployed to use those clusters in the feature importance selection. We opted for feeding the PCA components as well as the other variables into the feature importance selection algorithm because, although PCA combines uncorrelated variables with one another in such a way that each principal component will maximize variance, this does not mean that the components per se will be the most important classification features. Therefore, as a second step, we have used the same categorical four classes' dependent variable for a random forest based feature importance selection. In detail, we have used the Boruta algorithm, which is an improvement of the Random Forest feature selection model, also known as recursive feature elimination [31,32]. The Boruta algorithm adds randomness to the importance evaluation algorithm, so that the certainty about the importance of a given variable is increased. In short, a randomized copy of the variables is made at each iteration of the random forest importance computation. Thus, if a variable has a higher importance than the maximal importance of all randomized attributes it is retained. If there is some uncertainty, or if a variable has a lower importance it is rejected or discarded [32].

Third, a decision tree was used in order to define the discriminating path to the four classes of fitness and fatness. This classification model was used to visualize which independent variables would best split the data points into the four classes. However, classification was not limited to the decision tree. Another four classification models were used as well with the intent of testing which classification model would maximize the use of the selected independent variables, or features. The four alternative machine learning

classification models were Multiple Logistic Regression, Decision Tree, Naïve Bayes, and K-nearest neighbors. This step was necessary to test whether the features selected would effectively classify the data points. Finally, a fourth step, a mediation and moderation analysis [33] was conducted in order to investigate whether attenuation between CRF and FM% would occur when one of the variables extracted was used as covariate. We hypothesized that we would find attenuations, as previously shown in the literature [10,11], by means of variables linked to fat metabolism. The overall aim of this study was to use a data driven approach, employing machine-learning techniques, to generate new insights connecting fitness and fatness with demographic, blood lipids, insulin resistance, and inflammatory variables.

#### **2. Materials and Methods**

#### *2.1. Study Design and Participants*

The data analyzed in this study originated from two separate data collections conducted at Bangor University. Data from 81 apparently healthy participants (10 males and 71 females) were included in the analysis. All participants were informed about the study protocols and objectives, and provided written consent prior to the start of the studies. Study protocols were approved by the Ethics Committee of the School of Sports, Exercise and Health Sciences Department of Bangor University in conformity with the Declaration of Helsinki. The design of this study was purely observational.

#### *2.2. Body Composition, Fat Mass Percentage, Blood Markers and Cardiorespiratory Fitness Assessment*

Participants were pre-screened for cardiovascular diseases by means of the American Heart Association/American College of Sports Medicine Pre-Participation Questionnaire [34]. However, participants with elevated fasting levels of glucose, insulin and lipids were not per se excluded from this study. Body composition, fasting blood lipid profile and CRF (VO2max) were determined using standardized protocols described previously [21]. A cardiorespiratory fitness test was executed on a cycle ergometer (Corival 400, Lode, Groningen, The Netherlands), the protocol consisted of an incremental exercise test to exhaustion (1min at 50 + 20 W increments per minute). Oxygen uptake was measured breath by breath by means of a metabolic card (ZAN 600 CPET, Oberthulba, Germany). Fasting blood lipid profile (total Cholesterol, LDL and HDL), plasma insulin, plasma glucose, leptin and cytokines (IL-6, IL-10, and TNF-α) collection and analysis is also described in Sartor et al. [21]. Plasma glucose was analyzed by immobilized enzymatic assay (YSI 2300 STAT, Incorporated Life Sciences, Yellow Springs, OH, USA). Lipid profile was analyzed from plasma samples by optic enzymatic assay (Reflotron®, Roche Diagnostics, Mannheim, Germany). Plasma insulin was analyzed by ELISA (ultrasensitive human insulin ELISA kit, Mercodia, Uppsala, Sweden). Cytokines (IL-10, IL-6 and TNF-α) and adipokines were also analyzed from fasting plasma samples by ELISA (Bender MedSystems GmbH, Austria and BioVendor, Laboratoní medicína, Czech Republic, respectively). Insulin sensitivity and β-cell function were estimated using fasting plasma insulin and glucose by means of the Homeostatic model assessment 2 (HOMA2) [35].

#### *2.3. Classification Criteria*

Four classes were extracted from the database described above; a Higher-Fatness with Higher-Fitness (HFHF) group, a Higher-Fatness with Lower-Fitness (HFLF) group, a Lower-Fatness with Higher-Fitness (LFHF) group, and finally a Lower-Fatness with Lower-Fitness (LFLF) group. The grouping criteria were taken from Gallagher et al. [28] for fatness, and the American College of Sports Medicine guidelines [29] for fitness. The criteria are represented in Table 1.


**Table 1.** Classification criteria for body fat percentage and relative VO2max (mL/kg/min), age and sex.

#### *2.4. Data Analytics*

#### 2.4.1. Preprocessing

The full dataset collected at Bangor University premises was loaded into RStudio (Version 1.2.5033, 2009–2019 RStudio Inc., Boston, MA, USA). This initial dataset included 25 independent variables. A first missing data filter was applied and all variables with more than 70% missing data were discarded. After this step, 19 independent variables were retained. Two variables were converted into factorial variables, the classification variable as explained in Table 1 and the variable Sex. The retained variables were visualized to reveal imbalance. This visualization showed an imbalance towards females, as they represented 87% of our dataset. The imbalance was a consequence of the original research question of one data collection being confined to females. A zero- and near zero-variance predictors analysis was conducted, by means of nearZeroVar function (caret R package), to eliminate any independent variables that would not add anything in explaining variance (Table 2). However, no variables were rejected based on these criteria [36]. The preProcess function (caret R package) was used to center and scale the variables and missing data, were imputed using the bagImpute function which uses the bootstrap aggregating method [37]. Outliers were detected as values outside boxplot notches, using boxplot function (graphics R package). The notches were set as the median, plus or minus the standard error [38]. The detected outliers were excluded from the analysis.


**Table 2.** Zero- and near zero-variance predictors analysis.

BMI = Body Mass Index, Chol = Fasting Total Cholesterol, HDL = Fasting High Density Lipoprotein, LDL = Fasting Low Density Lipoprotein, TG = Fasting TriGlycerides, Fgluc = Fasting Glucose, BetacellF = β cell Function, InsSens = Insulin Sensitivity, InsRes = Insulin Resistance, TNFalpha = Tumor Necrosis Factor α, IL-6 = Interleukin-6, IL-10 = Interleukin-6, RER = Respiratory Exchange Ratio.

#### 2.4.2. Principal Component Analysis and Feature Selection

Once the data were pre-processed a principal component analysis was conducted to find what combination of variables would explain the variability of the data. The function PCA (FactoMineR R package) as described in [39] was used. Eigenvalues, which represent the amount of the variation explained by each principal component, were extracted by fviz\_eig. The number of retained components was set so that 70% of the total variance is explained. Correlation plots of all variables were produced using the corrplot function (corrplot R package). The importance of the twenty variables including five new Principal Components was evaluated by a recursive feature elimination technique based on the Boruta Random Forest method (Boruta R package) [32]. The Boruta function compares original importance attributes against importance achievable by shadow random variables, in iterations until convergence. The principal components were also included in the feature selection step, to test whether the most variation corresponded with the highest importance.

#### 2.4.3. Decision Tree

A decision tree was built using the nine variables selected by the Boruta algorithm, with the exclusion of the PCA dimensions. As first step, the class imbalance was compensated by means of weights for simple random sample (i.e., 1/probability). The decision tree was constructed using the rpart function (rpart R package) and vitalized by rpart.plot (rpart.plot R package). Tree depth was set as the smallest tree within one standard error of the minimum cross validation error [40].

#### 2.4.4. Classification Models

Multiple logistic regression, decision tree, naïve Bayes, and κ-nearest neighbors classification model were trained on our dataset by means of the train function (caret R package) as described in Kuhn [36]. The classes were the four subgroups (HFHF, HFLF, LFHF, LFLF) described above. In order to perform the multinomial logistic regression, the multinom method was selected within the train function. In order to evaluate the performance of each single classifier, accuracy tables and confusion matrices were generated, using the confusionMatrix in caret and visualized thanks to ggplot (ggplot2 R package) [36].

#### 2.4.5. Mediation and Moderation Analysis

Mediation analysis was conducted by means of the mediation R package [41]. Before analyzing, the mediation and moderation raw data for each variable were assessed for normality and linearity by means of quantile-quantile plots (qqnorm function, from the basic stats R package), centered, and scaled when required, as described earlier. Linear regressions models, via the lm function (stats R package), were built between the mediator and the independent variable (relative VO2max), and between the dependent variable (Fat Mass percentage) and the independent variable-mediator combined. The mediate function simulated the comparison between these two linear regressions, showing if the mediation would add a significant contribution in relating the independent and dependent variables. The mediation analysis resulted in the Average Causal Mediation Effects (ACME), the Average Direct Effects (ADE), and the combined effects (Total Effect), and the proportion mediated (Prop. Mediated). Moderation was executed by the gylma and stargazer R packages. A linear model was built between the dependent variable and independent variable plus the moderator, and between the dependent variable and the moderator plus the product.

#### *2.5. Statistical Analysis*

The descriptive statistics, means and standard deviations of all participants for the 15 included variables and for each of the four subgroups were analysed using the arsenal R package [42]. Data for the four subgroups were split using the filter function supported by the dplyr R package. One-way ANOVAs were performed to compare the four sub-groups and they were followed-up when appropriate both by the tableby function (arsenal R package). Significance level was set at 0.05.

#### **3. Results**

#### *3.1. Subgrouping and Difference Analysis*

As described in the method section, four subgroups were derived according to participants' CRF, body FM%, age, and sex. The subgroups sizes are not evenly distributed. Two subgroups HFHF and LFHF are rather small (*N* = 9, *N* = 6, respectively). In line with our intention to form four groups of different fatness and fitness levels, the ANOVA and follow-up showed significant differences between the two higher-fitness and lower-fitness levels. Moreover, the HFHF group and the LFHF groups also showed a significant difference in fitness, the lower in fatness being fitter (40.1 ± 2.9 mL/kg/min) than the higher in fatness (34.3 ± 4.3 mL/kg/min). As for the higher fatness/lower fatness split, this was fully achieved, as confirmed by the ANOVA and follow-ups (Table 3). As to be expected, BMI was significantly higher in the HFLF group compared with the LFHF and LFLF subgroups. There was a trend towards a higher BMI for the HFLF group when compared with the HFHF group, and a trend towards a higher BMI in the HFHF group compared with the LFHF group. It is to be noted that BMI does not fully reflect FM% (Table 3). Total fasting plasma Cholesterol levels showed significantly higher levels in the HFLF compared with the HFHF and LFHF groups. There was a strong trend towards a higher cholesterol level in the LFLF group compared with the HFHF group. The LFHF group showed higher HDL than the HFHF group. The LFLF group had a higher HDL level than the HFLF group. Moreover, there were two strong trends for a higher HDL in the LFLF group and the LFHF group versus the HFHF and the HFLF groups, respectively. LDL was higher in the HFLF group compared with the HFHF, LFHF, and LFLF groups. Finally, fasting plasma insulin was higher in the HFLF compared with the HFHF. Interestingly two, LFHF and LFLF, groups showed higher insulin values than the HFHF group (Table 3).


**Table 3.** Descriptive Statistics of Database, difference analysis and follow-up analyses.


HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower-Fitness group, VO2max = maximal oxygen uptake, BMI = Body Mass Index, HDL = Fasting High Density Lipoprotein, LDL = Fasting Low Density Lipoprotein, TNFalpha = Tumor Necrosis Factor α, IL-6 = Interleukin-6, IL-10 = Interleukin-6. Significant p-levels are highlighted in bold.

#### *3.2. Principal Component Analysis* the HFLF group compared with the HFHF, LFHF, and LFLF groups. Finally, fasting

the LFHF group versus the HFHF and the HFLF groups, respectively. LDL was higher in

*Int. J. Environ. Res. Public Health* **2021**, *18*, x 7 of 19

The independent variables, once filtered for missing data, were clustered by means of principal component analysis. Five principal component dimensions were found that explained 70% of the variance (Figure 1). Dimension 1 was dominated by glucose tolerance features, dimension 2 by Leptin and Sex, dimension 3 was constituted by lipid profile, dimension 4 by triglycerides and glucose, and, finally, dimension 5 by BMI and weight. (Figure 1). In Figure 2 the classification and the weight of the single individuals is shown when the first two components are put in relation. plasma insulin was higher in the HFLF compared with the HFHF. Interestingly two, LFHF and LFLF, groups showed higher insulin values than the HFHF group (Table 3). *3.2. Principal Component Analysis*  The independent variables, once filtered for missing data, were clustered by means of principal component analysis. Five principal component dimensions were found that

These five dimensions were further included in the feature selection process. Recursive feature elimination based on random forest showed that the stronger features in describing the four groups were IL-10, BMI, total cholesterol, HDL, LDL, dimension 1, beta cell function, dimension 4, IL-6, Age, dimension 3, and weight. In Figure 2 the interrelationship of the first two PCA components is shown and the four groups are clustered. Fitter groups tend to develop along dimension 1 while the less fit along dimension 2. explained 70% of the variance (Figure 1). Dimension 1 was dominated by glucose tolerance features, dimension 2 by Leptin and Sex, dimension 3 was constituted by lipid profile, dimension 4 by triglycerides and glucose, and, finally, dimension 5 by BMI and weight. (Figure 1). In Figure 2 the classification and the weight of the single individuals is shown when the first two components are put in relation.

**Figure 1.** Output of the principal component analysis: BMI = Body Mass Index, Chol = Fasting Total Cholesterol, HDL = Fasting High Density Lipoprotein, LDL = Fasting Low Density Lipoprotein, TG = Fasting TriGlycerides, Fgluc = Fasting Glucose, BetacellF = β cell Function, InsSens = Insulin Sensitivity, InsRes = Insulin Resistance, TNFalpha = Tumor Necrosis Factor α, IL-6 = Interleukin-6, IL-**Figure 1.** Output of the principal component analysis: BMI = Body Mass Index, Chol = Fasting Total Cholesterol, HDL = Fasting High Density Lipoprotein, LDL = Fasting Low Density Lipoprotein, TG = Fasting TriGlycerides, Fgluc = Fasting Glucose, BetacellF = β cell Function, InsSens = Insulin Sensitivity, InsRes = Insulin Resistance, TNFalpha = Tumor Necrosis Factor α, IL-6 = Interleukin-6, IL-10 = Interleukin-6, RER = Respiratory Exchange Ratio, Sex.0 = females, Sex.1 = males.

These five dimensions were further included in the feature selection process. Recursive feature elimination based on random forest showed that the stronger features in describing the four groups were IL-10, BMI, total cholesterol, HDL, LDL, dimension 1, beta cell function, dimension 4, IL-6, Age, dimension 3, and weight. In Figure 2 the interrela-

10 = Interleukin-6, RER = Respiratory Exchange Ratio, Sex.0 = females, Sex.1 = males.

*Int. J. Environ. Res. Public Health* **2021**, *18*, x 10 of 19

**Figure 2.** Clustering of the categorical variable, including the four fatness and fitness permutations. Relationship between the first principal component and the second principal component computed by PCA. The size of the icons for the single individuals shows their weight in classification. HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower - Fitness group. **Figure 2.** Clustering of the categorical variable, including the four fatness and fitness permutations. Relationship between the first principal component and the second principal component computed by PCA. The size of the icons for the single individuals shows their weight in classification. HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower -Fitness group.

#### *3.3. Classification Models*  The Random Forest based recursive feature elimination Boruta algorithm found *3.3. Classification Models*

twelve variables as certainly important in classifying the four fatness and fitness classes (Figure 3). Amongst these twelve are PCA dimensions 1,4 and 3, in order of importance. While the algorithm is uncertain about dimension 5 and discards dimension 2. IL-10, BMI, and cholesterol levels are clearly the most important variables. In Figure S1 the first 10 selected variables are shown as boxplot. Additionally, in Figure S1 linear correlations between variables are displayed, showing how the retained variables still carry most of the correlations. The Random Forest based recursive feature elimination Boruta algorithm found twelve variables as certainly important in classifying the four fatness and fitness classes (Figure 3). Amongst these twelve are PCA dimensions 1,4 and 3, in order of importance. While the algorithm is uncertain about dimension 5 and discards dimension 2. IL-10, BMI, and cholesterol levels are clearly the most important variables. In Figure S1 the first 10 selected variables are shown as boxplot. Additionally, in Figure S1 linear correlations between variables are displayed, showing how the retained variables still carry most of the correlations. *Int. J. Environ. Res. Public Health* **2021**, *18*, x 11 of 19

**Figure 3.** Recursive feature elimination; in green are depicted the variables that are certain. IL6. = Interleukin-6, IL-10 = Interleukin-6, RER = Respiratory Exchange Ratio, Sex.0 = females, Sex.1 = males, Dim.1 = Dimension 1 of the PCA, Dim.2 = Dimension 2 of the PCA Dim.3 = Dimension 3 of the PCA, Dim.4 = Dimension 4 of the PCA, Dim.5 = Dimension 5 of the PCA. When the twelve variables, including the PCA dimensions, selected by the Boruta **Figure 3.** Recursive feature elimination; in green are depicted the variables that are certain. IL6. = Interleukin-6, IL-10 = Interleukin-6, RER = Respiratory Exchange Ratio, Sex.0 = females, Sex.1 = males, Dim.1 = Dimension 1 of the PCA, Dim.2 = Dimension 2 of the PCA Dim.3 = Dimension 3 of the PCA, Dim.4 = Dimension 4 of the PCA, Dim.5 = Dimension 5 of the PCA.

other classification models.

classification performances. In fact, the Multiple Logistic Regression model showed a classification accuracy of 0.77 (95% CI: 0.6717, 0.8627), significantly higher than the No Information Rate (0.4691), and a κ-coefficient of 0.65, Figure 4. The Decision Tree model, displayed in Figure 5, although having the lowest accuracy (0.70, 95% CI: 0.5919, 0.8001) amongst the models generated here, still had an accuracy significantly higher than its No Information Rate (0.432), and an acceptable κ-coefficient (0.54) (Figure 4). The Naïve Bayes classifier showed the highest accuracy (0.88, 95% CI: 0.7847, 0.9392), significantly higher than the No Information Rate (0.58), and a moderate κ-coefficient equal to 0.79(Figure 4). Finally, the K-Nearest Neighbors classifier had an accuracy of 0.73 (95% CI: 0.6181, 0.8213), which was, however, not higher than the No Information Rate (0.76), with a rather weak agreement, a κ-coefficient of 0.47 (Figure 4). Overall, the latter performed worse than the

When the twelve variables, including the PCA dimensions, selected by the Boruta importance algorithm were used to generate the classification model, we found acceptable classification performances. In fact, the Multiple Logistic Regression model showed a classification accuracy of 0.77 (95% CI: 0.6717, 0.8627), significantly higher than the No Information Rate (0.4691), and a κ-coefficient of 0.65, Figure 4. The Decision Tree model, displayed in Figure 5, although having the lowest accuracy (0.70, 95% CI: 0.5919, 0.8001) amongst the models generated here, still had an accuracy significantly higher than its No Information Rate (0.432), and an acceptable κ-coefficient (0.54) (Figure 4). The Naïve Bayes classifier showed the highest accuracy (0.88, 95% CI: 0.7847, 0.9392), significantly higher than the No Information Rate (0.58), and a moderate κ-coefficient equal to 0.79 (Figure 4). Finally, the K-Nearest Neighbors classifier had an accuracy of 0.73 (95% CI: 0.6181, 0.8213), which was, however, not higher than the No Information Rate (0.76), with a rather weak agreement, a κ-coefficient of 0.47 (Figure 4). Overall, the latter performed worse than the other classification models. *Int. J. Environ. Res. Public Health* **2021**, *18*, x 12 of 19

**Figure 4.** Confusion Matrices, and accuracy of the four classification models. ACC = accuracy, HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower -Fitness group. **Figure 4.** Confusion Matrices, and accuracy of the four classification models. ACC = accuracy, HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower -Fitness group.

= Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group,

*Int. J. Environ. Res. Public Health* **2021**, *18*, x 13 of 19

**Figure 5.** Decision tree, where HDL = High Density Lipoprotein, LDL = Low Density Lipoprotein, IL-10 = Interleukin-10 and BMI = Body Mass Index are expressed in their original dimensions (mmol/L, mmol/L, pg/mL, respectively). HFHF = Higher-Fatness with Higher-Fitness group, HFLF **Figure 5.** Decision tree, where HDL = High Density Lipoprotein, LDL = Low Density Lipoprotein, IL-10 = Interleukin-10 and BMI = Body Mass Index are expressed in their original dimensions (mmol/L, mmol/L, pg/mL, respectively). HFHF = Higher-Fatness with Higher-Fitness group, HFLF = Higher-Fatness with Lower-Fitness group, LFHF = Lower-Fatness with Higher-Fitness group, LFLF = Lower-Fatness with Lower -Fitness group.

#### LFLF = Lower-Fatness with Lower -Fitness group. *3.4. Mediation and Moderation Analysis*

*3.4. Mediation and Moderation Analysis*  All selected variables were analyzed for mediation and moderation. As shown by the quantile-quantile plots in Figure 6, LDL and BMI did not require further scaling and/or centering and were the only two variables to show a significant partial mediation effect between CRF and FM% (Figure 7). Details of the causal mediation analysis are captured All selected variables were analyzed for mediation and moderation. As shown by the quantile-quantile plots in Figure 6, LDL and BMI did not require further scaling and/or centering and were the only two variables to show a significant partial mediation effect between CRF and FM% (Figure 7). Details of the causal mediation analysis are captured in Table 4. *Int. J. Environ. Res. Public Health* **2021**, *18*, x 14 of 19

**Figure 6.** quantile–quantile plots of the variables that showed partial mediation. **Figure 6.** Quantile–quantile plots of the variables that showed partial mediation.

**Figure 7.** Decomposed Mediation Analysis plot: ACME = Average Causal Mediation Effect, ADE =

Average Direct Effect, LDL = Low Density Lipoprotein, BMI = Body Mass Index.

**Figure 6.** quantile–quantile plots of the variables that showed partial mediation.

*Int. J. Environ. Res. Public Health* **2021**, *18*, x 14 of 19

**Figure 7.** Decomposed Mediation Analysis plot: ACME = Average Causal Mediation Effect, ADE = Average Direct Effect, LDL = Low Density Lipoprotein, BMI = Body Mass Index. **Figure 7.** Decomposed Mediation Analysis plot: ACME = Average Causal Mediation Effect, ADE = Average Direct Effect, LDL = Low Density Lipoprotein, BMI = Body Mass Index.

**Table 4.** Causal Mediation Analysis, Quasi-Bayesian Confidence Intervals.


LDL = Low Density Lipoprotein, BMI = Body Mass Index, ACME = Average Causal Mediation Effect, ADE = Average Direct Effect, Prop. Mediated = Proportion of the effect Mediated. Significant values: \*\*\* <0.001, \* <0.05. *N* = 81, Simulations: 1000.

#### **4. Discussion**

This present study embraces artificial intelligence as a tool to provide new insight into the fat but fit paradox [8]. Using unsupervised and supervised machine learning approaches to interrogate existing physiological data, this work indicates connection between markers of dyslipidemia, inflammation and cardiorespiratory fitness that reveal possible functional interaction of physiological systems underpinning the "fat but fit paradox".

#### *4.1. Descriptive Statistics in Relation to Fatness and Fitness*

We have created four classes, or groups, in line with population normative cut-off values [28,29]. Consistently, these groups differed significantly from one another in terms of fitness and fatness (Table 3). Fasting total cholesterol levels and LDL were significantly higher in the HFLF group, while HDL was higher in the groups with lower fatness. The decision tree depicted in Figure 5 shows how well HDL and LDL alone could differen-

tiate the HFHF group from the other groups. Although IL-10 did not show significant differences between the four groups, whereas IL-6 did, IL-10 seemed to be involved in the differentiation of individuals with lower fitness level in function of their fatness (Figure 5). Moreover, the Analysis of Variance amongst the four groups also showed differences in fasting insulin levels. Fasting insulin was the highest in the HFLF group and the lowest in the HFHF group (Table 3). This is of particular interest because it seemed to be associated with fitness rather than with fatness levels. Fitness has been shown to play an important role in protecting against glucose intolerance [43]. This may be related to the well-known effect of muscle contractile activity, hence exercise training, on insulin sensitivity [44].

#### *4.2. Machine Learning*

Principal component analysis clustered the various markers available in this study so that they could better explain the variance of the fatness and fitness categorical variable. This resulted in PCA Dimension 1, mainly composed of glucose tolerance indicators, such as fasting insulin, insulin sensitivity, and insulin resistance, as well as beta cell function derived from the HOMA2 model (Figure 1). However, supervised learning, namely the random Forest based feature selection algorithm, revealed that the importance of IL-10, cholesterol levels (i.e., HDL, LDL and total Cholesterol) along with BMI in classifying the four classes was greater than that of the above mentioned glucose tolerance PCA cluster. The interesting aspect of our approach is that our analysis clearly points towards dominant features, namely IL-10, LDL, HDL, BMI, for categorizing our four groups, in competition with other features, which are just as well known to be influenced by fatness and fitness. Besides the potential exercise dependent link between IL-10 and insulin/leptin sensitivity in the hypothalamus in animal studies [17], exercise was found to increase IL-10 levels in overweight-obese human subjects [20]. An interlink between fatness and IL-10, however, was found in obese subject after weight loss, revealing higher IL-10 levels [45]. Therefore, distinct features of our data could point towards an important discriminating function of IL-10 and LDL/BMI for fatness and fitness classification and could be linked to these findings. Moreover, exercise has been found to effect LDL as well as HDL levels [46].

#### *4.3. Partial Mediation*

Fatness and fitness are significantly inversely related [47]. This was confirmed by our data. In addition to this, however, we found that CRF is indirectly related to FM% through the mediation of LDL and BMI. Previous literature found that BMI could mediate CRF and cardio-metabolic risk in schoolchildren [48]. Another investigation in schoolchildren using a large dataset showed that CRF may have a beneficial effect on lipid profile, insulin metabolism and inflammation independent of fatness [49]. Our results seem to lead in the same direction. Specific effects of exercise training, of a high enough intensity, to promote aerobic capacity improvements have been linked to a decrease in concentration of atherogenic ox-LDL [46,50]. In addition, upregulation of fatty acid metabolism and transport through exercise dependent signaling pathways (particularly peroxisome proliferator-activated receptor) [51,52] and concurrent alterations in lipid profiles [53,54] are well described. Interestingly, IL-10 was found to be linked with LDL level as IL-10 was shown to induce uptake of LDL by fluid-phase endocytoses in macrophages leading to lowered LDL plasma levels [55].

#### *4.4. Implications*

We are aware that our study is retrospective. Thus, it provides a limited level of evidence. It is beyond the purpose of this study to accept or not the hypothesis that fitness plays a protective role in people with higher level of fatness. Yet the discriminating role that anti-inflammatory and cholesterol levels seem to make sense when addressing fatness and fitness may point in the direction of "healthy obesity" when the CRF level is high [4].

#### *4.5. Limitations*

The current study is based on data from 81 individuals. Several variables, such as systolic and diastolic Blood Pressure and C-Reactive Protein, had to be excluded from the analysis because of missing data. The observations and conclusions drawn from this study would need to be verified in a larger dataset. This study does not provide direct experimental evidence, but is merely observational and retrospective. These considerations need to be taken into account when evaluating our results and conclusions. Our dataset has more females than males, and although sex did not appear to play a key role in determining the classification, we cannot exclude that, with a higher number of males this factorial variable would or could have had a greater weight. Finally, by dividing our dataset into four classes we observed that these were not evenly distributed. This issue was partially mitigated by balancing, using class weights.

#### **5. Conclusions**

Our data analytics approach has shown a potential key role of IL-10 as well as HDL, LDL, total Cholesterol and BMI in the classification of people according to their fatness and fitness levels. Unsupervised learning showed that a cluster of glucose tolerance related variables explains the highest quote of the variance of the categorical variable. However, supervised learning did not select this PCA cluster. Mediation analysis showed that LDL and BMI partially explain the association between fitness and fatness. These results suggest that CRF and FM% may be interconnected by anti-inflammatory responses and cholesterol blood levels. This may be in line with the protective role of cardiorespiratory fitness suggested in recent years. However, large randomized controlled trials are needed to validate this hypothesis experimentally and conclusively.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/1660-460 1/18/4/1800/s1, Figure S1: Boxplot of selected variables, and correlation matrices before and after the selection.

**Author Contributions:** Conceptualization, H.-P.K., J.P.M. and F.S.; methodology, F.S.; software, F.S.; formal analysis, F.S.; investigation, H.-P.K., J.P.M. and F.S.; resources, H.-P.K.; data curation, H.- P.K.; writing—original draft preparation, F.S.; writing—review and editing, H.-P.K., J.P.M. and F.S.; visualization, F.S.; supervision, H.-P.K.; project administration, H.-P.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Schools of Sport Health and Exercise Sciences Research Ethics Committee of Bangor University (protocol code M15-15/16 and approval date 16/05/2016).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to restrictions related to the data protection regulations.

**Acknowledgments:** The authors would like to thank Themistoklis Marinos for helping with collection of data.

**Conflicts of Interest:** F.S. works for Royal Philips. All other authors do not have conflict to disclose.

#### **References**


### *Article* **Effects of a HIIT Protocol on Cardiovascular Risk Factors in a Type 1 Diabetes Mellitus Population**

**Jesús Alarcón-Gómez 1 , Joaquín Calatayud 2 , Iván Chulvi-Medrano 3, \* and Fernando Martín-Rivera 4**


**Abstract:** Cardiovascular complications are important causes of morbidity and mortality of Type 1 Diabetes Mellitus (T1DM) people. Regular exercise is strongly recommended to these patients due to its preventive action against this type of disease. However, a large percentage of patients with T1DM people present a sedentary behavior, mainly, because of the fear of a post-exercise hypoglycemia event and lack of time. High-intensity interval training (HIIT) is an efficient and safe methodology since it prevents hypoglycemia and does not require much time, which are the main barriers for this population to doing exercise and increasing physical conditioning. Nineteen sedentary adults (37 ± 6.5 years) with T1DM were randomly assigned to 6 weeks of either HIIT, 12 bouts first 2 weeks, 16 bouts in weeks 3 and 4, and 20 bouts in the last two weeks x 30-s intervals interspersed with 1-min rest periods, performed thrice weekly or to control group, which did not train. VO2max, body composition, heart rate variability (HRV), and fasting glucose were measured as cardiovascular risk factors. We suggest that the 6-week HIIT program used in the present study is safe since no severe hypoglycemia was reported and is an effective strategy in improving VO2max, body composition, HRV, and fasting glucose, which are important cardiovascular risk factors in T1DM people.

**Keywords:** type 1 diabetes; high-intensity interval training; exercise

#### **1. Introduction**

Type 1 Diabetes Mellitus (T1DM) is a chronic metabolic disease characterized by the insufficient production of endogen insulin caused by autoimmune β-cell destruction [1]. According to the International Diabetes Federation (IDF) and World Health Organization (WHO), in the world, 25–45 million adults (>20 years old) suffer from T1DM. In reference to children, adolescents, and young adults (0–20 years old), more than a million live with this pathology, with 130.000 new diagnosed cases per year. It was estimated that the number of people with T1DM in the world will increase a 25% by 2030 [2,3].

Microvascular (e.g., retinopathy, neuropathy, and nephropathy) and macrovascular complications (e.g., coronary arterial disease, peripheral artery disease, stroke, and heart failure) are important causes of morbidity and mortality from this disease [4]. In fact, people with T1DM are at a two-fold to eight-fold increased risk of cardiovascular disease, being this, the first cause of premature death in this population [5,6]. Therefore, brush over effective strategies for the prevention of cardiovascular comorbidities must be a primary concern in T1DM patients and health providers.

Regular exercise is strongly recommended for people living with type 1 diabetes to prevent cardiovascular episodes [7]. Nonetheless, the majority (>60%) of this population

**Citation:** Alarcón-Gómez, J.; Calatayud, J.; Chulvi-Medrano, I.; Martín-Rivera, F. Effects of a HIIT Protocol on Cardiovascular Risk Factors in a Type 1 Diabetes Mellitus Population. *Int. J. Environ. Res. Public Health* **2021**, *18*, 1262. https://doi.org/10.3390/ ijerph18031262

Academic Editors: Pedro R. Olivares, Amelia Guadalupe Grau and Jorge Pérez-Gómez Received: 16 November 2020 Accepted: 27 January 2021 Published: 31 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

does not achieve the current guidelines of exercise proposed by the American College of Sports Medicine (ACSM) and the American Diabetes Association (ADA) [8,9], which indicate at least 150 min of moderate to vigorous aerobic exercise per week and resistance training in 2–3 non-consecutive sessions, performing 1–3 sets of 10–15 repetitions, typically 50–75% of 1 repetition maximum (1RM) on 8–10 multijoint exercises [10].

The most recurrent pretexts that T1DM people state for not exercising are the lack of time, the fear of a hypoglycemia event, and loss of glycemic control due to inadequate knowledge about exercise variables management [11]. Those reasons make that few people with T1DM benefit from the improvement of aerobic capacity (VO2max), glucose regulation, body composition, endothelial function, blood lipid profile, and cardiac autonomic nervous regulation that physical exercise promotes and which are risk factors to develop cardiovascular disease [1,12–14].

The aforementioned barriers that T1DM people face may be overcome with highintensity interval training (HIIT), a training method that, despite being used since the early 20th century in sport performance, has been discovered to be an interesting tool for those with cardiometabolic diseases in the recent years [15]. HIIT involves repeated brief bouts of high intensity (>85% VO2max) intermitted by passive or active recovery periods, requiring lower exercise duration than moderate-intensity continuous training (MICT), also HIIT prevents the drop of glycemia typical of MICT, due to its anaerobic metabolism predominance [4]. There is also evidence to suggest that HIIT elicits at least the same cardiometabolic effects in healthy and pathologic population that MICT does [16,17]. These safe, effective, and time-efficient results are sufficient to consider HIIT as a beneficial form of training for the T1DM population.

There is little previous literature analyzing the effects of HIIT in T1DM. The trend in previous data shows a long-term benefit on cardiorespiratory fitness and glycemic control. However, the underlying mechanisms are not entirely clear, and positive HRV regulation and improved body composition may be the mechanisms causing this benefit.

Firstly, cardiac autonomic regulation can be monitored by Heart Rate Variability (HRV) which is the fluctuation in the time intervals between adjacent heartbeats [18]. HRV provides indirect insight into cardiovascular autonomic nervous system tone, corresponding to the balance between sympathetic and parasympathetic influences on the sinoatrial node [19]. A reduced HRV, which is related to a sympathetic modulation predominance, is associated with an increased risk of cardiovascular problems including sudden cardiac death [20,21]. The large fluctuations in blood glucose levels associated with T1DM tend to place this population at an autonomic control dysfunction and in a reduced HRV in comparison with their healthy counterparts [22]. Given that physical exercise positively affects cardiac autonomic function in people with type 2 diabetes [23] and HIIT has been shown as a promising strategy to improve HRV in healthy individuals and patients with metabolic syndrome [24], we aimed to examine the effect of a HIIT protocol on HRV of a population with T1DM. Secondly, body composition is a traditional cardiovascular risk factor, and in the same way, obesity and overweight are dramatic increased in T1DM people, in fact, almost 50% of patients with T1DM are either overweight or obese [1]. Insulin therapies and unhealthy lifestyles are the main mechanisms of weight gain in this pathological population [25]. Living with obesity or overweight has been widely linked to an enhanced risk of a cardiovascular accident [26], so preventing this situation is a key point in T1DM patients' healthcare. HIIT has shown interesting effects on body composition in healthy individuals [27] and disappointing results in obese and overweight people with T1DM [28,29], so it is important to expand the analysis of this topic.

Aerobic fitness and glucose regulation are the most studied cardiovascular risk factors in T1DM people after a period of HIIT [4,30–32] since this population shows reduced levels of VO2má<sup>x</sup> and irregular glucose control, but given there are only a few studies with these objectives, more researches are needed.

Therefore, the aim of this study was to investigate the long-term effects of 6-week highintensity interval training on HRV, body composition, fasting glucose, and aerobic fitness in a T1DM population since these variables are closely related to cardiovascular health and HIIT could be proposed as the solution to the problems that prevent this population from exercising.

#### **2. Materials and Methods**

#### *2.1. Participants*

We recruited 19 inactive adults (10 males and 9 females), clinically diagnosed as T1DM by one of the following criteria: HbA1c ≥ 6.5 % (≥48 mmol/mol), random plasma glucose ≥ 200 mg/dL (≥11.1 mmol/L), fasting plasma glucose ≥ 126 mg/dL (≥7.0 mmol/dL) or oral glucose tolerance test OGTT), 2-h glucose in venous plasma ≥ 200 mg/dL (≥11.1 mmol/L) [33]. The Valencian Diabetes Association (VDA) and social media announcement were the main recruitment methods. The following inclusion criteria were adopted: (1) aged 18–45 years, (2) duration of T1DM > 4 years, (3) HbA1C < 10% (4) no structured exercise training programs in the previous 6 months, (5) no diagnosed cardiovascular diseases. Subjects excluded from the study include those who smoke regularly, take any medication that affects heart rate, and those who had major surgery planned. Testing took place in the laboratory of the research group in prevention and health in exercise and sport at the University of Valencia. Participants were informed of the purposes, procedures, and risks involved in the study before giving their informed written consent to participate. The study procedures were in accordance with the principles of the Declaration of Helsinki and were approved by the local Institutional Review Board from the local university (H1421157445503).

#### *2.2. Experimental Design*

This is a randomized experimental, parallel design, open-label trial. The eligible subjects were randomly allocated by the researchers (www.randomizer.org) to the experimental (N = 11, 38 ± 5.5 years, 5 men and 6 women, height 1.68 ± 0.09 m, body mass 70.5 ± 7.4 kg and 20.5 ± 8.4 years diagnosed) or control group (N = 8, 35 ± 8.2 years, 4 men and 4 women, height 1.69 ± 0.07 m, body mass 72.05 ± 5.0 kg and 21.1 ± 6.5 years diagnosed), and stratified/classified by gender to ensure a balanced number of men and women in each group. They all were instructed not to modify their nutritional habits and not to perform any regular exercise program outside of the study, which was not supervised.

#### *2.3. Procedures*

Initially, participants attended the lab (7.00–9.00 am) after an overnight fast (>10 h) for the first assessment session. Subjects were instructed not to drink any caffeine or alcohol-contained products in the 24–48 h prior to the measurement to avoid any influence on HRV and body composition outcomes. The hour of the day that each subject completed each test was recorded, as well as the menstrual phase of each female participant with the aim of repeating the same conditions in the second measurement to block their influence in the results [34]. The same test protocols were performed exactly in the same way after the experimental period by both control and HIIT groups.

#### 2.3.1. Body Composition

An 8-electrodes bioelectrical impedance analysis scale (Tanita MC780MA, Tanita Corporation of America, Inc., Illionis, United States) with software GMON version 3.1.6. was conducted to measure body composition, specifically, fat mass (FM) and free fat mass (FFM). The participants wore light clothing and assumed a standing posture on their bare feet, then wait for the results printed from the device in accordance with the manufactures' instructions.

#### 2.3.2. Heart Rate Variability

Analysis of resting HRV measurement was conducted in a quiet dimly lit room with a controlled temperature (22 ± 1 ◦C). Emptying the urinary bladder was asked to the subjects before the beginning of the test. A Polar H10 heart rate (HR) sensor with the Polar Pro strap (Polar Electro, Kempele, Finland), previously moistened to increase the adherence to the skin, was worn around the participant's chest. HR monitor and a tablet with the validated Elite HRV App (Elite HRV LLC®, Asheville, North Caroline, USA) were connected via Bluetooth® [35].

In a supine position and following a 5-min rest, which was used to stabilize the signal without registering it, the HRV readiness began. During 5 min, the participant was instructed to relax and breathe at a natural rate while heart R-R intervals were recorded. Linear parameters in the time domain such SDNN (standard deviation of NN intervals), RMSSD (root mean square of successive RR interval differences and pNN50 (percentage of successive RR intervals that differ by more than 50 ms) and in the frequency domain: HF power and LF power, absolute power of the high and low-frequency band (0.15–0.4 Hz) and LF/HF power ratio (ratio of LF to HF power) were registered, but in order to analyze parasympathetic modulation changes and given the measurement characteristics, RMSSD and HF/LF ratio were the variables examined statistically [20]. HRV short-term recording characteristics were based in the Task Force of the European Society of Cardiology [36,37] used in previous researches with similar objectives [38,39].

#### 2.3.3. VO2max and Peak Power Output

Seven days later to the initial assessment, all the participants performed an incremental test on a cycle ergometer (Excite Unity 3.0, Technogym S.p.A, Cesena, Italia) to determine peak power output (PPO) and peak oxygen consumption (VO2peak) using a gas collection system (PNOE, Athens, Greece) that was calibrated in each test by means of ambient air [40]. Before starting the test, capillary blood glucose concentrations were checked by their own blood glucose monitoring devices. They were told to arrive at the institutional gym with a glycemic level > 100 mg/dL and less than 250 mg/dL in absence of ketones. If the glycemia was correct, the participant began the test normally. If glycemic values were low (<100 mg/dL), intake of 15–30g of fast-acting carbohydrates (CHO) was medically compulsory. Whenever patients presented hyperglycemic status (>250 mg/dL) without ketonuria (determined with test strip), a medically prescribed small corrective insulin dose was self-administered. In the presence of ketones, the exercise would be delayed. Glycemia was checked again until the level of blood glucose was optimum to start the test. In the same way, it was recommended that patients not exercise at the peak of insulin action [30].

The test consisted of a warm-up of 5 min at 40 Watts (W). After that, the workload was increased by 20 W every minute until volitional exhaustion. All the participants were verbally encouraged to give their maximum effort during the exercise. The test ends with a cooldown of 5 min at 40 W. Heart rate was continuously monitored by a Polar H10 (Polar Electro, Kempele, Finland). VO2peak was taken as the highest mean achieved within the last 15 s prior to exhaustion. Peak power output was registered to individualize the workloads in the experimental period training.

#### 2.3.4. Fasting Glucose

Before the intervention period, every morning during 28 days, the participants registered their fasting glycemia level by means of a blood sample obtained by a finger-stick. Subjects were asked to eat their normal diet the evening before each test. They were also instructed to conduct the blood glucose measurement at the same time, in the same position and with the same device, which must be one of the following: FreeStyle (FreeStyle Libre system; Abbott Diabetes Care, Alameda, CA) or Accu-Chek (Accu-chek glucometer, Roche, USA). These glucose monitors are approved by the Diabetes Technology Society and were accessible to the volunteers [41]. An online shared Excel (Microsoft Excel 2013®, Microsoft Corporation, Redmond, United States) file was used to facilitate data recording. The post-intervention analysis was conducted exactly in the same way and lasted exactly the same time as in the pre-intervention measurement (28 days).

#### *2.4. Training Protocol*

Training started the following week after the completion of the pre-experimental procedures. Participants of the experimental group trained three times per week for 6 weeks under the supervision of a researcher on a cycle ergometer (Excite Unity 3.0, Technogym S.p.A, Cesena, Italy). Heart rate while exercising was monitored with a Polar H10 (Polar Electro, Kempele, Finland) that was preconfigured with their heart rate zones. The HIIT protocol performed was a type 1:2, which means that the high-intensity intervals lasted half the time that the rest intervals did. The saddle height was always adjusted to the height of the subject's iliac crest. The training began with a 5-min warm-up at 50 W. Then, they performed repeated 30-s bouts of high-intensity cycling at a workload selected to elicit 85% of their individual PPO interspersed with 1 min of recovery at 40% PPO. The number of high-intensity intervals increased from twelve reps in weeks 1 and 2, to sixteen reps in weeks 3 and 4, to twenty reps in weeks 5 and 6. Training ended with a 5-min cooldown performed at 50 W. After the session, participants were told to check their glycemia level frequently and notify the investigators if a glycemia drop below 70 mg/dL occurred during the 24 h following the exercise [1].

All sessions were supervised by the investigators and in order to reflect real-world conditions, researchers did not advise about decreasing fast-acting insulin dosage or increasing carbohydrate consumption prior to each exercise session. Glycemic values were checked ten minutes before starting the training, and when values were out of range, researchers warned the patients and then they corrected the situation following their medically prescribed guidelines as such has been explained in pre-test procedures. Subjects assigned to the control group were asked to maintain their current lifestyle and dietary intakes during the study period.

#### *2.5. Statistical Analysis*

All variables were expressed as a mean and standard deviation (M ± SD) and were analyzed using a statistical package (SPSS Inc., Chicago, Illinois, USA). Normality assumption by Shapiro–Wilks was identified for each variable. A mixed factorial ANOVA (2 × 2) was performed to assess the influence of "*condition*" (i.e., control group vs. experimental group) and "*time moment*" variable (i.e., pre-intervention, post-intervention) over VO2max, HRV, body composition (FM and FFM), and fasting glucose. In the event that Sphericity assumption was not met, freedom degrees were corrected using Greenhouse-Geisser estimation. Post hoc analysis was corrected using Bonferroni adjustment. D Cohen and the associated CI were used to assess the magnitude of mean differences between control vs. experimental conditions. Significant differences were established at *p* < 0.05.

#### **3. Results**

The study randomized 21 patients with T1DM. At the end of the study, 2 participants from the control group had dropped out due to personal reasons not determined and pregnancy, respectively. Only completers were analyzed (Figure 1).

**Figure 1.** Flow diagram of inclusion of patients in the study.

#### *3.1. Adverse Events*

There were three mild hypoglycemia cases (67.9 ± 2.6 mg/dL) of 198 total trainings (1.5%), occurring immediately after exercise which only required a few minutes of rest and carbohydrate ingestion to be solved. No adverse cardiac events, respiratory events, or musculoskeletal injuries were reported in the experimental period. There were no episodes of hyperglycemia, nocturnal hypoglycemia, or episodes of diabetic ketoacidosis.

#### *3.2. Cardiovascular Risk Factor Outcomes*

#### 3.2.1. VO2max

A significant main effect of the interaction condition\*time moment was found (F = 72.18, *p* < 0.01, η<sup>p</sup> <sup>2</sup> = 0.81). The post hoc analysis showed significant changes between pre- and post-conditions in the experimental group (37.1 ± 4.1 vs. 40.4 ± 3.8 mL/min/kg); however, there were no significant changes in the control group (37.0 ± 5.5 vs. 37.2 ± 5.1), results are presented in Table 1.

#### 3.2.2. Body Composition

Body composition was improved in the intervention group after 6 weeks, through positive changes in FFM (F = 43.4, *p* < 0.01, η<sup>p</sup> <sup>2</sup> = 0.72) and FM (F = 60.0, *p* < 0.01, η<sup>p</sup> <sup>2</sup> = 0.78). In contrast, the post hoc analysis showed no significant changes in FFM (57.6 ± 9.8 vs. 57.9 ± 10.0 kg) and FM (15.4 ± 4.5 vs. 15.3 ± 4.6 kg) in the control group, as shown in Table 1.

#### 3.2.3. Heart Rate Variability

LF/HF ratio and rMSSD data obtained by means of short-term *Elite HRV* analysis are listed in Table 1. A significant main effect of the interaction condition\*time moment was found in the LF/HF ratio (F = 6.5, *p* < 0.05, η<sup>p</sup> <sup>2</sup> = 0.28). The post hoc analysis showed significant changes between pre- and post-conditions in the experimental group (37.8 ± 27.9 vs. 44.3 ± 27.7 ms). In contrast, the control group did not suffer any important change between test periods (40.0 ± 15.9 vs. 39.3 ± 16.5 ms). Similarly, LF/HF ratio (F = 10.5, *p* < 0.05, ηp <sup>2</sup> = 0.38) improved significantly. The post hoc analysis showed significant changes between pre- and post-conditions in the experimental group (2.6 ± 1.6 vs. 1.5 ± 0.9 ms<sup>2</sup> ), with no remarkable changes in the control group (2.1 ± 2.0 vs. 1.9 ± 2.2 ms<sup>2</sup> ).

#### 3.2.4. Fasting Glucose

A significant main effect of the interaction condition\*time moment was reported in the fasting glucose (F = 0.77, *p* < 0.05, η<sup>p</sup> <sup>2</sup> = 0.28). The post hoc analysis showed significant variations between pre- and post-conditions in the experimental group (135.8 ± 95.0 vs. 124.5 ± 15.6 mg/dL); nonetheless, there were no significant changes in the control group (131.8 ± 21.1 vs. 135.9 ± 25.0 mg/dL), results are presented in Table 1.

**Table 1.** Variables analysis before high-intensity interval training (HIIT) intervention and following training period in both groups.


Data are presented by mean ± standard deviation, ES: effect size (Cohen's d), \* *p* < 0.05 vs. baseline.

#### **4. Discussion**

The main results of our study indicate that a 6-week HIIT protocol is sufficient to improve HRV and body composition in a previous inactive T1DM population without clinical impairments. Furthermore, our data revealed an increase in VO2max and the long-term reduction in fasting blood glucose.

Noteworthy, only 3 of 198 total trainings resulted in hypoglycemia, and they were mild cases (69.7 ± 2.6 mg/dL), suggesting that HIIT prevents the blood glucose level drop, which is associated with catecholamine releasing and subsequent increase in hepatic glucose production, which offsets the effect of hyperinsulinemia [4,30–32].

#### *4.1. VO2max*

Previous studies have shown the ability of HIIT to improve overall cardiovascular fitness in T1DM individuals [30–32], in line with our results. Nevertheless, there is also a previous study that reports no changes in VO2max in T1DM people after a 12-week HIIT protocol. The methodological differences with those investigations must be taken into account when results are analyzed.

Six weeks of a 1:2 HIIT protocol have increased the VO2max by 8.9% in the present study, which is in line with a previous study reporting a 7% increase [42] after conduction a protocol with a similar time at high intensity. Other studies found a greater improvement in VO2max in T1DM people using HIIT conducted on a cycle ergometer, likely due to the use of higher intensity or volume. For instance, a previous investigation [30] with T1DM (HIIT group: N = 7), using the same 1:1 protocol (e.g., six weeks of three sessions per week increasing from 6 intervals to 10) at 100% VO2max intervals and at 50 W at rest intervals reported a 14% improvement in VO2max. Another study with nine sedentary T1DM volunteers [31] found a 19% improvement in VO2max after a 10-week HIIT protocol (3 sessions per week) performing ten 1-min bouts at 90% maximal heart rate (HRmax) interspersed with 1-min active recovery. Another example of greater improvements (18%) due to greater loading is a study [32] where sedentary T1DM people performed an 8-week HIIT protocol. This training method consisted of 20 min at 50% HRpeak in weeks 1–2, four 1-min intervals at 80% HRmax interspersed with 5-min active recovery at 50% HRpeak in weeks 3–4, and six 1-min intervals at 85% HRpeak with the same rest intervals in the last four weeks. In contrast, Lee and her research group showed no improvements in VO2max after a 12-week HIIT protocol, which consisted of 4 bouts of 4 min at 85–95% HRpeak interspersed with 3-min recovery intervals at 50–70% HRpeak applied in T1DM and overweight people [28], but caution is needed with those results since authors reported inaccuracies that affected to the methodological quality associated with the gas analyzer reliability.

These results have clinical importance given that low VO2max is a strong prognostic marker of cardiovascular disease in healthy [43] and T1DM people [44,45], which is especially important since they are at increased cardiovascular disease risk compared to non-diabetic counterparts [46].

#### *4.2. Body Composition*

Currently, a paucity of literature remains about the effects of HIIT on body composition of T1DM people. Farinha and coworkers reported a 3.3% increase in FFM, which concur with our results (3.4%). However, in this study, FM did not change, unlike the 6.4% obtained in the present study. Moreover, no changes in FM or lean mass were reported when HIIT was analyzed in obese or overweight T1DM people. These differences between studies may be due to different nutritional habits and baseline body composition levels of the participants. The rest of the investigations, previously mentioned, that analyze the long-term effects of HIIT on T1DM people [30,32,47], did not examine body composition changes. Instead, weight and the body mass index were evaluated, which from our point of view, are not valid variables to ensure the correct body composition measurement [48]. The improvement in body composition in T1DM participants after a HIIT protocol is a relevant finding since overweight and obesity are increasing alarmingly between T1DM patients and their prevention is crucial to reduce the cardiovascular risk in the population with this pathology [49]. However, more studies assessing HIIT on body composition measures are warranted before we can confidently prescribe HIIT as an effective training strategy for the prevention of overweight and obesity.

#### *4.3. Heart Rate Variability*

To the best of our knowledge, this is the first study to investigate the impact of HIIT on HRV in T1DM people. We found that participants who trained modified their nervous modulation of cardiac tissue by increasing parasympathetic activation through RMSSD (17.2%) and LF/HF ratio (42.3%), which may reduce cardiovascular risk [20,50], while the

non-exercise control group remained unchanged. Previous studies have inquired about HIIT effects on HRV, but in people with different pathologies or even healthy individuals and with different methodologies, so comparisons must be taken with caution. Evidence from those studies also suggests tendencies of higher HRV median values of RMSSD (9.6–22%) and lower LF/HF (7.05–30%) [24], in line with our results. It is important to stand out that HIIT protocols used in those investigations ranged from 2 to 24 weeks and were performed in cycle ergometer and treadmill. High-intensity intervals lasted from 8 s to 4 min and rest intervals from 12 s to 4 min, being active and passive. Training intensity was measured by different methods such as % of PPO, VO2max, and HRmax and HRV measurement were short- and long-term. Other investigations among sedentary healthy individuals [38] also found a reduction in LF/HF ratio of 8.7%, consistent with our findings, albeit rMSSD did not change, in contrast with our results.

These findings suggest that HIIT, regardless of the protocol type, has beneficial effects on HRV in healthy people and individuals with certain pathologies. Given the congruence with our results in T1DM and the lack of investigations in this cohort, more research is needed to confirm whether HIIT is an effective training strategy to improve HRV in T1DM people to reduce their cardiovascular risk.

#### *4.4. Fasting Glucose*

Our results revealed that fasting glucose was reduced by 7.8% in line with previous research with similar characteristics that reported a 7.5% reduction [31]. In the same way, a 11.2% mean glucose reduction was reported by Lee and coworkers, using continuous glucose monitoring during 14 days in overweight and obese volunteers [28]. Investigations that examined glucose behavior after a period of HIIT training, but in Type 2 Diabetes people and subjects with metabolic syndrome, also showed a median reduction in fasting glucose of 16 mg/dL (similar to 11mg/dL reported in the present study) [51]. In those investigations, different HIIT protocols were used, with 2–48 weeks and 3–5 sessions per week of total volume and high-intensity intervals and rest intervals (active and passive) ranging from 8 s to 5 min. Training intensity was always established above 80% of PPO, HRmax, or VO2max. Fasting glucose analysis was conducted predominantly by blood biochemistry.

Taking the aforementioned data, we may interpret that our results are consistent with previous investigations that applied an exercise intervention on T1DM people and other related pathologies that courses with impairment of the glycemia regulation (e.g., Type 2 Diabetes and metabolic syndrome). Thus, HIIT could be postulated as an effective training strategy to long-term reduce fasting blood glucose in T1DM people and consequently, reduce the cardiovascular risk in this population [52]. Nevertheless, more investigations are warranted to corroborate this hypothesis.

#### *4.5. Limitations*

The results of this study demonstrate cardiovascular risk factors benefits of 6-week HIIT in T1DM subjects. Nevertheless, caution is needed because this study has limitations that must be addressed: the absence of a healthy control group, the relatively small sample size, and the method used for fasting glucose measurement, which was done by the participants in their own home using their own personal glucose analyzer and it was not measured by the researchers or a certified clinical chemistry laboratory under controlled laboratory conditions using a single calibrated glucose analyzer. Moreover, we did not record insulin doses before and after the training period. This would be useful to determine whether insulin sensitivity is modified as reduced insulin exogenous administration is related to a decreased risk of cardiovascular complications in people with T1DM [30,53]. Body composition assessment method (bioimpedance) is a debatable procedure since there is evidence that suggests a decrease in skeletal muscle membrane potential in T1DM people that could disturb the bioimpedance results [54] and other investigations that support its use with this population [55]. Such limitations do not prevent this study from providing novel aspects for the research field and exercise prescription to T1DM.

#### **5. Conclusions**

In conclusion, a 6-week HIIT protocol 1:2 type, performing high-intensity intervals at 85% PPO and active rest intervals at 40% PPO in a cycle ergometer during 3 sessions per week, apart from being accessible and safe since participants were able to complete all the sessions with the intensity required without suffering any severe undesirable episode of hypoglycemia, was sufficient to improve VO2max, HRV, body composition and fasting glucose in a previously sedentary T1DM population. HIIT seems an interesting approach for reducing cardiovascular risk in T1DM individuals.

**Author Contributions:** Conceptualization, J.A.-G. and F.M.-R.; data curation, F.M.-R.; formal analysis, J.A.-G. and F.M.-R.; investigation, J.A.-G.; methodology, J.A.-G., J.C., I.C.-M., and F.M.-R.; supervision, F.M.-R.; writing—original draft, J.A.-G., I.C.-M., and F.M.-R.; writing—review and editing, J.A.-G., J.C., I.C.-M., and F.M.-R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Valencia University (H1421157445503).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Predictors of Athlete's Performance in Ultra-Endurance Mountain Races**

**Pedro Belinchón-deMiguel 1 , Pablo Ruisoto 2 , Beat Knechtle 3,4 , Pantelis T. Nikolaidis 5,6 , Beliña Herrera-Tapias <sup>7</sup> and Vicente Javier Clemente-Suárez 8,9, \***


**Abstract:** Background: In previous studies, ultra-endurance performance has been associated with training and psychological variables. However, performance under extreme conditions is understudied, mainly due to difficulties in making field measures. Aim: The aim of this study was to analyze the role of training, hydration, nutrition, oral health status, and stress-related psychological factors in athletes' performance in ultra-endurance mountain events. Methods: We analyzed the variables of race time and training, hydration state, nutrition, oral health status, and stress-related psychological factors in 448 ultra-endurance mountain race finishers divided into three groups according to race length (less than 45 km, 45–90 km, and greater than 90 km), using a questionnaire. Results: Higher performance in ultra-endurance mountain races was associated with better oral health status and higher accumulative altitude covered per week as well as higher positive accumulative change of altitude per week during training. In longer distance races, experience, a larger volume of training, and better hydration/nutrition prior to the competition were associated with better performance. Conclusions: Ultra-endurance mountain athletes competing in longer races (>90 km) have more experience and follow harder training schedules compared with athletes competing in shorter distances. In longer races, a larger fluid intake before the competition was the single best predictor of performance. For races between 45 and 90 km, training intensity and volume were key predictors of performance, and for races below 45 km, oral health status was a key predictor of performance. Psychological factors previously reported as ultra-endurance mountain race performance predictors were inconsistent or failed to predict the performance of athletes in the present research.

**Keywords:** psychology; odontology; nutrition; training; stress; running

#### **1. Introduction**

Participation in ultra-endurance mountain events has shown an exponential increase in recent years [1]. Specific research in this extreme endurance event has shown an increase in protein catabolism and muscle degradation [2,3], an increased erythropoiesis to compensate exercise-induced hemolysis [4], an accumulation of blood lactate below the anaerobic threshold [5], and an increase in the consumption of fat [6].

**Citation:** Belinchón-deMiguel, P.; Ruisoto, P.; Knechtle, B.; Nikolaidis, P.T.; Herrera-Tapias, B.; Clemente-Suárez, V.J. Predictors of Athlete's Performance in Ultra-Endurance Mountain Races. *Int. J. Environ. Res. Public Health* **2021**, *18*, 956. https://doi.org/10.3390/ ijerph18030956

Academic Editor: Ewan Thomas Received: 25 November 2020 Accepted: 20 January 2021 Published: 22 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Ultra-endurance mountain events are highly stressful, usually performed at 71% of maximum heart rate (HR) [7] and characterized by an intense sympathetic modulation evaluated on the basis of heart rate variability (HRV) [8]. Regarding the training programs, the literature has highlighted the importance of training schedule (speed and volume), the speed in half-marathon and marathon to attain 'top' performance in ultra-endurance races [9].

A higher performance in ultra-endurance events has been associated with two physiological factors, i.e., low body fat index [10,11] and an appropriate hydration and nutritional status [8,11,12]. In this respect, it was found that successful ultra-endurance mountain athletes drank enough to maintain an optimal performance during the race [12]. In addition, an appropriate fuel support is essential for performance, since an extreme negative energy balance was reported in previous ultra-endurance mountain events [8]. However, nutritional and fluid intake during ultra-endurance events presents a risk for oral health. In fact, the absence of oral health care and the presence of caries, dental erosion, and periodontal disease [13–16] are directly related to the extensive use of sports drinks, gels, and energy bars [17,18]. In this line, the negative effect of acid and high-carbohydrate foods aggravated by a decreased saliva flow rate during intense exercise [19,20] can negatively affect health and performance in ultra-endurance mountain athletes.

Previous literature on athletes' performance in these events has also focused on the analysis of ergogenic aids, showing the extensive consumption of nonsteroidal antiinflammatory drugs (NSAIDs) or central nervous system activators such as caffeine [1,21]. In addition, different psychological factors such as higher pain tolerance [22,23] or better stress management skills [24,25] have also been associated with better performance.

The role of multifactorial parameters in ultra-endurance mountain performance is still poorly known or understood. For this reason, we conducted the present research with the aim to analyze the role of training schedule, hydration, nutrition, oral health status, and stress-related psychological factors in athletes' performance in ultra-endurance mountain events. The initial hypothesis was that successful athletes would present a different training schedule, hydration, nutrition, oral health, and stress-related psychological profile compared with non-successful athletes.

#### **2. Materials and Methods**

#### *2.1. Participants*

A total of 448 ultra-endurance mountain race finishers were analyzed. According to the three types of ultra-endurance mountain races, based on race length, that athletes can perform, they were divided into three groups: G1, athletes involved in ultraendurance mountain races below 45 km (*n* = 234; 37.97 ± 7.31 years; 1.7315 ± 0.1341 m; 71.40 ± 10.38 kg); G2, athletes involved in ultra-endurance mountain races between 45 and 90 km (*n* = 79; 39.26 ± 6.92 years; 1.7535 ± 0.0731 m; 73.86 ± 10.61 kg); and G3, athletes involved in ultra-endurance mountain races longer than 90 km (*n* = 135; 41.09 ± 7.26 years; 1.7557 ± 0.0713 m; 71.45 ± 8.01 kg). The three groups did not differ in age, body mass index, or total years of experience as ultra-endurance athletes. The average rate of successful (finisher) and non-successful (non-finisher) athletes in G1 was 74%, in G2 61%, and in G3 40%. Participants were recruited when they presented the documentation attesting their participation in the races. The inclusion criterion was that they were official participants in the races, and the exclusion criterion was that participants reported a medical condition or were taking medication. We obtained written consent from each participant in accordance with the Declaration of Helsinki, and the study was approved by the University Bioethics Committee (CIPI/002/17).

Athletes participated in the following races: Valls d'Aneu Ultratrail, 92 km-long and with 7344 m of accumulated positive altitude change, Emmona Ultratrail, 130 km-long and with 10,194 m of accumulated positive altitude change, Canfranc-Canfranc Ultraendurance Mountain Race, 100 km-long and with 8848 m of accumulated positive altitude change, Great Trail of Peñalara, 114 km-long and with 5100 m of positive altitude change, Gruseg Trail, 51 km-long and with 2516 m of accumulated positive altitude change, and Extremadura Mountain Federation Circuit, including races from 6 km to 72 km and positive altitude changes from 910 m to 6133 m.

#### *2.2. Design and Procedure*

Data in the following self-reported variables were collected for all participants 24 h before the races at the race place. The questionnaire took about 10 min to complete. All participants in the races were invited to answer the questionnaire, and only those who accepted were registered. After the races, the race time was also recorded, as reported in the official classification of each race.

#### 2.2.1. Training, Hydration, and Nutrition Variables

We analyzed variables related to athletes' experience in mountain races (years), total race time in the last marathon (min) and half-marathon (min), training schedule (volume, intensity, and accumulative altitude change), expected race time (min), difference between expected and real time (min), stretching time per week (min), and stretching sessions' duration per week.

Regarding nutrition and hydration habits before the competitions, the following parameters were evaluated: intake of carbohydrate-based meals during the week of competition and fluid intake before the competition day (l). Finally, regarding the oral health status, the frequency of the following occurrences was recorded: grinding or clenching teeth while performing usual tasks, nibbling objects, biting nails, consuming vitamin C pills, having a wet pillow upon awakening, and burping repeatedly.

#### 2.2.2. Psychological Variables

We analyzed perceived stress, measured by the Perceived Stress Scale (PSS-14) [26], and general mental health status, measured by the Acceptance and Action Questionnaire, (AAQ-II) [27].

#### 2.2.3. Performance Race Variables

We analyzed the variables of total time in ultra-endurance mountain races (min), total distance (km), average running speed (km/h), accumulated positive and negative altitude (m), difficulty coefficient (total km × accumulated positive altitude change/1000), and the speed/difficulty coefficient ratio.

#### *2.3. Statistical Analysis*

Data were analyzed using the Statistical Package for the Social Sciences (SPSS) version 21 (SPSS Inc., Chicago, IL, USA). Kolmogorov–Smirnov tests were performed to test normality and homogeneity of each variable. A MANOVA was conducted to analyze differences between the three groups (G1, G2, G3) regarding the dependent variables previously described. Tukey tests were conducted for post hoc comparisons. In addition, the athletes in each group were divided in higher- and lower-performance subgroups by percentile 50 according to their race time, to analyze differences between performances in the variables analyzed. For this, an independent *t* test was used. The significance level was set at *p* < 0.05 in all analysis.

#### **3. Results**

No significant differences were found in any variable between G1 and G2 (Table 1). However, ultra-endurance mountain athletes in G3 recorded more experience, training sessions, hours per week, minutes per session, accumulated positive altitude per session and per week, compared with G1 and G2 athletes. No differences were found in psychological variables between the groups.


**Table 1.** Training and psychological variables in G1–G3 athletes.

Note: Group 1, race length < 45 km; Group 2, race length between 45 km and 90 km; Group 3, race length > 90 km.

Table 2 shows differences in performance predictors for athletes competing in the three ultra-endurance mountain race distances. For G1, better performance was associated with higher speed training per week, but not with differences in psychological or nutritional/hydration status. For G2, higher performance was associated with higher training speed per week and more training sessions per week, but no differences were found for the other variables. For G3, higher performance was associated with higher intake of fluids before the competition, with no differences for the training and psychological variables.

**Table 2.** Predictors of performance for athletes competing in ultra-endurance mountain races, based on race distance.



**Table 2.** *Cont.*

Note: Group 1, race length < 45 km; Group 2, race length between 45 km and 90 km; Group 3, race length > 90 km.

#### **4. Discussion**

The aim of the present study was to analyze the role of training, hydration, nutrition, oral health status, and stress-related psychological factors in athletes' performance in ultra-endurance mountain events. The initial hypothesis was partially confirmed, since differences in training, hydration, nutrition, and oral heath parameters were found for athletes depending on their performance and competition distance, but no differences appeared in stress-related psychological factors.

Previous research found that half-marathon and marathon race times predicted race performance in ultra-endurance mountain races [9]. In the current research, we found that G1 athletes and G2 athletes with a better performance had significantly faster marathon times. Better performance in mountain races was associated with faster half-marathon times, which is in line with previous research [9,28]. Higher performance in ultra-endurance mountain races for distances below 90 km was directly linked to three training-related variables, i.e., training intensity and higher speed in shorter races, as previously reported in the literature [29,30]; training volume and density (sessions per week), since a large training volume may be necessary to obtain physiological adaptations in aerobic and anaerobic threshold intensities—which are the intensity zones involved in ultra-endurance race performance [31]—in line with previous research in ultra-endurance mountain events [1,29], though recent studies propose that ultra-endurance runners would benefit from incorporating high-intensity interval training in a low-volume training program [32–35]; higher positive change of altitude accumulated per week. Previous studies have suggested that

the mechanical work performed on a positive slope may improve musculature, joint resistance, the resistance of ligamentous and tendinous tissues, and the effort tolerance of the locomotor system, which are critical for successful adaptation to the long-term efforts involved in ultra-endurance mountain events [36,37]. Therefore, it would be recommended to introduce more strength training sessions performed with a higher load in future training interventions for ultra-endurance mountain athletes [38].

Higher performance in ultra-endurance mountain races for distances over 90 km was linked to two main variables. Firstly, a higher fluid intake before the race, a result that highlights the importance of hydration in these races, consistent with previous studies that showed that the hydration status has a positive effect on thermoregulation and muscular function, factors crucial in extreme sport competitions [39]. It is important to note that a large fluid ingestion during ultra-endurance races could have a negative effect on performance, since inappropriate hydration can lead to hyper-hydration, and consequently to hyponatremia because of electrolyte imbalance, producing alterations in athletes' cardiovascular response and performance, or to dehydration which also compromises the athletes' health and performance. Secondly, the influence on performance of a larger experience in ultra-endurance mountain races is a factor that has been identified in earlier studies [1]. Interestingly, athletes with higher experience reported a more accurate self-perception of their performance, which indicates that athletes with a lower performance have a poorer knowledge of their own capacities [40].

Regarding the role of psychological variables in predicting athlete's performance, no differences between groups or performance based on perceived stress or general mental health were observed. One possible explanation is that most participants reported very low psychological stress levels and good general mental health, on the basis of the evaluation scale established for the general population (M = 18.51, SD = 7.05) [41]. Therefore, the collected data may have failed to detect effects because of the lack of variability of our sample, which is not representative of the general population. In addition, in the case of ultra-endurance mountain races below 45 km, better oral health parameters were associated with higher performance. This result is in line with previous research conducted in elite athletes [14,16], highlighting the importance of an optimal oral health status for correct physiological functioning, especially in individuals participating in these extreme races.

A limitation of the present study is that we analyzed distances of <45, 45–90, and >90 km; thus, the findings should be generalized with caution to athletes competing in longer distances. Another limitation is the use of a self-reported questionnaire. On the other hand, a strength of this study is that it analyzed a relatively large sample and examined the role of many performance predictors such as training, hydration, nutrition, oral health, and stress-related psychological factors. These findings are expected to have large practical applications for practitioners working with ultra-endurance runners, considering the increased number of annual ultra-endurance races and participants in these races [42].

Finally, we propose that more research should be conducted on the relationship between oral health and performance in athletes competing in ultra-endurance races, as well as on other factors such as type of diet, specific strength and mountain training programs, and mental strategies that could be adopted by high-performance runners, specifically during longer races.

#### **5. Conclusions**

Ultra-endurance mountain athletes competing in longer races (>90 km) have more experience and follow harder training schedules than shorter-distance athletes. In longer races, a larger fluid intake before the competition was the single best predictor of performance. For races between 45 and 90 km, training intensity and volume were key predictors of performance, and for races below 45 km, oral health status was a key predictor of performance. Psychological factors, previously reported as good predictors of ultra-endurance mountain race performance, had little influence in the present study.

**Author Contributions:** P.B.-d.M. and V.J.C.-S. did the data collection. P.B.-d.M. and V.J.C.-S. analyzed the data. P.B.-d.M., P.R., B.K., P.T.N., B.H.-T., and V.J.C.-S. wrote the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the University Bioethics Committee (CIPI/002/17).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** All study data are included in the present manuscript.

**Acknowledgments:** We want to acknowledge the contribution of Gran Trail de Peñalara, Alpinultras Circuit, Gruseg Trail and Extremadura Mountain Federation.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Acute Effects of Different Postactivation Potentiation Protocols on Traditional Rowing Performance**

**Alfonso Penichet-Tomas , Jose M. Jimenez-Olmedo \* , Luis Serra Torregrosa and Basilio Pueo**

Department of General and Specific Didactics, University of Alicante, 03690 Alicante, Spain; alfonso.penichet@ua.es (A.P.-T.); lst15@gcloud.ua.es (L.S.T.); basilio@ua.es (B.P.)

**\*** Correspondence: j.olmedo@ua.es

**Abstract:** Postactivation potentiation (PAP) describes an initial muscular activation with a submaximal or maximal load intensity that produces acute improvements in muscle power and performance in subsequent explosive activities. The objective of this study was to compare the effect of different PAP protocols in rowing performance. A crossover design involving seven rowers was used, in which two different PAP protocols were applied: PAP of maximal conditioning contractions (PAP MCC) on a rowing ergometer to provide greater transferability and, thus, enhance the magnitude of PAP stimuli on subsequent rowing performance; and PAP of maximal strength contractions (PAP MSC) in half squat and bench pull exercises, similar to the main exercises in rowing strength training, to perform a 20 s "all-out" test simulating a competition start. Student's *t*-test was used to compare means of the variables (*p* < 0.05). Effect size statistics were calculated using Cohen's *d*. The PAP MCC protocol resulted in significant differences, with an extremely large effect size in average power output (*p* = 0.034, *d* = 0.98) in the first 3 (*p* = 0.019, *d* = 1.15) and first 5 (*p* = 0.036, *d* = 0.91) strokes. This group also reached a greater number of strokes (*p* = 0.049, *d* = 2.29) and strokes per minute (*p* = 0.046, *d* = 1.15). PAP with maximal conditioning contractions in rowing warm-up enhanced subsequent rowing sprint and is an advisable strategy to potentiate performance at the start of rowing competitions and sprint regattas.

**Keywords:** sprint; postactivation potentiation; fixed seat rowing; performance

#### **1. Introduction**

Postactivation potentiation (PAP) describes an initial muscular activation with a submaximal or maximal load intensity that produces acute improvements in muscle power and performance in subsequent explosive activities [1]. This procedure involves concentric contractions of submaximal dynamic force close to a high percentage of maximal repetition (1RM) before the execution of an explosive movement with similar movement patterns. However, if the load induced by the PAP is poorly planned in relation to the volume or intensity of work, it can cause fatigue. Therefore, the balance between PAP and fatigue is vital because of its role in the magnitude of improvement in mechanical power [2]. Determination of the characteristics in relation to the components of the load (volume, intensity, density, and type of exercise) are key to the subsequent performance induced by the PAP [3].

The optimal intensity of the conditioning activities for competitive athletes varies between 75% and 90% of 1RM, and recovery time should be individualized because athletes differ in strength level, training experience, and muscle fiber structure, with the optimal recovery time being 6 min [4]. Stronger recreational level athletes would experience greater benefit with shorter breaks (5–10 min) than weaker athletes, who would need more time (15–20 min) [5]. Results of this study suggest that the principle of individualization is key to determining the rest time necessary for optimal activation performance following PAP protocols. Seitz et al. [6] observed increased force production with activation time

**Citation:** Penichet-Tomas, A.; Jimenez-Olmedo, J.M.; Serra Torregrosa, L.; Pueo, B. Acute Effects of Different Postactivation Potentiation Protocols on Traditional Rowing Performance. *Int. J. Environ. Res. Public Health* **2021**, *18*, 80. https://dx.doi.org/10.3390/ ijerph18010080

Received: 25 November 2020 Accepted: 18 December 2020 Published: 24 December 2020

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/ licenses/by/4.0/).

after PAP in elite athletes. Coaches must consider individual differences in the strength level(s) of athletes to adapt protocols and maximize performance effects. Esformes et al. [7] related the difference in muscle fiber activation according to the type of contraction with rest time after PAP. They concluded that it could positively influence power performance; however, in relation to the level of activation and type of contraction, there were only significant differences in isometric contractions with a long rest period (12 min), producing positive results for peak power output, peak force, maximum distance and rate of force development, and inducing a longer PAP period. On the other hand, the application of a previous PAP methodology involving maximal dynamic force with submaximal loads close to 1RM in half squat exercises can be very useful for improving performance in the ability to repeat high-intensity sprints [8]. Furthermore, PAP would be beneficial in events where the contribution of the aerobic energy system may be of high importance toward successful performance; however, research in this area is quite limited. Silva et al. [9] found that heavy strength exercise bouts improve 20 km cycling time trial performance with no alteration in pacing strategy. In rowing, PAP could improve the efficiency of the working muscles by increasing force production of muscle fibers if a specific pace is to be maintained [10].

Strength training in elite rowers represents 10–20% of the total training time. During the competitive season, rowers perform strength training with a load of between 85% and 95% of 1RM [11]. There is a strong relationship between a high manifestation of strength and greater performance [12]. Doma et al. [2] demonstrated that maximal dynamic contractions on a rowing machine increased power performance during the execution of a short sprint (10 s). In addition, the average power during a 20 s maximal effort test is an effective predictor of 2000 m Olympic rowing performance [13]. Nevertheless, traditional rowing is characterized, and differs from Olympic rowing, in technical execution because the rower is supported by a fixed seat in the coccyx area [14]. This implies that the degree of trunk amplitude both in the attack and final phases is greater than in Olympic rowing. Studies on traditional rowing have evaluated physiological factors as predictors of performance [15] as well as specific protocols for improving sports performance [16]. On the other hand, assessment and analysis strategies derived from Olympic rowing have been developed [17] to be able to adapt to the needs of traditional rowing [18], leading to comparative studies between both disciplines [19]. The scientific literature also includes studies in the area of traditional rowing on specific and conditional physiological requirements, quantifying the workload in official competition [20]. Accordingly, there are studies on the relationship between the characteristics of rowers and their sports performance [21], sports practice habits and quantification of training hours [22], as well as profile analysis of rowers using traditional rowing modalities based on the competitive level [14].

The PAP is considered to be an adequate specific warm-up by coaches in both day-today training or competition, and current scientific evidence shows that there is a positive effect on performance when using PAP in short sprint efforts [1]. In addition, the greatest response is manifested in experienced subjects in high-intensity training [23]. However, there is no current scientific evidence supporting intervention studies that have proposed the comparison of the effects of different PAP types on rowing sprint performance, a sport in which physical and physiological factors, such as aerobic and anaerobic power, influence as well as a high level of physical strength [24]. In addition, no study has analyzed the influence of PAP using a traditional rowing modality. Therefore, the objective of this study was to analyze and compare the acute effect of two different postactivation potentiation protocols in traditional rowing performance at the start of competition. The strength contraction with maximum loads in strength exercises of the main muscle groups in the rowing gesture are an effective PAP method for rowing performance. However, maximum effort power conditioning contractions will better prepare the rower as this PAP method increases neuromuscular performance and has a greater dynamic transfer because it involves a biomechanically similar exercise, and rowers perform more effective neural patterns enhancing a more specific gesture.

#### **2. Materials and Methods**

#### *2.1. Experimental Design*

A crossover design was used to compare the effect of both PAP protocols: PAP of maximal conditioning contractions (PAP MCC) and PAP of maximal strength contractions (PAP MSC). In this repeated measurement design, each participant performed the different protocols during different time periods (i.e., the participants crossed over from one protocol to another during the research process). The reason a crossover design was considered was that it could yield a more efficient comparison of treatments than a parallel design (i.e., fewer participants might be required in the crossover design to attain the same level of statistical power or precision as a parallel design). Hence, each participant served as his own matched control and performed both protocols: PAP MCC and PAP MSC [25]. Participants were tested three times in three consecutive weeks on the same day of the week at the same time. The first session was oriented to familiarize the athletes with the techniques and methods, and the next weeks' session addressed the 2 PAP protocols (Figure 1). The protocols were assessed in randomized order with a one-week interval in between.

**Figure 1.** Schematic representation of the study design.

#### *2.2. Participants*

Seven male rowers, competing at the national level, participated in the study. Inclusion criteria for the present study consisted of a minimum previous experience of 2 consecutive full years in traditional rowing, with a minimum of 10 months per year and an average of 12 h per week [17]. Rowers who did not meet the selection criteria were excluded from the study. Table 1 summarizes the anthropometric characteristics of the sample.

**Table 1.** Anthropometric characteristics.


BMI: body mass index; SD: standard deviation; CI: confidence interval.

The Ethics Committee at the University of Alicante gave institutional approval for this study, in accordance with the Declaration of Helsinki (IRB UA-2020-07-21). The subjects were informed about the study and gave their written informed consent. They were notified of the need for their commitment to the following requirements on the day of data collection and the previous day: not to perform high-intensity physical activity in the previous 24 h; not to consume any type of stimulant at least 5 h before testing; and not to eat any solid food at least 3 h before testing.

#### *2.3. Procedures*

Data collection was performed for 3 weeks on the same day of the week at the same time each week. The first week was used to perform the test of maximal repetition (1RM) in half squat and bench pull [26], to perform anthropometric measurements, and to familiarize the athletes with the 20 s "all-out" test. Anthropometric measurements were performed in basal conditions following standard protocols from the International Society for the Advancement of Kinanthropometry (ISAK) [27] using the following instruments: a TanitaBC-545N balance for body mass (0.1 kg) (Tanita Co., Tokyo, Japan) [28] and an astra stadiometer with a mechanical scale to measure height (0.1 cm). Body mass index (BMI) was computed as body mass (kg) divided by height squared (m<sup>2</sup> ) [29]. Skinfolds (sub-scapular, tricipital, bicipital, iliac crest, supra-spinal, abdominal, anterior thigh, and middle leg) were measured using a caliper calibrated to the nearest 0.2 mm (Holtain Ltd., Crymych, UK). Girth (relaxed arm, flexed arm, thigh, and calf) and breadth (humerus, stylion, and femur) measurements were performed using a flexible anthropometric steel tape to the nearest 0.1 cm (Holtain Ltd., Crymych, UK). Finally, body composition was calculated using the equations described by Withers, Craig, Bourdon, and Norton [30] for fat mass and the equation proposed by Lee et al. [31] for muscle mass.

After warm-up, the 1RM test consisted of two sets of 5 and 3 repetitions at approximately 50% and 70% 1RM, respectively. The subjects then had up to 5 attempts to obtain their 1RM [8]. A 3–5 min rest interval was adopted between trials. Each attempt consisted of 1 series of 1 repetition. If the attempt was successful, between 5% and 10% of the weight was added for the next attempt. When the subject could not perform the execution properly, until the subject was unable to lift the load, his previous attempt was his 1RM.

After performing the 1RM tests, the familiarization test was performed for proper technical execution of the 20 s "all-out" test. The subjects performed three sets of 20 s at maximal with a rest interval of 6 min to simulate the intensity on test days. The test simulated a start; as such, in the first 3 strokes, the shoulders started from a vertical position of the body (Figure 2). After the first 3 strokes, the objective would be to achieve the maximal possible power in 20 s with a correct wide stroke and with a maximal ratio of strokes per minute.


**Figure 2.** Differences in technique for (**a**) the first 3 strokes and (**b**) after the first 3 strokes.

During the second week, data were collected according to the crossover design in which the subjects were distributed in two groups: one performed the PAP MCC protocol and the other performed the PAP MSC protocol. In the third week, the groups rotated to perform the protocol they did not perform the week before. First, both groups performed 5 min of warm-up on the rowing machine (Concept 2 Model D, USA, Vermont) [11] with a coupling adapted for the reproduction of the traditional rowing stroke, programmed with a drag factor of 160, a given power (up to a maximal of 140 W or 2:15 / 500 m), and at a heart rate (HR) not exceeding 140 beats/min [17].

After the warm-up, each group performed a different method of PAP in each session before the 20 s "all-out" test, with a 6 min rest before and after the PAP protocol. The PAP MCC protocol consisted of performing a series of 20 s at maximal effort on the rowing machine before performing the 20 s "all-out" test. In Doma et al. [2], a 10 s maximal dynamic conditioning contraction protocol was executed on a rowing ergometer at maximal effort, followed by a 10 s "all-out" test. In the present study, the 20 s "all-out" sprint test was performed because it covered the 10 s time span and could yield additional information. In addition, a 20 s "all-out" test can be an effective predictor of the performance in a 2000 m test [13]. The rest interval between the PAP and the post-test was 6 min [32].

The athletes performed the PAP MSC protocol in the two exercises performed in the 1RM test, with a protocol of 4 series of submaximal approach (1 × 5 with the bar/2 min, 1 × 2 50% RM/4 min, and 1 × 1 85% RM) and 1 set of 3 repetitions at 90% of 1RM [33]. The rest interval prior to the 20 s "all-out" test was also 6 min [32]. The intensity of the exercise that elicits the greatest PAP effect should be individualized (60–100% 1RM) because it is dependent on the level of maximal strength [34]. The 20 s "all-out" test was performed on a rowing machine (Concept 2 Model D, Morrisville, VT, USA) [11] with a coupling adapted for traditional rowing fixing the seat. The measurement of performance variables was achieved using the Erg Data mobile device application in sync with the PM5 performance monitor of the rowing machine via Bluetooth Smart. Heart rate parameters were measured with the Polar M400 (Polar Inc., Kempele, Finland) with H7 Bluetooth® Smart Band [8].

#### *2.4. Statistical Analysis*

Results were analyzed using Statistical Package for Social Sciences (SPSS v.26 for Windows, SPSS Inc., Chicago, IL, USA), and values of each variable are expressed as mean and standard deviation. With the aim to determine whether the quantitative variables maintained the criterion of normality, a Shapiro–Wilk statistical test was performed. Fulfilling the criteria of normality, Student's *t*-test was used to compare the means of the variables (*p* < 0.05) pairwise between PAP MCC and PAP MSC. Cohen's *d* was used as a measure of the effect size of differences between protocols and interpreted according to Cohen's thresholds: small (*d* < 0.3), medium (*d* = 0.3–0.4), large (*d* = 0.5–0.6), very large (*d* = 0.7–0.9), and extremely large (*d* > 0.9) [35].

#### **3. Results**

When comparing performance parameters between PAP MCC and PAP MSC, the power outputs of the PAP MCC group were highest in the 20 s "all out" test. Significant differences with an extremely large effect size were noted in average power output (*p* = 0.034, *d* = 0.98), power output in the first 3 strokes (*p* = 0.019, *d* = 1.15), and in the first 5 strokes (*p* = 0.036, *d* = 0.91) (Table 2). Higher values and very large effect size were reached in maximal power output (*p* = 0.080, *d* = 0.74) and first stroke power output (*p* = 0.085, *d* = 0.84), although no significant differences were found.

**Table 2.** Results of the 20 s sprint test "all-out".


PAP: postactivation potentiation; MCC: maximal conditioning contractions; MSC: maximal strength contractions; SD: standard deviation; CI: confidence interval; W: watt; max: maximum; N: number; min: minute; HR: heart rate; \*: statistically significant (*p* < 0.05).

The total number of strokes (*p* = 0.049, *d* = 2.29) and the rate of strokes per minute (*p* = 0.046, *d* = 1.15) were significantly greater in the PAP MCC group than in the PAP MSC group, with an extremely large effect size.

Both average and maximum heart rates were also higher in the PAP MCC group than in the PAP MSC group, but no significant differences were found in heart rate difference. The effect size for average heart rate was extremely large (*p* = 0.050, *d* = 1.40), whereas there was a very large effect size for maximum heart rate (*p* = 0.225, *d* = 0.78).

#### **4. Discussion**

The objective of this study was to analyze and compare the acute effects of two different PAP protocols in traditional rowing performance using a crossover design such that each participant performed different protocols during different time periods. In the current study, we sought to determine the PAP protocol that would be most beneficial for rowing performance. The group with the highest values of power output reached in the 20-s "allout" test was the group that performed the PAP of maximal conditioning contractions (PAP MCC) in rowing ergometer compared with the group that performed the PAP of maximal strength contractions (PAP MSC) in half squat and bench pull. Doma et al. [2] found higher power values in Wmean, Wmax, and W1stroke with the rowers who performed dynamic conditioning contractions before the test. In the same way, isometric conditioning contractions on a rowing ergometer to the rowing warm-up appears to increase short-term rowing ergometer performance, especially at the start [10]. The effect of the postactivation potentiation of conditioning contractions, with similar movement patterns of rowing gesture, enhances power output. However, the type of conditioning contractions may have different effects on PAP and fatigue. Isometric conditioning contractions may induce central fatigue, while dynamic conditioning contractions may induce the opposite response [36].

Prior heavy load exercise will induce some activation adaptations to the nervous system that will help increase performance in the subsequent test(s) as long as the recovery time is adequate [8]. However, the PAP MSC group reached the lower values in all measured power variables. This may indicate that this type of PAP may not benefit rowing performance. The greatest adaptations will occur in experienced athletes performing highintensity training [23]; therefore, they would need a longer recovery time after PAP [5]. In a study by Esformes et al. [7], greater manifestation of the PAP MSC was also related to a shorter recovery time after the PAP. In this way, if the recovery time after the PAP is not sufficient, it will cause fatigue conditioning in subsequent efforts. Both the degree of PAP MSC and the recovery time after a PAP will be linked to performance optimization. Therefore, the degree of manifestation of the maximum dynamic strength must be taken into account individually to quantify the recovery time after the PAP correctly [6].

The increase in power achieved with the PAP MCC protocol may have explained the faster average stroke rate and the total number of the strokes, both in the present study performed in a 20 s "all-out" test and in other studies evaluated over 1000 m rowing ergometer time trial [10] and over a 10 s "all-out" test [2].

The mean and maximum heart rate values were higher in the PAP MCC group. When a very explosive test is carried out, HR increases very quickly when it comes from a 6 min recovery where it practically returns to a resting HR range. The PAP MCC group previously performed the PAP protocol, and the heart had already reached a working HR range. The same did not occur with the PAP MSC group because during the PAP protocol, they did not reach such high HR values. Therefore, when the test was performed, the HR did not increase so fast because HR is a component of late physiological response or due to a possible greater activation of the sympathetic system after conditioning contractions versus strength contractions [37].

Regarding limitations, findings from the present study should be interpreted with caution due its to the small sample size, although a crossover design was used. Furthermore, the results should be interpreted considering that the study was performed on a rowing start and not on competitive distance, an event in which contribution(s) of the aerobic

energy system may be highly important to successful performance. Future research aimed at confirming our results in a larger population and using a 2000 m rowing performance test, and not only in performance at the rowing start, would be interesting. Additionally, training using a 1RM load is inconsistent with safety rules. A better solution to describe the 1RM movement technique is to work at a specific rate of movement, describing the tempo of phases of movement (e.g., eccentric, transition, and concentric) [38].

#### **5. Conclusions**

The results suggested that postactivation potentiation with maximal conditioning contractions in rowing warm-up enhanced subsequent rowing sprint performance, simulating the start of a race in traditional rowing. It can be an interesting strategy to potentiate rowing performance if prescribed by coaches before activities that involve explosive starts; as such, rowing coaches should implement this form of training.

Coaches can prescribe maximal conditioning contractions in warm-ups before starting specific rowing training as well as in competitions. However, more research is needed to determine whether this PAP protocol and other methodologies can also improve 2 km rowing performance.

**Author Contributions:** Conceptualization, A.P.-T., L.S.T. and B.P.; data curation, A.P.-T. and L.S.T.; formal analysis, A.P.-T., J.M.J.-O. and B.P.; investigation, A.P.-T. and L.S.T.; methodology, A.P.-T., J.M.J.-O. and B.P.; project administration, A.P.-T.; resources, J.M.J.-O. and L.S.T.; software, A.P.-T. and B.P.; supervision, B.P.; validation, A.P.-T. and B.P.; visualization, L.S.T.; writing—original draft, A.P.-T.; writing—review and editing, A.P.-T., J.M.J.-O., L.S.T. and B.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

