Following the Affect and Health Behavior Framework (AHBF; Williams and Evans, 2014; Williams et al., 2019; Stevens et al., 2020), the consequences of exercise training on affective determinants of exercise behavior can be divided into the following three categories: (1) affective response, which refers to how one feels while performing or immediately after completing the exercise (e.g., core affect [valence and arousal]); (2) affect processing, which encompasses cognitive processing of previous or anticipated affective responses to exercise, in the form of automatic (e.g., affective association and implicit attitude) or reflective processing (e.g., affective judgments [affective attitude, exercise enjoyment]); and (3) affectively charged motivation, which is a motivational state that includes and/or has a basis in past affective responses to exercise, arising from automatic (e.g., hedonic motivation) as well as reflective (e.g., intrinsic motivation) processing pathways.

In the few previously published studies on affect-related responses to MICT compared with HIIT within a multi-week training program, exercise enjoyment (the second category) has been the most studied affective determinant to date (Heisz et al., 2016; Kong et al., 2016; Vella et al., 2017; Santos et al., 2021). Less frequently, constructs from different categories of the AHBF have been considered in combination. These studies examined changes in affective valence and intrinsic motivation (Gerber et al., 2018) or affective valence, exercise enjoyment, and affective attitude (Santos et al., 2021) in response to MICT versus HIIT.

Regarding affective attitude (the second category of the AHBF) the study by Santos et al. (2021) showed a constant pattern of positive change over a 2-week intervention, with participants in MICT reporting a more positive affective attitude than participants in HIIT throughout. With regard to affectively charged motivation (the third category of the AHBF), Thøgersen-Ntoumani et al. (2016) showed that intrinsic motivation for HIIT and MICT was positively linked to adherence but did not differ between training types in the middle of a 10-week program. Confirmatory evidence is provided by the study of Gerber et al. (2018), in which intrinsic motivation increased significantly from baseline to post-intervention within a 4-week period of either a sprint interval training (SIT) as a specific mode of HIIT, or a time-adjusted MICT without group difference.

In addition to repeated measurements of affective determinants through the application of multi-item questionnaires before, during, or after the intervention period, some studies repeatedly conducted in-situ measurements during and directly after exercise sessions to examine changes in task-related responses over the weeks. Based on the preliminary evidence, HIIT seems to result in greater (Heisz et al., 2016; Kong et al., 2016) or at least similar (Vella et al., 2017; Santos et al., 2021) exercise enjoyment (second category of the AHBF) compared with MICT. Moreover, different patterns of change emerged over the course of multi-week training programs. While two studies showed no difference between MICT and HIIT due to a constantly high level (Vella et al., 2017; 8 weeks) or an equally increasing level of exercise enjoyment (Santos et al., 2021; 2 weeks), two other studies found a difference in favor of HIIT due to a constant higher level (Kong et al., 2016; 5 weeks) or a progressive increase in exercise enjoyment over the weeks (Heisz et al., 2016; 6 weeks). Only one of these studies (Santos et al., 2021) and one additional study (Gerber et al., 2018) examined in-task affective valence (first category of AHBF). Santos et al. (2021) observed significantly lower peak negative values during exercise in HIIT compared with MICT across all three weeks with no significant change from the first to the last week of the intervention. Gerber et al. (2018) found no difference in affective valence during exercise between the MICT and HIIT groups across all weeks, but a significant decrease from the first to the last week of the intervention in both groups. However, in both studies, no interaction effect (time x training type) was observed, suggesting a similar pattern of change for both training conditions.

The preliminary evidence on changes in affect-related exercise determinants in response to a multi-week MICT or HIIT program reveals partly mixed results, with studies showing either no difference or favoring results for one type of training or the other. Importantly, this inconsistency in the findings - in addition to the methodological issues mentioned above - appears to be driven by the particular construct being studied and the survey method used for this purpose (pre-post questionnaires vs. in-situ measurements). Furthermore, the state of research leaves open whether the order of experiencing the two different exercise modalities matters for changes in affective exercise determinants.

This study aims to investigate training type- and sequence-dependent changes in different affect-related exercise determinants over a multi-week training program in insufficiently active adults. On this basis, the study can potentially generate evidence for the influence of different exercise modalities on affective determinants and thus on the maintenance of exercise behavior. Referring to the AHBF (Williams and Evans, 2014; Stevens et al., 2020), we examined three exercise-related categories of affective determinants: affective response to exercise, affect processing, and affectively charged motivation. Following the identified research gap, we first compared responses before and after a training period of either MICT or HIIT. Considering that previous exercise experiences of individuals influence their subjective experience as well as their reflective evaluation of different exercise regimens, in a second step we investigated the influence of the training sequence on changes in affective exercise determinants to contrast the two training types in an intra-individual comparison. For this purpose, the training program was switched following the first training period, resulting in two training sequences with reversed orders of training types (i.e., MICT - HIIT vs. HIIT - MICT).

In order to capture changes in affect-related exercise determinants over the intervention period, we used the classical approach of pre-post questionnaires particularly for affective attitude and intrinsic motivation. Additionally, we applied an innovative approach to compare changes in affective response to exercise in a more general manner and independent of the training regimen itself. Thus, the rationale was to go beyond task-specific responses to specific sessions with different exercise protocols (MICT/HIIT) by repeatedly recording changes in affective response to a standardized exercise session, which differs from the two types of training. For this purpose, we conducted in-situ measurements of affective valence and exercise enjoyment during and after a standardized vigorous-intensity continuous exercise session (VICE), respectively. This exercise intensity was chosen because the greatest variability in affect-related responses has been observed in this vigorous domain, especially in insufficiently active individuals (Ekkekakis et al., 2005, 2011).

Based on the preliminary evidence presented above (Thøgersen-Ntoumani et al., 2016; Gerber et al., 2018; Santos et al., 2021), we hypothesized that participants’ affective attitude and affectively charged motivational responses will show a similar pattern of positive change after the first period of HIIT or MICT. Regarding affective and enjoyment responses to exercise training programs, evidence indicates that compared with MICT, HIIT is experienced as less pleasurable but is reported post-exercise to be more enjoyable (Heisz et al., 2016; Kong et al., 2016; Vella et al., 2017; Gerber et al., 2018; Santos et al., 2021). However, to our knowledge, there have been no studies to date that have examined the change of in-task affect-related responses to a standardized VICE session as a result of different types of training. Therefore, the transferability of these findings to changes in a within-subject comparison with different training sequences (MICT - HIIT vs. HIIT - MICT) was investigated on an exploratory basis.

This study was part of the “Individual Response to Physical Activity” (iReAct) project, which is an interdisciplinary research network that investigates physiological and psychological responses to exercise at the individual level. The study protocol was approved by the Ethics Committee of the Medical Faculty, University of Tübingen (# 882/2017BO1). Information pertaining the current study is presented below. Further details on the registered trial (German Clinical Trials Register, # DRKS00017446) can be found elsewhere (Thiel et al., 2020).

The study used a 15-week, two-period sequential training intervention design in order to compare adaptive responses to two types of endurance training (MICT vs. HIIT) and their different sequential order (MICT - HIIT vs. HIIT - MICT). This means that participants started with either MICT or HIIT and then switched respectively to HIIT or MICT. The training programs were designed for a duration of six weeks each, so that – on the one side – physiological adaptations could occur. On the other side and from the psychological point of view of this study, this training period allows to study changes in affective exercise determinants based on subjective experiences and evaluations of the training program. In contrast to the wash-out period in clinical trials (e.g., Lim and In, 2021), we deliberately opted for a continuous sequence to allow evaluation of the training experience in direct comparison (i.e., within-person differences). After a baseline assessment, participants were randomly assigned to either the MICT - HIIT or the HIIT - MICT group. Randomization was computer-generated (nQuery 7.0) with a 1:1 allocation ratio to each sequence using mixed block sizes and two binary stratification factors: sex and maximal oxygen uptake (V̇O₂max).

For the purpose of this paper, we focused on changes in affect-related exercise determinants, which were collected in comprehensive assessments at baseline, between the two training periods (follow-up I), and after the completion of the second training period (follow-up II). A detailed overview of the within-subject design can be found in Figure 1.

Participants (men and women between 20 and 40 years of age) were recruited in six consecutive waves over a 2-year period (March 2018 to March 2020), primarily using the University of Tübingen and the University Hospital of Tübingen mailing lists. Interested individuals were asked to fill out the validated German version of the European Health Interview Survey - Physical Activity Questionnaire (EHIS - PAQ; Finger et al., 2015) to assess their physical activity levels. For participation in the study, participants had to be insufficiently active at the time of recruitment that is, not meeting the World Health Organization recommendations for moderate physical activity of at least 150 minutes per week. In addition, participants had to report less than 60 minutes per week of leisure-time exercise (including sports participation, aerobic activities, muscle strengthening) and no regular exercise engagement during the last six months.

To ensure that participants could complete the assessments and training intervention without risk and to avoid large heterogeneity in view of specific diseases or medication (e.g., severe previous illness), we opted for inactive, but healthy adults. For detailed inclusion and exclusion criteria, please refer to the study protocol (Thiel et al., 2020). Eligibility was checked during a telephone screening, as well as a medical screening prior to final enrollment in the study. Participants were provided with detailed information regarding the study procedure and associated risks prior to giving written informed consent.

The original power calculation for the whole interdisciplinary research project is described in the study protocol (Thiel et al., 2020) and was based on the primary endpoint regarding physiological adaptations (V̇O₂max). The targeted sample size was 60 subjects considering two degrees of freedom spent on the strata age and sex, resulting in effect sizes (difference of means divided by standard deviation) of 74.9% (type 1 error 0.05, two-sided, power 0.8, nQuery 7.0). As the calculated sample size was not reached, we further discuss this issue with respect to the present part of the study in the limitations section.

Affective attitude. To measure affective attitude toward exercise, we applied a questionnaire developed by Crites et al. (1994), whose German-language version was validated by Brand (2006). This instrument comprises four items based on the phrase “When I think about exercising, I feel…” Answers follow semantic differentials on a bipolar rating scale ranging from 1 (not relaxed/ not satisfied/ not happy/ not uncomfortable) to 7 (extremely relaxed/ extremely satisfied/ extremely happy/ extremely uncomfortable). The internal consistency of the items was acceptable (Cronbach’s α = .76), after previously reversing the fourth item (uncomfortable). Thus, the mean of the four items was used as the individuals’ affective attitude.

Intrinsic motivation. We used a validated German-language instrument for measuring the self-concordance of sport- and exercise-related goals (SSK-Scale; Seelig and Fuchs, 2006), considering only the 3-item subscale for the intrinsic mode of motivation as an indicator of affectively charged motivation. The internal consistency of this scale was good (Cronbach’s α = .89), so we used the mean of the three items as the individuals’ intrinsic motivation.

In-task affective valence. Core affective valence was measured every ten minutes during the standardized VICE session using the validated German version of the Feeling Scale (FS; Hardy and Rejeski, 1989; Maibach et al., 2020). The FS is a single-item, 11-point bipolar rating scale, ranging from -5 (very bad) through 0 (neutral) to +5 (very good) developed for the assessment of affective response along a displeasure-pleasure continuum. The in-task value was calculated as the average FS-rating between minute 20 and minute 50. The first FS-rating after minute 10 was not considered, since warm-up did not represent the targeted exercise at vigorous intensity.

Post-exercise enjoyment. Exercise enjoyment was assessed right after termination of the standardized VICE session. Referring to the single-item exercise enjoyment scale (EES; Stanley and Cumming, 2010), participants were asked to indicate how much they enjoyed the exercise session. Answers could be given on an analogue scale ranging from not at all (0) through neutral (50) to very much (100).

In laboratory visits at baseline, follow-up I, and follow-up II (see Figure 1), participants undertook an incremental step test to volitional exhaustion on a cycle ergometer (Ergoselect 200; Ergoline GmbH, Bitz, Germany) for determination of the V̇O₂max, peak power output (PO_peak), and lactate thresholds (first lactate turning point [LTP1] and second lactate turning point [LTP2]). The test began with a 2-min resting period on the bike, followed by 25-watt (W) step increments every three minutes, starting at 50 W for males and at 25 W for females, until task failure. Capillary blood lactate concentration ([La–]) was analyzed (Biosen S-Line; EKF, Cardiff, UK) by collecting capillary blood samples (20 μL) from the right earlobe before starting the test, during the last 20 seconds of each stage, and immediately after volitional exhaustion. Heart rate (HR) and electrocardiogram (ECG) were constantly monitored throughout the test (12-channel PC ECG; custo med GmbH, Ottobrunn, Germany). Breath-by-breath pulmonary gas exchange and ventilation (V̇E) were measured using a metabolic cart (MetaLyzer; CORTEX Biophysics, Leipzig, Germany). Further details on the laboratory setup and data processing can be found elsewhere (Mattioni Maturana et al., 2021).

Over a period of 15 weeks, participants underwent two training periods, starting with either HIIT or MICT and switching groups for the second period (see Figure 1). Each training period lasted six weeks and consisted of three weekly training sessions (on average). Minimum adherence was set at 15 out of 18 prescribed sessions in each training period. The training programs were designed with the goal that both exercise interventions would be matched for energy expenditure (Andreato, 2020).

Exercise protocols. MICT was prescribed as 60 minutes of continuous cycling at the power output (PO) corresponding to 90% of LTP1. Such exercise intensity was prescribed for participants to cycle within the moderate-intensity domain (Binder et al. 2008; Hofmann and Tschakert, 2017). HIIT started with a 10-min warm-up at a PO corresponding to 70% of HR_max, followed by 4x4-min intervals at a PO corresponding to 90% HR_max with a 4-min active resting period at 30 W between each interval. After the last interval, a 5-min cool-down period was performed at 30 W, totalizing 43 minutes of exercise. Such exercise intensities were chosen to ensure that participants would be within the severe-intensity domain during the load intervals (i.e., all the exercise intensities were above LTP2), allowing participants to reach 70% HR_max during the recovery periods.

Training monitoring. All exercise sessions were performed on calibrated cycle ergometers (ec5000; custo med GmbH, Ottobrunn, Germany) and supervised by trained personnel. During every session, participants’ HR and ECG were constantly monitored (3-channel ECG; custo med GmbH, Ottobrunn, Germany) to control for default training intensity and adjust for fitness changes over the weeks. After every training session, the exercise training data (i.e., second-by-second power output, cadence, and HR) were exported and stored for further processing (for details, see Mattioni Maturana et al., 2021).

Participants underwent three assessment phases (at baseline, follow-up I, and follow-up II; see Figure 1) during which they were asked to complete a paper-pencil questionnaire on affective determinants of exercise behavior. Moreover, participants performed a 60-min standardized VICE session on a calibrated cycle ergometer (Ergoselect 200; Ergoline GmbH, Bitz, Germany), with the following exercise intensity values based on the incremental step test performed at each assessment phase: After a 10-min warm-up at a PO corresponding to 90% of LTP1, participants cycled at a constant PO corresponding to the midpoint between LTP1 and LTP2 (i.e., vigorous-intensity domain) for 50 minutes. During the VICE session, in-situ assessments were performed to delineate affect-related responses to acute exercise. Participants’ responses were recorded via smartphone (Google Nexus 5; LG Group, Seoul, South Korea) with the movisensXS application (movisens GmbH, Karlsruhe, Germany). During exercise, A3 posters were used as visual reference and participants’ responses were recorded by the investigator so that the participants could concentrate on the exercise itself. The post-exercise survey was conducted independently by the participants with smartphone in hand.

Descriptive statistics were generated and intraclass correlation coefficients (ICCs) were calculated for the outcome measures. Furthermore, estimates of the within-person variability across the three standardized VICE sessions as well as the between-person variability in the affect-related outcomes were calculated. For the trajectory analysis across the training program, violin plots were created for each outcome based on the derived type- and sequence-descriptive statistics (see Supplementary Table 1). Effect sizes (d) and confidence intervals (95% CI) were calculated for repeated measures within training groups (baseline vs. follow-up I; follow-up I vs. follow-up II) or sequences (baseline vs. follow-up II) (Morris and DeShon, 2002; see Supplementary Table 2). Following the conventions suggested by Cohen (1988), effect sizes were interpreted as small (d = 0.2), medium (d = 0.5), and large (d = 0.8). Reported were those changes in affective determinants whose 95% CI do not contain zero.

We fitted a mixed model for repeated measures with the three levels group sequence (MICT - HIIT vs. HIIT - MICT), subject (ID), and session (follow-up I and follow-up II), including the fixed effects training type (MICT vs. HIIT), training period (period 1 vs. 2), and training sequence according to a 2x2 cross-over design and a random intercept on the subject level. Models were calculated separately for each of the four affect-related constructs (Model AA: affective attitude, Model IM: intrinsic motivation, Model AV: affective valence, and Model EE: exercise enjoyment) as dependent variables. The respective baseline value was included in the model as a covariate. Due to low numbers of units on the subject level, simple covariance structures (scaled identity) had to be chosen to reach convergence in all the models. Following the study protocol, a modified intention-to-treat (ITT) population was used: Subjects who did not contribute outcome measurements for either of the two periods were excluded. For subjects who contributed data for at least one period, multiple imputation (MI) was applied under the assumption of a missing-at-random (MAR) mechanism. As our data were approximately normal, we applied the Amelia package for longitudinal imputations under the use of the expectation-maximization (EM) algorithm generating m = 25 sets (Honaker et al., 2011).

Although all convergence criteria were fulfilled, we faced non-positive-definite hessian matrices on some of the MI datasets for one of the outcomes (affective valence). Despite choosing a simple covariance structure and increasing the number of step-halvings or the number of Fisher scoring steps, we could not resolve this problem and therefore applied a sensitivity analysis running the model on the reduced number of MI sets with valid hessians only (m = 18).

All analyses were conducted with R (R version 4.1.1 and RStudio 1.4.1717, PBC, Boston, MA, USA) and SPSS (Version 26, IBM Corp, Armonk, NY, USA). Given the four affective outcome measures, we set the significance level of α = 0.0125 to adjust for multiple testing using the Bonferroni correction to control for the family-wise error rate.

Data were analyzed from 40 previously insufficiently active healthy adults (72% female) aged between 20 and 40 years (M = 27, SD = 6). An overview of participants’ demographic, anthropometric, and physical activity characteristics at baseline can be found in Table 1. A total of 58 participants were assessed for eligibility, 49 of which were included in the randomization process and nine of which were excluded during medical screening (Consort Flow Diagram, Figure 2). During the baseline assessment, five participants dropped out for different reasons. Two other participants did not complete the first training period due to illness and thus inability to complete the minimum adherence. The nine participants of the last recruitment wave (MICT - HIIT group: n = 5; HIIT - MICT group: n = 4) had to terminate the second training period due to COVID-19 restrictions but were still included in the intention-to-treat analysis (for missing values, see Table 2). Two non-native speakers included in deviation from the study protocol were subsequently excluded from this data analysis because comprehension of the questionnaires could not be guaranteed.

Descriptive between-person and within-person statistics of the study variables can be found in Table 2. The grand mean of each outcome was in the upper half of the respective scale. The empirical range of person means varied from 2.50 to 6.92 for affective attitude, from 1.67 to 5.67 for intrinsic motivation, from -1.58 to 3.92 for affective valence, and from 11.67 to 83.50 for exercise enjoyment, indicating substantial between-person variability. The ICCs indicated that in affective attitude and intrinsic motivation 76% of the total variance referred to between-person differences, while in affective valence and exercise enjoyment 50% or 56% could be attributed to within-person differences with a range of 0.00 to 3.54 or 1.41 to 38.21. Thus, it can be noted that the between-person stability is higher for the questionnaire data than for the in-situ measurements.

For the two questionnaire measures, there were somewhat different trajectories over time with respect to training types and sequences. A moderate increase in mean affective attitude from baseline to follow-up II was observed for both training sequences (d_{MICT - HIIT} = 0.52, 95% CI: 0.08 – 1.34; d_{HIIT - MICT} = 0.70, 95% CI: 0.69-1.97) with somewhat higher values in the MICT - HIIT group (Figure 3A). In the HIIT - MICT group, a moderate increase was already observed after the first period of HIIT (d = 0.50, 95% CI: 0.55-1.84), followed by a small increase after the second period of MICT (d = 0.34, 95% CI: 0.07-1.35). However, Model AA revealed no significant fixed effect for training type (T = -0.01, p = 0.994), training period (T = -0.58, p = 0.566), or training sequence (T = 0.21, p = 0.837) on affective attitude measures (see Table 3). With regard to mean intrinsic motivation, no substantial changes could be observed for both training sequences (Figure 3B). Accordingly, Model IM revealed no significant fixed effect for training type (T = -0.44, p = 0.663), training period (T = -0.67, p = 0.506), or training sequence (T = 0.54, p = 0.590) on individuals’ intrinsic motivation (see Table 3).

For the in-situ measurements, there were different changes in affect measures between the training sequence groups. In the MICT - HIIT group, mean affective valence - assessed during VICE - showed a moderate to large increase from baseline to follow-up I (d = 0.61, 95% CI: 0.09-1.36) and follow-up II (d = 0.91, 95% CI: 0.26-1.76). In contrast, no substantial changes could be observed in the HIIT - MICT group (Figure 3C). Conformingly, Model AV revealed a significant fixed effect for training sequence (T = 2.56, p = 0.011) on affective valence (see Table 3). However, the fixed effect for training type (T = -2.00, p = 0.045) missed the significance level (α = .0125). Mean exercise enjoyment assessed right after termination of VICE - showed no substantial changes from baseline to follow-up I for both training sequences. In contrast, enjoyment values in the HIIT - MICT group decreased to a large extent from baseline to follow-up II (d = -0.90, 95% CI: -1.93 - -0.47) (Figure 3D). However, Model EE revealed no significant fixed effect for training type (T = -1.10, p = 0.272), training period (T = 0.21, p = 0.834), or training sequence (T = 1.17, p = 0.242) on exercise enjoyment (see Table 3).

The strength of the current study is the within-subject design that allows comparison of two different training types and sequences (MICT vs. HIIT; MICT - HIIT vs. HIIT - MICT) with data from both pre-post questionnaires and in-situ measurements during a VICE session before, between, and after two 6-week training periods. We standardized exercise intensities relative to metabolic landmarks both during training periods and prior to VICE sessions with reference to a physiological framework (Binder et al., 2008) and captured different affective exercise determinants in accordance with the AHBF (Williams and Evans, 2014). Further, we applied a multilevel modeling approach to account for the nested data structure. Finally, by recruiting adults who did not achieve the recommendations for health promoting physical activity, the current findings are of direct relevance to a segment of the population that is particularly in need of interventions for promoting exercise.

Limitations of the current study are also important to consider. As a consequence of the elaborate study design with 15-weeks of training and three extensive diagnostic blocks that did not allow for extended absences, only 40 participants could be recruited. Moreover, there are missing values because of an early termination of the last survey wave due to the COVID-19 pandemic. Consequently, MI was applied, resulting in non-positive-definite hessian matrices for affective valence in some of the MI datasets. However, sensitivity analyses showed stable results for testing the hessian matrices, and alpha adjustment was used to reduce the risk of alpha error for the present substudy. Other limiting factors include limited external validity due to the highly standardized ergometer training, possible retention effects within a multi-week training program associated with improvements in affect-related exercise determinants in both training sequences, and limited generalizability of the results due to the relatively young and healthy sample studied here. Taking all this into account, observed effects have to be interpreted with caution and should be replicated in future studies with less structured exercise modalities, a control group without exercise intervention, and larger as well as more diverse samples (e.g., in terms of transferability to people with chronic diseases).

Furthermore, we examined only the reflective constructs in the second and third categories of the AHBF. Since dual-process models such as the DMT (Ekkekakis, 2003) – and more recently the ART (Brand and Ekkekakis, 2018) – postulate that a combination of both the automatic affective valuation (type-1 process) and the reflective evaluation (type-2 process) of exercise determine whether an individual either remains in a state of physical inactivity or starts and maintains exercise, future studies should additionally examine the effects of MICT and HIIT on automatic constructs (e.g., implicit attitude, hedonic motivation). In a next step, the relevance, specific contribution, and interplay of various affective determinants in influencing future exercise behavior should be determined (Stevens et al., 2020).