Chronotype at the beginning of secondary school and school timing are both associated with chronotype development during adolescence

The misalignment between late chronotypes and early school start times affect health, performance and psychological well-being of adolescents. Here we test whether, and how, the baseline chronotype (i.e. chronotype at the beginning of secondary school) and the school timing affect the magnitude and the direction of the developmental change in chronotype during adolescence. We evaluated a sample of Argentinian students (n = 259) who were randomly assigned to attend school in the morning (07:45 a.m.–12:05 p.m.), afternoon (12:40 p.m.–05:00 p.m.) or evening (05:20 p.m.–09:40 p.m.) school timings. Importantly, chronotype and sleep habits were assessed longitudinally in the same group of students along secondary school (at 13–14 y.o. and 17–18 y.o.). Our results show that: (1) although chronotypes partially align with class time, this effect is insufficient to fully account for the differences observed in sleep-related variables between school timings; (2) both school timing and baseline chronotype are independently associated with the direction and the magnitude of change in chronotype, with greater delays related to earlier baseline chronotypes and later school timings. The practical implications of these results are challenging and should be considered in the design of future educational timing policies to improve adolescents’ well-being.

www.nature.com/scientificreports/ H2 Both school timing and baseline chronotype are independently associated with the developmental change in chronotype. Previous cross-sectional studies showed that chronotype is progressively delayed during adolescence until it reaches a peak at the end of this developmental stage [7][8][9][10][11][12] . Under comparable environmental cues (e.g. light-dark cycle), this chronotype delay should have limits: either because of the magnitude of the change and at what age this 'delaying process' starts and ends during adolescence, or because the limits imposed by the intrinsic mechanism of the circadian clock 17,[59][60][61][62][63] . Importantly, individuals will not be entrained to the environment outside these theoretical upper (and lower) limits of chronotype (Supp. Fig. 1). The exact value of the upper limit is unknown and might depend on different environmental and social factors (e.g. light exposure, geographical longitude or latitude, culture, etc.). Beyond that, depending on the limit value, the magnitude and/or the direction of the developmental change in adolescents' chronotype will (or will not) be affected. For example, an extremely late upper limit will have no effect on adolescents' chronotype (because adolescents will not reach that upper limit, even at the end of secondary school).
Four alternative scenarios appear when considering the previously mentioned upper limits and including school timing and baseline chronotype as independent variables in a correlation model explaining the magnitude and/or the direction of the developmental changes of chronotype during adolescence.
Box Figure 1. Four theoretical scenarios of the association of school timing and/or baseline chronotype (in 1st year of high school) with ΔChronotype (MSFsc 5th year-MSFsc 1styear). (a) Only school timing is associated. (b) Only the baseline chronotype is associated. (c) Baseline chronotype and school timing are additively associated. (d) Baseline chronotype and school timing interact. The graphs on each scenario represent the expected developmental change in chronotype (i.e. ΔChronotype) as a function of the baseline chronotype (i.e. 1st year MSFsc) for each school timing. A zero value on the vertical axis indicates no change in chronotype from 1st to 5th year. Positive or negative values indicate that chronotype is delayed or advanced, respectively, in 5th compared to 1st year. Each colored line represents the linear relation between ΔChronotype and baseline chronotype for each school timing. Grey arrows represent ΔChronotype for three representative baseline chronotypes (early, intermediate and late), the base of the arrows represent students' chronotype in their 1st year (i.e. the baseline chronotype) and the arrowheads represent students' chronotype in their 5th year. www.nature.com/scientificreports/ (a) The school timing affects the magnitude (not the direction) of ΔChronotype independently of each student's baseline chronotype (Box Fig. 1a). Earlier school timings will exert more pressure on the ageassociated delay in chronotype. Thus, the magnitude of ΔChronotype will be smaller in students attending the morning school timing than in students in the other school timings. Note that in this scenario the age-associated expected change does not exceed the upper limit of chronotype (Supp. Fig. 1a). (b) Only the baseline chronotype, and not school timing, is related to the magnitude (but not the direction) of ΔChronotype (Box Fig. 1b). Students with later baseline chronotypes will reach the upper limit of the possible chronotypes range before their peers with earlier baseline chronotypes. Consequently, the magnitude of ΔChronotype will be smaller in students with later baseline chronotypes, irrespective of school timings. Moreover, as students became older, chronotypes would be delayed for all school timings. Note that in this scenario the theoretical upper limit has to be lower than in scenario 1, affecting students' ΔChronotype (Supp. Fig. 1b). (c) School timing and baseline chronotype are associated with the magnitude and direction of ΔChronotype.
Here, both phenomena act together but independently (Box Fig. 1c). On the one hand, later baseline chronotypes will experience a smaller ΔChronotype due to reaching the upper limit of possible chronotypes. On the other hand, school timing will also affect ΔChronotype: earlier school timings will exert higher pressure and, consistently, students attending school earlier will experience smaller ΔChronotype. Note that the school timing effect would take place considering the existence of the upper limit and, in the most extreme cases, it would lead to negative ΔChronotype (i.e. morning attending students with late baseline chronotypes will not delay, or will even advance, their chronotype, experiencing an earlier chronotype in their 5th year compared to their 1st) (Supp. Fig. 1c). (d) Not only school timing and baseline chronotype but also their interaction are associated with the magnitude and the direction of ΔChronotype (Box Fig. 1d). In this scenario, the age-associated expected change in chronotype is within the range of possible chronotypes. Each school timing differentially affects ΔChronotype depending on the baseline chronotype. In particular, while the magnitude of the pressure exerted by morning and afternoon school timing is larger for later chronotypes, the evening school timing is late enough to not exert any pressure, regardless of baseline chronotype (Supp. Fig. 1d).
H3 Age-related changes in both social jetlag and sleep duration are associated with ΔChronotype and modulated by school timing. We predict that larger chronotype delays would be associated to higher levels of social jetlag and shorter sleep duration on weekdays, with stronger associations in the morning school timing.

Results
Mean chronotype and SDw depend on school timing and age, while SJL depends only on school timing (H1). To evaluate how school timing and age longitudinally affect chronotype during adolescence (data distribution in Supp. Fig. 2), we ran a linear mixed-effect model with chronotype (i.e. MSFsc) as the dependent variable, including school timing (morning, afternoon or evening), age (1st or 5th year) and their interaction as fixed effects, and students' id as a random effect (Supp. Table 1, Supp. Table 2). As in our previous cross-sectional results 57 , we found a main effect of school timing (F 2,256 = 29.697, P < 0.0001, partial η 2 = 0.188, 90% confidence interval (CI) = 0.119-0.256). Morning-attending students presented earlier chronotypes than both afternoon-and evening-attending students (Fig. 1a, Supp. Table 3), suggesting that school timing affects students' biological time, improving its alignment to the school timing where students were randomly assigned. We also found a significant main effect of age (F 1,256 = 41.921, P < 0.0001, partial η 2 = 0.141, 90% CI = 0.081-0.207), with earlier chronotypes in 1st year. Importantly, a significant interaction between school timing and age (F 2,256 = 12.062, P < 0.0001, partial η 2 = 0.086, 90% CI = 0.036-0.142) reveals that chronotype' changes throughout adolescence are modulated by school timing. At 1st year, adolescents' chronotype only slightly differed between school timings, but this difference gets larger by their 5th year (Fig. 1a). Consistently with our previous crosssectional study, school timing modulates how adolescents' chronotype changes with age.
To evaluate whether the observed modulation was sufficient to fully, or only partially, align students' chronotype with their school schedules, we assessed the effects of age and school timing on both social jetlag (SJL) and sleep duration (SD) levels.
Second, we ran a mixed effect model for Sleep Duration (SD) including school timing, age and type of day (weekday or free day) (data distribution in Supp. Fig. 4 and sleep timings in Supp. Table 7) as fixed effects and students' id as a random effect (Supp. www.nature.com/scientificreports/ CI = 0.276-0.379). This last effect indicates that students sleep less on weekdays, as expected. Additionally the interactions between type of day and age (F 2,768 = 10.117, P = 0.002, partial η 2 = 0.013, 90% CI = 0.003-0.029) or school timing (F 2,768 = 90.161, P < 0.0001, partial η 2 = 0.190, 90% CI = 0.150-0.230) were also significant. Conversely, the interaction between age and school timing (F 2,768 = 1.998, P = 0.136, partial η 2 = 0.005, 90% CI = 0.000-0.015) and the triple interaction between age, school timing and type of day (F 2,768 = 1.368, P = 0.255, partial η 2 = 0.004, 90% CI = 0.000-0.012) were not significant. On weekdays, students sleep less when they are older and morning-attending students sleep less than adolescents with later school schedules (Fig. 1c). Instead, students do not differ in their sleep duration, despite their age or school schedule, on free days (Fig. 1c, Supp. Table 10). Thus, morning-attending students present very short SDw (i.e. high levels of sleep loss) in their 1st year of school and this situation aggravates as adolescence progresses. The difference in SDw between school timings was not compensated by napping (Supp. Fig. 5 [71][72][73] . Altogether, the results presented here support that school timing modulates only partially adolescents' chronotype with both SJL and SDw levels depending on the school timing to which students were randomly assigned at the beginning of secondary school. The developmental change in chronotype during adolescence is associated with both school timing and baseline chronotype (H2). In the previous section, we showed that mean changes in chronotype depend on age and school timing. However, the association between 1st and 5th year's chronotypes is low-to-moderate in all school timings (morning: t = 3.462, P = 0.001 r = 0.344, 95% CI = 0.149-0.514; afternoon: t = 1.769, P = 0.080 r = 0.182, 95% CI = − 0.022-0.372; evening: t = 4.439, P < 0.0001 r = 0.461, 95% CI = 0.261-0.623) (see also Supp. Fig. 6). Here, we propose that baseline chronotype (i.e. 1st year chronotype) might be related to this lack of stability. Consistently, we contrasted the four scenarios previously described (Hypotheses box H2, Box Fig. 1, Supp. Fig. 1) to evaluate whether baseline chronotype and school timing are associated with the developmental changes in chronotype (i.e. age-related changes in chronotype or ΔChronotype).
Our results are consistent with scenario c (Hypotheses Box H2): both school timing and baseline chronotype are additively associated with ΔChronotype during adolescence, with no interaction between them. Even though morning-attending students experienced, on average, a lower delay in their chronotype from 1st to 5th year (compared with their afternoon-and evening-attending peers), overall, students with earlier baseline chronotypes exhibited larger delays and those with later chronotypes showed smaller delays or advances, regardless of their school timing.
Age-related changes in SJL and SDw are associated with ΔChronotype and school timing (H3). Later chronotypes are associated with higher levels of social jetlag (SJL) and a lower sleep duration on weekdays (SDw), especially when attending school in the morning 39,65-69 . Here we explored whether the individual changes in SJL or SDw during adolescence depend on ΔChronotype and/or school timing. We ran a linear regression model with the age-related changes on SJL (i.e. ΔSJL = SJL 5th year-SJL 1st year) (Supp. Fig. 7) as the dependent variable and ΔChronotype and school timing as predictors (Supp. Table 16, Supp. Table 17). We found significant main effects of both school timing (F 2,253 = 4.493, P = 0.012, partial η 2 = 0.034, 90% CI = 0.004-0.075) and ΔChronotype (F 2,253 = 235.795, P < 0.0001, partial η 2 = 0.482, 90% CI = 0.413-0.543) (similar results were found when using SJL sleep corrected formula 70 ). In brief, the more delayed the chronotype becomes from 1st to 5th year, the bigger the change in SJL. For example, if a hypothetical afternoon-attending student exhibits a ΔChronotype equal to the mean change for their school timing (ΔChronotype = 61 min, e.g. from 05:00 to 06:01), their SJL will increase by just 2 min. However, another student, with a 1 h-larger ΔChronotype (e.g. from 05:00 to 07:01, i.e. 121 min), would increase their SJL on 35 min along secondary school. Importantly, the interaction between ΔChronotype and school timing was significant (F 2,253 = 7.021, P = 0.001, partial η 2 = 0.053, 90% CI = 0.014-0.100). The association between ΔSJL and ΔChronotype was progressively weaker the later the school timing, even though the comparison between afternoon and evening school timings was not significant (Fig. 3a, Supp. year, respectively). No significant differences were found between 1st and 5th year at any school timing. Post-hoc comparisons, P < 0.017 (P < 0.05, Bonferroni corrected). (c) School timing and age affect sleep duration on weekdays (SDw) but not on free days (SDf). On weekdays, adolescents sleep less in their 5th year regardless of their school timing, and in the morning school timing regardless of their age. Students sleep more on free days than on weekdays independently of their age and school timing. No differences were found between school timings and age on free days. The interaction between school timing and age was not significant. The asterisk (*) indicates significant difference in sleep duration between 1st and 5th year across school timings, which was found on weekdays but not on free days. Post-hoc comparisons, P < 0.0038 (P < 0.05, Bonferroni corrected www.nature.com/scientificreports/ compared with their afternoon-and evening-attending peers (slope comparisons: morning vs. afternoon: t = 2.767, P = 0.017; morning vs. evening: t = 3.552, P = 0.001).
Age-related changes in SDw also showed interindividual differences (Supp. Fig. 8), even though changes on mean SDw were similar when comparing school timings (Fig. 1c). We ran a linear regression model with the age-related changes in SDw (ΔSDw = SDw 5th year-SDw 1st year) as the dependent variable, and ΔChronotype and school timing as predictors (Supp. Table 19 Supp. Table 20). As expected, the main effect of school timing was Figure 2. Developmental changes in chronotype (ΔChronotype) during adolescence is associated with both school timings and students' baseline chronotype. (a) Baseline chronotype tertile affects the change in Chronotype from 1st to 5th year. Those students on the earliest tertile delayed their chronotype by their 5th year, independently on school timing (from 04:11 to 05:08, from 04:47 to 06:38, from 05:08 to 06:58, for morning-, afternoon-and evening-attending students respectively). Students on the latest tertile of MSFsc, on the other hand, did not delay or even advanced their chronotypes (from 07:20 to 06:07, from 07:34 to 07:29, from 07:59 to 08:03, for morning-, afternoon-and evening-attending students respectively). (b) ΔChronotype correlates with baseline chronotype and school timing. Students with a late baseline MSFsc experienced a lower chronotype change from 1st to 5th year. Particularly, the slope of the relationship between ΔChronotype and baseline chronotype indicates that, when the baseline chronotype is 1-h later, the age-related changes on chronotype are lower: 43 min (95% CI = − 53 to − 32 min), 49 min (95% CI = − 61 to − 38 min) and 32 min (95% CI = − 44 to − 20 min) for morning-, afternoon-and evening-attending students, respectively. N = 259. ΔChronotype = MSFsc 5th year-MSFsc 1st year. Baseline chronotype = MSFsc in 1st year. Color indicates school timing: yellow, green and blue for morning, afternoon and evening, respectively. www.nature.com/scientificreports/ non-significant (F 2,253 = 1.433, P = 0.241, partial η 2 = 0.011, 90% CI = 0.000-0.037), indicating that SDw change similarly in different school timings (Fig. 1c). We found a significant main effect of ΔChronotype (F 1,253 = 8.196, P = 0.0046, partial η 2 = 0.031, 90% CI = 0.006-0.075) and, importantly, a significant interaction between ΔChronotype and school timing (F 2,253 = 7.852, P < 0.001, partial η 2 = 0.058, 90% CI = 0.017-0.108), indicating that school timing modulates the effect of ΔChronotype on age-related changes in SDw. In particular, afternoon-and evening-attending students with larger delays in their chronotype throughout adolescence exhibit less shortening, or even a lengthening, of their SDw (afternoon: b = 0.191, 95% CI = 0.013-0.369, t = 2.108, P = 0.036; evening: b = 0.515, 95% CI = 0.278-0.756, t = 4.281, P < 0.0001) (Fig. 3b). To illustrate, an average afternoon-attending student (ΔChronotype = 61 min) would decrease their SDw by 44 min, while a peer with a 1 h-larger ΔChronotype (i.e. 121 min) would decrease their SDw by 32 min. Note that the corresponding slope is the difference between 44 and 32 min, which is 12 min. On the other hand, morning-attending students with the greatest delays in their chronotypes by their 5th year, showed a tendency to shorten their SDw the most, although the slope was not different from zero (b = − 0.109, 95% CI = − 0.310-0.093, t = − 1.062, P = 0.289). Despite the fact that both the slopes for afternoon-and evening-attending students did differ from zero, only evening and morning slopes significantly differ between them (morning vs. evening: t = − 3.950, P < 0.001) (Supp. Table 21). Even though one would expect that age-related chronotype delays in morning-attending adolescents would be strongly associated www.nature.com/scientificreports/ with a comparable increase in SJL and decrease in SDw 3,54,74-76 , our results show that SJL increases accordingly with the chronotype delay while SDw did not decrease as much as expected.

Discussion
Here we achieved three related and complementary aims. First, we reproduced and strengthened our previous cross-sectional results 57 on how school timing and age affect chronotype and sleep. Second, we showed that the magnitude and the direction of the age-associated change in chronotype is associated with both school timing and baseline chronotype (i.e. chronotype of adolescents in their 1st year of secondary school) consistent with scenario c (Hypotheses box H2, Box Fig. 1c). We proposed a mechanism that is consistent with our results (i.e. the existence of an upper limit), but our data is not sufficient to test which is this limit or if other mechanism is involved. Third, we found that the developmental changes in chronotype are associated with social jetlag and sleep duration age-related changes. Consistently with our previous data 57 , we found that students' chronotypes were partially aligned with their school timing. Chronotype depends on both school timing and age, as well as on their interaction: the midpoint of sleep on free days (MSFsc) is later in older adolescents and later school timings, with larger differences between school timings for older adolescents. Social jetlag (SJL) is higher and sleep duration on weekdays (SDw) is shorter when school timing is earlier, especially for older students attending school in the morning. Most results were consistent between both studies, reinforcing our conclusion that school timing, as a social cue, partially modulates adolescents' internal timing.
Importantly, our longitudinal design allowed us to further analyze the low-to-moderate stability of chronotype during adolescence reported in previous longitudinal studies [9][10][11] . We found that ΔChronotype correlates with baseline chronotype. A previous study reported that chronotype development was associated with the interaction between age and circadian preferences 9 finding similar results. However, baseline chronotype as a factor related to the low stability of chronotype along development has not been previously reported and, thus, our approach and results contribute to understand how chronotype changes during adolescence.
Previous works have studied how chronotype is affected either by school timing 7,8,62 or by age 7-12,60 during adolescence, but here we analyzed these two factors together and longitudinally. Our results are consistent with our prediction that baseline chronotype and school timing are independent and additive associated with adolescents' chronotype ( Supplementary Fig. 1c, Scenario 3): 1-later school timings are related to later chronotypes, with a stronger association in older adolescents, and 2-earlier baseline chronotypes experience a bigger ΔChronotype. According to our model, while a morning-attending student with a baseline chronotype equal to 05:46 (i.e. the mean baseline chronotype for this school timing) practically does not change their chronotype during secondary school, 1-h later baseline chronotypes (i.e. 06:46) would advance their chronotypes 45 min by 5th year. Thus, morning-attending adolescents who have late baseline chronotypes at the beginning of secondary school might experience smaller delays or even advance their chronotype with age. As summarized in Scenario 3, getting older is not necessarily associated with later chronotypes.
Here we propose a mechanism that includes the existence of limits to ΔChronotype during adolescence. Particularly, an upper limit associated with either the developmental stage achieved at the end of secondary school 59,60 or with the entrainment mechanism of the circadian clock 62 might explain the association of both school timing and baseline chronotype with ΔChronotype. Previous works showed that chronotype variability among adolescents depends on different factors, including genetics, culture, light exposure, schedules and age 2,[4][5][6][7][8][12][13][14][15][16][17]57 . On the one hand, advanced pubertal stages have been associated with later chronotypes 59,80 students with later baseline chronotypes might be the ones presenting the most advanced pubertal stages at the beginning of secondary school. If this case, they would reach the upper limit before their peers and, consistently, they would have a smaller delay in chronotype between 1st and 5th year than they peers who started secondary school at lower pubertal stages. On the other hand, the upper limit might be associated with the entrainment mechanism. To be entrained to the external 24 h light-dark cycle, humans have to be exposed to light at specific times of the day 2,17,62 . Consistently, the interindividual variability exists but has limits and the range of chronotypes does not cover the 24 h (i.e. some theoretically possible chronotypes might not be compatible with entrained rhythms 62 ). Although humans can invert their wake-sleep cycle to be active at night and sleep during the day, as individuals who work night shifts, these subjects do not exhibit stable entrained rhythms 61,63,81 . As chronotype is delayed throughout adolescence 7,8,10,11 , students with later baseline chronotypes would reach the upper limit of stable entrainment 82 before their earlier peers. Furthermore, in our setting, this scenario is especially plausible because Argentinian adolescents exhibit particularly late baseline chronotypes 7,8,53,57,82,83 . Based on our data, we cannot disambiguate whether the upper limit of chronotype exists and depends on the developmental stage and/or on the entrainment mechanisms of the circadian clocks. A possible concern associated with the correlation between the changes observed in MSFsc and its baseline value is that it might be partially/fully explained by the regression towards the mean 84 . This effect occurs because there exists random errors when a variable is assessed within-subjects. Thus, if the initial measurement were an extreme value it is highly probable that the next time we assess the same variable in the same subject the value would be closer to its true mean. This could lead to an artificial association between baseline values and the change observed in the variable. Consistently, the magnitude of the effect depends on the magnitude of within-subject variability and on the precision of the measurement instrument used. Although we acknowledge that this artifact might be influencing our results (especially considering that there is scarce literature validating MCTQ in adolescents), the effect of the regression towards the mean, if present, would only partially affect our results and it does not detract them from its validity. First, MCTQ is a reliable instrument to assess adolescents' chronotype: similar results (magnitude and direction) associated with age and school timing were obtained using MCTQ 8,54,57,85,86 and other instruments (e.g. DLMO, actigraphy, MEQ, etc.) 7 www.nature.com/scientificreports/ good when assessing adolescents and young adults 89 and ΔChronotype does not change when adolescents are evaluated and reevaluated with one year difference; furthermore, no differences were found between early and late types 90 . That is, variability within-subjects seems to be low. Moreover, similarly to the results presented here, when the period between test and retest during adolescence is longer, the midpoint of sleep derived from actigraphy showed higher within-subject variability 9,10 . Altogether, these findings support the idea that our results are due to a development effect that is built during adolescence and not due to regression towards the mean derived from MCTQ lack of reliability. Finally, we reported a difference of almost 2 h in ΔChronotype when comparing earlier and later chronotypes on different school timings (Fig. 2a). This magnitude is biologically significant and it is highly improbable that it would be observed only because of regression towards the mean, especially when well established effects acting on chronotype during adolescence, such as age and school timing effects, rarely reach this magnitude [8][9][10]41,54,57,87 . Future work is needed to fully understand the causes of the association we found between baseline chronotype and ΔChronotype during adolescence. Importantly, our results show not only that ΔChronotype correlates with baseline chronotype and School timing but also that ΔChronotype is associated with age-related changes in SJL and SDw. As expected, large delays in chronotype were associated with an increase in SJL and this association progressively weakens from morning to evening school timings (Supp. Discussion). In contrast, a delay in chronotype was associated with an increase in SDw in both afternoon and evening school timings, with a steeper association for the latter, and we found no association in the morning (Supp. Discussion). Thus, a better alignment between adolescents' internal timing and school timing seems to be beneficial in terms of sleep duration for afternoon-and evening-attending adolescents but not for their morning-attending peers. The latter was surprising because, although most adolescents shorten their SDw from 1st to 5th year, the difference on SDw does not depend on their ΔChronotype. A possible explanation for this result is that morning-attending students were already sleep-deprived in their 1st year and, consequently, their SDw might not be further shortened because of homeostatic reasons. Thus, the effect of a delay in Chronotype in morning-attending students is mostly absorbed by an increased in SJL levels and not by a shortening on SDw.
This study has several limitations. First, chronotype and sleep-related variables were self-reported through standardized questionnaires. Consistently, we cannot rule out a bias in students' answers, but they are highly improbable because students were blind to our experimental hypotheses. However, objective assessment of sleep and chronotype, such as actigraphy, could be more suitable. Second, our analyses are based on regressions, which do not allow us to establish causality but only association among variables. Third, we did not have access to other predictors that might modulate chronotype and its developmental change, such as pubertal stage, socioeconomic status, the usage of medications, the presence of illnesses, etc. Fourth, in this longitudinal study we only have data from two time points, one at the first and another at the last year of secondary school. The inclusion of additional time points (e.g. in the middle of secondary school) would be preferable but it was impossible due to operative reasons. Finally, the lack of assessment of chronotype and sleep habits before the beginning of secondary school does not allow us to unequivocally affirm that the initial point was completely balanced between school timings, even considering the random assignation.
This research also has some important strengths. First, the longitudinal design allows us to study the developmental changes during adolescence. Second, the sample size of our study is one of the highest among similar studies 53,54,56,91,92 . Third, as in our previous study, we worked with three different school schedules, including an evening school timing (17:20 -21:40). Finally, the random assignation of students to a particular school schedule at the beginning of their secondary school, allowed us to assume no differences in chronotype and sleep habits between school timings before starting secondary school.
Our results have several practical implications. First, we found that an early morning but also the afternoon school timing is associated with unhealthy sleep habits in adolescents. Consistently, a practical implication when thinking about better school start times would be that later morning school starting times might help but may not be enough for adolescents to have healthy sleep. This is especially relevant in populations exhibiting particularly late chronotypes, such as the adolescents from Argentina, Uruguay and Spain 54,93 . In these situations, an evening school timing might be at least considered by the educational community and/or policy makers. Many families and even the educational community believed that the morning school timing is 'the most favorable school timing' , but our work and several others support the idea that it is not the case: afternoon (or evening) school timing might be a more equitable and even preferable environment where early chronotypes do not present an advantage over late ones 53,55,57 . Second, we show that school timing modulates chronotype and sleep habits during adolescence and, then, the undesirable conditions or behaviors associated with eveningness reported in the literature (e.g. depressive feelings or substance use) might be associated with the lack of alignment between chronotype and school timing (previous studies only include students that attend school in the morning 10,11,[94][95][96][97][98][99]. Thus, future studies should include the effect of school timing. Third, the low-to-moderate stability observed in chronotype during adolescence suggests that chronotype is a malleable target for interventions 10,11,[94][95][96][97][98][99] : knowing that earlier baseline chronotypes would exhibit larger delays, provides us new insights to help design interventions addressing adolescents sleep health and behavior.
Finally, the association between baseline chronotype and the magnitude and direction of ΔChronotype reported here go against the most parsimonious and intuitive notion that all adolescents would similarly delay their chronotype. Chronotype does not necessarily delay during adolescence and it is associated with baseline chronotype and school timing. These results might modify the previously suggested policies to improve the alignment between school timing and adolescents' internal timing. For example, the assignment of school timing based only on baseline chronotype would not be as beneficial for adolescents' sleep health and academic performance [53][54][55]57 as expected. Of course, more evidence is needed to shed light on this matter and to understand their practical implications, but this novel finding adds knowledge to the field and opens a new range of possibilities and questions. Exploring the underlying mechanisms of the association of both school timing and baseline www.nature.com/scientificreports/ chronotype effects with how chronotype changes during adolescence will lead us to a better understanding of how we can help adolescents to reach healthier sleep habits.

Methods
Ethical approval. The study and all the methods included on it were conducted in accordance with relevant guidelines and regulations, including the ethical recommendations for human chronobiological research 100 and Argentinian national regulations. The study was not invasive of the integrity of the participants and the data was collected during regular school hours. The study was approved by the institutional Ethical Committee of the Universidad Nacional de Quilmes (Verdict #4/2017) and by the head of school. Written informed consent was obtained from the head of the school. Students provided oral informed consent to participate.
Participants. This

Procedure.
A crucial aspect of our experimental setup is that in this particular school, the school timing (morning, 07:45-12:05; afternoon, 12:40-17:00; evening, 17:20-21:40) is set by a lottery system at the beginning of the secondary school and maintained during the whole secondary school, as described in depth in our previous study 57 . Briefly, in June 2015 students in their first school year filled a questionnaire including demographic information (date of birth and self-defined gender) and the Spanish version of the MCTQ 24 . MCTQ includes questions about sleep habits and results in a local time point (MSFsc) where earlier times (i.e. low values) indicate early chronotypes and late times (i.e. high values) indicate late chronotypes 24 . Data collection was performed during students' typical school hours (morning, afternoon and evening school timings). Data collection and analysis were not performed blind to the conditions of the experiments. The exact same procedure was applied in both June 2015 (during their first school year) and July 2019 (during their last school year).

Measurements.
For each student on each school year, we obtained a chronotype index: the sleep-corrected midpoint of sleep time on free days (MSFsc) 24 , social jetlag (SJL = MSF -MSW. Note that we use the difference and not the absolute difference because we are interested in the direction of the misalignment between social and biological timings) and sleep duration on both week (SDw) and free days (SDf). From these measurements we also calculated the ΔChronotype (i.e. developmental change in chronotype, MSFsc 5th year -MSFsc 1st year ), the ΔSJL (i.e. SJL 5th year -SJL 1st year ) and the ΔSDw (i.e. SDw 5th year -SDw 1st year ).
Not all the variables were obtained for all students. Missing values occurred when a variable could not be calculated because of incomplete information (i.e. when a student did not complete all of the MCTQ questions). The data from a student was only included if the information was enough to calculate at least MSFsc, SJL, SDw and SDf. Missing data were omitted from the analyses. Statistical analysis. All statistical analyses were performed using the R system for statistical computing (v.4.0.2; R Core Team, 2020).
We ran linear mixed-effect models to determine whether school timing (morning, afternoon or evening) and age (1st or 5th school year) were associated with MSFsc or SJL. For sleep duration, the linear-mixed model included type of day of the week (weekdays or free days), school timing (morning, afternoon or evening) and age (1st or 5th school year). The same analysis was perform for total sleep duration (nocturnal sleep + naps). Students ID was included as a random effect in every model. P-values were computed using lmerTest package 101 .
We ran a generalized linear models to test whether the developmental change in chronotype depends on school timing (morning, afternoon or evening) and baseline chronotype (i.e. chronotype in 1st year) and to evaluate whether age-related changes in SJL and SDw depend on school timing (morning, afternoon or evening) and on the developmental change in chronotype.
Normality of the residuals of the models was check using Kolmogorov-Smirnov tests. Student's t-tests were used to perform post-hoc pairwise comparisons for categorical variables. We used an alpha level of 0.05 for all of the statistical tests. When applicable, we used Bonferroni correction for multiple comparisons (corrected P < 0.05). Partial η 2 effect sizes were computed using sjstats package version 0.18.0.

Data availability
The data and code that support the findings of this study are available from the corresponding author on request.