Behavioural assessments
5CSRTT
Due to the large amount of parameters assessed in the different manipulations of the 5CSRTT, only the most outstanding effects of WMI and sex are described in the results (see Table S1–S11 for the complete results). In Tables 1, 2 and 3, an overview is presented of significant results.
Table 1 Effect of manipulation on test performance in the 5CSRTT compared to baseline. Baseline parameters for each manipulation can be found in supplementary tables S1–S11 Table 2 Effect of white matter injury (WMI) compared to control animals on test performance in the 5CSRTT. Baseline parameters for each manipulation can be found in supplementary tables S1–S11 Table 3 Effect of sex on test performance in the 5CSRTT. Baseline parameters for each manipulation can be found in supplementary tables S1–S11 Training and motivation
Animals were trained until they responded stably (criteria: accuracy > 80% and omissions < 20) in the task. Stability for both criteria was reached in 49 training days. Performance was comparable for control and WMI animals as well as for male and female rats (see Table S1). Together, these results indicate that control and WMI animals of both sexes do not differ in acquisition of the 5CSRTT and thus have a similar starting point. No differences in accuracy or the latency to respond and obtain the reward were found between WMI and control animals and males and females after the ad libitum food test, indicating similar motivation to perform the task in all groups (see Table S2).
Most prominent effects: omissions, interactions of treatment and sex
The most prominent effects on 5-choice task performance compared to baseline were found on the amount of omissions made when the ITI was short (Table S3), short and variable (Table S4) or varied around the mean (both longer and shorter; Table S5).
Animals omitted more trials in the 2 s ITI and short variable ITI test compared to baseline (2 s ITI: Ftest(1,44) = 53.05, p < 0.001; short variable ITI: Ftest(1,44) = 44.15, p < 0.001, see Fig. 1A and B).
When the ITI was shortened to 2 s, WMI animals differed from control animals in the amount of omissions (Ftest*treatment*sex(1,44) = 5.10, p = 0.03; Ftreatment*sex(1,44) = 9.72, p < 0.001, see Fig. 1A). Under baseline conditions, WMI and control animals of both sexes did not differ in their omissions (baseline: Ftreatment*sex(1,44) = 2.06, p = 0.16). However, a short ITI resulted in a different response in males and females (test: Ftreatment*sex(1,44) = 9.97, p = 0.003), male control animals omitted more trials compared to female control animals (t(26) = 2.85, p = 0.008) whereas male and female WMI animals did not significantly differ in their omitted trials (t(18) = − 1.80, p = 0.09; see Fig. 1A). In addition, both male and female WMI animals did not differ in their omissions to their controls (during the short ITI: males WMI vs. control: t(23) = 2.49, p = 0.02, females WMI vs. controls (t(21) = − 2.03, p = 0.06, Bonferroni-corrected p-value: p < 0.025).
When the time between two trials was short and variable (see Fig. 1B), both males and females made more omissions when the ITI was short and variable compared to baseline (Ftest*treatment*sex(1,44):5.91, p = 0.02; Ftreatment*sex(1,44) = 9.52, p = 0.004; males: Ftest(1,23) = 34.11, p < 0.001; females: Ftest(1,21) = 13.25, p = 0.002). In addition, male WMI rats omitted less trials than control males in test conditions (Ftreatment(1,23) = 5.69, p = 0.03; Ftest*treatment(1,23) = 1.69, p = 0.21), whereas this was not observed in females (Ftest*treatment(1,21) = 4.53, p = 0.05; Ftreatment(1,21) = 4.04, p = 0.06).
Omissions made in the test where the ITI varied around the mean did not differ from baseline (Ftest(1,44) = 1.31, p = 0.26, Fig. 1C). However, an interaction effect of treatment and sex was found in the amount of omitted trials (Ftreatment*sex(1,44) = 5.48, p = 0.02, see Fig. 1C). Male WMI rats omitted less trials compared to control males in both baseline and test conditions (Ftreatment(1,23) = 4.37, p = 0.048), whereas WMI and control females omitted trials equally both under baseline and test-conditions (Ftest(1,21) = 3.55, p = 0.07; Ftest*treatment(1,21) = 3.55, p = 0.07; see Fig. 1C and Table S5).
Sex differences: perseverative responses and accuracy
In general, female rats made more perseverative responses compared to males under baseline conditions irrespective of experimental condition. The long variable ITI manipulation and the presence of a loose distractor resulted in a different behavioural pattern between the sexes (long variable ITI: Ftest*sex(1,44) = 4.63, p = 0.04; loose distractor: Ftest*sex(1,44) = 6.00, p = 0.02) whereas in the short variable ITI condition (short variable ITI: Fsex(1,44) = 5.39, p = 0.03), the WMI males made less perseverative responses.
In the long variable ITI condition (for full results and statistics, see table S7), when the sexes were analyzed separately, both males and females made less perseverative responses (males: Ftest(1,23) = 7.55, p = 0.01; females Ftest(1,21) = 10.94, p = 0.003, see Fig. 1D) compared to baseline but no effect of treatment was found (males: Ftreatment(1,23) = 0.67, p = 0.42; females Ftreatment(1,21) = 2.54, p = 0.13). In addition, the males showed a difference in response due to treatment (Ftest*treatment) = 7.55, p = 0.01, Fig. 1D): under baseline conditions male WMI rats showed a trend towards making less perseverative responses compared to control males (males baseline: control vs. WMI t(24) = 3.46, p = 0.08; males test: control vs. WMI t(24) = 0.31, p = 0.58). Female rats did not significantly show this interaction effect (Ftest*treatment) = 0.09, p = 0.76).
In the short variable ITI test, no effect of test was found in the amount of perseverative responses (see Table S4 for full results and statistics). When analyzed separately per sex, male WMI rats made less perseverative responses compared to control rats (Ftreatment(1,23) = 4.69, p = 0.04). Although the same pattern is observed in females rats, this was not significant (Ftreatment(1,21) = 3.25, p = 0.09, see Fig. 1E).
When analyzed per sex, in the presence of a loose distractor in the cage (for full results and statistics, see table S11), females tend to increase their perseverative responses compared to baseline (females: Ftest(1,21) = 3.70, p = 0.07); males Ftest(1,23) = 2.14, p = 0.16; see Fig. 1F). No effect of treatment or an interaction of test and treatment were found in both sexes (see table S11).
In the variable short stimulus duration condition, males tended to be less accurate during the short stimulus duration trials (Fsex(1,44) = 3.74, p = 0.06) and the sexes reacted differently to the test conditions (Fsex*test(1,44) = 2.45, p = 0.048). Post hoc analysis per sex showed that male rats were less accurate during the short stimulus duration trials (Ftest(1,23) = 10.29, p < 0.001) whereas female rats were equally accurate during test and baseline trials (Ftest(1,21) = 1.80, p = 0.14, see table S9).
Impulsivity: long 7 s ITI, long variable ITI and mean variable ITI
The ITI was increased to a fixed time of 7 s or a long variable time (between 5 and 13 s) or differed around the mean (between 3 and 7 s) in order to assess the animal’s ability to withhold a response (i.e. impulsivity) (for full results and statistics, see table S5-S6-S7).
The manipulations resulted in expected and previously reported effects, i.e. increased premature responses and decreased accuracy without an effect on perseverative responses, the latency to a correct response or the latency to obtain a reward. In the variable long ITI test, in addition to increased premature responses and a reduced accuracy, also more trials were omitted, less perseverative responses were made and the latency to make a correct response was increased, without the latency to obtain the reward being affected. When the ITI varied around the mean, premature responses increased and it took longer to make a correct response. The accuracy, omissions, perseverative responses and the latency to collect a reward did not change.
On average, treatment and sex did not affect performance. The only differences observed were in the long variable or mean variable ITI condition, where WMI animals were faster in collecting the reward, both under baseline as well as test conditions (long variable ITI: Ftreatment(1,44) = 10.61, p = 0.002; mean variable ITI: Ftreatment(1,44) = 4.74, p = 0.04). No other effects of WMI were found in the long- and mean-variable ITI manipulation (see table S5 and S7). Males and females showed a difference in the latency to make a correct response in the long variable ITI manipulation (Ftest*sex(1,44) = 4.63, p = 0.04). Under test conditions, males took longer to make a correct response whereas for females this was only a trend (males: Ftest(1,23) = 68.49, p < 0.001; females: Ftest(1,21) = 3.30, p = 0.08). Under baseline conditions, males and females performed similar (see Table S7).
Sustained attention: 2 s ITI, short variable ITI, a fixed or variable short stimulus duration and distractors
To investigate attentional performance, animals were subjected to a fixed short ITI (2 s), a short variable ITI (between 1 and 5 s), a fixed and variable short stimulus duration (fixed: stimulus duration is 0.5 s; variable: stimulus duration is alternating every 20 trials between 1.2 and 0.5 s) and two distractors (distractor 1: a fixed hurdle in the middle of the box and distractor 2: a wooden block that can be moved around the cage).
Tasks that place a high demand on attention are reported to result in decreased accuracy, increased omissions and longer response latencies without affecting premature responding. In line with this expectation, manipulations that place a high demand on attention made animals less accurate (short stimulus duration: Ftest(1,44) = 102.86, p < 0.001; short variable stimulus duration: Ftest(1,44) = 8.17, p < 0.001; distractors: fixed block: Ftest(1,44) = 10.69, p = 0.002; loose block: Ftest(1,44) = 12.46, p < 0.001, see Tables S8-S9-S10-S11), with the exception of an increase in accuracy by shortening the ITI to 2 s (Ftest(1,44) = 34.67, p < 0.001) and no difference compared to baseline in the short variable ITI condition (see table S3 and S4 for full results and statistics).
In addition, the manipulations indeed resulted in more omitted trials (see Tables S8-S11). Furthermore, animals made either less (2 s ITI and short variable ITI, Table S3-S4), more (short stimulus duration and loose block distractor, Table S8 and S11) or did not differ (variable short stimulus duration or the fixed block distractor, Table S9-S10) in the amount of premature responses.
The latency to a correct response was in most test conditions longer compared to baseline conditions in all animals (2 s ITI: Ftest(1,44) = 100.09, p < 0.001; short variable ITI: Ftest(1,44) = 34.34, p < 0.001, and both the distractors: fixed block: Ftest(1,44) = 20.37, p < 0.001, loose block: Ftest(1,44) = 19.17, p < 0.001, Table S3-S4 and S10-S11). A fixed but not variable short stimulus duration decreased the time it took the animals to make a correct response (short stimulus duration (Ftest(1,44) = 66.66, p < 0.001, see Table S8-S9).
The latency to obtain the reward was longer when a distractor was present (fixed block: Ftest(1,44) = 16.31, p < 0.001, loose block: Ftest(1,44) = 21.82, p < 0.001, see Table S10-S11), but did not differ with shorter (and variable) ITIs and stimulus durations (see Table S6-S9).
With a few exceptions, no effects were found of WMI and sex on performance. A fixed distractor made control animals less accurate compared to WMI animals (Ftreatment(1,44) = 5.25, p = 0.03, see Table S10). In the variable stimulus duration test, animals with WMI made less premature responses compared to control animals (Ftreatment(1,44) = 3.99, p = 0.05, Table S9). Females in the 2 s ITI test made less premature responses compared to males, especially in baseline conditions since there were hardly any premature responses in the test condition (Fsex(1,44) = 6.74, p = 0.01; Ftest*sex(1,44) = 6.41, p = 0.02; post hoc: baseline male vs. female: F(1,46) = 7.11, p = 0.01; test male vs. female F(1,46) = 1.51, p = 0.23). No effects on premature responses were found for the short variable ITI manipulation (see Table S3-S4).
Both distractors differentially affected the latency to a correct response in male and female rats (fixed block: Fsex*test(1,44) = 4.42, p = 0.04; loose block: Fsex*test(1,44) = 6.21, p = 0.02; Fsex*test*treatment(1,44) = 4.77, p = 0.03). Post hoc testing per sex showed that in the fixed and loose block manipulations, males take longer to make a correct choice (fixed block: Ftest(1,23) = 21.22, p < 0.001; loose block: Ftest(1,23) = 22.52, p < 0.001), whereas females did not differ or tended to be faster between baseline and test conditions in their response time (fixed block: Ftest(1,21) = 3.12, p = 0.09, see Table S10; loose block: Ftreatment(1,21) = 3.55, p = 0.07 see Table S11). The fixed block distractor differentially affected the latency to a reward in males and female rats (fixed block: Fsex*test(1,44) = 4.42, p = 0.04). Post hoc testing per sex showed that male rats took longer to collect the reward when the distractor was present but female rats did not (males: Ftest(1,23) = 67.26, p < 0.001; females: Ftest(1,21) = 0.97, p = 0.34, see Table S10).
Probabilistic reversal learning test
To assess cognitive flexibility in control and WMI animals, a probabilistic reversal learning test was used. Due to the large amount of parameters assessed, only the most outstanding effects of WMI and sex are described in the results (see Table S12-S20 for the complete results).
Over the course of 5 consecutive 120-trial sessions, the animals improved in performance in this task, they increased the proportion of trials that animals pressed the high probability lever (Fday(4,176) = 13.97, p < 0.001) (Fig. 2A). WMI and sex did not affect performance of the animals (Ftreatment(1,44) = 2.58, p = 0.12; Fsex(1,44) = 1.72, p = 0.20)(Fig. 2A and Table S12).
Animals of both groups successfully developed a strategy by pressing the initially high probability lever > 50% of the trials, switched to the other lever (initially low probability but now high probability of receiving a reward) when the probabilities changed and switching back when the contingencies changed back again (Fsession(2,88) = 48.00, p < 0.001; Fig. 2B/C, detailed statistics on the performance of animals during trials 1–40, 41–80 and 81–120 during sessions 1–5 can be found in table S12).
Reaction time in the task improved over days (Fdays(4,172) = 8.62, p < 0.001) and male rats where always faster than females (Fdays*sex(4,172) = 2.41, p = 0.05, see Fig. 2D). Reaction time differed between sexes and treatment (Fsex(1,43) = 16.45, p < 0.001; Ftreatment(1,43) = 3.87, p = 0.03; Fsex*treatment(1,43) = 4.62, p = 0.02). Post hoc testing revealed that female WMI rats were faster compared to female control rats (tfemale(20) = 2,96, p = 0.008) whereas males were equally fast in both treatment conditions (tmale(23) = − 0.30, p = 0.72). Both in control and WMI conditions, male rats were faster compared to female rats (tcontrol(26) = − 5.05, p < 0.001; twmi(17) = − 2.18, p = 0.04; Table S13).
Win-stay performance increased over days (Fdays(4,172) = 27.31, p < 0.001), as well as the number of pellets animals earned (Fdays(4,172) = 5.55, p < 0.001). No other effects were found for these parameters (see tables S14: win-stay and S15: pellets won). Lose-shift behaviour differed for the WMI and control animals over days (Fday*treatment(4,172) = 3.21, p = 0.01). Post hoc analysis showed that only on day 3, WMI animals shifted away more from the lever they were pressing after a loss compared to control animals (tday3(45) = − 2.00, p = 0.05), and on the other days, no differences were found (See table S16).
Finally, we fitted the trial-by-trial data to a Q learning model (Verharen et al. 2019) to assess the effects of WMI on the component processes underlying probabilistic reversal learning performance. A comparison of the best-fit model parameters revealed that animals demonstrated an increase in reward learning parameter α + over days (Fdays(4,120) = 4.00, p = 0.004), but there was no difference between groups, indicating a stronger impact of positive (rewarding) feedback on behaviour in later sessions in all animals. Post hoc comparisons showed that there was a reduced learning rate on day 1 compared to day 5 (t1vs5(45) = − 0.30, p = 0.01); i.e. all animals adapted their behaviour more strongly in response to positive feedback on day 5 as compared to day 1. No other effects on reward learning were found (Table S17). Learning from negative feedback (punishment learning rate α-) did not differ over days or between sexes or groups (Table S18). In addition, perseveration, measured as stickiness parameter π in the computational model (indicating a preference for the lastly chosen lever, independent of outcome), was not different over days or between sexes and groups (Table S19) and was close to 0. This indicates that the side chosen in the last trial did not affect the choice of the animal in the upcoming trial much, and thus, choices were based mostly on the value representation of the two options rather than perseverative behaviour on the same lever. The exploit/explore parameter (β), measuring the extent to which animals consistently chose the highest valued option, did not differ either over days or between sexes or groups (Table S20). Together, these results indicate that WMI and control animals behaved similarly in the task.
Absence of myelin deficits in PFC subregions
The prefrontal cortex is involved in cognitive functioning such as working memory, decision-making and inhibitory response control and attentional set-shifting (Dalley et al. 2004). Therefore, we explored possible myelination deficits in 3 subregions of the medial prefrontal cortex (mPFC) (dorsal prelimbic (dPrL), ventral prelimbic (vPrL) and the border of vPrL and infralimbic (IL) subregions) (Fig. 3). Figure 3A–C shows full slice scans, zoom-in and insets indicating the locations of the micrographs in these subregions of the mPFC. No differences were observed in myelination in the PFC subregions between control and WMI animals at the age of 9 months, i.e. after testing in the 5CSRTT (L1 prelimbic: t(13) = 1.06, p = 0.31; L2 infralimbic: t(13) = 1.53, p = 0.15; L3 dorsal penduncular: t(13) = 1.51, p = 0.16, Fig. 3D–G).
To rule out a possible positive effect of extensive training on myelination, a group of animals was subjected to the double-hit model and sacrificed at P30, the time point we previously observed myelination deficits in the sensory cortex. The data in Fig. 3H–K show that no significant differences in MBP + area were observed in WMI animals when compared to control animals in any of the 3 areas in the prefrontal cortex at P30 (L1 prelimbic: t(10) = 1.87, p = 0.10; L2 infralimbic: t(10) = 2.21, p = 0.05; L3 dorsal penduncular: t(10) = 1.41, p = 0.19). If anything, the WMI group showed a trend towards more myelination in the infralimbic cortex at this time point.
To verify that our double-hit model of WMI, using fetal inflammation plus postnatal hypoxia, resulted in myelination deficits in the sensory cortex as described earlier (van Tilborg et al. 2018a, b), we checked myelination at P30 in the sensory cortex (Fig. 4). Figure 4A–C shows full slice scans, zoom-in and inset indicating the locations of the micrographs in layer V of the sensory cortex. Figure 4D–E confirms that WMI animals have ~ 30% loss of cortical MBP + area compared to control rats at P30 in this brain area (t(10) = 1.06, p = 0.02). Next, we assessed whether the myelination deficits observed at P30 in the sensory cortex were still present in this area at 9 months of age. At this later time point, we did not observe differences in myelination between control and WMI rats anymore in the sensory cortex (t(13) = 2.01, p = 0.07; Fig. 4D/F), indicating a transient, delayed deficit in myelination after fetal inflammation plus postnatal hypoxia which restored over time.