Motivation under aversive conditions is regulated by a striatopallidal pathway in primates

All behavioral experiments were conducted under room temperature, and the monkeys performed the tasks in a soundproof, dark booth to minimize external noise and visual distractions. We trained the two monkeys to perform a variant of the Approach–Avoidance decision-making task (Ap–Av task, Figure 1B).³⁶ Precue-period: To initiate each trial, the monkeys were required to make a saccade to a white square fixation point (precue) at the center of the screen and keep the fixation for the specified time in order to continue the task. The required fixation time varied across sessions, ranging from a minimum of 0.8 seconds to a maximum of 5.2 seconds. Cue-period: Following the precue-period, a visual cue consisting of red and yellow bars appeared in the same location as the precue. The length of the red bar was directly proportional to the amount of reward delivery (water or isotonic drink), while the yellow bar length was directly proportional to the duration of the air-puff delivery. In each trial, the length of both bars was independently and pseudo-randomly varied across 1 to 100 steps. The monkeys were required to start a gaze at the cue within the time limit (2.0–3.0 s) and maintain the gaze during the cue presentation period (1.0–1.5 s). Response-period: After the cue-period, two choice targets (a white cross and a white square) appeared: one on the left and the other on the right of the cue. The placement of the targets was randomly altered in each trial for MK#2. The monkeys were required to indicate their decisions by making a saccade to either target within the target presentation period (2.0–4.0 s). If they gazed at the cross (+) target for the specified time (0.3–0.5 s), the choice was considered “approach”. If they gazed at the square (□) target for the specified time (0.3–0.5 s), the choice was considered “avoidance.” Outcome-period: When “approach” was chosen, a specific amount of water indicated by the length of the red bar was given to the monkeys. After the reward delivery, a pre-indicated duration of air-puff was delivered to the monkeys’ faces based on the length of the yellow bar shown during the cue. When “avoidance” was chosen, a fixed amount of minimal reward was given in order to maintain the monkeys’ willingness to perform the task. An inter-trial interval (ITI) was inserted after each trial for the specified time. If the monkeys successfully completed each trial, the current trial was counted as a 'successful trial'. However, if the monkeys made incorrect eye movements, such as ignoring the visual guidance (omission) or not maintaining eye fixation on it (break), the trial was terminated and categorized as a 'fixation error', depending on the phase in which the error occurred. In addition, as control to establish conditions sufficient to alter decision-making, we implemented “Rew+ / Air–” control sessions in MK#2 (Figure 5B). In these sessions, the reward per unit length of the red bar increased, and the air-puff per unit length of the yellow bar decreased during post-trials compared to pre-trials.

We also trained the monkeys to perform the Approach–Approach decision-making task (Ap–Ap task, Figure 1B). The procedure of the Ap–Ap task was identical to that of the Ap–Av task, except that both the red and yellow bars represented rewards. When the monkey chose the cross (+) target, a specific amount of water indicated by the length of the red bar was delivered as a reward. When the monkey chose the square (□) target, a specific amount of water indicated by the length of the yellow bar was delivered as a reward. We adjusted the amount of reward per unit length of the red bar to be 1.0–2.0 times greater than that of the yellow bar.

A pump (Diaphragm Precision Micro Injection Pump Smooth Flow Pump Q Series; QI-100-TT-P-S, Tacmina Corporation) regulated the amount of reward delivered through the water tube. This amount was controlled proportionally to analog signals (DC 4–20 mA) sent from a computer operating the task. An oil-free scroll air compressor (SLP-07EED, Anest Iwata) produced the compressed air used for air-puff. The delivery duration of the air-puff was regulated by a solenoid valve (AB31-02-3, CKD) that was activated by a solid state relay (phototriac; applied output load: 10 A at 24–240 VAC; with zero cross function and screw terminal; G3NA-210B-UTU, Omron). This relay received analog signals of DC 5 V from the task computer, transmitted via transistor-transistor logic (TTL). Eye movements were monitored using an infrared light camera and eye-tracking software (ViewPoint Eye Tracker; Monocular; 400Hz; MCU400, Arrington research). The tasks were controlled through a NIMH-MonkeyLogic system (National Institute of Mental Health). Analog signals of the eye movements were collected with task event markers in real time through Plexon's OmniPlex system (Plexon).

After training the monkeys MK#1 and MK#2 in the tasks, we conducted a sterile surgery to implant a plastic recording chamber on each monkey's skull using bone cement and ceramic screws. Prior to the surgeries, we performed magnetic resonance (MR) imaging with T1-weighted turbo spin echo (0.3 Tesla, 1 mm slice thickness; 7 Tesla, 0.3 mm slice thickness) to identify stereotaxic coordinates of target brain regions. For both MR imaging and surgeries, we anesthetized the monkeys with intramuscular injections of ketamine (5–8 mg/kg) and atropine (0.005–0.025 mg/kg) with xylazine (0.2 mg/kg). Anesthesia was maintained by inhalation of 1–2.5% sevoflurane or 1–2% isoflurane with 1.5 L O₂. Daily antibiotic injections were administered starting one day before surgery and continued for at least one week afterward. After full recovery, a second surgery was performed under anesthesia and aseptic conditions, using the same procedures as described above, to remove the skull over the target regions.

After the chamber implantation, we performed in vivo single-unit recordings of neuronal activities in the VS and VP while the monkeys performed both the Ap–Ap and Ap–Av tasks in alternating blocks. The transition between each block of these tasks occurred without prior notification after a preset number of successful trials. All recordings were conducted under room temperature conditions, and the monkeys performed the tasks inside a soundproof, dark booth to minimize external visual and auditory stimuli.

We carefully inserted a sterile glass-coated electrode with an impedance of 0.5–2.5 MΩ (Alpha-Omega) through a hole, spaced at 1 mm intervals, on a plastic grid placed over the recording chamber, using an oil hydraulic micromanipulator (MO-97A, Narishige) and determined its location through the aid of MR images. Digitized and amplified neural signals were collected along with task event markers in real time through the OmniPlex system (Plexon). We classified neural signals into single-unit activities by using offline Sorter (Plexon). Action potentials were aligned and classified according to task event markers, using NeuroExplorer (version 5.445; Plexon). For further detailed analyses of firing rates and spike widths, we employed MATLAB 2024b (MathWorks). Data from fixation error trials, including omissions and fixation breaks, were excluded from analysis.

We utilized a type of inhibitory DREADDs known as hM4Di (i.e., a mutated inhibitory G-protein-coupled human muscarinic receptor 4), which suppresses neuronal activity upon activation and effectively silences the targeted neurons. We produced a mosaic adeno-associated viral vector (AAV2.1-CaMKIIα-hM4Di-IRES-AcGFP) optimized for high transgene expression in the primate brain as described previously³⁸ with the aim of introducing hM4Di receptors into GABAergic MSNs within the VS. We employed CaMKIIα (Calcium/calmodulin-dependent protein kinase II subunit α) promoter that enabled hM4Di receptors to express specifically in MSNs of the VS.³⁹ We conducted a surgery for the viral vector injections under anesthesia and sterile conditions in a fully equipped operating room. We injected AAV2.1-CaMKIIα-hM4Di-IRES-AcGFP (2.0 x 10¹³ gc/ml for monkey MK#1; 2.5 x 10¹³ gc/ml for monkey MK#2) bilaterally into the VS (Figure 1A; Table S1). The target regions were estimated using MR imaging and in vivo single-unit recordings, based on the relative position of holes on the recording grid. For monkey MK#1, the injections were manually performed into the right and left hemispheres through the grid holes, using a beveled needle of a 10 μL Hamilton microsyringe. For monkey MK#2, the injections were conducted into the right and left hemispheres through a stainless cannula connected to a 25 μL Hamilton microsyringe via fused silica tubing, controlled by a syringe pump (KD Scientific; WPI). The total amount of viral aliquots injected for monkey MK#1 was 7 μL into the left hemisphere and 8.2 μL into the right hemisphere (1.0–2.2 μL per track, four tracks for each hemisphere). Immunohistochemical analysis of hM4Di-AcGFP expression in MK#1 revealed that several injection tracks did not result in detectable receptor expression (Figures 1A and S4). Despite this partial expression, chemogenetic manipulation produced clear behavioral effects in MK#1. Therefore, in MK#2, we adjusted the total injection volume to 3 μL per hemisphere (1.5 μL per track, two tracks for each hemisphere) in order to achieve a comparable level of receptor expression. We waited at minimum 10 weeks after the viral vector injections to ensure valid expression of the hM4Di receptors.⁹²

In order to silence the hM4Di receptors, we employed deschloroclozapine (DCZ), a novel and highly selective actuator.⁴³ We dissolved DCZ (HY-42110A, MedChemExpress) in dimethyl sulfoxide (DMSO; 13408-64; CAS# 67-68-5, Nacalai Tesque) at a concentration of 1 mg DCZ per mL of DMSO. These stock solutions were stored in a –80°C deep freezer. We prepared a fresh working solution on the day of usage by diluting the stock solution in saline to concentrations of 342.02 or 684.04 nM. We used an injectrode (tungsten electrode–silica capillary assembly; Unique Medical Corporation) consisting of a tungsten electrode (diameter, 200 μm; 0.2–1.5 MΩ) and a silica capillary tube (outer diameter, 150 μm), enclosed in a polyimide tube (outer diameter, 480 μm) and connected to a 25 μL Hamilton microsyringe controlled by a syringe pump (KD Scientific; WPI). This injectrode enabled us to infuse DCZ into the VP while recording neuronal activities from this region. We placed the injectrode in the VP, 1.0–1.5 mm deeper than the ventral edge of the anterior commissure (AC–VP border) (Figures S1A–S1C). We maintained a consistent task type throughout the week for the monkeys to perform, either the Ap–Av or Ap–Ap task, and infused DCZ, or vehicle, once a week when their performance consistently stabilized over two consecutive days. After the monkeys completed a set number of pre-trials (150 for MK#1; 100 for MK#2), we locally infused the working solution of DCZ into the VP. For monkey MK#1, the DCZ infusion was performed bilaterally, delivering 1.5–2.0 μL per hemisphere at a rate of 0.15–0.2 μL/min (Table S2). However, immunohistochemical analysis of hM4Di-AcGFP expression in MK#1 revealed asymmetric expression between hemispheres, raising the possibility that the observed behavioral effects in MK#1 may have resulted predominantly from unilateral modulation. To examine the impact of unilateral manipulation, DCZ infusion in MK#2 was carried out unilaterally, with 0.7–1.0 μL delivered into a single hemisphere at a rate of 0.1 μL/min. The task resumed approximately 5–13 minutes after the infusion was completed. As previously reported,⁴³ we also verified that DCZ administration alone was not associated with significant alterations in goal-directed behavior in the non-DREADD control monkey (n = 1; MK#2), in which bilateral DCZ infusion into the VP was performed prior to viral vector injection (342.02 nM, 2.0 μL per hemisphere; Figures S1D–S1F).

For histological identification, we inserted a monopolar stainless steel electrode (120 mm length with PIN; 1–1.5MΩ; UESLGESEXN1M, FHC) to make iron deposits in the VP region of monkey MK#1, where DCZ had been infused via the injectrode (Figure S1C). Using the same procedure as in in vivo single-unit recordings, we positioned the electrode at the AC–VP border. Electrolytic deposition of iron ions occurred at the electrode tip by applying a negative current of 10 μA for 10 s. Following the deposition, we waited approximately for 3 min. On the day following the deposition, monkey MK#1 was administered a deep anesthesia using an overdose of secobarbital sodium (24 mg/kg, i.v.). Subsequently, transcardial perfusion was performed, first using 0.1 M phosphate-buffered saline (PBS; pH 7.4), followed by 10% formalin in 0.1 M PBS. Following an overnight immersion in 10% formalin at 4°C, the brain underwent a gradual saturation process with sucrose in 0.02% sodium azide (Sigma) in 0.1 M PBS. This process consisted of increasing sucrose concentration from 10% over five nights, then to 20% over four nights, and finally to 30% over ten nights. The brain was frozen with dry ice and sliced into 40-μm-thick coronal sections using a freezing sliding microtome. Sections were preserved in 0.02% sodium azide in 0.1 M PBS.

To visualize immunoreactive signals of hM4Di-AcGFP, sections underwent a series of steps. After rinsing 3 times for 5 min in 0.01 M PBS, sections were pre-treated with 3% H₂O₂ for 20 min. After 5 times rinse in 0.01 M PBS containing 0.2% Triton X-100, the sections were pre-treated in tyramide signal amplification (TSA) blocking reagent (NEL700A001KT, AKOYA Biosciences) in 0.01 M PBS for 30 min, and then incubated with rabbit anti-GFP antibody (1:1000; G10362; AB_2536526, Invitrogen) in the same fresh medium for 60 min at RT and subsequently overnight at 4°C. After 5 times rinse in PBS, the sections were incubated with anti-rabbit polymer HRP (GTX83399, GeneTex) solution for 1 h. After rinsing 5 times, the sections were treated in TSA biotin working solution (NEL700A001KT, Akoya Biosciences) in TSA blocking reagent for 15 min. The sections were rinsed 5 times and then incubated in TSA blocking reagent containing Streptavidin Alexa Fluor 488 (1:1000; S11223, Invitrogen) for 60 min. We rinsed the sections 3 times in 0.01 M PBS, mounted onto glass slides coated with gelatin, and air-dried. We then coverslipped the sections using ProLong Diamond Antifade Mountant with DAPI (P36971, Invitrogen). Immunoreactive signals in the regions of interest were observed and digitally imaged with a confocal fluorescence microscope (BZ-X800; Keyence Corp., IL, USA) fitted with 2×, 4×, 10×, and 40× objectives. The images were viewed and processed using BZ-X analyzer software (Keyence Corporation, IL, USA) and exported as TIFF images.

Behavioral and electrophysiological experiments were replicated in two monkeys (MK#1 and MK#2) across multiple sessions. Because the experimenter performed real-time electrophysiological recordings and DCZ infusions, blinding during data collection was not feasible. However, behavioral and neuronal data were analyzed using automated scripts without reference to condition identity to minimize bias. Sample sizes were not predetermined by formal power analysis. Numbers of animals (two macaques), sessions, and recorded units were determined based on feasibility in nonhuman primate experiments. The within-subject, session-paired design increased statistical power, and all key effects were replicated across both animals and multiple sessions.

All statistical analyses were performed at the session level. The experiments followed a weekly paradigm in which each DCZ session was counterbalanced and flanked by control sessions conducted on the preceding and following days. Because task performance was relatively stable from day to day but exhibited slower fluctuations across weeks, each DCZ session was paired with the immediately preceding control session to minimize week-to-week drift (within-subject, matched task conditions).

Given the modest number of sessions and the possibility that the assumption of normality might not hold, both parametric and nonparametric tests were performed. Paired t-tests were used for within-subject comparisons between control and DCZ sessions, and two-sample t-tests were used for independent comparisons. In parallel, nonparametric Wilcoxon signed-rank tests were applied to confirm the robustness of these results, and Wilcoxon rank-sum tests were used for between-session summary statistics where appropriate. All tests were two-sided, and exact p-values are reported. Statistical significance was defined as P < 0.05.

To quantify motivational changes induced by VS–VP pathway manipulation, we analyzed fixation errors occurring during the precue period (precue-FEs; Figures 2B–2E, S2A, and S2B), which reflect failures to initiate goal-directed behavior. Because session lengths (total trial numbers) varied across control and DCZ sessions, the accumulated number of precue-FEs in each session was computed up to the minimum trial count of each control–DCZ pair to ensure matched session lengths. For each monkey, the total accumulated precue-FEs were then compared between control and DCZ sessions (Figures 2C and 2E, left).

To examine how the effect of DCZ evolved over time within a session, trial progression was normalized to a common 0–100% scale, where 0% and 100% corresponded to the first and last trials up to the pairwise minimum trial count. This truncation and normalization procedure aligned data across sessions and minimized potential influences of differences in session length or total trial number. The cumulative number of precue-FEs was then plotted as a function of normalized trial progression, and trajectories were compared between control and DCZ sessions (Figures 2C and 2E, right). See also Figures S2C and S2D for cue-FEs, response-FEs, and total number of FEs.

To determine whether behavioral initiation was influenced by previous trial history, we analyzed how initiation failures (i.e., precue-FEs) depended on the outcome and valuation of the preceding trial (lag-1). For each trial, the following predictors were extracted from the immediately previous trial: (i) previous error (binary: 1 = error, 0 = successful), (ii) previous reward magnitude, and (iii) previous aversive magnitude. Cues were pseudo-randomized, ensuring that these variables were statistically independent of the offer of the current trial.

Within each session, the probability of an initiation failure on trial t was modeled using a logistic regression including the three lag-1 predictors:

Statistical significance for each coefficient was assessed with a two-sided Wald test (P < 0.05). To provide a model-free summary of the serial-error effect, we computed the difference in conditional probabilities between consecutive trials:which quantifies how often initiation failures tended to repeat.

Finally, repetitive errors (consecutive error-to-error sequences) were counted for each session and compared between control and DCZ conditions using Wilcoxon signed-rank tests. All analyses were conducted separately for the Ap–Av and Ap–Ap tasks, and results were pooled from the two monkeys for population summaries (Figure 3).

Reaction time (RT) during the response period was defined as the latency from target onset to target selection. Each DCZ session was paired with a corresponding control session recorded within the same week and under the same task condition (Ap–Av or Ap–Ap).

For the choice-dependent RT analysis (Figures 4A and 4B), RTs were categorized according to the chosen target: in the Ap–Av task, selection of the cross (+) target indicated acceptance of the combined offer (approach, Ap), whereas selection of the square (□) target indicated rejection of the offer (avoidance, Av). In the Ap–Ap task, cross and square targets corresponded to the reward magnitudes indicated by the red and yellow bar lengths, respectively. For each session, mean RTs were calculated separately for each choice type. The effects of DCZ infusion were evaluated by comparing the mean RTs for each choice type between DCZ and control sessions using the Wilcoxon signed-rank test (two-sided). This procedure was conducted separately for the Ap–Av and Ap–Ap tasks to determine whether VS–VP pathway suppression differentially affected approach- and avoidance-related responding.

To examine how RT modulation depended on cue offers, RTs were visualized as two-dimensional offer matrices (Figures 4C and 4D). Matrix analyses were performed both for individual monkeys and on pooled data, to characterize how the pattern of RT modulation varied across cue offers. In the Ap–Av task, the x- and y-axes represented the offered reward amount and air-puff duration, respectively, whereas in the Ap–Ap task, they represented the two offered reward magnitudes associated with the cross and square targets. After pooling all trial-wise RT data across sessions, we computed local mean RTs on a two-dimensional offer grid by convolving the data with a two-dimensional square kernel (20% × 20% window). This procedure effectively smoothed the data and reduced sampling noise, providing a continuous map of RT modulation across all offered conditions. For each offer cell, two-sample t-tests were applied to compare pre- versus post-infusion trials. Regions showing significant differences (P < 0.05) were highlighted on the heatmaps. The decision boundary estimated from the behavioral choice model (see STAR Methods section choice model) was overlaid on the Ap–Av RT maps to relate RT modulation to cue combinations that favored Av choices (left of the boundary) or Ap choices (right of the boundary). The boundary itself was derived independently of the RT data. All analyses were implemented in MATLAB 2024b (MathWorks).

Pupil diameter was recorded together with eye position using an infrared eye-tracking system (ViewPoint Eye Tracker; Monocular; 400 Hz; MCU400, Arrington Research). Analyses focused on the precue period, defined as the interval from the onset of precue fixation to cue presentation.

The change in Ap and Av choices between the pre- and post-trials after DCZ infusion was quantified using the following method (Figure 5A). Each decision was mapped onto a decision matrix and convolved with a 20 × 20 point square window. After convolution, each choice datum was stacked in each cell of the 100 × 100 point decision matrix. We calculated the t-statistics for each point using the stacked choice data and employed Fisher's exact test to determine statistically significant differences (P < 0.05 at each point) between the two blocks. The increase in Av choices between the pre- and post-trials was represented by the total area of points in the decision matrix showing a significant increase in Av (%ΔAv). Similarly, the increase in Ap choices was represented by the total area of points exhibiting a significant increase in Ap (%ΔAp).

To characterize the choice pattern, we performed econometric modeling (Figure 5C). Specifically, the probability of choosing the cross target (p_AP) was calculated using the logistic function p_AP = 1 / (1 + exp(–f(x, y))). The function was parameterized as f(x, y) = ax + by + c, where x and y represent the lengths of the red and yellow bars, respectively. The coefficients a, b, and c were determined through generalized linear regression fitted to the behavioral choices in the pre-trials. We performed a normalization procedure on the choice data in both the pre- and post-trials. For the dataset that produced the decision boundary ax + by + c = 0, we applied the procedure described previously^36,50 to align it with the standard decision boundary a₀x + b₀y + c₀ = 0. The standard decision boundary was set through the points (20, 0) and (50, 100). For data to the left of the decision boundary, x was resized as x' = x ^∗ X_a / X_b, where X_a = –(c + by) / a and X_b = –(c₀ + b₀y) / a₀. For data to the right of the decision boundary, x was resized as x' = 100 – (100 – x) ^∗ X"_a / X"_b, where X"_a = 100 + (c + by) / a and X"_b = 100 + (c₀ + b₀y) / a₀. This procedure ensured that the decision boundary derived from the transformed data corresponded to the standard decision boundary. We applied this normalization procedure to the choice data in the pre- and post-trials independently. After normalization, session data were pooled per monkey, mapped onto the decision matrix, and convolved with a 20 × 20 point square. We conducted a t-test at each time point to evaluate the differences between pre- and post-trials. The resulting t-values reflect the magnitude of change in Ap and Av choices (Figure 5D).

To assess neuronal correlates of motivational initiation, analyses of VS and VP activity focused on the precue period, during which monkeys were required to maintain fixation on the central precue until cue presentation (Figures 6A and S5). Each recording session consisted of alternating blocks of Ap–Ap and Ap–Av tasks, allowing the same neuron to be recorded over multiple unsignaled task transitions. Only stably isolated units from sessions containing at least two unsignaled task transitions were included, enabling within-cell adaptation analysis.

For each neuron, the mean firing rates during the Ap–Av and Ap–Ap tasks were compared. Differences were expressed as t-values and plotted as histograms to visualize population distributions (Figure 6B). The population mean of t-values was tested against zero to determine whether overall activity showed systematic task-related bias. For classification, the same trial sets used for population analyses were compared, and the task preference of each neuron was expressed as the t-value of the firing rate difference (Ap–Av – Ap–Ap). Neurons with significantly higher firing during Ap–Av trials were defined as “Ap–Av type,” and those with higher firing during Ap–Ap trials as “Ap–Ap type” (t-test, P < 0.05; Figure S6). Neuron counts and recording locations for significantly biased populations are shown in Figures S4C and S4D. See Figures S7A–S7H for firing activity of these task-specific biased populations across variables: trial number (low, 0–50%; high, 50–100%), reward magnitude (low, 0–50%; high, 50–100%), air-puff magnitude (low, 0–50%; high, 50–100%), and choice target.

For each task-specific group, trial-by-trial firing rates were ratio-normalized by dividing activity by the mean firing rate of that neuron across all trials, allowing comparison across neurons with different baseline activity levels. Normalized precue activities were computed for each of the four alternating task blocks and visualized as box plots showing the median, interquartile range, and outliers beyond 1.5 × IQR (Figures 6C and 6E).

To characterize adaptation dynamics during block transitions, trial-wise mean firing rates were computed for each neuron (Figures 6D and 6F). For each post-transition trial, firing rates were compared with the mean of the 30 pre-transition trials (t-test, P < 0.05). Trials showing significant differences were marked individually (black circles), and consecutive significant trials were clustered (two consecutive, pink; ≥3 consecutive, pink-shaded intervals).

To examine how neural activity reflected previous-trial outcomes, we analyzed correlations between each neuron’s firing rate during the precue period and the outcome of the immediately preceding trial (0 = successful, 1 = error; Figures 7 and S7I). Errors included those occurring during the precue, cue, and response periods; similar results were obtained when the analysis was restricted to precue fixation errors. For each neuron, the mean firing rate within the precue period (from fixation onset to cue onset) was computed across all trials and correlated with the binary outcome variable using Pearson’s correlation coefficient (r). Statistical significance was determined with a two-sided test (P < 0.05). The distribution of r values across neurons was tested against zero using a Wilcoxon signed-rank test.

For neurons showing significant correlations, time-resolved firing activity was further analyzed by aligning spike trains to cue onset and computing normalized firing rates in sliding 100 ms bins (50 ms step). Firing rates at each time bin were normalized by dividing by the mean firing rate during the cue period. This normalization enabled comparison of relative modulations across neurons with different baseline activity levels and facilitated the detection of sustained effects extending into the cue period. Population averages were then computed across neurons within each condition, separately for trials following error and compete outcomes. Periods showing significant differences the two conditions were identified using two-sided t-tests (P < 0.05) and are indicated by shaded yellow bands in Figure 7.