During passive fixation, LC neurons were weakly active, with no spikes measured on more than a third of all trials and otherwise a median (interquartile range [IQR]) firing rate of 1.8 [0.9–3.6] sp/s measured during 1.1 s of stable fixation (Figure 2A). For each session, we divided trials into six groups based on the given LC neuron’s firing rate: trials with firing rate = 0 (LC_zero), 1, 2, 3, and ≥4 sp/s formed five groups, and all trials with firing rate >0 formed a sixth group (LC_non-zero). We then assessed ACC neural activity measured at the same time, conditioned on the LC group.

The mean and variance of ACC spike counts, along with their Fano factor (the ratio of variance over mean), increased steadily as a function of the duration of the counting window. These trends showed an apparent, small relationship with simultaneously measured LC spiking activity, such that the mean, variance, and Fano factor measured in the ACC were, on average, slightly higher for LC_zero versus LC_non-zero trials (Figure 2B–D). However, these relationships were not statistically reliable when considering data from both monkeys combined together and only in one case (Figure 2G) when considering data separately for each monkey (sign-rank test for H₀: median difference between LC_zero and LC_non-zero conditions computed per ACC unit, or ANOVA test for an effect of LC firing rate group, p > 0.05 in all cases; Figure 2B–G). Across ACC units, differences in Fano factor for LC_zero versus LC_non-zero trials were associated strongly with differences in spike-count variance and more weakly with differences in spike counts (Figure 2—figure supplement 1).

In addition, the mean, variance, and Fano factor of LC spike counts also increased steadily as a function of the duration of the counting window. However, none of these trends showed a statistically significant relationship to simultaneously measured ACC spiking activity (sign-rank test and ANOVA, p > 0.05 in all cases; Figure 2—figure supplement 2).

During passive fixation, pairs of neurons in ACC had coordinated spiking activity that varied considerably in sign and magnitude across pairs but also, as has been reported previously, depended systematically on the size of the time window used to count spikes (de la Rocha et al., 2007). Specifically, pairwise spike-count correlations (r_sc) in ACC had median [IQR] values of 0.03 [–0.01 0.07] using 200 ms counting windows and systematically increased and became more variable across pairs with larger counting windows up to 1 s (0.05 [–0.03 0.14]; ANOVA test for effect of window size, p = 0.0243).

These ACC spike-count correlations also depended on the spiking activity of LC neurons measured at the same time. Specifically, the correlations were largest when (1) they were measured using relatively large time windows; (2) for ACC pairs that exhibited relatively large, positive correlations independent of LC firing; and (3) when the simultaneously recorded LC neuron was not active (Figures 3 and 4). For the example ACC pair illustrated in Figure 3A–C, r_sc was near zero when computed using relatively small time windows and then increased steadily with increasing bin sizes. The magnitude of these increases in r_sc with bin size was larger on trials in which the simultaneously measured LC neuron was not spiking versus spiking (Figure 3C).

The population of ACC pairs exhibited similar trends, with r_sc values tending to be smaller on LC_non-zero versus LC_zero trials, particularly for larger time windows and for ACC pairs with non-negative LC-independent r_sc values. To visualize these effects, we first divided ACC pairs into terciles, according to their LC-independent r_sc values (Figure 4). Under these conditions, ACC r_sc values from the upper two terciles of LC-independent r_sc values were reduced by up to ~50% when computed on trials with at least one or more LC spike versus trials with no LC spikes (Figure 4D–F). The magnitude of these reductions was not related in a consistent manner to the magnitude of LC_non-zero activity (ANOVA test for group effect of LC firing rates > 0). These relationships were not evident on shuffled trials, supporting the idea that the trial-by-trial relationships that we identified were not spurious reflections of the spiking statistics of each region considered separately (Figure 4—figure supplement 1). Furthermore, these LC-linked differences in ACC r_sc values did not result simply from changes in ACC neuron firing rates because the two measures were unrelated (Figure 4—figure supplement 2A). We found a weak but statistically reliable relationship between LC-linked changes in ACC r_sc and changes in ACC Fano factor (Figure 4—figure supplement 2B).

In contrast, spike-count correlations (r_sc) from simultaneously measured pairs of LC neurons did not depend systematically on concurrently measured ACC activity. In general, ACC-independent LC r_sc values were roughly similar to LC-independent ACC r_sc values, increasing steadily and becoming more variable as a function of the counting-window size; for example, median [IQR] values were 0.02 [0 0.09] for 200 ms windows and 0.06 [–0.02 0.19] for 1 s windows. However, these values did not show reliable differences when compared on trials with high versus low ACC firing, analyzed in the same way as the LC-linked ACC r_sc values (Figure 4—figure supplement 3). Moreover, the distributions of these ACC-linked LC r_sc values did not appear to come from the same (shifted) distribution as the LC-linked ACC r_sc values for all five time bins (Kolmogorov–Smirnov test for H₀: both sets of values come from the same distribution, p < 0.0189 in all five cases). Thus, differences in single-unit LC activity were associated with differences in coordinated activity patterns of pairs of neurons in ACC, but differences in single-unit ACC activity were not associated with differences in coordinated activity patterns of pairs of neurons in LC.

To further examine relationships between coordinated activity in ACC and activation of LC-linked arousal systems, we examined both relative to changes in pupil diameter. During passive fixation, there is a quasi-periodic fluctuation of the pupil (Lowenstein and Loewenfeld, 1969; Pong and Fuchs, 2000; Joshi et al., 2016). Consistent with our previous findings (Joshi et al., 2016), the timing of LC spiking was related to the phase of these ongoing pupil fluctuations, such that LC spiking tended to be higher preceding dilation versus constriction. These modulations showed a systematic precession with dilation phase that corresponded to a delay of ~270 ms from the maximum modulation of LC activity to the relevant pupil change (Figure 5A; median of intercepts from linear regression of per-time-bin peaks of individual LC PETHs; bootstrapped 95% confidence interval from regression for each LC neuron = [177–353 ms]; for examples of pupil cycles and measurement epochs, see Figure 5—figure supplements 1 and 2, respectively).

Combined with our current findings that pairwise correlations in ACC tended to be smaller when LC is active during passive fixation, these results predicted that pairwise correlations in ACC should also vary systematically with the phase of ongoing fluctuations in pupil diameter. We found such a relationship between pupil phase and ACC r_sc values. Specifically, over a wide range of pupil phase, decreases in ACC r_sc followed increases in LC spiking (the fitted line in panel B is shifted to the right relative to the fitted line in A; note also that increases in ACC r_sc, seen as the bright yellow band in the lower-left corner of panel B, occurred even earlier and are examined in more detail in Figure 9). These modulations showed a systematic progression with pupil phase over roughly the same time frame as LC firing, implying that modulations of both LC firing and ACC r_sc had relatively fixed temporal relationships to pupil fluctuations and therefore to each other, as well (Figure 5B; Wilcoxon rank-sum test for H₀: equal LC PETH and ACC r_sc slopes, p = 0.23).

We further examined relationships between LC and ACC neural activity in the context of external events that can cause a startle response. We showed previously that a brief, loud sound played on randomly selected fixation trials (‘beep trials’) can elicit a transient pupil dilation as well as transient responses from individual LC neurons (Joshi et al., 2016; evoked responses from an example LC neuron are shown in Figure 6A, and the LC population average from three monkeys is shown in Figure 6B). The beep stimulus also elicited a consistent, albeit weaker, response in the ACC (evoked responses from an example ACC neuron are shown in Figure 6D, and the ACC population average from two monkeys is shown in Figure 6E). The averaged ACC response included a slight excitation and then inhibition relative to baseline within 200 ms of stimulus presentation. Although we found some individual cases with a biphasic response (e.g., Figure 6D), the average response likely resulted from the range of heterogeneous responses of individual ACC neurons (Figure 6—figure supplement 1).

The startle events also caused transient reductions in the trial-to-trial variability (measured as the Fano factor) of spiking activity of individual neurons in both LC and ACC, with a slightly more sustained effect in the ACC (Figure 6C and F). This reduction in neuronal variability following the onset of a stimulus, also referred to as quenching, has been reported previously for neurons in a range of other cortical regions and for a range of task conditions (Churchland et al., 2010).

In the ACC, we found no evidence that the startling sound caused systematic changes in coordinated activity when measured independently of the LC response. Specifically, spike-count correlations were highly variable across ACC neuron pairs, with no statistically reliable, systematic differences when compared before versus after the beep (comparing ACC r_sc before versus after beep for spikes measured in 1 s bins; Mann–Whitney U-test, p > 0.05 for both monkeys).

In contrast, we found systematic relationships between the beep-induced changes in coordinated activity in the ACC and the simultaneously measured LC response. Specifically, LC neurons did not show a characteristic transient response on every presentation of a startle stimulus: the median fraction of trials with a response that included a transient increase and then decrease in spiking activity was 54% for monkey Sp and 43% for monkey Ci (Figure 7—figure supplement 1). Therefore, we tested if and how arousal-linked changes in ACC correlated activity depended on whether or not the simultaneously recorded LC neuron responded transiently to the startling sound. Specifically, we computed ACC r_sc in time bins of different sizes, separately for 1 s preceding and 1 s following the sound. ACC r_sc tended to be larger after versus before the beep stimulus, but only for the subset of trials in which the beep also elicited the characteristic LC response (Figure 6A and B), for both monkeys (Figure 7A). These modulations did not reflect simply transient differences in LC firing rates but rather specific differences between the presence and absence of the stimulus-evoked LC response (Figure 7B). These modulations of ACC pairwise correlations also did not reflect LC-driven differences in the stimulus-driven quenching of response variability in ACC, which was similar on trials with and without LC-evoked responses (Figure 7—figure supplement 2).

We showed previously that the magnitude of beep-evoked LC responses is correlated positively with the size of the simultaneously evoked change in pupil diameter (Joshi et al., 2016). We reproduced those results in our current data set (Figure 8—figure supplement 1). We extended those results by showing that these pupil-LC relationships also relate systematically to beep-evoked changes in ACC r_sc. Specifically, we measured beep-evoked changes in ACC r_sc separately for trials grouped by LC and pupil responses, focusing on how large or small pupil responses related to the evoked-LC versus no-evoked-LC differences in ACC r_sc shown in Figure 7. In general, the beep-evoked changes in ACC r_sc that required an LC-evoked response were larger on trials in which the evoked LC response was accompanied by a large versus small pupil dilation (Figure 8A). These effects were not evident when considering baseline, not evoked, pupil size (Figure 8B).

To better understand the relative timing of firing patterns (both individual neuron firing rate changes and paired-neuron r_sc changes) in LC and ACC, we measured their time courses relative to two external events with well-defined timing: onset of stable fixation on no-beep trials and onset of the sound on beep trials (Figure 9). For no-beep trials, LC firing rates increased after the fixation point turned on and before stable fixation was acquired (Figure 9A). This epoch also included elevated ACC r_sc (particularly when the simultaneously recorded LC neuron was not active, consistent with the results shown in Figure 4) that preceded the elevation in LC firing (Figure 9B, gray lines). Both LC firing and ACC r_sc tended to stabilize after stable fixation was attained. These results are consistent with the possibility that LC activation prior to fixation onset drives (or at least coincides with) a lasting decrease in ACC r_sc during the remainder of the trial.

Following a beep, there was a different pattern of relationships between LC firing and ACC r_sc. Specifically, we divided trials that did (Figure 9C and D) or did not (Figure 9E and F) have a beep-evoked LC response (like in Figures 7 and 8). In general, the beep caused a transient increase in ACC r_sc, and this increase tended to occur more rapidly on trials with (Figure 9D, gold shaded area) versus without (Figure 9F, gold shaded area) a concomitant LC response (Mann–Whitney U-test, for H₀: the linear rate of increase of ACC r_sc in this period was larger when LC neuron did versus did not respond, p = 0.0297). Note that the LC transient response (Figure 9C) peaked at the beginning of the epoch in which the ACC r_sc started to increase, which is consistent with the idea that the LC-evoked response drives (or at least coincides with) a faster rate of ACC r_sc increase (steeper slope in the gold region in Figure 9D compared with F). The evoked LC response was also associated with a more reliable and larger increase in ACC r_sc later in the trial (compare Figure 9D and F; Mann–Whitney U-test for H₀: ACC r_sc 500–1000 ms after the beep was the same on trials with versus without a beep-evoked LC response, p = 0.0350 for both monkeys’ data pooled together) but was not associated with differences in ACC r_sc in the time just before the beep (i.e., ACC r_sc did not anticipate whether or not the beep would evoke an LC response; Mann–Whitney U-test for H₀: ACC r_sc –500–0 ms before the beep was the same on trials with versus without a beep-evoked LC response, p = 0.0693 for both monkeys’ data pooled together).

Each monkey was implanted with a recording cylinder that provided access to LC+ (the LC and adjacent, NE-containing subcoeruleus nucleus; Sharma et al., 2010; Paxinos et al., 2008; Kalwani et al., 2014), inferior colliculus (IC), and superior colliculus (SC). The detailed methodology for targeting and surgically implanting the recording cylinder and then targeting, identifying, and confirming recording sites in these three brain regions is described in detail elsewhere (Kalwani et al., 2014; Joshi et al., 2016). Briefly, the LC was targeted initially using custom (Kalwani et al., 2009) and/or commercial software (Brainsight). Tracks were then refined using electrophysiological recordings and microstimulation in brain regions dorsal to LC. Neurons in the intermediate layers of SC (SC_i) exhibited spatial tuning on a visually guided saccade task and could elicit saccades via electrical microstimulation (Robinson, 1972; Sparks and Nelson, 1987). IC units exhibited clear responses to auditory stimuli. Activity in the trigeminal mesencephalic tract (me5), located immediately dorsal to the LC, showed distinct activity related to orofacial movements such as sipping. LC+ units had relatively long action potential waveforms, were sensitive to arousing external stimuli (e.g., door knocking), and decreased firing when the monkey was drowsy (e.g., eyelids drooped; Aston-Jones et al., 1994; Bouret and Sara, 2004; Bouret and Richmond, 2009). Nonoptimal tracks also helped with mapping; for example, a more medial track could miss me5 but lead to the trochlear decussation with characteristic ramp-and-hold activity related to downward saccades. Likewise, tracks that encountered IC but not SC_i were likely too lateral and often missed me5 and LC. Sites were verified using MRI and assessing the effects of systemic injection of clonidine on LC+ responses in monkeys Oz and Ci and by histology with electrolytic lesions and electrode-tract reconstruction in monkey Oz (Kalwani et al., 2014). Recording and microstimulation in these brainstem-targeting tracks were conducted using custom-made electrodes (made from quartz-coated platinum-tungsten stock wire from Thomas Recording) and a Multichannel Acquisition Processor (Plexon, Inc).

ACC cylinders were placed at Horsley–Clarke coordinates 33 mm anterior-posterior (AP), 8 mm lateral (L), in the left hemisphere for monkey Sp and in the right hemisphere for monkey Ci. For ACC recordings, we targeted the dorsal bank of the anterior cingulate sulcus, ~4–6 mm below the cortical surface. ACC tracks were planned and refined using MRI and Brainsight software, as well as by listening for characteristic patterns of white and gray matter during recordings in an initial series of mapping experiments. Recordings were conducted using either custom-made single electrodes or multicontact linear electrode arrays (8- and 16-channel V-probe, Plexon).

For each brain region, we recorded and analyzed data from all stable, well-isolated units that we encountered. Neural recordings were filtered between 100 Hz and 8 kHz for spikes (Plexon MAP). Spikes were sorted offline (Plexon offline sorter). Electrical microstimulation in SC_i consisted of biphasic (negative-positive) pulses, 0.3 ms long, 100 ms in duration, and delivered at 300 Hz via a Grass S-88 stimulator through a pair of constant-current stimulus isolation units (Grass SIU6) that were linked together to generate the biphasic pulse.

For each recorded neuron, we considered spiking activity only during stable fixation, defined as a 1.1 s window that began 1 s after attaining fixation in which the monkey’s gaze remained within a square window 0.2° per side centered on the fixation point. To assess ACC spiking activity patterns conditioned on LC activation, we divided trials within each session based on whether the single LC neuron from which recordings were being made either did (LC_non-zero) or did not (LC_zero) produce at least one action potential during stable fixation. We also divided session trials into four additional groups of trials in which the LC neuron fired 1, 2, 3, and ≥4 spikes.

We used 10 bin sizes ranging from 100 ms to 1 s, spaced logarithmically, to count spikes. The mean spike count, variance of spike counts, and the Fano factor (the ratio of the variance to the mean of spike counts) for each neuron were calculated across trials for each bin size. Pairwise spike-count correlations (r_sc) were calculated for each bin size as follows. First, trial spike counts from each neuron in the pair were z-scored, and trials on which the response of either neuron was >3 standard deviations different from its mean were removed to avoid effects of outlier responses (Kohn and Smith, 2005; Smith and Kohn, 2008). Then, the MATLAB function corrcoef was used to obtain the Pearson correlation coefficient. Shuffled estimates were made by calculating r_sc from pairs of neurons with spike-count vectors generated from randomly selected trials for each neuron.

ACC r_sc values were calculated for all trials for each pair of well-isolated neurons, independently of spiking in LC and also separately for trials divided into groups depending on spiking in LC (LC_zero and the five LC > 0 groups). The LC-independent r_sc values collected from all pairs and measured using ACC spikes counted using each time bin were divided into terciles that corresponded broadly to pairs that were negatively correlated (tercile 1), uncorrelated (tercile 2), or positively correlated (tercile 3). This analysis allowed us to assess whether changes in r_sc in one region associated with spiking in the other region depended on the pairs being correlated to begin with or not.

Data and Matlab code for all figures in this manuscript are available at here copy archived at swh:1:rev:22b8c94380c057201fb1fcc56c6037b70c56021c; Joshi and Gold, 2021.