Task-evoked pupil dilation and BOLD variance as indicators of locus coeruleus dysfunction

Abstract

Pupillary responses during cognitive tasks are linked to functioning of the locus coeruleus (LC). The LC is an early site of abnormal tau deposition, which may contribute to key aspects of Alzheimer’s disease (AD) pathophysiology. We previously found attenuation of pupillary responses to increases in cognitive load in individuals with mild cognitive impairment (MCI), suggesting pupillary responses may provide a biomarker of early risk for AD associated with LC dysfunction. The LC modulates cortical activity through two modes of operation: tonic and phasic. Early LC damage has been predicted to result in a state of persistent high tonic LC activity that may disrupt task-related phasic activity. To further examine whether pupillary responses are associated with early LC dysfunction, we measured pupil dilation during a digit span task as a measure of phasic activity, and low frequency BOLD variance (LFBV) during resting-state fMRI in key nodes of the ventral attention network as a measure of cortical reactivity related to LC tonic activity in 358 middle-aged men. Individuals with greater LFBV in ventral attention network nodes, i.e., higher tonic brain activity at rest, showed a smaller increase in pupil dilation from low to moderate cognitive loads. Thus, higher tonic LFBV activity at rest was related to reduced task-appropriate phasic dilation increases. The results support predictions from prominent models of LC functioning in which early LC dysfunction leads to persistent high tonic rates of activity during rest and lower signal-to-noise of phasic responses during task performance. Taken together with previous findings of early AD pathophysiology in LC and reduced phasic dilation responses to increased cognitive load in individuals with MCI, the present results suggest that pupillary responses may index early LC dysfunction and should receive further study as a potential biomarker of risk for AD.

1. Introduction

Increasing evidence suggests that the pathological processes underlying Alzheimer’s disease (AD) begin years, if not decades, prior to the onset of behavioral symptoms (Aizenstein et al., 2008; Bateman et al., 2012; Bennett et al., 2006; Jansen et al., 2015). With this knowledge, a shift toward secondary prevention strategies is underway in which intervention is initiated during this extended prodromal state (Dubois et al., 2014; Golde et al., 2011; Sperling et al., 2011). There is a great need for early screening tools that are easily implemented, cost effective and come with a relatively low subject burden to enable more widespread use. Pupil dilation measured during a cognitive task may be one such tool. Individuals show increasing pupil dilation that scales with task load until cognitive capacity is exceeded, at which point there is an abrupt decline in pupil dilation (Geva et al., 2013; Granholm et al., 1996).

We previously found differences in pupillary responses recorded during a digit span task in late middle-aged adults based on individual differences in cognitive ability and mild cognitive impairment (MCI), a state presumed to reflect prodromal AD (Granholm et al., 2017). The digit span task elicits a well-characterized pupil dilation response that can be examined for load-dependent changes (Granholm et al., 1996). Individuals with lower capacity (shorter maximum spans) exhibited greater pupil dilation (i.e., required more effort) at lower loads (3- and 6-digit spans) but they had a more marked decrease in dilation at high loads when capacity was exceeded. Individuals with single-domain amnestic MCI, who performed similarly to cognitively normal participants, showed greater compensatory effort (i.e. greater dilation) at lower loads. In contrast, individuals with multi-domain MCI, who performed worse than the other groups, demonstrated a flat dilation profile in which phasic cognitive effort did not differ across loads, suggesting that compensatory mechanisms had been overwhelmed even at low loads. However, the neural mechanisms underlying these differences remain unclear.

Additional motivation for the utility of pupil dilation measurements in the study of MCI and early AD comes from findings that the pupil dilation is tightly coupled with activity in the locus coeruleus (LC) norepinephrine (NE) pathways (Alnaes et al., 2014; Gilzenrat et al., 2010; Koss, 1986; Murphy et al., 2014; Rajkowski et al., 1993, 1994). Both pupil dilation and activation of noradrenergic neurons in the LC have been shown to increase in a correlated manner with mental (or cognitive) effort (Varazzani et al., 2015), and there is evidence of a causal, although potentially indirect, relationship of LC activation with pupil dilation (Joshi et al., 2016). Dysfunction in the LC is relevant to the study of AD because recent disease models suggest that the LC is the earliest site of tau deposits, which may appear prior to age 30 (Braak and Del Tredici, 2011; Braak and Del Tredici, 2012; although see Crary et al., 2014). Indeed, deposits of neurofibrillary tangles and extensive tissue loss occur in the LC and have been found to correlate with disease duration and symptom severity (Bondareff et al., 1987; German et al., 1992; German et al., 1987; Grudzien et al., 2007). Our previous findings link differences in pupil dilation to MCI, which is thought to represent an early stage of AD. The current study attempts to further link differences in pupil dilation to dysfunction in the LC, a structure affected by AD pathology early in the disease course.

In order to provide converging evidence that differences in pupil dilation reflect dysfunction in the LC, we were particularly interested in whether an altered pupil response would be associated with changes in cortical activity measured by fMRI, since both may be related to LC functioning and are found to be disrupted in preclinical AD (Ewers et al., 2011; Granholm et al., 2017; Sperling et al., 2011). Converging evidence suggests that the LC-NE system plays a major role in modulating neural activity in the ventral attention network (VAN) (Corbetta et al., 2008; Coull et al., 1999; Dosenbach et al., 2006; Geva et al., 2013; Sara, 2009; Sara and Bouret, 2012). The VAN is a right-lateralized network and includes the right temporo-parietal junction (TPJ), anterior cingulate (ACC), and inferior prefrontal cortex (iPFC). The ACC and iPFC have highly recurrent connections with the LC, and the LC modulates task focus through its ability to drive activity within the TPJ (Aston-Jones and Cohen, 2005; Aston-Jones et al., 2002; Samuels and Szabadi, 2008; Walz et al., 2013). During normal functioning, fMRI activity in the VAN is correlated with activity in the LC as well as with pupillary responses (Alnaes et al., 2014; Murphy et al., 2014; Raizada and Poldrack, 2007).

According to the adaptive gain theory (Aston-Jones and Cohen, 2005), an influential model of LC function, the LC modulates neural activity and behavior through two modes of functioning: tonic and phasic. Increased tonic activity facilitates exploration by boosting the gain of VAN neurons, increasing their baseline firing rate. In contrast, the phasic mode facilitates task performance by increasing the signal-to-noise ratio, suppressing responses to irrelevant stimuli while allowing evoked responses to task-relevant stimuli (Aston-Jones and Cohen, 2005; Corbetta et al., 2008; Sara and Bouret, 2012).

To examine whether altered activity in the VAN is associated with LC dysfunction as would be predicted by the adaptive gain theory, we measured associations between pupil dilation during a cognitive task and low-frequency BOLD variance (LFBV) during resting-state fMRI. Although the LC can facilitate exploration through high tonic activity, it demonstrates low levels of tonic activity during quiet waking when little vigilance or task-engagement is required (Aston-Jones and Bloom, 1981; Rajkowski et al., 1994; Rajkowski et al., 1997; Rasmussen et al., 1986). In the context of a resting-state scan with no behavioral demands, the LC should presumably enter a state of low tonic activity with only small fluctuations in activity. Early stages of damage to the LC may result in a paradoxical increase in both LC firing rate and NE metabolism (Chiodo et al., 1983; Elrod et al., 1997; Szot et al., 2006) resulting in a persistent state of high tonic activity. A state of high tonic activity has been proposed to underlie activation of the VAN and sensitivity to environmental stimuli as seen during stimulus-driven states in fMRI studies of attention (Corbetta et al., 2008). Therefore, higher tonic LC activity may result in greater VAN cortical reactivity, i.e., greater LFBV, due to external stimuli. While low frequency BOLD signal arising from LC activity during resting-state fMRI has been studied in the context of functional connectivity (Bar et al., 2016; Zhang et al., 2016), we focused on the degree of fluctuation in activity, measured by LFBV, as an index of cortical reactivity brought on by levels of tonic activity.

When functioning normally during cognitive tasks that require focused attention, the LC switches to the phasic mode (Aston-Jones and Cohen, 2005; Nieuwenhuis et al., 2005; Rajkowski et al., 2004). In the event of LC dysfunction, as in preclinical or prodromal AD, the coordinated and temporally specific activity required for phasic responses to behavioral challenges is likely to be impaired (Aston-Jones and Cohen, 2005). The pupil dilation response in the digit span task is well-characterized (Granholm et al., 1996), and we have previously shown that it is sensitive to differences in cognitive status (i.e., MCI) (Granholm et al., 2017). In relation to disrupted LC functioning, a lack of proper phasic modulation may manifest as attenuated (or “flattened”) increases in pupil dilation response as task demands increase.

Thus, if early LC dysfunction leads to both a paradoxical increase in tonic LC activity and a reduction in phasic LC activity, as described above, we hypothesized that an association would be found between high levels of LFBV during a resting-state fMRI scan (due to difficulty maintaining low tonic activity) and attenuated increases in pupil dilation response during a digit span task (suggesting inefficient adaptation to changing task demands during phasic state activity).

2. Methods

2.1. Participants

Participants were drawn from wave 2 of the Vietnam Era Twin Study of Aging (VETSA) (Kremen et al., 2013; Kremen et al., 2006). VETSA participants comprise a national, community-dwelling male sample that is similar to American men in their age range with respect to health and lifestyle characteristics based on Center for Disease Control and Prevention data (Kremen et al., 2013; Kremen et al., 2006; Schoeneborn and Heyman, 2009). The study was approved by the Institutional Review Boards at the University of California, San Diego (UCSD), Boston University (BU), and the Massachusetts General Hospital (MGH). Imaging took place at UCSD and MGH.

Pupillary responses were assessed in 979 individuals at UCSD and BU after excluding participants with a self-reported history of glaucoma in either eye, penetrating eye wounds to both eyes, eye surgery to both eyes that involved the muscle, or use of cholinesterase inhibitors or anticholinergic eye drops. Of these participants, valid pupillary response data were obtained for 954 individuals (in 25 cases data could not be collected due to either technical failure of the equipment or the inability of the subject to complete the task without excessive blinking).

The VETSA 2 MRI component included 447 individuals using standard MRI exclusion criteria (e.g., no metal in the body). Of these, resting-state fMRI (rs-fMRI) data were collected the day after neuropsychological testing and pupillometry sessions on 422 individuals. Forty-four individuals were excluded for excessive motion (mean framewise displacement >.4 mm). The final sample included in the present study was composed of 358 participants with both pupillometry and fMRI data. The sample had a mean age of 61.9 years (SD=2.65), was primarily Caucasian (89.9%), and had a mean education of 13.8 (SD=2.06) years. MCI status was determined as described previously in greater detail (Granholm et al., 2017; Kremen et al., 2014) according to the Jak/Bondi actuarial-neuropsychological approach (Bondi et al., 2014; Jak et al., 2009; Jak et al., 2015; Kremen et al., 2014). Thirty individuals were classified as amnestic single domain MCI (aMCI), 15 as non-amnestic single domain MCI (naMCI), and 9 as multiple domain MCI (mMCI).

2.2. Task-evoked pupil dilation

Pupillometry methods have been described previously (Granholm et al., 2017). Briefly, handheld NeurOptics PLR-200 pupillometers were modified by the manufacturer to record pupil diameter from one eye at 30 Hz for up to 15 seconds in a viewing tube. The device is the size of a television remote with recording optics inside one end of a 1.5-inch viewing tube that surrounds the eye. Ambient light was blocked from reaching the tested eye by the viewing tube; participants closed and held their hand over their other eye.

Pupillary responses were recorded during blocks of trials of 3 (low load), 6 (moderate/near capacity load), and 9 (high/overload) digits presented aurally on a laptop computer at the rate of one per second. Participants heard “Ready” one second before the first digit and “Repeat” one second after the last digit. Each trial was inspected for artifacts in a graphic display on the device. Trials were administered until two clean trials (<50% of data contained artifacts) were recorded or four trials were attempted per digit-span condition. Pupil diameter samples were averaged for each second of recording, which corresponds to the presentation of digits at one-second intervals. The primary dependent variable was change relative to baseline during the second immediately following the last digit presented. Baseline pupil size during the first second of each trial was regressed from the pupil change score, to remove individual differences in tonic pupil size.

2.3. Task performance

To determine each individual’s short-term memory capacity, we used the maximum span length correctly recalled on the Digit Span subtest of the Wechsler Memory Scale–III (WMS-III; Wechsler, 1997). WMS-III Digit Span was administered prior to digit span trials during pupillometry during the course of neuropsychological testing. Because these data were from wave 2 of the study, we also accounted for practice effects by adjusting wave 2 WMS-III Digit Span scores by the expected increases in scores due to repeated testing. We did not conduct pupillometry in wave 1. As in the study of Ronnlund et al. (2005), for each test, we calculated the difference score as the mean score of participants who underwent cognitive testing at wave 1 minus the mean score of those who took the tests for the first time at wave 2 (attrition replacements). We calculated attrition effects for each test as the mean scores of wave 1 participants who returned for wave 2 minus the mean score of all participants at wave 1. Practice effects were then calculated as the difference score minus the attrition effect. These practice effect values were subtracted from wave 2 scores of returning participants.

2.4. MRI Acquisition

Images were acquired at UCSD and MGH. At UCSD, images were acquired with a GE 3T Discovery 750 scanner (GE Healthcare, Waukesha, WI, USA) with an eight-channel phased array head coil. The imaging protocol included a sagittal 3D fast spoiled gradient echo (FSPGR) T₁-weighted volume (TE=3.164msec, TR=8.084msec, TI=600msec, flip angle=8°, FOV=256cm, matrix=256×256, in-plane resolution=1×1mm, slice thickness=1.2mm, slices=172,) used for registration purposes and to determine cortical thickness and surface area. Rs-fMRI scans were collected with aT₂*-weighted echo-planar imaging sequence (EPI; TE=30msec, TR=3000msec, flip angle=85°, FOV=256, matrix=64×64, in-plane resolution=4×4mm, slice thickness=3 mm). Thirty-two slices were acquired in an interleaved order to provide whole-brain coverage. One hundred volumes were acquired (50 forward and 50 reverse phase encoding directions). At MGH, images were acquired with a Siemens Tim Trio, (Siemens USA, Washington, D.C.) with a 32-channel head coil. The imaging protocol included a 3D magnetization-prepared rapid gradient-echo (MPRAGE) T₁-weighted volume (TE=4.33msec, TR=2170msec, TI=1100msec, flip angle=7°, FOV=256, matrix=256×256, in-plane re solution=1×1mm, slice thickness=1.2mm, slices=160). Rs-fMRI scans were collected with a T₂*-weighted echo-planar imaging sequence (TE=30msec, TR=3000msec, flip angle=85°, FOV=216, matrix=72×72, in-plane resolution=3×3mm, slice thickness=3mm). Forty-seven slices were acquired in an interleaved order to provide whole-brain coverage. One-hundred twenty-four volumes were collected.

2.5. MRI Processing

Structural MRI images were processed to account for effects of partial voluming on measures of LFBV and to generate anatomical regions of interest (ROIs). The processing methods have been described previously (Kremen et al., 2010; McEvoy et al., 2015) and were based on the FreeSurfer 5.1 software package (Dale et al., 1999; Fischl and Dale, 2000; Fischl et al., 1999; Fischl et al., 2004). Surface models were reviewed for accuracy, and white matter and brain masks were manually edited as necessary, in alignment with standard, objective editing rules. The surface was then divided into 66 distinct cortical regions (33 per hemisphere) according to the Desikan-Killiany atlas (Desikan et al., 2006; Fischl et al., 2004). Cortical thickness was extracted from each ROI and standardized with a z-transform.

Rs-fMRI scans were processed as follows: The first 4 volumes were discarded to allow for equilibration of the T₁-weighted signal. B₀ distortions were corrected using the reversing gradient method (Chang and Fitzpatrick, 1992; Holland et al., 2010; Morgan et al., 2004) and gradient nonlinearity distortions were corrected for each frame (Jovicich et al., 2006). Slice timing correction and motion correction was performed using AFNI (Cox, 1996), and the resulting head motion time courses were later used to regress out motion-related signal. Mean framewise displacement was calculated (Power et al., 2012; Van Dijk et al., 2012) for later use as a covariate in group analyses (Satterthwaite et al., 2012). Automated registration between functional and structural images was performed using mutual information (Wells et al., 1996) with coarse pre-alignment based on within-modality registration to atlas brains. Rs-fMRI time courses were normalized with the mean of each voxel. Linear regression was used to remove quadratic trends, motion parameter timecourses, cerebral white matter and whole brain signals, and their first derivatives. Time courses were band-pass filtered between 0.01 and 0.08 Hz. We then sampled the time courses at a fixed distance of 1 mm from the white-gray boundary (into the gray matter) onto the cortical surface for each individual subject. The average time course was calculated for each cortical surface-based ROI. Based on our a priori hypothesis of key brain regions belonging to the LC-NE system (i.e., temporo-parietal junction [TPJ], anterior cingulate cortex [ACC], and inferior prefrontal cortex [iPFC]), we examined LFBV in the supramarginal gyrus (SMG), caudal ACC, and pars opercularis (PO; in the iPFC) ROIs defined by the Desikan-Killiany cortical parcellation (Desikan et al., 2006), respectively. As a negative control, we also assessed the lingual gyrus (LG), a region not particularly associated with the LC-NE system. We did not examine functional connectivity between nodes because we were primarily interested in how the LC modulates activity in each of these regions individually, and LFBV provides an index of fluctuating activity in the tonic state.

2.6. Statistical Analysis

Analyses were conducted using linear mixed effects models in R v3.2.1 (R Core Team, 2014). Subjects nested within twin pair were included as random effects to control for repeated measures within subject and correlations within dyads. Differences in pupil dilation were assessed across repeated measures of load (3, 6, or 9 digits). Separate models were run for each hypothesized ROI and the negative control ROI. Task load and LFBV of a given ROI, along with their interaction, were included as fixed effects. To control for differences in task performance, we included the z-transformed maximum forward span (from the WMS-III Digit Span subtest), and its interaction with load as additional fixed effects. The interaction terms of task load with LFBV were of primary interest because they indicate the degree to which change in pupil dilation from one level of task difficulty to another differs as a function of brain activity. Change between levels of load may be more sensitive to LC system disruption than pupil dilation within a given level of load because the former reflects appropriate modulation of activity in response to changing behavioral contexts. Three models were tested to examine our hypothesized ROI; therefore, the significance threshold was adjusted using a Bonferroni correction (p<0.0167). Significant associations were determined using the type III test of fixed effects. Where interaction terms were significant, we assessed contrasts of adjacent levels of load to better understand the underlying patterns of change in pupillary response.

We included variables that might be related to either rs-fMRI or pupillary responses. To account for anticholinergic effects of medications not already leading to exclusion on pupillary responses, a composite score reflecting the cumulative anticholinergic effect of all other medications was computed (Granholm et al., 2017). APOE genotype was included as a covariate on the basis of presence or absence of an ε4 allele, a risk factor for AD. The standardized cortical thickness of each ROI tested was included to control for effects of atrophy or partial-voluming on functional measures of LFBV. Additional participant-level covariates were included to control for age, pupillometry device (two devices per site), scanner (one per site), handedness (right vs. left to control for potential differences in lateralization), and mean framewise displacement during the rs-fMRI scan (i.e. motion). The full fixed effects model is shown below; subject nested within case was also included as a random effect to control for non-independence of twin data:

$${\mathit{\text{PupilDilation}}}_{\mathit{\text{Load}}}=\mathit{\text{Load}}+{\mathit{\text{LFBV}}}_{\mathit{\text{ROI}}}+{\mathit{\text{Load}}}^{\ast }{\mathit{\text{LFBV}}}_{\mathit{\text{ROI}}}+\mathit{\text{MaxForwardSpan}}+\mathit{\text{Medication}}+\mathit{\text{APOE}}\mathit{4}+{\mathit{\text{CortThickness}}}_{\mathit{\text{ROI}}}+\mathit{\text{Age}}+\mathit{\text{PupilDevice}}+\mathit{\text{Scanner}}+\mathit{\text{Handedness}}+\mathit{\text{Motion}}$$

3. Results

3.1. Participant Characteristics and Their Relationship to LFBV

Only 13 participants were unable to complete 2 trials of all load levels with a clean pupil scan (i.e., with minimal blinks). The average maximum forward span on the WMS-III digit span subtest was 6.86 (SD=1.89). Maximum digit span was significantly associated with LFBV in the ACC [t₍₃₄₅₎=2.226, p=0.027], but not in either of the other two ROIs. There were 79 (22.07%) APOE-ε4 carriers. There were no significant relationships between LFBV and age or APOE-ε4 status in any ROIs. LFBV was significantly higher in the SMG in individuals with MCI compared to those who were cognitively normal [F_(3,5.19)=5.276, p=0.001], driven by marginally greater LFBV in aMCI compared to cognitively normal participants [t₍₃₃₅₎=1.763, p=0.079].

3.2. Pupil Dilation and LFBV

At any individual level of load (3, 6 or 9), there were no significant relationships between LFBV and pupil dilation for any of the ROIs. However, LFBV was related to the slope of (i.e., change in) pupil dilation across levels of load. After correction for multiple comparisons, there was a significant interaction of LFBV and load on pupil dilation in the SMG [F_(2,942)=6.959, p<0.001], ACC [F_(2,942)=4.568, p=0.011], and PO [F_(2,942)=8.779, p<0.001] (Figure 1). These interactions were primarily driven by smaller changes in pupil dilation between loads 3 and 6 for participants with higher LFBV [SMG: t₍₉₄₂₎=−2.384, p=0.017; ACC: t₍₉₄₂₎=−2.359, p=0.019; PO: t₍₉₄₂₎=−.626, p=0.009]. The interaction between LFBV × load was not significant in the negative control region LG [F_(2,942)=0.155, p=0.86], suggesting a degree of regional specificity in these effects.

Figure 1.

Model predicted change in pupil dilation as a function of low frequency BOLD variance (LFBV). Quartiles are shown for ease of interpretation, but LFBV was included as a continuous variable in the models. Error bars represent within-subject standard error. There was a significant interaction (p<0.05) between load and LFBV in all three regions of interest. This effect was driven by a steeper slope between loads 3 and 6 in individuals with lower LFBV. Note that negative values of change in pupil diameter do not represent constriction, but are a byproduct of mean centering when adjusting for baseline pupil diameter.

3.3. Relationship of pupil dilation with other covariates

The two-way interactions between load × MCI status and load × maximum digit span have been reported previously (Granholm et al., 2017). The three-way interaction of LFBV × load × MCI status was not significant in any ROI, indicating that the interactions between load and pupil dilation reported above did not differ in MCI subjects. No other covariates were significantly related to pupil dilation.

4. Discussion

Consistent with our hypotheses, the degree to which pupil responses increase across task loads was related to the variance of rs-fMRI BOLD fluctuations within key nodes of the VAN. We focused on variance in the BOLD signal rather than correlations between regions due to the role of the LC in modulating gain in cortical activity (Aston-Jones and Cohen, 2005). As predicted participants with higher LFBV demonstrated an attenuated increase in pupil dilation between low and moderate levels of task demands (3 and 6 digits), driven by somewhat higher pupil responses at low loads and somewhat lower pupil responses at moderate loads. This flatter pupil dilation slope may reflect two processes, both of which are consistent with less efficient LC function. Slightly greater dilation at the low load may indicate that additional effort is needed to compensate for inefficiency or reduced capacity. However, it does not necessarily mean that people have reached their capacity limit. As seen in Figure 1, dilation was significantly greater at the moderate load compared with the low load. Instead, the flatter slope associated with higher LFBV suggests that individuals with higher LFBV failed to fully ramp up effort at the moderate load. In effect, they had a narrower dynamic range of response. Put differently, higher LFBV was associated with poorer modulation in response to increasing task demands.

The altered pupillary responses and heightened LFBV is consistent with impaired function of the LC and subsequent NE release. LC dysfunction should result in changes to both tonic and phasic states (Aston-Jones and Cohen, 2005). Level of tonic activity is likely to determine LFBV during the resting-state scan whereas phasic responses likely contribute to pupil dilations during the digit span task (Aston-Jones and Bloom, 1981; Rajkowski et al., 1994; Rajkowski et al., 1997; Rasmussen et al., 1986; Usher et al., 1999). Damage to the LC has been shown to result in compensatory increases in LC firing rate and NE metabolism (Chiodo et al., 1983; Elrod et al., 1997; Szot et al., 2006) that may result in a state akin to the tonic mode of LC firing (Aston-Jones and Cohen, 2005; Weinshenker, 2008). In this “persistent tonic state”, activity in the VAN may resemble stimulus-driven states of high cortical reactivity in which there are relatively high levels of fluctuating activity that may persist even in situations requiring little vigilance or task engagement such as a resting-state scan.

Our results are consistent with this theory. Higher LFBV may represent increased tonic activity due to LC dysfunction, and a decreased dynamic range in the pupillary response profile during the digit span task may represent a corresponding disruption in phasic activity. During the phasic mode, increased tonic LC firing rate would produce a higher baseline level of activity and individuals may require more effort at low loads. Although phasic activations may still occur, changes in task-related signals are obscured due to a decreased signal-to-noise ratio. That is, the LC shows reduced differentiation in pupil response with increasing task difficulty between low and moderate loads. Consistent with that idea, we previously showed that this reduced differentiation was more severe in more advanced multiple-domain MCI compared with single-domain MCI (Granholm et al., 2017).

Controlling for maximum digit span means that change in pupillary dilation from low to moderate cognitive loads cannot simply be explained by differences in short-term memory ability. Despite evidence of disrupted LC function, participants with higher LFBV were still able to perform the task successfully, especially at lower loads. This is consistent with the neuromodulatory role of the LC, which facilitates optimal performance and coordinated network functioning by way of NE release through diffuse cortical projections rather than serving as the putative neural substrate of short-term and working memory (Aston-Jones and Cohen, 2005; Coull et al., 1999; Raizada and Poldrack, 2007; Sara, 2009). It may be that even sub-optimal functioning of the LC-NE system and the VAN is sufficient for the digit span task, and performance decrements may only become apparent on more difficult tasks. Alternatively, other compensatory mechanisms may help maintain performance, such as increased cortical sensitivity to NE (Harik et al., 1981; Raskind et al., 1999; Szot et al., 2006; Szot et al., 2007), compensatory reorganization of the LC-NE functional network (Jacobs et al., 2015), or recruitment of additional cortical networks outside of the LC-NE system as is commonly found in fMRI studies of aging and preclinical AD (Cabeza, 2002; Clément and Belleville, 2010; Elman et al., 2014; Huang et al., 2012; Morcom et al., 2007; Mormino et al., 2012; Reuter-Lorenz and Cappell, 2008).

A benefit of the relatively narrow age range of the VETSA sample is that the current results are not likely to reflect age-related differences. This is supported by a lack of significant association between age and either pupil dilation or LFBV. Previous research has found age-related decreases in BOLD variability, and that greater variability is beneficial (Garrett et al., 2010, 2013; Grady and Garrett, 2014). The current findings do not necessarily conflict with these reports given the regional specificity of BOLD variability effects observed here and recent findings that BOLD variability demonstrates age-related decreases in some brain regions and increases in others (Nomi et al., 2017). Additionally, there are important differences in the behavioral contexts of fMRI acquisition paradigms between studies that may influence the expected state of LC firing (i.e., fixation blocks interspersed with cognitive tasks versus a task-free resting-state scan). In fact, the greater dynamic range in pupil dilation considered to reflect appropriate modulation of LC activity in the current study is consistent with the greater modulation of BOLD variability between behavioral states previously found in young compared to older subjects (Garrett et al., 2013). Thus, it appears that appropriately modulating activity between behavioral contexts is a prime indicator of healthy functioning.

The current study does have several limitations. While we did not have a direct measure of LC activity, a tight link between pupil dilation and LC activity has been shown in humans as well as through single cell recordings in animals (Alnaes et al., 2014; Aston-Jones and Cohen, 2005; Gilzenrat et al., 2010; Joshi et al., 2016; Murphy et al., 2014; Rajkowski et al., 1993, 1994; Samuels and Szabadi, 2008). This link has resulted in the use of pupillometry as a well-validated proxy of mental effort in research over the course of the past few decades (Laeng et al., 2012). Further evidence suggests a strong link between activity in the VAN with both pupillary response and the LC-NE system (Alnaes et al., 2014; Corbetta et al., 2008; Nieuwenhuis et al., 2005; Nieuwenhuis et al., 2011). We based our association between high LFBV and high tonic activity on the proposed link between LC tonic mode and increased sensitivity to environmental stimuli within the VAN (Corbetta et al., 2008). However, the relationship between LC activation and expected BOLD signal in the cortex is complex. Single cell recordings during tonic activation of LC neurons and subsequent release of NE demonstrate that modulatory effects differ on a cell-by-cell basis depending on brain region, frequency and strength of stimulation, and behavioral context (Devilbiss and Waterhouse, 2004; Devilbiss and Waterhouse, 2011; Polack et al., 2013; Poulet and Petersen, 2008). This is further complicated by the relationship between cortical synchrony with BOLD signal, which can differ by region and frequency band (Conner et al., 2011; Winterer et al., 2007). The relationships between pupil dilation and LFBV found in the current study appear to be regionally specific, and thus do not reflect global properties of brain activity. We note that, at least in the SMG, participants with MCI showed higher LFBV, suggesting that higher variability in this region is indicative of disrupted or inefficient functioning. Further work is necessary to elucidate the complex relationship between single-cell recordings of LC activity and resulting cortical BOLD signal.

Similarly, we were not able to directly assess pathology within the LC in these participants. Although there may be multiple factors driving differences in pupil dilation and LFBV, there is reason to suspect that AD-related processes – the presence of abnormal tau in particular – play a role. A study of 2,332 autopsy examinations across all age ranges found AD-related tau inclusions in nearly every case, and there is evidence that the earliest site of abnormal tau is in the LC (Braak and Del Tredici, 2011; Braak et al., 2011). These findings have been proposed to reflect AD pathological processes that originate in the brainstem and progress across the lifespan (Braak and Del Tredici, 2012; although see Crary et al., 2014). Although there may be little or no detrimental effects on neural functioning in the initial stages, the LC is likely to be one of the first brain regions affected when conformational changes result in more toxic forms of tau. Evidence that abnormal tau can precipitate formation of Aβ plaques and exacerbate inflammatory responses provides additional avenues through which early LC dysfunction may have downstream effects on cortical functioning (Chalermpalanupap et al., 2013; Heneka et al., 2002; Heneka et al., 2010). Therefore, altered activity within the LC may represent both a functional outcome and key contributor to disease progression. However, further work is necessary to determine whether pathology within the LC is specific to AD-related processes.

With the continued development of tau PET tracers, it may soon be feasible to include in vivo imaging measures of tau deposits in future studies focusing on the LC. However, the relatively low resolution of PET imaging and off-target binding in the brainstem likely preclude direct measurement of tau within the LC itself. We also do not know which of the current participants may go on to develop AD, but we will be able to examine this in future waves of the study. On the other hand, we did show that pupillary responses differentiated cognitively normal and MCI groups (Granholm et al., 2017), and individuals with MCI tended to have greater LFBV in the current analysis. Lastly, we did not collect task-based fMRI or resting-state pupil dilation recordings. Therefore, we were unable to assess the relationship between patterns of tonic and phasic activity within modality.

We previously found that pupil dilation response differs based on MCI status. The current study suggests that differences in pupil dilation are related to dysfunction in the LC by linking differences in task-evoked pupil responses with cortical activity of regions belonging to the LC-NE system. We propose that these findings reflect the ability to appropriately modulate LC activity given the behavioral context. That is, LC dysfunction likely results in disruptions to both tonic and phasic modes of activity. Recent evidence that abnormal tau is present in the LC much earlier than previously thought suggests that impaired LC functioning may reflect early phase AD-related processes. The relationship between pupillary response and LFBV is consistent with the idea that the effects of any such processes in the LC may be propagated to cortical functioning. That these changes in pupillary response are detectable in the absence of substantial performance decrements suggests that this easily implemented measure of pupillary response may ultimately have utility as a marker of impaired LC functioning and as an additional early screening tool for AD-related risk. However, further work is necessary to determine the specificity of pupil dilation responses in relation to LC dysfunction and AD-related risk. Finally, the results provide a strong rationale for the value of further study of LC integrity in conjunction with fMRI examination of LFBV during cognitive task performance.

Highlights.

• Pupil dilation during a task may index locus coeruleus (LC) phasic activity

• Rs-fMRI BOLD variance in a network modulated by LC may index LC tonic activity

• Higher BOLD variance was associated with attenuated dilation responses during task

• Consistent with persistent tonic state and decreased phasic modulation by LC

• Results support the study of pupil dilation as a potential biomarker for AD risk