Technical Comments

Response to Comment on “Modafinil Shifts Human Locus Coeruleus to Low-Tonic, High-Phasic Activity During Functional MRI”

See allHide authors and affiliations

Science  16 Apr 2010:
Vol. 328, Issue 5976, pp. 309
DOI: 10.1126/science.1177948


Contrary to the arguments put forth by Astafiev et al., our new analyses and high-resolution replication show that modulation of brain activity by the cognitive enhancing drug modafinil indeed localizes to the locus coeruleus. The authors’ critical stance regarding blood oxygen level–dependent imaging in the brainstem does not follow from their data.

We confirmed a role of the brainstem locus coeruleus (LC) in human cognition through brain imaging of healthy humans treated with the cognitive-enhancing drug modafinil (1). Astafiev et al. (2) question the extent of the brain activation in our study based on published postmortem studies of the human LC, and their results from a novel in vivo neuromelanin-sensitive magnetic resonance imaging (MRI) method. We contend that the authors neglect several critical issues in their evidence and ours.

Astafiev et al. (2) describe the active brain clusters we identified as located “more anterolateral and superior” to their neuromelanin-positive loci. However, human brain atlases (35) would locate these in the inferior colliculus or off of the subcortical neuraxis altogether. This is clearly not the case, as evidenced by our published figures (1). Some of this ambiguity may arise from their idiosyncratic use of some directional indicators (e.g., “dorsocaudal axis” combines two planes and does not specify poles of an axis). Despite their implication, the LC proper (i.e., A6 nucleus) does extend rostral to the fourth ventricle [see figure 10.1.C in (4), figure 51 in (5), figure 1C in (6) and figure 18 in (6)]. The relatively lateral extensions of our clusters likely represent the A7 nucleus, which is contiguous with the LC core and often included in the LC complex [figure 9 in (6), figures 17 and 18 in (7)].

We also disagree that the size of our activated pontine region is inconsistent with an LC source. Assuming [from (8)] that the “true” area of LC activation is contained within a tube 32.8 mm2 in the plane orthogonal to the neuraxis and 16 mm long, then applying an 8-mm smoothing kernel (as we did) expands this homogeneous cluster to 3675 mm3. This volume expansion results from smoothing a thin structure, where surface area predominates over internal volume. Consequently, one could observe up to 459 active voxels (resliced to 8 mm3) in each hemisphere, larger than our reported cluster sizes. Therefore, we disagree that our reported results greatly exceed what could plausibly arise from the LC using our methods.

In addition, neuromelanin-sensitive imaging in (8) and the data of Astafiev et al. (2) likely underestimate in vivo LC volumes from our study for several reasons. First, the subjects in (8), aged 60 to 104 years (versus 33 years on average in our study), exhibited strong age-related declines in LC volume and cell number, consistent with the literature. Second, postmortem tissue shrinkage may lead to underestimates of in vivo brain volumes. Third, there is good evidence that many tyrosine hydroxylase-immunoreactive LC neurons contain no pigment (9, 10). Finally, neuromelanin contrast in MRI is close to the limit of detection (1113). Underestimates of LC extent by neuromelanin may be particularly manifest at the margins, where cell density is lower. Accordingly, to date there is no published evidence that the three-dimensional boundaries of the LC can be reliably ascertained by neuromelanin-sensitive in vivo imaging.

With these limitations in mind, we reevaluated our published data with the use of a recently available probabilistic LC mask (Figs. 1B and 2B), derived from neuromelanin-sensitive MRI in 44 healthy adults (14). These results clearly show that (i) in our original data, experimental effects are observed within this LC mask (Figs. 1C and 2C); (ii) when these data are not subject to explicit spatial smoothing, these effects persist within the LC mask (Figs. 1D and 2D); and (iii) in results from a new study sample (n = 9), using high-resolution functional MRI, the reported effects are once again observed within the LC mask (Figs. 1E and 2E).

Fig. 1

Axial views of (A) template brain, (B) LC mask from (14), (C) original drug effect masked to LC, (D) original drug effect in unsmoothed data, masked to LC, and (E) results in new sample using high-resolution echo planar imaging. Clusters in (C) to (E) are all within LC mask. Color bar is t statistic. See Fig. 2 for more details.

Fig. 2

Three-plane views of results summarized in Fig. 1. Group mean beta values are at right of contrast maps, for Placebo and Drug conditions, left (black) and right (green), respectively. Color bar is t statistic. Throughout panels (C) to (E), significant clusters are each conjoined to LC mask from (14) and observed in rostrodorsal pons, adjacent to midline, at inferolateral bank of cerebral aqueduct or fourth ventricle. (A) Template brain (in sagittal, axial, and coronal views, respectively) depicting brainstem region isolated to depict experimental results in (C) to (E). (B) LC mask, derived as probabilistic map by neuromelanin-sensitive MRI in 44 healthy adults (14). (C) Statistical parametric map, depicting significant pontine deactivation in Drug minus Placebo contrast, for cue-related BOLD signal change, from study reported in (1), masked to LC mask in (B). Maximum for drug effect is observed at coordinates –2, –36, –16 (left LC) and 4, –36, –16 (right LC). The following tests of beta values are statistically significant: DrugCue, P = 0.003 (two-tailed, one-sample t test); PlaceboCue versus DrugCue, P = 0.013 (two-tailed paired t test). (D) Results from data in (C), but without explicit spatial smoothing step. Maximum for drug effect is observed at coordinates –2, –36, –18 (left LC) and 4, –36, –16 (right LC). The following tests of beta values are statistically significant: DrugCue, P = 0.023 (two-tailed, one-sample t test); PlaceboCue versus DrugCue, P = 0.044 (two-tailed paired t test). (E) Identical contrast as in (C) and (D) above, in new sample (n = 9). Task and treatment design, and image acquisition/processing/analytic methods are identical to that reported in (1), except that echo-planar MR images are acquired with voxel size 2 mm isotropic and at more oblique angle off axial plane. Maximum for drug effect is observed at coordinates –4, –38, –30 (left LC). The following test of beta values is statistically significant: PlaceboCue versus DrugCue, P = 0.034 (two-tailed paired t test).

Another issue raised by Astafiev et al. (2) is the general validity of blood oxygen level–dependent (BOLD) imaging in the brainstem. The authors compare a BOLD signal time series, from a neuromelanin-localized brainstem slice, to that of a nearby fourth ventricle region. They suggest that the result implies an artifact detected in the pons. However, this result is ambiguous. It is equally likely that valid adjacent pontine activity is manifest in the fourth ventricle through partial volume effects. These possibilities are not disambiguated, and its “artifactual” nature is purely conjectural. Therefore, it is a weak argument that our data, which follow strong predictions from pharmacological and physiological studies in animal models, are suspect on this basis.

A critical distinction absent from the comment by Astafiev et al. (2) is that the key event in our task demanded no motor response, which was associated with their event-related signal change. We reported the effects of increased synaptic norepinephrine (NE) on brainstem BOLD response to the symbolic value of a cue representing a task-relevant rule. This is fully consistent with our working model of LC function. Accordingly, we found no drug effects within the ventricle [figures 1 and 2 in (1)]. These results indicate that this pontine region differs from the ventricle in BOLD response properties.

Astafiev et al. also fail to consider our pharmacological evidence. The dorsal pons contains NE cells only in the defined A4 to A7 nuclei (6), which contain the highest NE transporter (NET) concentrations in the brain (15). Modafinil inhibits NET and the dopamine transporter (DAT), with indirect effects on other transmitters but without any evidence whatsoever to indicate direct binding to elements of these other transmitter systems (16). Therefore, it seems highly improbable that our drug effects, observed where they were, could arise from a neurochemical system other than the catecholamines.

To summarize, our paper (1) reported changes in pontine activity consistent with the known pharmacological effects of modafinil and an important animal model of locus coeruleus function. Although we applaud the efforts of Astafiev et al. (2) to encourage development of refined imaging methodologies, we maintain that our reported experimental effects are consistent with an LC source and that the data shown in their comment are ambiguous: Neuromelanin MRI may underestimate the spatial extent of the LC, and Astafiev et al.’s BOLD time series may represent valid LC activity measured in the ventricle. Furthermore, this motor response–related signal change is irrelevant for our results, which were cue-related and unassociated with a motor response. Additional interrogation of our data reveals experimental effects within an independently derived LC mask (also observed in a new sample), and fails to confirm any nonparenchymal signal. In relation to our study, Astafiev et al. appear to be speculating on their interesting but ambiguous data and neglecting important details, including limitations in their methods and the pharmacological nature of our work.

References and Notes

  1. This work was supported by UL1 RR024146 from the National Center for Research Resources (NCRR) to M.J.M., and MH059883 to C.S.C.
View Abstract

Navigate This Article