Empty sella syndrome as a window into the neuroprotective effects of prolactin

BACKGROUND To correlate structural integrity of visual pathway white matter tracts with prolactin levels in a patient who demonstrates downward herniation of the optic chiasm secondary to medical treatment of a prolactinoma. METHODS A 36-year-old woman with a prolactinoma presented with progressive bilateral visual field defects nine years after initial diagnosis and medical treatment. She was diagnosed with empty-sella syndrome and instructed to stop cabergoline. Hormone testing was conducted in tandem with routine clinical evaluations over one year and the patient was followed with diffusion magnetic resonance imaging (dMRI), optical coherence tomography (OCT), and automated perimetry at three time points. Five healthy controls underwent a complementary battery of clinical and neuroimaging tests at a single time point. RESULTS Shortly after discontinuing cabergoline, diffusion metrics in the optic tracts were within the range of values observed in healthy controls. However, following a brief period where the patient resumed cabergoline (of her own volition), there was a decrease in serum prolactin with a corresponding decrease in visual ability and increase in radial diffusivity (p<0.001). Those measures again returned to their baseline ranges after discontinuing cabergoline a second time. CONCLUSIONS These results demonstrate the sensitivity of dMRI to detect rapid and functionally significant microstructural changes in white matter tracts secondary to alterations in serum prolactin levels. The inverse relations between prolactin and measures of diffusion and visual function provide support for a neuroprotective role of prolactin in the injured nervous system.


INTRODUCTION
Crush and stretch injuries to the optic nerve, tract and chiasm in animals reveals that demyelination is a primary process in the progression of delayed axonal degeneration (1-4). Human studies looking at compression of the optic chiasm have confirmed these findings and also suggested the possibility of rapid vision recovery (5,6). Unlike primary axotomy, in which axonal connections are immediately severed, delayed axonal degeneration is a potentially reversible process. Several mechanisms likely contribute to recovery, including remyelination (7), microtubule reorganization, shifts in ion channel permeability and restoration of glial-neuronal connections at the paranodes (8).
Left unhindered, delayed axonal degeneration and its associated sequela lead to irreversible cell death (1-3, [9][10][11]. Prolactin, a hormone synthesized in and released by the anterior pituitary gland with well-established roles in lactation, demonstrates promise as a mediator of delayed axonal degeneration and is known to control important mechanisms in the central nervous system (12,13). In the retina, prolactin serves as a neurotrophic factor required for maintaining homeostasis during both injury (14) and physiologic aging (15), with receptors prominent in the ganglion cell layer as well as the outer and inner nuclear layers (16). In white matter, high prolactin levels signal oligodendrocyte proliferation, modulate neurotransmission (17) and promote white matter repair after demyelination in mice (18). A specific disease model -compression of retinofugal fibers by large prolactin-secreting pituitary tumors (i.e. prolactinomas) provides a window with which to observe the potential mechanisms through which prolactin exerts a neuroprotective role in the injured human brain (19). Patients with prolactinomas often experience a stereotyped loss of vision in the temporal hemifields secondary to mass effect, in addition to multiple endocrinopathies (20). Treatment using dopamine agonists (e.g. cabergoline), increases the available substrate capable of binding D2 receptors on lactotroph cells in the anterior pituitary gland and inhibits prolactin release (21). A secondary effect of cabergoline is reduction in tumor size (21). As such, it is used as a first line treatment for prolactinomas. Surgical decompression is reserved for cases of dopamine agonist-resistant prolactinomas or acute changes in tumor size that necessitate immediate intervention to prevent blindness. Dopamine agonist therapy for large prolactinomas can rarely cause symptomatic empty-sella syndrome. A syndrome characterized by both downward herniation of the optic chiasm into an empty sella turcica (the bony cave that houses the pituitary gland) and delayed secondary vision loss. Only 21 cases are known to be reported in the literature, a large majority of them treated with either reduction or cessation of dopamine agonist therapy (22)(23)(24)(25)(26)(27), or with surgical management to untether or elevate the optic apparatus (28)(29)(30)(31)(32)(33)(34).
With cessation of dopamine agonist therapy, all reported patients demonstrated improvement of visual outcomes without any changes in either the level of herniation or appearance of the optic chiasm. How vision returns despite persistent downward herniation of the chiasm in this small cohort of patients has not previously been studied.
Here we investigate mechanistic hypotheses of the effect of prolactin on glial-neuronal function in an observational study of a rare patient with symptomatic empty sella syndrome. Both delayed axonal injury and retinal health were studied longitudinally at three time points over a year, using a combination of diffusion magnetic resonance imaging (dMRI) and optical coherence tomography (OCT), respectively -and compared with a cohort of healthy control participants. Indexed measures of diffusion have been used as an in vivo proxy for myelin integrity (3,(35)(36)(37)(38). Specifically, an increase in radial diffusivity (RD) without changes in axial diffusivity (AD) indicates a breakdown in myelin, while decreased AD indicates axonal degeneration (37). These properties have been leveraged in human subjects to characterize microstructural changes occurring across the length of the optic tract for numerous pathologies that include pituitary macroadenomas (5), optic neuritis (39), and glaucoma (40). Additionally, retrograde degeneration of retinal ganglion cells can be characterized as a function of retinal nerve fiber layer (RNFL) and ganglion cell complex (GCC) thickness as measured by OCT (41)(42)(43).
Notably, the patient under study temporarily resumed cabergoline on her own volition prior to her second research visit. In retrospect, that event provided a natural contrast for observing the effects of varying levels of serum prolactin on measures of visual function, white matter integrity, and retinal health that otherwise would not be possible to observe. White matter integrity and retinal health were correlated with both serum levels of prolactin and ophthalmologic assessment of visual function (Humphrey perimetry) -allowing us to infer a causal relation between prolactin and maintenance of the coupled structure-function relationship between white matter integrity and visual ability. Secondly, this study demonstrates that diffusion MRI metrics are sensitive to detect early changes in white matter associated with varying levels of serum prolactin.

Participant Recruitment
This research was conducted as part of an ongoing pituitary tumor research study approved by the Research Subjects Review Board at the University of Rochester (RSRB00071763). Patient AJ, a 36-year old woman with a large prolactinoma was recruited as part of this study. See supplemental text, S1 for a detailed clinical history. Five healthy control participants (n = 10 total hemifields, mean age 35.8 ± 11.88 SD) were also recruited and reviewed by an ophthalmologist (ZW) to confirm eligibility. Exclusion criteria included glaucoma, diabetic retinopathy, history of central retinal artery occlusion, optic disc drusen, multiple sclerosis, stroke, and previous head trauma. Supplemental Table 1 displays basic demographic information for all study participants. All participants gave written consent for participation in the study.

Measurement of Serum Prolactin Levels
Serum prolactin was obtained via laboratory blood draw by a trained phlebotomist and analyzed by the University of Rochester Clinical Laboratories as part of the patient's routine clinical care. The prolactin was measured via Electrochemiluminescence Immunoassay with a reference range of 4.8-23.3ng/ml. Approximately 10 laboratory draws were performed over the course of a year (see supplemental Figure 1) and data from each research visit was paired with the closest obtained laboratory value for analysis. Control participants each had a one-time blood draw obtained at a single University of Rochester Medical Center outpatient clinic on the same date as their OCT or dMRI testing, and analyzed by the same clinical laboratory as patient, AJ.

Ophthalmologic Evaluation
Formal Ophthalmologic Testing: Automated 24-2 Humphrey perimetry (Zeiss HFA II-i series) and three-dimensional macular cube OCT (Zeiss Cirrus HD-OCT model 5000, 512x128 scan protocol with 6x6x2mm volumes) were performed for each eye -right eye (OD) and left eye (OS). Peripapillary retinal nerve fiber layer thickness (pRNFL) was additionally obtained using an optic disc cube 200 × 200 protocol. Testing was performed three times over a year for patient AJ, and at a single time point for all control participants. Crawford and Howell's modified t-test was used to compare retinal thickness measurements and mean deviation for each eye at each time point for patient AJ to the control population (44). analyzed with respect to visual hemi-field (i.e. the averaged thickness of homonymous hemiretinas) for comparison with optic tract integrity. To account for contributions from nasal versus temporal halves, retinal measures were weighted 53% contribution from nasal hemi-retina and 47% for temporal hemi-retina (46). The relations between weighted retinal thickness by hemi-field and corresponding optic tract diffusion metrics across all participants were evaluated using linear regression. Processed OCT data from Zeiss was unavailable for one healthy control due to acquisition artifact and thus excluded from the GCC analyses (controls n = 8 total hemifields).

Magnetic Resonance Imaging Acquisition and Processing
MRI Acquisition and Analysis: Scanning was performed at the University of Rochester Center for Advanced Brain Imaging and Neurophysiology on a 3T Siemens MAGNETOM Prisma scanner with a 64-channel head coil. T1 weighted images were acquired at the start of each session with a MPRAGE pulse sequence (TR=2530 ms, TE=3.44 ms, flip angle=71°, FOV=256×256 sq mm, matrix =256×256, resolution=1 cu mm, 192 sagittal slices). Diffusion MRI data were acquired using a single-shot echo-planar sequence (65 diffusion directions, echo spacing=0.66ms, EPI factor=172, b=0, 1000, 3000s/sq mm, 96 slices, resolution=1.5 cu mm, 68 non-diffusion weighted volumes). Three non-diffusion weighted volumes were collected at the same resolution with reversed phase-encode blips to estimate the susceptibility-induced off-resonance field as implemented in FMRIB software library, or FSL (47,48). FSL utilities were used to reduce motion artifacts and eddy current distortions and perform brain extraction (49).
Probabilistic tractography of the optic tracts was performed on the preprocessed b=1000 files, using two fibers per voxel and Bayesian estimation (50-52), following techniques previously described (5,53). All data in the main text are reported at a threshold of 2% for radial (RD) and axial (AD) diffusivity. See Supplemental Text, S2 for additional discussion on tractography, threshold determination, and supplemental analyses related to fractional anisotropy and mean diffusivity.

Structural Analysis of the Optic Tracts
Optic tract cross-sectional area (CSA) was approximated using T1-MPRAGE scans of both patient AJ and controls, assuming an elliptical shape. Measurements were made using Horos, an open source DICOM viewer, freely available for download at: www.horosproject.org. Height and width were evaluated just posterior to the optic chiasm. One trained researcher (RH) obtained measurements for all participants. A Welch's t-test was done to compare right and left optic tract CSA of this study's controls with the respective CSA measured by Andrews et al., 1997, with no statistically significant difference between the data sets (54). The Crawford and Howell's modified t-test was subsequently used to compare the patient's right and left optic tract CSA at each time point with the healthy controls (44). Linear regression analyses were used to relate CSA with prolactin and diffusion MRI metrics.

Clinical Presentation: Patient AJ, a 36-year-old Woman with Empty Sella Syndrome
Patient AJ presented to the UR Medicine Pituitary Program at the University of Rochester with complaints of progressive bilateral visual field defects, photosensitivity and bilateral ocular pain nine years after initial diagnosis and subsequent medical treatment of a large prolactin-secreting pituitary tumor ( Figures 1A-B). Clinical MRI demonstrated both a significant reduction in tumor size compared with her initial diagnosis (9 years earlier) and downward herniation of the chiasm into an empty sella turcica ( Figure 1C). Ophthalmologic examination revealed bitemporal visual field defects on Humphrey perimetry (AJ0, Figure 1D) with intact visual acuity (20/20 in both OS and OD) and normal pRNFL (OD, 75 µm; OS 88 µm) (AJ0, Figure 1E). Prolactin levels were measured at 17. supplemental text S1 for a detailed clinical history.  Table 4). Notably, this pattern was not significant (all p > 0.05), and thus consistent with values for patient AJ at the border of normal physiologic thickness (75 µm for pRNFL). Prior work indicates that values below the threshold of 75 µm are associated with a low probability of recovery of visual function following injury to the optic tracts (42,55). Reduced thickness was primarily in the nasal hemiretinas bilaterally ( Figure 1E).

Ophthalmologic Evaluation of Visual Function and Retinal Thickness
GCC data were also correlated with mean deviation, demonstrating a significant relation between retinal thickness and visual function across all study participants (r = 0.63 and p = 0.015; Figure 2D). These data allow us to infer a causal relation between GCC and Humphrey perimetry. It is important however, to note that the stability of retinal thickness measures for patient AJ observed over the duration of the study, suggests that the transient decline in visual function at time point 2 (AJ2) cannot be primarily explained by macrostructural changes within the retina. Raw data for all retinal layers is provided in the supplemental text.

Diffusion MRI is Sensitive to Alternations in Serum Prolactin Levels
Diffusion MRI was obtained on all study participants, and at three separate time points for patient AJ in tandem with ophthalmologic and laboratory evaluations. At the onset of the study (AJ1), and shortly after initial discontinuation of cabergoline, AJ's diffusion metrics within the optic tracts were within the range established by the healthy controls sample ( Figure 3A-B

Optic Tract Size does not Correlate with Measures of Diffusion or Hormonal Function
A core finding described above is that diffusion indices of myelination track prolactin levels in patient AJ. In order to rule out the possibility that the observed effects on diffusion indices are derivative of macrostructural changes, such as thinning of the optic tracts secondary to increasing traction on the chiasm, we measured optic tract size for all healthy controls (n = 10 optic tracts) and patient AJ across all time points. The range of right optic tract CSA values was 6.22-9.96 mm 2 (mean=8.47 mm 2 ) for healthy controls and 9.56-10.02 mm 2 (mean=9.81mm 2 ) for patient AJ. In the left optic tract, values ranged from 6.27-11.00 mm 2 (mean=9.18 mm 2 ) for controls and 9.83-10.35mm 2 (mean=10.08mm 2 ) for patient AJ. These data are consistent with previously published studies (54) and there were no right-to-left discrepancies. No statistical differences were identified between optic tract CSA in patient AJ (at any time point) and healthy control participants (p = 0.14, using Welch's t-test). The tight range of CSA values at each time point demonstrates stability of optic tract size and suggests that increased traction or physical deformation of the optic tract is not the primary factor in the patient's worsened clinical exam at AJ2. This is further supported by the fact that no correlations were identified between optic tract size and measures of AD, RD, prolactin or GCC, all p > 0.34.

DISCUSSION
Here, we demonstrate a stereotyped pattern of diffusion MRI changes in response to varying levels of serum prolactin in a rare, isolated, model of nerve traction injury. Specifically, increased RD indexed a significant decrease in serum prolactin. Important to this finding, is that no relations were present between axial diffusivity, levels of prolactin, retinal thickness or visual function. These data confirm, in-vivo, previous observations suggesting a neuroprotective role of prolactin in the human brain (12,13,17,18,56) and are consistent with the suite of diffusion MRI signals that have been shown to index the integrity of myelin (3,(35)(36)(37)(38). Adding to the specificity of these stereotyped diffusion MRI changes is the demonstration of concordant changes in visual ability, indicating disruption of a tightly coupled structure-function relationship between vision and white matter integrity with decreasing serum levels of prolactin.
Notably, these patterns were present in the absence of macrostructural changes to the optic tracts, as assessed by CSA and are supported by stable measures of retinal thickness and axial diffusivity. Given the specificity of these findings it is reasonable to infer that serum levels of prolactin causally affect the coupled structure-function relation between white matter integrity (i.e. myelination) and visual ability. Additionally, these data support the use of diffusion MRI as an independent index of the effects of prolactin on white matter in the human brain, both for future observational studies and for future interventional studies.

Radial diffusivity as an index for glial-neuronal interactions that support myelin health. Prior work has documented
a pattern of reduced RD after surgical decompression of the optic chiasm that is disproportionate to changes in AD (5). This pattern is thought to characterize rapid remyelination in the human brain (i.e., days to weeks) following chiasmatic decompression. The current study demonstrates this same pattern of diffusion changes following periods of hyperprolactinemia rather than decompression -and does so in the setting of chronic injury from persistent traction.
These data suggest that in the setting of an isolated stretch injury, radial diffusivity indexes the glial-neuronal interactions that support myelin sheath integrity and preservation of function. Key to this conclusion is that despite persistent traction on the optic chiasm throughout the study, white matter injury, as assessed with diffusion MRI, was only made apparent after AJ resumed cabergoline on her own volition -an event that resulted in decreased levels of serum prolactin. This is an important distinction given what is currently known about both neuronal stretch injury (8,57) and prolactin's role in neural recovery (13,17,18). Once a nerve is stretched, a stereotyped pattern of microstructural changes leads to the development of axonal swelling and correlated changes in the microtubular component of the axonal cytoskeleton, particularly microtubule loss at nodes of Ranvier and at internodal regions (8).
More severe injury models describe diffuse patterns of injury with permeabilization of the axolemma and subsequent formation of peri-axonal spaces, myelin inclusions, and reactive axonal swellings culminating in secondary axotomy (57).
There is a growing body of evidence in both in-vitro and in-vivo models demonstrating prolactin's role in mediating glial-neuronal interactions, including improved astrocyte viability and decreased astrogliosis (12). Critical for inferring microstructural properties from diffusion MRI data is the b-value used during image acquisition: in our case b=1000. At this b-value, diffusion patterns are most sensitive to changes in the extra-axonal compartment (58).

Recruitment of oligodendrocyte precursor cells to the injured optic tract can effectively decrease radial diffusivity and
is consistent with known properties of prolactin (18). Recovery of both the number and density of microtubules, which has been shown to occur as quickly as 4 hours after an optic nerve stretch injury (8) can also reduce radial diffusivity and represents a potential area of future interest with respect to possible associations with prolactin. These physical changes, which can be measured in the extra-axonal space, likely accompany upregulation of neurotransmission that improves signal conduction and information transfer to the striate cortex. Increased signal conduction alone would be unlikely to provide microstructural changes large enough to be measured at the resolution of diffusion tensor imaging.
Data from this study, therefore, add to the argument that prolactin influences a complex network of interactions important for maintaining the health of oligodendrocytes, cytoskeletal structures, and levels of myelination -which would otherwise be susceptible to secondary axonal injury.

Stability of Retinal Thickness Measures.
Recovery after retinal thinning, as measured by OCT varies with pathology (41-43, 45, 55, 59, 60). Optic neuritis patients experience recovery of visual function with increasing VEP amplitudes despite continued loss of pRNFL thickness and significant optic nerve atrophy over 12 months (61). Klistorner and colleagues suggest that this discrepancy is driven by a combination of both remyelination and neural reorganization.
When used to measure recovery following nerve decompression (e.g. as treatment for a pituitary macroadenoma), pRNFL and photopic negative response are sensitive to detect changes at three months (41)(42)(43)55). These changes lag behind both improvements in visual function and diffusion MRI indices and are relatively stable at the individual level. In chiasmal compression by pituitary tumors, GCC thickness analysis had greater correlation to mean deviation, a measure of visual defect, than RNFL analysis (62). Additionally, some patients had GCC thinning but no abnormalities in RNFL or mean deviation of visual defect on perimetry testing (63), suggesting increased sensitivity of GCC to injury. As such, this study primarily focused on GCC.
Here, we demonstrate preservation of retinal ganglion cell thickness despite persistent traction injury to the optic nerves, chiasm and tracts over the course of a year, including a transient period (AJ2) of worsening white matter injury and decreased vision. Only one other case with a similar pattern of preserved pRNFL thickness despite recorded deficits on Humphrey perimetry and injury to the optic chiasm has been reported in the literature -a patient with a large, compressive prolactinoma and hyperprolactinemia that was evaluated prior to starting cabergoline (64). While it is possible that the stability of retinal thickness measurements in patient AJ is secondary to exposure to high levels of prolactin -this pattern is complicated by a known delay (approximately 6-8 weeks) in measurable response to pathology, which warrants further investigation. Our findings indicate that radial diffusivity is a sensitive marker that is dynamically modulated in the central nervous system coincident with rapidly changing serum prolactin levels.
The role of cabergoline. Given that the transient normal prolactin level measured in patient AJ was precipitated by use of cabergoline, it is possible that cabergoline has an independent effect on radial diffusivity and visual function.
Chuman and colleagues suggest that cabergoline drug toxicity directly results in vision loss, and that cessation of therapy subsequently leads to vision recovery (22). While plausible, this pattern of delayed vision loss is not typically seen in patients treated with high doses of cabergoline (65), and has not been reported in the absence of empty sella syndrome. Further investigation in animal models relating the impact of cabergoline on D2 receptors in white matter may help to better separate differences between the changes related to prolactin versus those related to dopamine agonists alone. A second hypothesis proposes that in the absence of cabergoline, undetectable tumor regrowth leads to untethering of the of the optic chiasm and a subsequent return of visual function. Two issues arise in this context.
First, no patient has been reported to demonstrate improvement in the degree of optic chiasm herniation, at least across all case reports of which we are aware and for which cessation of medical therapy was the treatment strategy (22)(23)(24)(25)(26)(27).
Second, the amount of deformation of the optic chiasm poorly correlates with visual function, as demonstrated in both compressive pituitary tumor patients (66), and individuals with an incidental finding of primary empty sella syndrome as an anatomic variant (67). As such, we suggest that it is unlikely that an "undetectable" change in tethering is responsible for the significant reduction in radial diffusivity and improved visual function observed with hyperprolactinemia in this study.

CONCLUSIONS
In summary, we demonstrate in a single patient with empty sella syndrome secondary to dopamine agonist therapy, that increasing serum levels of prolactin correlate with improved visual function and a reduction in radial diffusivity.
These data support, in-vivo, a neuro-protective role for prolactin in the injured human brain, confirming work previously conducted in animal models, and establishes radial diffusivity as an important index for tracking the white matter impact of varying levels of serum prolactin. These findings offer a non-invasive means of measuring the effectiveness of novel therapies targeting prolactin as a mediator of neuroprotection in the human brain.