Seasonal rewiring of the songbird brain: an in vivo MRI study

Authors


Dr G. De Groof, as above.
E-mail: geert.degroof@ua.ac.be

Abstract

The song control system (SCS) of songbirds displays a remarkable plasticity in species where song output changes seasonally. The mechanisms underlying this plasticity are barely understood and research has primarily been focused on the song nuclei themselves, largely neglecting their interconnections and connections with other brain regions. We investigated seasonal changes in the entire brain, including the song nuclei and their connections, of nine male starlings (Sturnus vulgaris). At two times of the year, during the breeding (April) and nonbreeding (July) seasons, we measured in the same subjects cellular attributes of brain regions using in vivo high-resolution diffusion tensor imaging (DTI) at 7 T. An increased fractional anisotropy in the HVC–RA pathway that correlates with an increase in axonal density (and myelination) was found during the breeding season, confirming multiple previous histological reports. Other parts of the SCS, namely the occipitomesencephalic axonal pathway, which contains fiber tracts important for song production, showed increased fractional anisotropy due to myelination during the breeding season and the connection between HVC and Area X showed an increase in axonal connectivity. Beyond the SCS we discerned fractional anisotropy changes that correlate with myelination changes in the optic chiasm and axonal organization changes in an interhemispheric connection, the posterior commissure. These results demonstrate an unexpectedly broad plasticity in the connectivity of the avian brain that might be involved in preparing subjects for the competitive and demanding behavioral tasks that are associated with successful reproduction.

Introduction

One of the most important developments that has taken place in neuroscience in the past 25 years is the realization that the brain is not the fixed structure it was thought to be but rather displays extensive dynamic changes. These changes constitute what is commonly called neuroplasticity. Understanding their specific nature and control mechanisms represents a critical step towards a full understanding of brain functioning.

Some of the most dramatic brain structural modifications are the seasonal changes affecting a connected set of brain nuclei controlling singing behavior, the song control system (SCS) in oscine songbirds. In most temperate zone species, reproduction is a seasonal phenomenon. In a specific group of birds belonging to the order Passeriformes behaviors associated with reproduction, such as singing, are performed at higher rates during the breeding season (Phillmore et al., 2006). In parallel, a seasonal variation in the volume of song control nuclei has been observed (e.g., Nottebohm, 1981; Kirn et al., 1989; Brenowitz et al., 1991, 1998; Bernard & Ball, 1995, 1997; Smith et al., 1995; Smith, 1996; Caro et al., 2005). Due to the magnitude of these changes, sometimes as large as a 99% increase (Nottebohm, 1981), seasonal variation in the brain of songbirds has emerged as one of the best model systems for the study of naturally occurring brain plasticity (Tramontin & Brenowitz, 2000; Ball et al., 2004; Brenowitz, 2004).

Most of the gray matter structures known to be plastic are part of the SCS (Nottebohm, 1981; Hill & DeVoogd, 1991; Tramontin & Brenowitz, 1999, 2000; Tindemans et al., 2003; Van Meir et al., 2004, 2006; Thompson & Brenowitz, 2005). Recently, improvements in in vivo magnetic resonance imaging (MRI) techniques have made it possible to assess neuroplasticity of the SCS (Tindemans et al., 2003; Van der Linden et al., 2004; Van Meir et al., 2004, 2006). Diffusion tensor imaging (DTI) has also been implemented for the analysis of the songbird brain (De Groof et al., 2006) in the form of a high resolution variant of the method used to assess changes in the brain during human brain development (Huppi & Dubois, 2006; Cascio et al., 2007; Lebel et al., 2008) or neurodegeneration (Ulug et al., 1999; Rose et al., 2000; Taber et al., 2002). DTI is an MRI method that uses water diffusion as a highly sensitive marker of the microarchitecture of cellular membranes. The noninvasive in vivo nature of the technique allows for longitudinal studies in individuals following changes with time.

In the present study, we repeatedly imaged by DTI the brain of nine individual male starlings during the breeding and the nonbreeding season. We focused on potential changes in the fiber tracts within and outside the SCS because this aspect of brain structure has been poorly, if at all (see however Holloway & Clayton, 2001), investigated by histological techniques. We have demonstrated in this way an unexpected seasonal plasticity in the wiring of the brain both within and outside the SCS.

Materials and methods

Experimental setup

Nine male European starlings (Sturnus vulgaris; ∼75 g) were obtained from wild-caught stock maintained in large outdoor aviaries at the Drie Eiken Campus (University of Antwerp). They were temporarily housed during the experiments in two indoor cages (1.40 × 2.20 × 2.10 m) under an artificial light–dark cycle reproducing the photoperiod observed at the corresponding time of the year. The experiments were conducted during the breeding season (April 7–26, 2004) in Belgium, and repeated outside this season (July 19–August 2, 2004) when birds have become photorefractory and are no longer stimulated by long-day photoperiods so that their gonads are fully regressed (Dawson, 1983). Between experiments, birds were relocated in the large outdoor aviaries at the University of Antwerp. Food and water were available ad libitum. All birds were individually marked with a numbered metal ring and color bands. All experimental procedures were performed in accordance with the European guidelines for the care and use of laboratory animals (86/609/EEC) and were approved by the Committee on Animal Care and Use at the University of Antwerp, Belgium.

DTI–MR protocol

Birds were anaesthetized as previously described (Van Meir et al., 2004) with an initial intramuscular (chest) injection of 5 mL/kg of a mixture containing 0.33 mL xylazine (Rompun; 20 mg/mL), 2.10 mL ketamine (Ketalar; 50 mg/mL) and 4.33 mL saline solution. During the whole experiment starlings were kept anaesthetized with this mixture at a rate of 0.15 mL/h through a chest catheter (Micro-Flo, 27GA; DKS Overscan, Milano, Italy) and body temperature was continuously monitored and automatically regulated within a narrow range of 40–41°C. The bird’s head position was fixed by a nonmagnetic stereotaxic beak-bar and head-holder combined with a circular receive-only surface coil (diameter 24 mm) and a transmit head coil (Helhmoltz; diameter 45 mm). Imaging was carried out on a 7 T horizontal bore MR microscope (MRRS, Guildford, UK), provided with shielded gradients (8 cm width, maximal strength 400 mT/m; Magnex Scientific Ltd., Oxfordshire, UK).

First, pilot images were acquired in three orthogonal directions. DTI-MR data were then obtained, consisting of twenty-four adjacent sagittal slices (thickness 0.4 mm) covering the entire right hemisphere of the brain. Diffusion-weighted (DW) spin-echo images were acquired with diffusion gradients applied in six noncollinear directions (diffusion gradient strength 69 mT/m for each direction, time diffusion gradient δ = 12 ms, interval between onsets diffusion gradients Δ = 20 ms). The diffusion gradient scheme used was the Basser scheme (Basser et al., 1994). We decided not to use the single-shot echo-planar imaging (SS EPI) method applied in most human clinical studies because of the associated increase in susceptibility gradients in high-field conditions. Furthermore, an SS EPI protocol could not guarantee the higher spatial resolution achieved by spin echo sequence (in our case 0.1 × 0.1 mm). Additional image parameters were: field of view (FOV) 25 mm, spectral width 25 kHz, TE 43 ms, TR 2200 ms and acquisition matrix 256 × 128 (image matrix 256 × 256). To minimize the effects of residual motion we used multiple averages (Nav = 14). Each DTI experiment took ∼ 8.5 h. Following the experiment the starlings recovered under an infrared light, after which they were returned to their aviary. All starlings recovered fully after each MR experiment.

DTI-MR data processing and statistical analyses

Any motion during the measurement was corrected for by coregistering the six DW images to the non-DW image (b0-image) by maximization of mutual information (MIRIT; multimodality image registration using information theory; Maes et al., 1997). The b-matrices were calculated using analytical expressions (Mattiello et al., 1997) incorporating diffusion gradients and image gradients [b-value (diffusion gradients only) = 788 s/mm2]. Diffusion tensor images and fractional anisotropy (FA) maps were calculated using ExploreDTI (Leemans et al., 2005) implemented in MATLAB code (The Mathworks Inc., Natick, MA, USA). Diffusion anisotropy (derived from the diffusion tensor) describes how variable the diffusion is in different directions within a given voxel and is commonly quantified via FA (Pierpaoli & Basser, 1996). FA was computed on a voxel-by-voxel basis using the following equation (Le Bihan et al., 2001).

image

with λ1, λ2 and λ3 being the eigenvalues of the diffusion tensor and <λ> the average of the three eigenvalues. The eigenvalues of the diffusion tensor, λi, can be interpreted as the diffusivity in the axial and radial directions of a fiber tract: axial diffusivity (AD) = λ1 and radial diffusivity (RD) = (λ2 + λ3)/2 (Song et al., 2005).

Regions of interest (ROIs)

Considering that the FA maps show the best neuroanatomical contrast in starlings (De Groof et al., 2006), ROIs were delineated on these maps (see Fig. 1) using AMIRA software (Amira software version 3.1; Mercury Computers Systems, San Diego, CA, USA), a program designed to analyze imaging data. Paired t-tests (Wilcoxon signed-ranks test) were performed on the ROI data. Differences were considered significant for P < 0.05.

Figure 1.

 Regions of interest. (A–E) Sagittal FA maps (from 0.4 mm midsagittal to 3.2 mm midsagittal) of one individual starling in spring, on which the ROIs are depicted. These ROIs were drawn with the help of the AMIRA program based on our previous DTI study (De Groof et al., 2006) and on atlases of the avian brain (Stokes et al., 1974). (F) Schematic drawing of the song control system and its anatomical connections (note the LMAN→RA projection and the HVC→X projection both run along the LaM, in red). Image resolution, 0.1 × 0.1 mm. Abbreviations: CoA, commissura anterior; CO, chiasma opticum; CoP, commissural posterior; DLM, nucleus dorsolateralis anterior thalamis, pars medialis; DM, dorsomedial nucleus of the intercollicular complex; LaM, lamina mesopallialis; LFS, lamina frontalis superior; LMAN, lateral magnocellular nucleus of the anterior nidopallium; LPS, lamina palliosubpallialis; nXIIts, nucleus nervi hypoglossi pars tracheosyringealis; OM, tractus occipitomesopallialis (medial telencephalic part); PAm, nucleus parambigualis; RA, robust nucleus of the arcopallium; RAm, nucleus retroambigualis; X, Area X. Scale bar, 10 mm.

The neuroanatomical nomenclature used in the present paper is based on the recently revised nomenclature for the avian brain (Reiner et al., 2004). All regions were outlined on the FA maps based on the information provided in this description of the avian brain (Reiner et al., 2004) and on the previously published DTI study of the starling brain (De Groof et al., 2006). One additional structure not described in our previous study was also outlined here. It is the dorsomedial nucleus of the intercollicular complex (DM) and was delineated on a slice positioned 3.2 mm lateral from the midsagittal plane (Fig. 1E). DM was identified as an ovoid area located between tectum opticum and nucleus intercollicularis.

Correlations between FA, changes in FA and volume

Relationships between individual differences in FA and seasonal changes in FA were analyzed with the Pearson’s product moment correlation coefficient (r). Correlations were considered significant for P < 0.05. In each case, the correlation coefficient r and the associated probability (P) are provided.

DTI fiber tracking

Fiber tracking was applied to the diffusion tensor data sets, using custom-written software (ExploreDTI; Leemans et al., 2005) as previously described (De Groof et al., 2006). The FA threshold for beginning the tracking procedure was set to 0.1 and the FA threshold for stopping the procedure was set to 0.01. The ROI for ‘seeding’ the tracking algorithm (seedpoint) was chosen manually on the FA maps. The color code of the fibers represents the orientation of the principal diffusion direction (green, dorsal–ventral direction; red, rostral–caudal direction; blue, medial–lateral direction).

T2-data protocol and data-processing

In order to obtain quantitative information for the tissue T2 within the brain, a spin echo multislice (FOV, 25 mm; 24 sagittal slices covering the right hemisphere; slice thickness, 0.4 mm) multiexperiment was performed (acquisition matrix 256 × 128; TR, 2000 ms; 10 averages) at the same position and orientation as the DTI images. This included two experiments in which the T2 weighting was gradually increased during the two experiments (TE1, 18 ms; TE2, 56 ms). The calculation of quantitative maps in which the grey level of each pixel represents the fit parameter T2 was carried out using in-house software developed in IDL (Interactive Data Language). All data in the text are represented by their mean ± SD.

Histology

The brains of five starlings (three killed by decapitation in the spring during the breeding season and two in the summer when birds are photorefractory) were fixed by immersion in 4% paraformaldehyde and then, after cryoprotection with 20% sucrose, were cut in sagittal sections on a cryostat at 0.2 mm thickness. Sections were then postfixed for 20 min in 10% paraformaldehyde, rinsed in phosphate-buffered saline (PBS) and stained for myelin for 10 min in Luxol blue (0.1% in acetic acid) at 65°C. After three rinses in distilled water, coloration was differentiated in a Li2CO3 solution and fixed for 30 s in 70% norvanol, and sections were rinsed again several times in water. A counter-coloration of cell bodies was then performed by dipping sections for 4 min in 0.5% periodic acid and then in a hematoxylin solution for 5 min. After further rinses in distilled and then tap water, sections were mounted on microscopic slides and coverslipped in a gelatin-based medium.

Results

FA maps provide a detailed view of myelinated fiber tracts

In agreement with our previous DTI study (De Groof et al., 2006), FA maps provided high resolution very detailed views of the starling brain with a particularly sharp identification of fiber tracts due to their highly polarized 3-D organization. Fig. 1 provides a series of sagittal slices organized in a medial to lateral order (from 0.4 to 3.2 mm from the midline) illustrating the different fiber tracts that were analyzed quantitatively in a seasonal context.

The SCS is organized into two pathways. Both pathways start from the nucleus HVC (used as the proper name; formerly called the high vocal center). The main descending motor pathway is essential for song production and includes the robust nucleus of the arcopallium (RA). The anterior forebrain pathway (AFP) is essential for song learning and for the maintenance of song in relation to auditory feedback (Brainard & Doupe, 2000). This pathway includes Area X (used as proper name) and the lateral magnocellular nucleus of the anterior nidopallium (LMAN).

The AMIRA software was used to delineate the ROIs covering the major song control nuclei (RA, Area X and LMAN, with exception of HVC which is not visible on the FA maps) and all the largest fiber tracts. These included major connections within the SCS including the HVC projection to RA, the lamina mesopallialis (LaM), which includes a large part of the projection of HVC to Area X (Nottebohm et al., 1976) as well as a long section of the projection from LMAN to RA (Nottebohm et al., 1982; Bottjer et al., 1989) and the projection of RA to the DM which is incorporated in the tractus occipitomesopallialis (OM; Wild, 1993), as well as the fiber capsule surrounding Area X and RA. The distribution of these fiber tracts along the medial to lateral axis of the brain is clearly visible in these images. A schematic presentation of these connections and their anatomical relationship with the major lamina separating the layers of the telencephalon is also presented in Fig. 1F. Other fiber tracts not directly connected to the SCS were also clearly visible; these included the lamina frontalis superior (LFS) and lamina palliosubpallialis (LPS), the commissura anterior and posterior (CoA and CoP) and the optic chiasma (CO).

Interpretation of DTI-MR data

DTI is a powerful MRI method allowing detection with high sensitivity of microscopic differences in tissue properties. Water diffusion in tissues is highly sensitive to differences in the microarchitecture of cellular membranes. The apparent diffusivity will increase in a voxel when the average space between membrane layers increases, and conversely smaller spaces will lead to lower diffusivities. This makes this technique particularly sensitive to structural changes affecting fiber tracts in the brain. The noninvasive in vivo nature of the technique also allows for repeated measures and thus longitudinal studies following changes in time in focal subjects. The power of repeated measures, allowing use of subjects as their own controls, combined with statistical analyses of 3-D images and with the multiple levels of cellular information that can be deduced from the DTI parameters allows location of anatomical and cellular plasticity on a whole-brain scale, i.e. within and outside the SCS.

A meaningful voxel-by-voxel interpretation of a diffusion tensor image requires simplification of the tensor data (i.e. the three eigenvalues, λ1,2,3; see Fig. 2). Therefore tensor information is usually distilled into scalar maps. The two most common measures used are mean diffusivity (MD) and FA. The MD is the average of the three eigenvalues (λ1,2,3) of the tensor (Fig. 2) and reflects the average diffusion (in any direction) of the whole voxel. MD changes correlate with isotropic changes in water diffusion, reflecting changes in the ratio between intra- and extracellular volume that are due to either cell swelling or shrinkage (movement of water in or out of the cells) but also to dendrite branching (see Beaulieu, 2002; Mori & Zhang, 2006 for reviews). FA on the other hand is a measure of the preference of the diffusivities in a certain direction. It is calculated from the three eigenvalues (see Materials and methods section) and is quite sensitive to changes in the diffusivities. A change in FA, however, is not always straightforward in interpretation because it depends on all three eigenvalues. Therefore several recent studies have suggested that specific combinations of the eigenvalues [e.g. RD, (λ2 + λ3)/2] could be used to describe diffusion changes in pathology. Within white matter RD appears to be modulated mostly by myelin content (see Fig. 2A and B), whereas the axial diffusivity (AD, λ1) is more specific to the number and coherence of axons in the voxel (Song et al., 2002; also Fig. 2C).

Figure 2.

 The diffusion tensor and its relation to the microstructure of tissues. (Left) Ellipsoidal representation of the diffusion tensor field with a T2-weighted background image. Note that every voxel (position r) of the data set is uniquely defined by three eigenvalues λi(r) of the diffusion tensor D(r). (Right) Examples of shapes of the tensors associated with different microstructural arrangements within the pixel. The red path is a schematic representation of random diffusion of a proton. A is an example of an unmyelinated fiber tract, B and C are myelinated fibers but B has coherence of the axons while C does not. Note both A and C have approximately the same FA (same cigar shape) but the eigenvalues (λi) are quite different, underlining the need for investigation of all DTI parameters.]

Many published studies have focused primarily on the diffusion anisotropy (usually the FA measure), which may not be sufficient to characterize tissue changes. For example, white matter pathology often causes decreases in the anisotropy, which may result from increased RD (diffusion perpendicular to the axons), reduced AD (diffusion parallel to the axons), or both. An increased RD has been correlated with myelin loss in studies on animal models of experimental demyelination and remyelination (Song et al., 2005). Measurements of the MD may provide a better understanding of how the diffusion tensor is changing. In this study we quantified and analyzed all DTI parameters but also the T2 measure (which provides information on regional changes in total water content) in order to have a picture as complete as possible of underlying cellular changes in white matter.

Improved connectivity in the vocal motor circuit during the breeding season

It has long been known and well documented that the numbers of HVC neurons projecting to RA increases during the breeding season (Rasika et al., 1994; Hidalgo et al., 1995; Tramontin & Brenowitz, 1999; Van Meir et al., 2004, 2006). However, changes in this fiber tract per se have not been specifically quantified and, furthermore, seasonal changes in connectivity have never been studied in the entire vocal motor pathway. The present longitudinal DTI data demonstrated significant seasonal changes in multiple nuclei and their connections in this circuit including the HVC-RA connection, RA and the fibers surrounding this nucleus (RA-projecting HVC fibers forming a shell around RA; Konishi & Akutagawa, 1985), the medial telencephalic part of the tractus occipitomesencephalicus (OM; which contains RA fibers projecting towards the DM of the mesencephalon; Wild, 1993) and DM itself (Table 1). While the connection between HVC and RA displayed a decrease in FA (due to an AD decrease and an RD increase; Table 1) between the breeding and nonbreeding season, the medial part of OM (in the telencephalon) containing the connection between RA and DM displayed a decreased FA due to increased RD (λ2 + λ3)/2), a proven correlate for increased myelination (Song et al., 2002, 2005). Our data therefore suggest an upregulation of myelination during the breeding season that should support an improved conduction of electrical signals (Waxman, 1977). In order to verify our observation five subjects were killed for histology and with a Luxol–hematoxylin coloration we could indeed confirm that OM in starlings had more myelin during than outside the breeding season (Fig. 3).

Table 1.   DTI and T2 metrics for the vocal motor control pathway from breeding (April) and nonbreeding (July) starlings
ROIFractional anisotropy (FA)Mean diffusivity, MD (μm2/ms)Axial diffusivity, AD (μm2/ms)Radial diffusivity, RD (μm2/ms) T2 (ms)
  1. Values are mean ± SD of n = 9 subjects in each case. FA is a dimensionless value that varies between 0 (isotropic) and 1 (anisotropic). DM, dorsomedial nucleus of the intercollicular complex; HVC, formerly called the high vocal center; OM, tractus occipitomesencephalicus; RA, robust nucleus of the arcopallium; ROI, region of interest. *Values that changed seasonally, P < 0.05 (Wilcoxon signed-ranks test; April vs. July).

HVC-RA Tract
 April0.22 ± 0.030.57 ± 0.030.70 ± 0.040.50 ± 0.0239 ± 3
 July0.16 ± 0.02*0.57 ± 0.020.66 ± 0.03*0.53 ± 0.03*38 ± 2
Fiber capsule of RA
 April0.29 ± 0.020.65 ± 0.020.85 ± 0.030.54 ± 0.0238 ± 2
 July0.25 ± 0.02*0.65 ± 0.020.83 ± 0.01*0.56 ± 0.0238 ± 2
RA
 April0.18 ± 0.020.70 ± 0.020.81 ± 0.030.63 ± 0.0238 ± 2
 July0.15 ± 0.02*0.70 ± 0.020.81 ± 0.030.65 ± 0.0238 ± 2
OM
 April0.67 ± 0.030.59 ± 0.011.13 ± 0.050.31 ± 0.0236 ± 2
 July0.61 ± 0.04*0.63 ± 0.061.13 ± 0.050.37 ± 0.06*37 ± 3
DM
 April0.22 ± 0.020.63 ± 0.030.77 ± 0.030.56 ± 0.0333 ± 2
 July0.20 ± 0.02*0.62 ± 0.020.74 ± 0.03*0.55 ± 0.0333 ± 2
Figure 3.

 Seasonal changes in the connectivity of the telencephalic vocal motor circuit. The upper row presents sagittal FA maps (right is caudal) of the same subject in different seasons. The second row contains two sections in the sagittal plane (right end of section is caudal) from two starlings collected in April or July and stained with Luxol to identify myelinated fiber tracts. Notice that in April there were more myelin fibers in OM than in July; the fiber tract between HVC and RA (arrowheads) also contains more fibers and the size of RA is larger in April than in July. The third row of images presents fiber-tracking with RA as the seed point. The inserts show the individual diffusion tensors of the OM tract. Notice that they are broader in July than in April. This is due to the increase in λ2 and λ3 (increase in RD in July; see Table 1).

Improved wiring of the AFP during the breeding season?

Within the AFP, the LaM, which contains the connections between HVC and Area X and between LMAN and RA (see Fig. 1F; also Nottebohm et al., 1982) also displayed a higher connectivity during the breeding season as demonstrated by a λ1–induced change in FA [both FA and AD (AD = λ1) are higher in the spring; Table 2]. LMAN displayed the same high AD during the breeding season, probably because the fibers from HVC to Area X enter X from its dorsal margin and cross LMAN, as demonstrated in canaries (Nottebohm et al., 1976). Additionally, an increase in T2 was observed in LMAN from the breeding to the nonbreeding season. This in sharp contrast with changes detected in the neighboring nucleus Area X where at the same time a decrease in T2 was detected (Table 2). Previous studies indicated that the volume change in Area X during the breeding season can be attributed to a change in neuron size and spacing (Thompson & Brenowitz, 2005). This increased spacing between cells and larger neuron size are probably responsible for the higher T2 during the breeding season. This increase in free water in one nucleus associated with a decrease in an adjacent nucleus supports the anatomical specificity of these changes. Whilst the cellular bases of the changes observed in Area X probably involve variations in cell spacing, these cellular bases are not known for LMAN; however, DTI data suggest the existence of changes in the opposite direction, changes which should be investigated using histological techniques.

Table 2.   DTI and T2 metrics for the anterior forebrain pathway (AFP) from breeding (April) and nonbreeding (July) starlings
ROIFractional anisotropy (FA)Mean diffusivity MD (μm2/ms)Axial diffusivity AD (μm2/ms)Radial diffusivity RD (μm2/ms) T2 (ms)
  1. Values are mean ± SD of n = 9 subjects in each case. FA is a dimensionless value that varies between 0 (isotropic) and 1 (anisotropic). Area X, used as proper name; LaM, lamina mesopallialis; LMAN, lateral magnocellular nucleus of the anterior nidopallium; LPS, lamina palliosubpallialis; ROI, region of interest. *Values that changed seasonally, P < 0.05 (Wilcoxon signed-ranks test; April vs. July).

Area X
 April0.16 ± 0.030.54 ± 0.030.62 ± 0.040.50 ± 0.0334 ± 2
 July0.15 ± 0.020.54 ± 0.020.61 ± 0.020.50 ± 0.0233 ± 1*
Fibre capsule of X
 April0.27 ± 0.040.53 ± 0.020.69 ± 0.030.47 ± 0.0334 ± 2
 July0.23 ± 0.03*0.53 ± 0.020.66 ± 0.03*0.48 ± 0.0234 ± 1
LMAN
 April0.18 ± 0.020.56 ± 0.040.66 ± 0.040.51 ± 0.0431 ± 2
 July0.14 ± 0.03*0.55 ± 0.030.62 ± 0.04*0.51 ± 0.0233 ± 1*
LaM
 April0.28 ± 0.020.53 ± 0.020.70 ± 0.020.45 ± 0.0235 ± 2
 July0.25 ± 0.02*0.53 ± 0.020.68 ± 0.03*0.46 ± 0.0235 ± 1
LPS
 April0.23 ± 0.020.55 ± 0.020.68 ± 0.030.48 ± 0.0234 ± 2
 July0.21 ± 0.02*0.55 ± 0.020.67 ± 0.030.49 ± 0.0235 ± 1

We also identified seasonal changes in FA (decrease) and AD (decrease; Table 2) in the fibers surrounding Area X that either originate dorsally from LMAN or travel from the ventral edge of Area X to the dorsolateral nucleus of the medial thalamus (DLM; Vates & Nottebohm, 1995), and are all part of the anterior forebrain pathway (see Fig. 1F).

FA values of Area X and RA measured during the breeding season were correlated with FA values in LaM during the same season (r = 0.683, P = 0.043 and r = 0.752, P = 0.019 respectively) and also with the seasonal FA changes [100 × (FAspring– FAsummer)/FAspring] in LaM (r = 0.879, P = 0.002 for Area X and r = 0.772, P = 0.015 for RA respectively). Seasonal changes [100 × (FAspring– FAsummer)/FAspring] in FA values in RA were correlated with those obtained in Area X (r = 0.679, P = 0.044). These coordinated changes suggest functional relationships between the connectivities in the anterior forebrain pathway and the vocal motor system.

Increased myelination in the optic chiasma during the breeding season

The optic chiasma displayed a pronounced decrease in FA between spring and summer (Table 3) due to an increased RD in the nonbreeding season (Fig. 4, Table 3). An increased RD is a correlate for decreased myelination (see Fig. 1B). Other DTI parameters did not change between April and July in the optic chiasma (Table 3). The other main nuclei of the visual system were also investigated (nucleus rotundus, optic tectum, visual Wulst and nucleus geniculatus lateralis pars dorsalis principalis) but no seasonal changes in diffusivity could be detected in these structures (data not shown).

Table 3.   DTI and T2 metrics for the commissures and optic chiasm from breeding (April) and nonbreeding (July) starlings
ROIFractional anisotropy (FA)Mean diffusivity MD (μm2/ms)Axial diffusivity AD (μm2/ms)Radial diffusivity RD (μm2/ms) T2 (ms)
  1. Values are mean ± SD of n = 9 subjects in each case. FA is a dimensionless value that varies between 0 (isotropic) and 1 (anisotropic). CoA, anterior commissure; CO, optic chiasma; CoP, posterior commissure; ROI, region of interest. *Values that changed seasonally, P < 0.05 (Wilcoxon signed-ranks test; April vs. July).

CoA
 April0.68 ± 0.050.60 ± 0.021.15 ± 0.080.31 ± 0.0428 ± 3
 July0.70 ± 0.070.60 ± 0.041.17 ± 0.050.30 ± 0.0628 ± 2
CoP
 April0.61 ± 0.060.60 ± 0.011.08 ± 0.060.35 ± 0.0333 ± 1
 July0.61 ± 0.030.63 ± 0.03*1.15 ± 0.05*0.37 ± 0.0335 ± 2
CO
 April0.63 ± 0.020.58 ± 0.031.07 ± 0.060.33 ± 0.0227 ± 3
 July0.51 ± 0.08*0.63 ± 0.061.01 ± 0.050.44 ± 0.08*29 ± 1
Figure 4.

 Seasonal changes in the optic chiasma. The left panel presents a midsagittal colour-coded FA map (right is caudal) of one male starling in the breeding season. The middle panel illustrates at higher magnification the region of the CO of the same subject at the two different seasons. The colours define the main diffusion direction in each voxel (red: rostral – caudal; green: dorsal – ventral; blue: medial – lateral; see schematic representation by arrows at the top left part of the figure). Schematic representations of the diffusion tensors have also been overlaid on the CO. Note the broader tensors in July although the main diffusion direction (the colour coding) of the tensors does not differ between seasons. Abbreviations: CO, optic chiasma; CoA, anterior commisure; CP, posterior commisure; Cb, cerebellum. Scale bar, 10 mm.

Interhemispheric fiber tracts

The mean diffusivity, MD, in the CoP was lower in April than in July (due to a significant reduction in the AD; see Table 3), suggesting that this fiber tract during the breeding season is tighter (e.g. less extracellular space between axons; Table 3). However, the shape of the tensors in CoP remained the same (i.e. no change in FA between seasons; Table 3), suggesting a similar coherence of axons in this commissure between seasons. These differences were not present in the CoA, in which no seasonal change could be detected (see Table 3 for detail).

Discussion

We demonstrate here by DTI that the connectivity in the songbird brain exhibits an extreme seasonal plasticity that is not only limited to the SCS. Based on the comparative analysis of the different DTI parameters we were also able to deduce some of the cellular aspects that contribute to this plasticity (See Table 4). In the few cases in which anatomical data obtained by histology are available, these data confirm the conclusions drawn from the DTI results, which strongly supports the validity of the present in vivo approach.

Table 4.   Summary table of changes in DTI or T2 parameters from the breeding (April) to nonbreeding season (July) with the currently accepted interpretation of the results
ROIFractional anisotropy (FA)Mean diffusivity, MDAxial diffusivity, ADRadial diffusivity, RDT2Interpretations
Axonal projections decreased* DemyelinationOther from our dataOther from the literature
  1. The ↑ and ↓ indicate increases and decreases in the non-breeding season, respectively. *When FA and AD both decreased, the interpretation was that axonal projections decreased. When RD increased, the interpretation was that there was demyelination. Area X, used as proper name; CO, optic chiasma; CoP, posterior commissure; DM, dorsomedial nucleus of the intercollicular complex; HVC, formerly called the high vocal center; LaM, lamina mesopallialis; LMAN, lateral magnocellular nucleus of the anterior nidopallium; OM, tractus occipitomesencephalicus; RA, robust nucleus of the arcopallium; ROI, region of interest. Stocker et al. (1994); §review by Tramontin & Brenowitz (2000); Thompson & Brenowitz (2005); **Hessler & Doupe (1999); ††Olveczky et al. (2005).

HVC-RA tract  YesYes  
RA fiber capsule   Yes   
RA       Demyelination and decrease in neuronal density§, soma size§ and synaptic traits§
OM    Yes  
DM   Yes   
Area X      Extracellular and/or intracellular fluid decreasedNeuronal density and soma size decreased
Fibre capsule of X   Yes   
LMAN  Yes Extracellular water, cell spacing increased or cell size decreasedOr axonal activity** decreased
LaM   Yes  Or axonal activity†† decreased
CoP     Extracellular space between axons increased 
CO    Yes  

Our results corroborate established findings about plasticity of the HVC-RA pathway in songbirds. The adult HVC continues to incorporate new RA-projecting neurons that replace older dying cells (Paton et al., 1985; Kirn & Nottebohm, 1993). Elevated sex steroids appear to decrease the turnover and increase the survival of HVC neurons, thus increasing their numbers during the breeding season (Rasika et al., 1994; Hidalgo et al., 1995; Tramontin & Brenowitz, 1999). There are consequently fewer RA-projecting HVC neurons during the nonbreeding season and this is reflected in the λ1 decrease that was indeed detected by DTI (decrease in AD). The decrease in axonal connections was also confirmed at a qualitative level by our histological analysis of Luxol-stained sections.

In addition, the increase in RD observed during the nonbreeding season in this tract has been previously correlated with myelin loss in studies on animal models of experimental demyelination and remyelination (Song et al., 2005). In canaries specifically, testosterone has been shown to enhance the myelination of both HVC and RA (Stocker et al., 1994). This concordance between DTI and histological results can therefore be considered a proof of principle of DTI technique in songbirds.

In the lower part of the vocal motor circuit (e.g. the OM tract), DTI parameter changes were limited to an increase in RD. The fact that the first eigenvalue (AD) remained unchanged indicates that variation in the DTI signal is purely reflecting a decrease in myelination between April and July that may facilitate singing. Male European starlings indeed sing at increased rates during the breeding season and this song is primarily involved in mate attraction (Eens, 1997). The medial telencephalic part of the OM tract harbors projection fibers from RA to DM and the vocal motor nucleus (XIIts; Wild, 2004, 1993). RA neurons show a seasonal plasticity in spontaneous firing rate (Meitzen et al., 2007) and this plasticity is affected by steroid hormones (Park et al., 2005). In this context, it is also important to note that a high concentration of 5α-reductase, a testosterone-converting enzyme, is present in myelin membranes in mammals (Poletti & Martini, 1999), and gonadal sex steroids modulate brain myelination in mammals (Celotti et al., 1987; Melcangi et al., 1988) and canaries (Stocker et al., 1994). Recently it has been shown that electrical activity in axons acts on surrounding astrocytes which in turn affect myelinating oligodendrocytes (Ishibashi et al., 2006). This activity-dependent mechanism could thus regulate the seasonal myelination that was detected here by DTI in OM as a function of the discharges originating from RA. These data are thus consistent with the previous work indicating that electrical activity promotes myelination of central nervous system axons into postnatal life.

It should also be noted that in many cases the effects of testosterone in the songbird brain are mediated by neural conversion to estradiol by the enzyme aromatase. High levels of aromatase are expressed in the brain of male songbirds, including starlings (Riters et al., 2001), especially in regions adjacent to song control nuclei (Saldanha et al., 2000; Schlinger, 2002). Aromatase expression is high near HVC, and HVC expresses estrogen receptors (Shen et al., 1995; Bernard et al., 1999; Fusani et al., 2000; Soma et al., 2004). Thus, the conversion of testosterone to estrogens might be a critical event for the seasonal growth of this song control nucleus. In this way aromatase and estrogens might also play a role in the control of seasonal changes in axonal projections from HVC neurons to RA.

An increased axonal connectivity (increased FA) was also present in DM during the breeding season and the CoP appeared tighter in the breeding season (decreased MD). Together with the fact that the CoP includes the connections between DM and the nucleus uvaeformis (Striedter & Vu, 1998) and that this connection is one of those involved in synchronization of the premotor activity in both hemispheres (Schmidt et al., 2004), these observations suggest that the observed improved interhemispheric connectivity might serve the mentioned synchronization.

Unlike RA-projecting HVC neurons, the X-projecting HVC neurons and the RA-projecting LMAN neurons remain stable during life (Alvarez-Buylla et al., 1988). The change in AD observed in the corresponding structures (LaM, the lamina containing both connections, the capsule around Area X, and LMAN) could thus be attributed to two nonexclusive causes: (i) a seasonal change in the number of axons, due to axonal retraction followed by re-growth of the axons originating in HVC-neurons; or (ii) an increased activity of these axons during the breeding season. Axonal retraction of HVC neurons has been suggested in a recent study (Thompson et al., 2007). A retraction of the Area X-projecting HVC axons could alter the coherence of the fiber tract in which these axons as well as the LMAN-to-RA-projecting axons (which may or may not retract) reside. As shown in Fig. 2, coherence of axons can alter the AD.

Another potential mechanism that could explain the change in AD in these parts of the AFP is a change in axonal activity. Such a change should be accompanied by an increase in microtubular density in axons, a change in axonal diameter and the accompanying increase in axoplasmic flow as described in the developing white matter of the human brain (Wimberger et al., 1995; Partridge et al., 2004). A dynamic manganese-enhanced MRI experiment (Van Meir et al., 2004) showed that a treatment with testosterone of female starlings leads to a specific increase in the amount of Mn2+ that is transported from HVC to Area X per unit of time. Mn2+ is transported via fast axonal transport, which is activity-dependent and associated with vesicles transported along microtubules.

We also demonstrate here a correlation between aspects of the plasticity observed in the motor pathway (FA change in RA) and the AFP (FA change in Area X). A similar correlation was observed in a previous MRI study using manganese as a dynamic contrast agent. Mn2+ was injected into HVC of a group of female starlings and its transport towards RA and Area X was followed in vivo seasonally in the same starlings. This Mn2+ transport decreased between spring and summer. A high Mn2+ accumulation in Area X was correlated with a high accumulation in RA, which suggested unexpected functional relationships between the two types of HVC projection neurons (Van Meir et al., 2006).

Recent studies have shown that the AFP is not only needed for song learning but also has a function in adult songbirds, presumably controlling vocal variability (Olveczky et al., 2005). In adult zebra finches, it has been demonstrated that LMAN and Area X neurons fire vigorously during singing (Jarvis & Nottebohm, 1997; Hessler & Doupe, 1999), despite the fact that they are not required for normal adult song production (Hessler & Doupe, 1999). Area X-projecting HVC neurons are also known to fire actively during singing in these species (Rosen & Mooney, 2006; Kozhevnikov & Fee, 2007). These electrophysiological studies are consistent with the present demonstration of correlated changes in the motor and anterior forebrain pathways and together strongly support the notion that the singing-related AFP activity represents in part a copy of the premotor signals also sent to the motor pathway especially during the breeding season.

A more unexpected result of the present DTI study was the detection of seasonal changes in myelination in the CO as attested by the increased radial diffusivity in the summer. Male starlings become sensitive to females and start producing female-directed song during the breeding season. Seasonal plasticity in the visual system might thus mediate changes in visual acuity that could potentially play a role in the control of reproductive behavior. Such functional changes in the visual system of songbirds have not been documented to the best of our knowledge but should definitely be investigated based on the morphological changes detected here.

The underlying mechanisms that might cause these changes also need to be considered. In mammals the enzyme 5α-reductase, which catalyzes the conversion of testosterone into 5α-reduced metabolites, is expressed in high concentration in the optic chiasma (Celotti et al., 1987; Melcangi et al., 1988) and it is especially concentrated in myelin membranes (Poletti & Martini, 1999). A seasonal peak in 5α-reductase has been observed in the telencephalon of starlings during the breeding season. Interestingly, 5α-reduced metabolites of testosterone also seem to play a role in myelination (Magnaghi et al., 1999, 2004). Thus, if the changes in 5α-reductase activity detected in the telencephalon are also present in the optic chiasma, they presumably represent the proximal mechanism underlying the structural changes that were detected here by DTI.

In summary, we demonstrate here broad seasonal changes in the organization of white matter structures intimately associated with the SCS. These results extend in a major way our view of the seasonal plasticity affecting the avian brain (Nottebohm, 1981; Tramontin & Brenowitz, 2000; Ball et al., 2002; Brenowitz, 2004). This plasticity is particularly prominent in the SCS but also seems to affect the visual system, possibly fine-tuning its performance during the breeding season. Further investigations using DTI in conjunction with anatomical and electrophysiological approaches are now in order to further explore the causal relationship between seasonal changes in steroids, brain plasticity and (singing) behavior.

Acknowledgements

The authors wish to thank Professor Dr Marcel Eens for providing the birds. This research was supported by Concerted Research Actions (GOA funding) from the University of Antwerp and grants from the Research Foundation–Flanders (FWO, project no. G.0420.02) to A.V.dL. and by BOF-KP funding from the University of Antwerp to V.V.M. G.D.G. was supported by a PhD fellowship of the Research Foundation–Flanders (FWO) and V.V.M. is Postdoctoral Fellow of the Research Foundation–Flanders (FWO). This work was also supported by grants from the National Institutes of Health (R01 NS35467) and the Belgian Fonds de la Recherche Fondamentale Collective (FRFC; 2.4562.05) to J.B. This research was carried out in the frame of the European-funded NoE EMIL (LSHC-CT-2004–503569) and DiMI (LSHB-CT-2005–512146).

Abbreviations
AD

axial diffusivity

AFP

anterior forebrain pathway

Area X

used as proper name

CO

optic chiasma

CoA

anterior commissure

CoP

posterior commissure

DLM

dorsolateral nucleus of the medial thalamus

DM

dorsomedial nucleus of the intercollicular complex

DTI

diffusion tensor imaging

DW

diffusion-weighted

FA

fractional anisotropy

HVC

formerly called the high vocal center

LaM

lamina mesopallialis

LMAN

lateral magnocellular nucleus of the anterior nidopallium

MD

mean diffusivity

MRI

magnetic resonance imaging

OM

tractus occipitomesencephalicus

RA

robust nucleus of the arcopallium

RD

radial diffusivity

ROI

region of interest

SCS

song control system

Ancillary