Natural variability in bee brain size and symmetry revealed by micro-CT imaging and deep learning

Abstract

Analysing large numbers of brain samples can reveal minor, but statistically and biologically relevant variations in brain morphology that provide critical insights into animal behaviour, ecology and evolution. So far, however, such analyses have required extensive manual effort, which considerably limits the scope for comparative research. Here we used micro-CT imaging and deep learning to perform automated analyses of 3D image data from 187 honey bee and bumblebee brains. We revealed strong inter-individual variations in total brain size that are consistent across colonies and species, and may underpin behavioural variability central to complex social organisations. In addition, the bumblebee dataset showed a significant level of lateralization in optic and antennal lobes, providing a potential explanation for reported variations in visual and olfactory learning. Our fast, robust and user-friendly approach holds considerable promises for carrying out large-scale quantitative neuroanatomical comparisons across a wider range of animals. Ultimately, this will help address fundamental unresolved questions related to the evolution of animal brains and cognition.

Fig 1. Flowchart of the steps to perform large-scale quantitative comparative analyses of bee brain size and organisation using micro-CT imaging and the Biomedisa segmentation platform.

After sample preparation and volume reconstruction (grey boxes), the micro-CT scans are segmented with AVIZO 2019.1 (red boxes) in combination with Biomedisa (blue boxes). The volumes are then measured with AVIZO 2019.1 and statistically analysed with R Studio (yellow box). For the honey bee dataset, the loop within the white box represents the training process for 3, 7, 12, 18, and 26 brains. Each step involved a different number of segmented and manually corrected 3D images, specifically 4, 5, 6, and 8 respectively. Upon completing the loop, the remaining 84 brains were segmented and manually corrected. All processing times provided are average times per brain.

Fig 2. Surface renderings of an example of CT-scanned honey bee head and reconstructed brain neuropils.

(A) Frontal view of the head of a forager bee (ID 79, hive H4). (B) Surface rendering of the head with the mandibles removed. (C) Overlay of the head and reconstructed neuropils. (D) Frontal cross-section of the tomogram with the segmentation boundaries of the mushroom bodies (MB), central complex (CX), antennal lobes (AL), medullae (ME), lobulae (LO) and other neuropils (OTH). (E) Frontal view of the reconstructed MB (dark blue), CX (sky blue), AL (light sky blue), ME (beige), LO (red) and OTH (orange). (F) Dorsal view of the reconstructed neuropils. (B), (C), (E) and (F) were created with ParaView Glance integrated in Biomedisa.

Fig 3. Segmentation accuracy of Biomedisa’s semi-automatic and automatic segmentation of honey bee CT scans.

(A) Average Dice scores (red) and average symmetric surface distances (ASSD, blue) of the automatic segmentation results for an increasing number of 3D training images. (B) Semi-automated segmentation accuracy (Dice score) of the AVIZO interpolation and the Biomedisa interpolation as well as segmentation accuracy of the automatic segmentation for uncropped 3D image data, using Biomedisa’s auto-cropping and manually cropped 3D image data. Boxplots show median volumes (intermediate line) and quartiles (upper and lower lines). Pink triangles display mean volumes. For performance tests of the automatic segmentation, the 84 three-dimensional honey bee test images were split into 30 three-dimensional validation images and 54 three-dimensional test images (see “Methods”).

Fig 4. Variation in brain and neuropils volumes (mm³) for honey bees (N = 110).

(A) Total brain. (B) Antennal lobes (AL). (C) Mushroom bodies (MB). (D) Optic lobes (OL). (E) Medullae (ME). (F) Lobulae (LO). (G) Central complex (CX). (H) Other neuropils (OTH). Boxplots show median volumes (intermediate line) and quartiles (upper and lower lines). Pink triangles display mean volumes. Coloured symbols show mean (± s.d. when available) of neuropil volumes described for forager honey bees in other studies: using confocal microscopy (square): Brandt et al. [30] (N = 20 bees - light blue); Steijven et al. [37] (N = 10 - kaki); Haddad et al. [36] (purple); using nuclear magnetic resonance microscopy (star): Haddad et al. [36] (N = 8 - deep blue); using stereo microscopy (diamond): Gowda & Gronenberg [28] (N = 7 - yellow); Mares et al. [38] (N = 25 - turquoise); Maleszka et al. [39] (N = 30 - burgundy); Withers et al. [24] (red); Durst et al. [40] (N = 12 - grey); using light microscopy (cross): Gronenberg & Couvillon [29] (N = 121 European and Africanized honey bees - dark green); using CT scan (point): Greco and Stait-Gardner [41] (N = 10 - coral). For the total brain volume, comparisons with other studies are not shown because some studies did not provide total brain volume information, while others measured different neuropils compared to our study.

Fig 5. Correlation between neuropil volumes and total brain volume (mm³) for honey bees (N = 110).

(A) Antennal lobes (AL). (B) Mushroom bodies (MB). (C) Optic lobes (OL) (=ME+LO). (D) Medullae (ME). (E) Lobulae (LO). (F) Central complex (CX). (G) Other neuropils (OTH). Regression lines displayed with 95% confidence intervals. Pearson correlation coefficient (r) and p-value are given. Strong correlations (r>0.40) and significant correlations (p<0.05) are displayed in bold. (H) Linear correlations for the different neuropils (y-axis not given: differs for each neuropil). The grey dashed line indicates true isometric correlation (slope = 1).

Fig 6. Variation in brain volume and absolute volume of neuropils (mm³) between honey bee colonies (population A - purple, N = 6 hives; population B - green, N = 3 hives).

(A) Total brain. (B) Antennal lobes (AL). (C) Mushroom bodies (MB). (D) Optic lobes (OL). (E) Medullae (ME). (F) Lobulae (LO). (G) Central complex (CX). (H) Other neuropils (OTH). Boxplots show median volumes (intermediate line) and quartiles (upper and lower lines). Statistical comparisons (p-values) for the neuropil volumes between hives were obtained from the F-test following LMMs and are displayed in bold when significant (see sample sizes in the Methods and statistical details in Table 1).

Fig 7. Variation in brain volume and neuropils volumes (mm³) and correlation between neuropil volumes and total brain volume for bumblebees (N = 77).

(A) Total brain. (B) Antennal lobes (AL). (C) Mushroom bodies (MB). (D) Optic lobes (OL). (E) Medullae (ME). (F) Lobulae (LO). (G) Central complex (CX). (H) Other neuropils (OTH). Boxplots show median volumes (intermediate line) and quartiles (upper and lower lines). Pink triangles display mean volumes. Coloured symbols show the mean of neuropil volumes described for bumblebees in another study using CT scan: Smith et al. [14] (N = 38 bees - blue).

Fig 8

Comparisons between left and right volumes (mm³) for paired neuropils in honey bees (A−E) and bumblebees (F−J). (A and F) Antennal lobes (AL) (N = 110 honey bees, N = 77 bumblebees). (B and G) Mushroom bodies (MB) (N = 59 honey bees, N = 36 bumblebees). (C and H) Optic lobes (OL) (N = 110 honey bees, N = 77 bumblebees). (D and I) Medullae (ME) (N = 110 honey bees, N = 77 bumblebees). (E and J) Lobulae (LO) (N = 110 honey bees, N = 77 bumblebees). Boxplots show median volumes (intermediate line) and quartiles (upper and lower lines). Statistical differences (p-values) for the neuropil volume between left and right side were obtained with Student’s paired samples t-Test, and are displayed in bold when significant.

Fig 9. Correlations between right-left volumes (mm³) between AL, OL and MB.

(A and D) Correlations between AL right-left and OL right-left volumes (N = 110 honey bees, N = 77 bumblebees). (B and E) Correlations between MB right-left and AL right-left volumes (N = 59 honey bees, N = 36 bumblebees). (C and F) Correlations between MB right-left and OL right-left volumes (N = 59 honey bees, N = 36 bumblebees). Pearson correlation coefficient (r) and p-value are given. Strong correlations (r>0.40) and significant correlations (p<0.05) are displayed in bold.