In vivo measurements of NAD(P)H state over the central brain and across MB regions.

Scale bars: 50 µm. Axes indicate anterior (A), dorsal (D) and right (R) anatomical orientations. The circled dot and cross indicate the directions pointing toward and away from the viewer, respectively. (a) Illustration of the main steps of the mapping of NAD(P)H decay parameters over the central brain. FLIM images were spatially binned using a disk-shaped kernel (see Methods) and the resulting curves were fitted to extract decay parameters. The parameter histograms are derived from the whole 3D image. Each step is illustrated by a single horizontal section of the example 3D image. (b) Heatmaps of ffree, τbound and τmean over 3 horizontal slices of a single image. Contours of the MB, extracted from the MB-DsRed image, are superimposed in white. Voxels with photon counts under 500 are shown in gray. The cortex and neuropil are marked on the first slice of the τbound heatmap with red letters “c” and “n,” respectively. (c) Average of all transverse sections of the MB-DsRed image corresponding to panel b (gray), with colored segmentation regions superimposed on the left MB. White dashed lines denote the depth of the horizontal sections shown in panel b. (d) Horizontal view of a left MB showing the maximum intensity projection of the KC-specific marker with superimposed segmentation regions. The circled dot indicates the direction pointing towards the viewer. (e) Within-subject averages of ffree and τbound in the segmented MB regions, for 5 flies. Gray horizontal lines indicate the cross-subject mean values. No significant variations of ffree were observed between the considered regions (repeated measures ANOVA, F3,12 = 1.7, p = 0.21). Significant variations of τbound were observed between areas (repeated measures ANOVA, F3,12 = 14.8, p = 2.4·10-4). Three post-hoc comparisons were performed to compare the somata and calyx region with the other areas. It showed that τbound was significantly higher in the somata and calyx (n = 5, µ = 3.67 ± 0.23 ns) compared to the peduncle (n = 5, µ = 3.21 ± 0.07 ns; paired samples Student’s t-test, t4 = 4.7, Bonferroni-adjusted p = 2.7·10-2), the vertical lobe (n = 5, µ = 3.32 ± 0.10 ns; paired samples Student’s t-test, t4 = 4.0, Bonferroni-adjusted p = 4.8·10-2) and the medial lobes (n = 5, µ = 3.18 ± 0.09 ns; paired samples Student’s t-test, t4 = 4.3, Bonferroni-adjusted p = 3.8·10-2).

NAD(P)H states in the somata and calyx regions of the MB, for naive and conditioned flies.

Scale bars: 25 µm. (a) Illustration of the main steps of the computation of the metabolic profile in the KC somata region. The fluorescence intensity of the KC-specific cytosolic marker MB-DsRed was simultaneously recorded with the NAD(P)H FLIM image. The marker image was segmented using thresholding to isolate the MB. Next, we used a MB template with previously defined masks delineating the KC somata region (purple) and the calyx (turquoise). This template was registered to the segmented marker image. The registered masks could be used to isolate the target area (the somata region in this illustration) in the FLIM image. The masked FLIM image was spatially binned using a disk-shaped kernel (see Methods) and the resulting decays were fitted to obtain ffree and τbound values in the target area. Each step is illustrated by a single horizontal section of the example 3D image. (b) Left: Horizontal sections of the average spatial distribution of ffree in the somata area and calyx in naive flies. The average distribution was obtained by computing maps of ffree in the posterior MB region, registering them to the MB template based on the corresponding MB-Dsred images and averaging the results. Axes indicate anterior (A), dorsal (D) and right (R) anatomical orientations. Top right: Average of ffree across all transverse sections of the average spatial distribution, with white dashed lines indicating the depth of the horizontal sections. The circled cross indicates the direction pointing away from the viewer. Bottom right: Within-subject averages of ffree in the somata and calyx regions. ffree values in the somata region (n = 17, µ = 0.682 ± 0.021) were significantly lower than in the calyx (n = 17, µ = 0.746 ± 0.023; paired samples Student’s t-test, t16 = 7.9, p = 6.1·10-7). (c) Top: Schematic representation of the aversive conditioning protocol. Control flies underwent unpaired training, receiving electric shocks, followed by sequential exposure to odors A and B after a 2-minute break. Conditioned flies underwent paired conditioning, receiving electric shocks while exposed to odor A, followed by exposure to odor B. Bottom: Within-subject averages of ffree in the somata and calyx regions for conditioned flies. In the somata region, ffree values for the flies subjected to unpaired stimuli (n = 15, µ = 0.662 ± 0.016) were not significantly different than for the ones subjected to paired conditioning (n = 16, µ = 0.658 ± 0.013; Student’s t-test, t29 = 0.8, p = 0.43). In the calyx, ffree values for the flies subjected to unpaired stimuli (n = 15, µ = 0.709 ± 0.020) were not significantly different than for the ones subjected to paired conditioning (n = 16, µ = 0.706 ± 0.017; Student’s t-test, t29 = 0.4, p = 0.67).

NAD(P)H states around somata of KC subtypes in naive and conditioned flies.

(a) Illustration of the main steps of the computation of the metabolic profile around the somata of a given KC subtype. Standard fluorescence intensity image of the subtype-specific nuclear marker was simultaneously recorded with the NAD(P)H FLIM. The marker image was thresholded to remove background noise and the resulting segmentation was used to mask the FLIM image. The masked FLIM image was spatially binned using a disk-shaped kernel (see Methods) and fitted to obtain ffree values in the target area. Each step is illustrated by a single horizontal section of the example 3D image. Scale bar: 25 µm. (b) Within-subject averages of ffree near the somata of α/β, α’/β’ and γ KCs show significant variations (one-way ANOVA, F2,40 = 7.1, p = 2.4·10-3). ffree is significantly higher near α/β neurons (n = 13, µ = 0.730 ± 0.019) than near γ neurons (n = 15, µ = 0.701 ± 0.022; Student’s t-test, t26 = 3.6, Bonferroni-adjusted p = 3.7·10-3). The values of ffree for α’/β’ neurons (n = 15, µ = 0.715 ± 0.018) were not significantly different from those near α/β (Student’s t-test, t26 = 2.0, Bonferroni-adjusted p = 0.16) or γ neurons (Student’s t-test, t28 = 1.9, Bonferroni-adjusted p = 0.21). (c) Within-subject averages of ffree near the somata of α/β and γ KCs after odor conditioning. Near α/β somata, ffree is significantly lower after paired conditioning (n = 19, µ = 0.718 ± 0.020) compared to the control condition (n = 18, µ = 0.734 ± 0.025; Student’s t-test, t35 = 2.0, p = 4.98·10-2). No statistically significant difference was observed near γ somata between conditioned (n = 12, µ = 0.700 ± 0.017) and control flies (n = 11, µ = 0.693 ± 0.011; Student’s t-test, t21 = 1.0, p = 0.31).

Examples of fit and distribution of 2I* for Dataset 1.

For each subject’s image, all voxels within the segmented MB regions were included. On the left, the distributions of the 2I* quality parameter are displayed, with the dashed line marking the empirically defined threshold for poor fits (2I* = 200). On the right, three post-binning decay curves randomly selected from each image are shown, along with their corresponding bi-exponential model fits. Below each decay plot, the residuals (differences between the observed and fitted decays) are displayed.

Distributions of ffree and τbound in different segmented MB regions for the subjects of Dataset 1.

The dashed line indicates the mean value of each distribution. (a) ffree distributions. (b) τbound distributions.

Establishment of a reproducible MB template.

(a) Illustration of the main steps of the preprocessing pipeline. This process was applied to the 165 MB images obtained from flies expressing fluorophore DsRed in the cytoplasm of all KCs (Dataset 3). (b) Illustration of the template building process. An initial template is obtained by averaging all preprocessed images. The images are then registered to this intermediate template. The resulting registered images and transformations are averaged. A transformation proportional to the inverse of the mean transformation is applied to the average registered image to get an updated template image. The cycle was repeated several times to obtain the final template. (c) Examples of preprocessed MB images and final template. The left and right columns show dorsal and lateral maximum intensity projections of each image. (d) Views of the MB template image generated through volume rendering with ParaView software (Ahrens et al., 2005). Axes indicate anterior (A), dorsal (D), ventral (V) and right (R) anatomical orientations. View-aligned axes orientations depicted with dot (towards viewer) and cross (away from viewer). Scale bar: 25 µm. (e) Schematic illustrating the process of evaluation of the amount of displacement required to match MB images to different target images, namely the template, the raw average or a randomly selected image. (f) Mean displacements resulting from the registration of all images of the dataset to different target images. Registration to a single image was performed 5 times on different randomly chosen images.

Establishment of masks for the somata and calyx regions.

Illustration of the processes used to obtain masks isolating the somata and calyx regions. Each step is illustrated by a slice of 3D image at a fixed depth. Both masks are defined over the previously generated template image of the cortical region of the MB. The somata mask was established using the images of Dataset 4. After registering the images to the template, based on the cytosolic marker channel, the nuclear marker channel was averaged and thresholded. The calyx mask was obtained by thresholding the template image and excluding the voxels belonging to the somata region.

Evaluation of intra-subject and intra-hemisphere similarities in the spatial distribution of MB neuronal subtypes.

(a) Schematic illustrating the method used to assess differences between templates generated from single or mixed hemispheres. This process used images preprocessed as for template building (Sup. Figure 3a), downsampled by a factor 2 (using linear interpolation and anti-aliasing filtering) to reduce the computational cost. Step 1: The complete set of preprocessed images is randomly divided into two non-overlapping half sets, arranged in random order. In the mixed hemispheres case, the half sets are created by including both left and right images indiscriminately. In the inter-hemispheres case, the first and second half exclusively contain left and right MB images, respectively. Step 2: A template is constructed using the first N images from each half set, where N represents the subset size. Step 3: Step 3: The two resulting templates are compared by registering one to the other using rigid registration and computing the normalized image distance between the aligned templates. This procedure was repeated with 8 different random splits performed for each case. (b) Comparison of the normalized image distances between templates constructed from subsets of varying sizes, consisting of either mixed or separate hemispheres. In the 30-image case, the mixed-hemisphere distances (n = 8, µ = 11.7 ± 2.0 %) were significantly lower than inter-hemisphere distrances (n = 8, µ = 14.4 ± 1.6 %; Student’s t-test, t14 = 2.8, p = 1.4·10-2). (c) Overlapped maximum intensity projections of two templates obtained using all available left MB images (in red) and right MB images (in green), respectively. The right MB template was aligned to the left one through rigid registration. Axes indicate anterior (A), dorsal (D) and right (R) anatomical orientations. View-aligned axes orientations depicted with dot (towards viewer) and cross (away from viewer). 25 µm scale bar. (d) Mean projections of the determinant of the Jacobian determined from the displacement field obtained through the registration of the right template onto the left template. Values superior to 1 indicate that the left template should be expanded to match the right one. Values inferior to 1 indicate that it should be compressed. Voxels located outside the segmented left MB image were ignored in the average. Pixels representing the mean value of ignored voxels only were set to 1.

Influence of memory conditioning on average τbound values in the somata and calyx region.

(a) Left: Horizontal sections of the average spatial distribution of τbound in the somata area and calyx in naive flies. The average distribution was obtained by computing maps of τbound in the posterior MB region, registering them to the MB template based on the corresponding MB-Dsred images and averaging the results. Axes indicate anterior (A), dorsal (D) and right (R) anatomical orientations. Right: Average of ffree across all transverse sections of the average spatial distribution, with white dashed lines indicating the depth of the horizontal sections. The circled cross indicates the direction pointing away from the viewer. (b) Within-subject averages of τbound in the somata region and in the area encompassing the somata and calyx. τbound values were significantly higher in the KC somata region (n = 17, µ = 3.94 ± 0.19) than in the calyx (n = 17, µ = 3.31 ± 0.09 ns; dependent-samples Student’s t-test, t16 = 11.9, p = 2.2·10-9). (c) Within-subject averages of τbound in the somata and calyx regions, for flies subjected to unpaired and paired stimuli. In the somata region τbound values for the flies subjected to unpaired stimuli (n = 15, µ = 4.01 ± 0.16 ns) were not significantly different than for the ones subjected to paired conditioning (n = 16, µ = 3.95 ± 0.26 ns; Student’s t-test, t29 = 0.8, p = 0.41). In the calyx region, τbound values for the flies subjected to unpaired stimuli (n = 15, µ = 3.24 ± 0.20 ns) were not significantly different than for the ones subjected to paired conditioning (n = 16, µ = 3.24 ± 0.11 ns; Student’s t-test, t29 = 0.0, p = 0.99).

Lateralization of ffree spatial distribution in the somata area.

Left: Illustration of the different types of image-to-image comparisons on the MBs of two subjects. Right: Comparison of single-hemisphere maps of ffree in the somata region. The ffree map distance was computed between all pairs of registered maps. We compared inter-subject/inter-hemisphere differences (n = 227, µ = 5.6 ± 0.9) with with inter-subject/intra-hemisphere differences (n = 224, µ = 5.6 ± 0.9) and intra-subject differences (n = 14, µ = 5.2 ± 1.0). We found that local differences were not significantly different when comparing the same or different hemispheres in different subjects (Student’s t-test, t(449) = 0.2, Bonferroni-adjusted p = 1.0), suggesting the absence of hemisphere-specific spatial distribution of ffree values. We found that local differences were not significantly different either when comparing different hemispheres in the same or different subjects (Student’s t-test, t(236) = 1.54, Bonferroni-adjusted p = 0.25), suggesting that subject-specific spatial distribution of ffree are not shared between both hemispheres.

Relationship between ffree and τbound in different datasets.

Each dot represents the averages values of ffree and τbound for a single subject. ρ indicates the Pearson correlation coefficient. The dotted lines indicate least squares affine fits. (a) Relationship between average ffree and τbound values of segmented MB regions, for all subjects in Dataset 1. (b) Relationship between average ffree and τbound values of the somata region, for all subjects in Dataset 2. (c) Relationship between average ffree and τbound values in the different neuronal subtypes, for all subjects in Dataset 6.

Estimations of τbound and ffree on simulated decays.

Each dot represents the values of ffree and τbound estimated from a single simulated decay. ρ indicates the Pearson correlation coefficient. The dotted line indicates the least squares affine fit. (a) Estimations from 5,000 decays simulated by uniformly sampling ffree between 0.65 and 0.75 and τbound between 3 and 4.5 ns, independently. (b) Estimations from 5,000 decays simulated by sampling correlated ffree and τbound couples on the same ranges, with a defined correlation of -0.86.

Observation of average spatial distribution of KC subtypes.

(a) Main steps of the computation of the average spatial distributions of the 3 main KC subtypes using Dataset 3. Each row shows the processing of a set of images of flies expressing a KC-specific marker (displayed in gray) and a subtype-specific nuclear marker (displayed in green). (b) Schematic illustrating the method used to assess differences between average spatial distributions of the different subtypes. (c) Comparison of the normalized image distance between average spatial distributions of identical or different subtypes, computed from different numbers of images. The graph reports mean and standard deviations. The inter-subtype distances were compared with the intra-subtype distances for the averages obtained from 20 images. The inter-subtype distances (n = 60, µ = 19.3 ± 1.6 %) were significantly higher than the intra-α/β (n = 20, µ = 8.6 ± 1.2 %; Student’s t-test, t78 = 26.8, Bonferroni-adjusted p = 1.3·10-40), the intra-α’/β’ (n = 20, µ = 14.9 ± 0.7 %; Student’s t-test, t78 = 11.5, Bonferroni-adjusted p = 5.9·10-18) and the intra-γ distances (n = 20, µ = 14.6 ± 0.7 %; Student’s t-test, t78 = 12.4, Bonferroni-adjusted p = 1.0·10-19). (d) Horizontal slices (left) and posterior projection (upper right) of the subtype proportion map. The proportions of red, green, and blue indicate the proportion of α/β, α’/β’ and γ neurons, respectively. The correspondence between color and proportions is indicated by the ternary diagram (lower right). Axes indicate anterior (A), dorsal (D) and right (R) anatomical orientations. The circled cross indicates the direction point away from the viewer. 25 µm scale bar. (e) Distribution of the average proportions of each subtype over all voxels. The dashed lines indicate the overall proportion of the subtypes, given by the number of neurons of the considered subtype against the total number of KCs, using the average neuron counts reported by Aso et al. (Aso et al., 2009).

Measure of intra-subject and intra-hemisphere similarities of the spatial distribution of the MB neuronal subtypes.

Left: Illustration of the different types of image-to-image comparisons on the MBs of two subjects. Right: Image-to-image distances based on 165 MB images of Dataset 3. The intra-subject distances were measured between all pairs of left and right MBs belonging to the same subject (n = 48, n = 36 and n = 42 values for α/β, α’/β’ and γ, respectively). The inter-subject/intra-hemisphere distances were measured between all pairs of images of the same hemisphere belonging to different individuals (n = 813, n = 609 and n = 790 values). The inter-subject/inter-hemisphere distances were measured between all pairs of images of different hemispheres belonging to different individuals (n = 816, n = 598, n = 785 values). Statistical tests were performed in order to assess possible intra-subject or intra-hemisphere similarities. It revealed that α/β spatial distributions between left and right hemisphere images are more similar if they belong to the same subject than if they belong to two different subjects (Student’s t-test, Bonferroni-adjusted p = 1.4·10-3, t838 = -3.4). No other statistically significant effect was observed.

Measures of τbound near the somata of KC subtypes in naive and conditioned flies.

(a) Within-subject averages of τbound near the somata of α/β, α’/β’ and γ KCs. No significant variation is observed (one-way ANOVA, F2,40 = 1.5, p = 0.23) between α/β (n = 13, µ = 3.85 ± 0.22 ns), α’/β’ (n = 13, µ = 3.83 ± 0.22 ns) and γ KCs (n = 13, µ = 3.97 ± 0.26 ns). (b) Within-subject averages of τbound near the somata of α/β and γ KCs after odor conditioning. Near α/β somata, τbound is higher after paired conditioning (n = 19, µ = 3.82 ± 0.18 ns) than in the control condition (n = 18, µ = 3.70 ± 0.22 ns) but this difference is not statistically significant (Student’s t-test, t35 = 1.9, p = 6.1·10-2). No statistically significant difference was observed either near γ somata between conditioned (n = 12, 3.92 ± 0.28 ns) and control flies (n = 11, 3.92 ± 0.25 ns; Student’s t-test, t21 = 0.0, p = 0.97).

Influence of Ldh knockdown on ffree and τbound around the somata of γ neurons.

These plots compare flies with an Ldh knockdown mutation (Ldh-KD) with control flies. Left: Within-subject averages of ffree near the somata of γ KCs. No significant difference was observed between controls (n = 12, µ = 0.697 ± 0.011) and mutated flies (n = 12, µ = 0.684 ± 0.039; t22 = 1.1, p = 0.27). Right: Within-subject averages of τbound near the somata of γ KCs. No significant difference was observed between controls (n = 12, µ = 4.00 ± 0.16 ns) and mutated flies (n = 12, µ = 3.95 ± 0.31 ns; t22 = 0.52, p = 0.61).

Average values of τfree across the different datasets.

τfree was estimated from FLIM images of the different datasets, using the same preprocessing and masks as in the main results. These estimations were obtained by applying the same fitting procedure as in the other analyses, without fixing τfree. Each value represents the within-subject average. (a) Average τfree values in the somata and calyx region (µ = 0.437 ± 0.012 ns), the peduncle (µ = 0.479 ± 0.020 ns), the vertical lobe (µ = 0.484 ± 0.014 ns) and the medial lobes (µ = 0.470 ± 0.014), for all subjects in Dataset 1 (n = 5). (b) Average τfree values in somata region (µ = 0.415 ± 0.017 ns) and in the calyx (µ = 0.435 ± 0.018 ns), for all subjects in Dataset 2 (n = 17). (c) Average τbound values near the somata of α/β (n = 13, µ = 0.386 ± 0.012 ns), α’/β’ (n = 15, µ = 0.0402 ± 0.016 ns) and γ neurons (n = 15, µ = 0.398 ± 0.017 ns), for all subjects in Dataset 6.

Composition of dataset 2.

Composition of dataset 3.

Composition of dataset 4.

Composition of dataset 5.

Composition of dataset 6.

Composition of dataset 7.

Composition of dataset 8.

Composition of dataset 9.