Introduction

Time-memories, the association of experiences or behavioral activities with the time-of-day1,2,3,4, allow animals to anticipate recurrent events and prepare appropriate behavioral responses in advance5. Accordingly, animals that naturally monitor different food sites per day likely evolved advanced capabilities to form time-memories. For example, honey bees, which are central place foragers, have been shown to learn to visit different feeder locations at different times of the day1,6,7. Further observations under natural conditions revealed that the foragers restrict their daily visits to the time-of-day the flowers provide highest nectar volumes8. Odor learning experiments finally indicated that honey bees are capable to distinguish events in time if they are separated by just 45 min9.

Learning and recalling behavioral actions with respect to the time-of-day require functional connections between brain areas involved in memory formation and the circadian clock10,11. Studies in flies and mice confirmed that circadian clock mutants fail to establish time-memories3,4. However, for Drosophila, which can establish a basic form of time-memory —such as retaining spatial and temporal memory independently, but not yet shown to associate specific times with specific locations4— so far only indirect connections between the clock and mushroom bodies (MBs), the major learning and memory center in insects, have been reported12.

For honey bees, we recently showed that foraging at a known feeder and approaching a known feeder without food reward induces the expression of the neuronal activity-regulated transcription factor (TF) Egr1 in the small- and large-type KCs of the MBs. In case the feeder is not rewarded Egr1 expression declines after 1 h coinciding with the cessation of flight activity. Comparing Egr1 expression in groups of foragers that were active over the whole day with groups of foragers that were trained to forage at a feeder for two hours demonstrated that the Egr1 expression in the latter group is synchronously upregulated only during the respective 2 h, whereas in the former group expression levels are high during the whole day. Moreover, time-trained foragers even showed Egr1 expression around the training time when they were prevented from leaving the colony by artificial rain and caught inside the hive. In-situ hybridization studies indicated that this Egr1 expression, which was not induced by foraging activity or environmental stimuli, was restricted to small-type KCs (sKCs)13. Based on these findings we hypothesize that (i) the sKCs, which are the most prominent multimodal neurons in the MBs, play an important role in time-memory associated anticipatory activity, (ii) anticipatory expression of Egr1 in the sKCs might be involved in activating transcriptional responses facilitating foraging-related neuronal processes including, for example, learning and memory processes, and (iii) the sKCs are functionally connected with the circadian clock.

To explore time-training associated gene expression dynamics and possible functional connections between the circadian clock and KCs in more detail we now combined feeder time-training experiments with a time-series RNA sequencing (RNA-seq) of sKC-enriched tissue samples and single molecule fluorescent in-situ hybridization (smFISH). Using these experimental approaches, we asked: (i) Do genes in the sKCs show a transcriptional response associated with the training time, (ii) do activated, i.e. Egr1 expressing, KCs also express pdfr (receptor for the major clock neuromodulator in insects12,14 indicating input from the central pacemaker (CP), the master circadian clock, and (iii) do KCs express clock genes similar to cells in the vertebrate hippocampus?

Results

Feeder time-training is associated with a training time specific transcriptional response in sKCs

We trained free-flying honey bee (Apis mellifera) foragers to visit a feeder from 09:00 to 11:00 for 10 consecutive days. Foragers visiting the feeder on at least 3 out of the last 4 training days were considered as time-trained15 and collected in liquid nitrogen on day 11 at 7 different time-points [06:00, 10:00, 14:00. 18:00, 22:00, 02:00, 06:00, (Fig. 1, A and B)]. During the collection the feeder was not rewarded to omit reward induced changes in gene expression (Fig. 1, A and B, also see Methods). For the collection of sKC-enriched tissue samples we dissected the cell body area of the sKCs inside the four MB calyces of each brain using laser capture microdissection. RNA extracts of 11–12 brains were pooled for one RNA-seq (Fig. 1C). In total we sequenced 27 sKC-enriched RNA samples (2 colonies x 2 forager groups per colony x 7 time-points) as one sample was lost to technical issues.

For all the samples, we consistently detected reads of 9229 genes (filtering criteria = minimum count value of 5 in one or more replicates per time-point, Table S3). Gene ontology (GO) analysis of all detected genes (DGs) revealed enrichment of TFs, ribosome and initiation factors, SNARE interactions in vesicular transport and ER-Golgi transport and protein kinases, among others (FDR < 0.01; Figure S1A).

The following differential gene expression analysis using a pairwise comparison test (p.adj < 0.05) identified 995 differentially expressed genes (DEGs) (Fig. 1D and Table S4). Importantly, grouping the 995 DEGs according to peak expression values (Fig. 1E) revealed that 862 (86.6% of DEGs) genes show maximum (n = 484) or minimum (n = 378) peak expression at 10:00, the collection time-point which is one hour into the feeder training time (Fig. 1, D and E). GO analysis of the DEGs suggests an enrichment of ecdysteroid kinase, motor protein domains, and G-protein coupled receptor gene sets (FDR < 0.01; Figure S1B). In comparison, MetaCycle algorithm identified only 152 (1.6%) genes showing a significant daily oscillation at BH.Q.<0.05 (Fig. 1F, Table S5), and 686 (7.4%) genes at BH.Q.<0.3 (Table S6). GO analysis with 686 genes indicated enrichments of a few gene sets (e.g. transferase, choline/ethanolamine kinase) and none with 152 genes at FDR < 0.05 (Figure S1C). 124 and 416 genes overlapped between DEGs and circadian genes at BH.Q.<0.05 and BH.Q.<0.3, respectively (Fig. 1G). As a result, the temporal expression dynamics in the sKCs of time-trained honey bee foragers primarily align to the training time and not the daily circadian rhythm (Fig. 1, D to G). The high ratio of DEGs showing peak expression during the training time suggests that the transcriptional response is linked to the learned foraging schedule, and is regulated by Egr1 and/or other TFs13,16,17.

Fig. 1
figure 1

sKC-enriched transcriptome of time-trained foragers. (A and B) Schematic illustration of the feeder time-training paradigm (A) and collection time-points (B). Collection at 10:00 from the unrewarded feeder ensured sampling of individuals expressing learned time-memory1,6. (C) Overview of experimental pipeline for RNA-seq with sKC-enriched tissue samples. For details regarding (A) to (C) see Methods. (D) Heatmap showing temporal expression pattern of all DEGs. The color scale shows gene-specific Z-scores from DESeq2-normalized expression, with red indicating higher and blue lower expression. (E) Bar chart summarizing time of max (red) and min (blue) peak expressions for each of the 995 DEGs. The genes for day 11, 06:00 and day 12, 06:00 are combined as no genes were found to be differentially expressed between these two time-points (pairwise comparison test, p.adj < 0.05). This chart highlights that most genes [862 out of 995 (86.6%) DEGs] show peak expression at 10:00; one hour into the training time (feeder not rewarded). (F) Heatmap for expression dynamics of 152 genes showing circadian oscillation (daily periodicity) in MetaCycle algorithm, BH.Q. < 0.05. For both heatmaps (i.e., D and F) consecutive two rows for each collection time-point represent replicates from same colony, and next two columns from a second colony, and each column represents one gene of the corresponding dataset. Behavioral experiments with the two colonies were performed separately. (G) Pie charts showing the ratio of DEGs (p.adj < 0.05) and oscillating genes among detected genes with temporal expression differences, at BH.Q. < 0.05 (left panel) and at BH.Q. < 0.3 (right panel) for oscillating genes. See also Figure S1.

Full size image

The transcriptional response of the sKCs includes Egr1 downstream genes and canonical memory-related genes

We collated two gene libraries to explore in more detail whether Egr1 is involved in the observed transcriptional response and whether neural signaling genes involved in learning and memory processes are regulated during the training time. Our list of 149 candidate Egr1 downstream (Egr1 DS) genes was generated from data by Khamis et al. 201518 (see methods). A list of 272 neural signaling (NS) genes including neurotransmitters, receptors, second messengers and transcription factors was compiled on a selection of published data19,20,21,22,23,24,25(also see table S7).

Among the total DGs, we found 134 (90% of the) Egr1 DS genes, and 229 (84%) NS genes (Fig. 2A and Figure S2, A to I). Further, 39 out of the 134 Egr1 DS genes, and 40 of the 229 NS genes showed significant daily changes in expression (Fig. 2B), and most of these DEGs exhibited peak expression at 10:00 (35 out of 39 Egr1 DS- and 31 out of 40 NS- genes; Fig. 2, C and D).

Among the 35 Egr1 DS DEGs with highest expression at 10:00, 4 genes are involved in dopamine/ecdysteroid signaling (e.g. DopEcR, E74), 10 genes in intracellular signaling cascades (cac, mpk 3, meng-po), 5 genes in modulating transcription, and 3 genes in synaptic signaling (Fig. 2C). In addition, 3 of the genes peaking at 10:00 (Ap-1, SIK2, Inos) have been reported to function in modulating the core molecular clock in both insects and vertebrates26,27,28. In the group of 31 NS DEGs with peak expression at 10:00, we found 12 genes functioning in neurotransmitter and biogenic amine signaling (e.g. nAChRa, mGlutR1, TyrR, Octbeta2, Dop1), 6 genes involved in neuropeptide signaling, 6 genes of intracellular signaling cascades, 6 TFs (e.g. AP-1, Egr1, Hr38, Creb), and 2 synaptic genes (Fig. 2D).

As genes of the dopamine/ecdysteroid signaling pathway have been shown to play an important role in age-related onset of foraging, and learning and memory during foraging13,16,18,29, we additionally screened all DGs for genes of this pathway. Altogether, we found 9 genes of the ecdysteroid signaling pathway, 4 of them (DopEcR, E74, Egr1, Hr38) showing changes in gene expression at training time and 5 genes (EcR, Usp, E75, Br-c, Mblk1/E93) without daily changes in expression (Figure S2A). Expression of all these genes, except DopEcR, in sKCs or large-type KCs (lKCs) have been reported before30. Interestingly, in Drosophila ecdysteroid-signaling (e.g., E75) is also involved in modulating the molecular clock31. In addition to dopamine/ecdysteroid signaling, tyramine and octopamine pathways also influence foraging behavior, learning, and sensory responsiveness in honey bees32,33. While dopamine/ecdysteroid pathways show the strongest link to time-memory, these other biogenic amines likely contribute to the complex regulation of foraging and deserve further exploration.

To summarize, screening the RNA-seq data for selected set of genes demonstrates that several genes involved in canonical molecular pathways underlying learning and memory in honey bees and other insects, e.g. octopamine, dopamine and glutamate signaling, cAMP-PKA intracellular messenger cascade and activation of transcription34,35 showed training-time related peak expression. These genes represent a promising group of candidate time-training sensitive genes. In addition, our results (Fig. 2C and Figure S2A) substantiate a specific role of the dopamine/ecdysteroid signaling pathway and the TF Egr1 in foraging-related preparatory neuronal processes13,16,18,29. Both likely act in combination with several other TFs (e.g. AP-1, Hr38).

Times-series RNA-seq indicates clock gene expression in sKCs

Encouraged by the results of the candidate gene search and the detection of per expression in honey bee MB single-cell RNA-seq studies36, we also screened the list of DGs for clock genes. Indeed, we found all 8 core clock genes annotated for honey bees (cry2, per, clk, cyc, tim2, cwo, pdp1 and vri) and pdfr, the major output of the central circadian pacemaker (CP). Moreover, cry2 exhibited a significant expression difference among the collection times (p.adj < 0.05, peak expression at 22:00, Figs. 1D and 2E) and significant daily oscillation in gene expression (MetaCycle, BH.Q.< 0.05, Figs. 1F and 2E). The pattern of circadian oscillation and the peak expression time at 22:00 are consistent with whole brain mRNA expression data for honey bees trained to forage at a feeder in the morning37. In addition to the clock genes and pdfr, we found 9 other neuropeptides and neuropeptide receptors (AstA, AstC, Dh31,

Fig. 2
figure 2

Expression patterns of selected genes. (A) Venn diagram showing the overlap among total no. of detected genes, and those from the libraries prepared for Egr1 DS genes (bioinformatically predicted and provided by authors18) and NS genes (literature search). (B) Venn diagram showing overlap among total no. of DEGs, and detected genes of Egr1 DS- and NS- libraries. (C and D) Heatmaps showing the temporal expression of Egr1 DS- and NS- library DEGs. The genes are separated into different categories based on literature review. In (C): Dop/EcR, dopamine-ecdysteroid signaling (the role of Ddc in ecdysteroid pathway in not yet ascertained and thus not included in the count of Dop/EcR signaling genes in main text); circa, genes involved in molecular circadian clock; NT, neurotransmission; IS, intracellular signaling; TF1, transcription factors; SY1, synaptic genes (most of them are involved in synaptic vesicle release); others, genes that could not be designated into a functional category. In (D): NTr, receptors of neurotransmitters; NPr, receptors of neuropeptides; BAr, receptors for biogenic amines; SM, second messenger cascade genes; TF2, transcription factors; SY2, synaptic genes (all of them are involved in synaptic vesicle release); NTe, neurotransmitter enzymes; NP, neuropeptide genes. Corresponding bar diagrams indicate that 89.74% Egr1 DS DEGs and 77.5% NS DEGs show peak expression at 10:00. (E) (i) Heatmap showing expression of all annotated canonical clock genes. Right panel: Cosinor plot showing expression of 5 clock genes characterized for whole brain in37. (ii) Heatmap showing expression of clock-related neuropeptide receptors (predicted in18). See also Figure S2 and Table S1. In (E) only cry2 and ASTC-R are DEGs.

Full size image

Dh44, ASTA-R, ASTC-R, dhr) reported to be involved in clock signaling in Drosophila18,39, Fig. 2E and Figure S2J).

Spatiotemporal expression dynamics of Egr1 and pdfr indicate a functional interaction between the clock and sKCs

To explore functional interactions between the circadian clock and the time-training associated transcriptional response in the sKCs, we examined temporal relations between time-training induced behavioral responses13,15,40,41 and expression dynamics of Egr1 and pdfr. We asked (i) do sKCs get activated, i.e. show Egr1 expression, in anticipation of the training time, (ii) do these sKCs also co-express pdfr indicating a direct modulatory input from the CP, and (iii) do Egr1 and pdfr expression in sKCs develop with the experience of training days demonstrating a correlation between progress of time-training and molecular processes in these neurons.

Again, we trained foragers for 10 days to visit a feeder from 09:00 to 11:00, but this time we provided food at the feeder on the collection day. This way we could monitor the spatiotemporal expression of Egr1 and pdfr during the anticipatory phase when the foragers were still inside the colony (08:00), when they started foraging (09:00), i.e. the first feeder visit, and one hour into continuous foraging [(10:00), also see Figure S3A)]. In time-trained individuals collected at 08:00, Egr1 and pdfr expression was already activated but largely restricted to the sKCs region (Fig. 3A, Figure S3B). At 09:00, expression of both genes extended over the whole area of sKC, and some parts of the other KCs (Fig. 3B, Figure S3C). Finally, at 10:00, after one hour of continuous foraging, Egr1 and pdfr expression was activated over the whole cell body area of the MB calyces (Fig. 3C, Figure S3D). For all the three times-points, we detected significant but not absolute colocalization of Egr1 and pdfr, measured as high degree of overlap between the fluorescent signals of the two probes (Fig. 3D; mean Pearson’s r ≥ 0.5, but ≠ 1 for Manders’ M1 and M2 ~ 1.0; Figure S3E). A quantitative analysis confirmed that the area showing this colocalization both in the region of sKC and other KCs increase across the three time-points (Fig. 3E, Figure S3F). mRNAs of Egr1 and pdfr were detected in- and out-side of DAPI staining indicating that both transcripts are transported outside the nucleus likely for the synthesis of the respective proteins [(Figure S3G), and for specificity and efficiency of RNAscope see Figure S3H]. This tightly synchronized spatial expression pattern of Egr1 and pdfr is preferentially restricted to the KCs and not in other brain regions, for example, CP and central complex (Figures S4 and S5).

To summarize, our smFISH studies indicate a correlation between the spatiotemporal expression pattern of Egr1 and pdfr and the temporal dynamics of behavioral responses associated with feeder time-training. Foragers time-trained for 10 days show an anticipatory activation of Egr1 and pdfr expression in a subgroup of sKCs one hour before foraging which must be initiated by internal signals. In contrast, Egr1 and pdfr co-expression in the lKCs, which is primarily activated during continuous foraging, might be regulated by a combination of internal and external signals16,42.

Co-expression of major clock genes in circumscribed cell groups of major memory centers

Finally, as our RNA-seq study detected clock gene expression in the sKC-enriched tissue samples we asked which sKCs express clock genes, and whether they co-express the clock genes per, cry2, and pdfr, or not. We used newly generated custom-made RNAscope probes for per, and cry2. We detected widespread low expression levels of per, cry2 and pdfr over the whole population of KCs inside the MB calyces. Additionally, we detected small, confined clusters of cells with increased RNA expression levels and stronger co-expression of the two clock genes (see orange arrows in Fig. 4A) in each of the three KC compartments within one MB calyx, i.e. the sKCs group, and in each of the two areas of lKCs adjoining the sKCs (Fig. 4, A and B; left panel, and Figure S6, A and B). Only these clusters showed changes in cry2 expression, being low at 10:00 and high at 22:00, consistent with the oscillation of cry2 in our RNA-seq data and matching the pattern of daily cry2 oscillation in morning-trained foragers (Figure S6C37). In each examined brain the location of the clusters in the MB was mirror symmetric between the brain hemispheres. Among the four examined brains, the location of the sKC cluster as well as lKC clusters showed interindividual variation (Fig. 4A and Figure S6A).

Fig. 3
figure 3

Egr1 and pdfr expression dynamics in MB KCs of time-trained foragers. (A) RNAscope images showing expression pattern of Egr1 (green) and pdfr (magenta) primarily restricted to the sKCs (dashed yellow line) at 08:00. (B) RNAscope images showing Egr1 (green) and pdfr (magenta) are co-expressed in more KCs compared to (A) i.e., in more sKCs, and beyond sKCs at 09:00. (C) RNAscope images showing co-expression of Egr1 (green) and pdfr (magenta) is expanded to mostly all KCs apart from a specific KC area at 10:00. For (A), (B) and (C) right panel shows magnified image of one calyx (as indicated by marked area of left panel). Brain section depth: Mid, 150 to 300 μm. Observations were consistent across 3 analyzed individual brains, and no. of sections per brain (n) = 15 for each of the three time-points. For representative images of all three individuals see Figure S3. The high degree of co-expression in panel (D) already suggests colocalization between the Egr1 and pdfr puncta. (D) Scatter dot plot (mean and SEM) showing Pearson’s correlation coefficient (r) for time-points, 08:00, 09:00, and 10:00, suggesting high degree of correlation between Egr1 and pdfr (r ≥ 0.5 ≠ 1). (E) Box-whiskers plot showing percentage (%) of co-expressed (co-exp.) pixels for Egr1 and pdfr in the sKC area (blue) and other KCs (pink) at 08:00 (8), 09:00 (9) and 10:00 (10). The trendline joins the means (yellow stars) of each column. Test for linear trend in one-way ANOVA, p < 0.05 stands significant (****, p < 0.0001). For both (D) and (E) different calyces at high magnification were used for quantification across N = 3 at equivalent brain depth and each single dot represents data from a different calyx. The respective KC areas were identified by DAPI staining. See also Figures S3 to S5.

Full size image

In addition to the KCs, we also found per, cry2 and pdfr expression in neuron groups adjacent to the central complex neuropils, the brain neuropils in insects involved in spatial navigation, time-compensated sun compass orientation and dance communication43,44,45. Expression of pdfr occurred in groups of cell bodies posterior to the protocerebral bridge (PB). The labeled cell bodies are arranged above each other like those of the central complex columnar neurons that are part of the sun-compass system45, (Fig. 4B; right panels). Co-expression of per and cry2 occurred in cell groups in close vicinity of the PB and central body (Fig. 4B; right panels), but whether they are associated to the central complex requires more detailed studies. Finally, we found co-expression of per, cry2, and pdfr in the three distinct groups of cell bodies located between the protocerebrum and the optic lobes that had been previously identified as the PER-expressing dorsolateral neurons (DLN) and lateral neurons 1 and 2 (LN1, LN2) of the CP (Fig. 4C46). This identification of CP neurons by our probes and the similarity in the expression pattern of clock genes between the CP and the KCs substantiate our detection of per, cry2 and pdfr expressions in the KCs.

Current anatomical descriptions of the CP and other neurons expressing clock genes in honey bee brains are based on immunostaining studies with a polyclonal antibody against PER46. These studies detected PER expressing cells in several brain areas but not in the higher-order integration centers (Table S2). Now, our studies with riboprobes for per and cry2 detected additional per and cry2 co- expressing cell groups in the MBs and the vicinity of the central complex neuropils. This co-expression of per and cry2 in combination with the circadian oscillation of cry2 in the KCs (Fig. 4 and Figure S6, A to F) raise the question whether KCs might exhibit functional molecular clocks.

Discussion

How animal generate time-of-day memories and how they use these to plan their behavioral activities is still an open question. Now our findings provide first insights into molecular processes associated with the expectation of foraging activity in time-trained honey bee foragers. In particular, our experiments document: (i) anticipatory co-expression of the neuronal activity-regulated TF Egr1 and the receptor for pigment dispersing factor (pdfr) in the sKCs, (ii) synchronized peak-level expression of 862 genes including Egr1 DS genes and canonical memory-related genes during the trained foraging time, and (iii) co-expression of per and cry2 in circumscribed groups of sKCs and lKCs as well as cells associated with the central complex suggesting circadian or non-circadian function of clock genes in higher order brain neuropils in time-trained foragers. Future studies are needed to clarify the specific function of these molecular processes in foraging-related time-memory. In particular, establishing a causal role for Egr1 in regulating the observed transcriptional response will require future studies employing refined or novel gene knockdown techniques that can be effectively implemented within naturalistic behavioral paradigms such as multi-day time-training in free-flying foragers.

Fig. 4
figure 4

Potential local molecular clocks in MB KCs and central complex. (A) RNAscope images for per (yellow), cry2 (green), and pdfr (magenta) and co-expression (white) in the MB calyces. sKC region, orange dotted line. Arrowheads: orange, spatial clustering of cells showing high transcript abundance; cyan, per (yellow) and cry2 (green) co-expression in basal sKCs indicated earlier with immunohistochemistry with PER in46. Right panel: High magnification image showing cells with high transcript abundance spatially clustered. DAPI, blue. (B) RNAscope images showing clock gene expressing cells in memory centers. Left major panel: per (yellow), cry2 (green) and pdfr (magenta) expressions in MB. A larger region of the MB shown in (A) is used for representation here. Right major panel: pdfr expression in central complex (left), and expressions of per (yellow) and cry2 (green) and co-expression of per and cry2 (cyan) in cell groups in close vicinity of central complex (right). Whether there are cells showing co-expression of pdfr, along with per and cry2 is still unclear. Since the pdfr labeled cell bodies resemble that of the columnar neurons thus for clarity the two panels (i.e., for pdfr and clock gene expressions) are shown separately. DAPI, blue. (C) RNAscope images showing expression of per (yellow), cry2 (green), and pdfr (magenta) and co-expression (white) in CP neurons. DAPI, blue. Left panel: Schematic representation of identified CP neurons in honey bee foragers, adapted from Fuchikawa et al., 201746. See also Figure S6 and Table S2.

Full size image

Honey bee colonies evolved an elaborate organization of collective foraging in which scouts search for new food sources and recruits exploit those food sources visiting them as long as they provide sufficient food reward47. Moreover, the recruits schedule their daily foraging activity according to the times the flowers provide the most nectar15,41. Such time-trained foragers show characteristic anticipatory behaviors before they start foraging involving leaving their resting place, aggregating near the hive entrance and early flights toward the food source41. Stimulation with a learnt feeder scent during this anticipatory phase activates the respective route memory48. Accordingly, we hypothesize that the anticipatory activity of sKCs and anticipatory co-expression of Egr1 and pdfr likely play a role in the preparation of foraging activity including reactivation of navigational memory and initiation of molecular processes needed for memory reconsolidation. A first comparison of Egr1 and pdfr co-expression in the brain of a forager trained for 2 days with that in the brains of foragers trained for 10 days suggests that this co-expression might develop with training days (Figure S3, E and F). If this trend holds, pdfr expression might be involved in a tighter control of the temporal activity of these cells by the CP48. However, as this observation for day 2 is based on a single forager, future studies are needed to explore this in more detail. Importantly, pdfr expression may require a threshold of reward expectation to sustain a peak at the trained time—conditions met in the RNAscope experiments with rewarded bees. In contrast, reduced Egr1 levels in unrewarded bees sampled in the RNA-seq study at 10:00 likely dampen pdfr expression [Figure 2, C and E (ii)], providing a plausible explanation for the observed discrepancy between these datasets.

Two basic mechanisms have been proposed to explain how the circadian clock orchestrates time-of-day related learning and memory processes: (i) synchronization of the learning performance with the daily foraging activity entrained to the day-night or feeding cycles41,50,51, and (ii) by association of an event or behavior with a specific time-of-day (“time-stamp hypothesis”11). In accordance with the entrainment hypothesis, studies in mice and Drosophila preferentially focused on demonstrating temporal correlations between daily variation in learning performance with circadian oscillations in the expression of memory-related genes4,52,53. Similarly, an earlier behavioral study in honey bees suggested a relation between learning performance and the subjective time-of-day, independent of the timing of the light-dark cycle54, a finding that suggests an involvement of daily oscillations of memory-related genes. In contrast to those findings, the results of our earlier Egr1 experiments13,16 and the current RNA-seq study indicate that feeder time-training in honey bees is associated with a transcriptional response in the sKCs coinciding with the trained foraging time (Table S1). First, 862 (86.6%) out of the total 995 DEGs at p.adj < 0.05 showed maximum or minimum peak expression at 10:00, the collection time-point which is one hour into the feeder training time, whereas only 152 DEGs (1.6%) exhibit a significant daily oscillation at BH.Q.<0.05. Secondly, 35 of the 39 differentially expressed Egr1 DS genes and 31 of the 40 differentially expressed neural signaling genes showed peak-expression during the training-time. Thirdly, Egr1 expression after time-training for several days does not show varying daily oscillations but training-time specific expression13. Although we consider our results as strong evidence for the time-stamp hypothesis of time-memory, we recognize that both mechanisms, entrainment and time-stamp association, are not necessarily exclusive and both might be active in honey bees. Thus, our restricted time-training might have modulated clock-driven daily oscillation of gene expression in such a way that they appear as a function with a peak expression during training-time. Future studies focusing on the detailed effects of different time-training schedules, e.g. multiple feeder times, on the temporal modulation of gene expression in the sKCs will help to resolve this issue. This would also clarify which subset of the 862 DEGs is crucial in time-memory processes. Further, parallel sampling from non-foraging bees at the trained time would help determine the extent to which the DEGs observed at 10:00 reflect anticipatory, time-memory-driven Egr1 activation versus foraging-related transcriptional activity.

In addition to the transcriptional response associated with time-training, our RNA-seq study also detected the expression of all 8 so far annotated honey bee clock genes and the cycling of cry2 in sKC-enriched tissue samples. Furthermore, these results are corroborated by smFISH imaging showing co-expression of per and cry2 in circumscribed groups of sKCs and lKCs and daily changes in cry2 expression levels in those cells similar to those in the CP. Interestingly, a recent RNA-seq study of Drosophila flies, kept under LD and DD conditions, also detected expression of major clock genes in KCs, but none of the detected cycling matched at all that of the canonical cycling of clock genes in the CP52. Thus, we consider our results as first evidence that in honey bees some groups of MB neurons might exhibit functional molecular clocks. Certainly, more detailed studies are needed to confirm our findings. Although our RNA-seq results and imaging data confirm each other, we also like to mention that our findings differ from earlier PER immunostaining studies46. We detected most of the neurons marked by the PER antibody, including CP neurons (Table S2), but only our per riboprobe labeled KCs. It is unclear whether this difference is a consequence of the molecular techniques or the time-training of the foragers in our experiments. Future studies employing simultaneous RNAscope codetection of per transcripts and immunostaining for PER protein in time-trained bees could clarify this observed discrepancy between per mRNA and PER protein expression in KCs. Interindividual differences in the groups of KCs that were labeled by the riboprobe suggest that many or most of the KCs might be capable to express per and cry2 and may develop functional molecular clocks in association with the establishment of time-memories. Unfortunately, our experiments do not allow any conclusion about the specific function of these neurons, except that they present a separate functional module in addition to that of the Egr1/pdfr interaction. Memory neurons expressing clock genes (e.g. Per and Clock) were first reported in the hippocampus of mice twenty years ago55, but the function of the clock genes in these cells and the function of the cells in the neural circuits of the hippocampus are still unresolved. Currently two hypotheses are discussed: either the clock genes simply act as TFs involved in learning and memory processes or they are part of a functional molecular clock, and the cells act as local clocks56,57,58,59. Local clocks in memory centers like the hippocampus would allow autonomous timekeeping independent of the brain’s CP58,59 and/or provide time-of-day information that could be combined with other contextual features of an event like the color or odor of a food reward11. Given that our smFISH experiments indicate the cycling of cry2 in the labeled neurons, we favor the idea that these neurons might function as local clocks, but this must be confirmed by future studies. As genomic and genetic manipulations of brain processes in honey bees are still in their infancies, a combination of different time-training schedules and in-situ hybridization studies could generate detailed correlative data that will allow more specific hypotheses on the function of these cells.

In addition to the MB KCs, smFISH imaging detected pdfr, per, and cry2 expression in cell groups associated with the central complex, the insect brain neuropil involved in spatial orientation, path integration, time-compensated sun compass orientation and dance communication60. Connections between the circadian clock and the central complex have been proposed for a long time44,60, and our findings echo results of an earlier immunostaining study in monarch butterflies, which detected CRY2-positive axonal projections in the central complex61. However, given the different methods it is unclear whether the two studies labeled homologous neuron populations.

Current studies in insects and vertebrates indicate large variations in the anatomical connections and physiological interactions between the clock and memory centers among animal species12,46,58. These differences likely correspond with the capability to form time-memories or even more generally the capability to represent time52,58. Certainly, honey bees are champions of time perception and memory, and this capability likely evolved in the context of optimizing foraging efficiency62. Interestingly, comparative studies on the evolution of large MBs and molecular diversification of KCs, as found in honey bees, indicate that this evolution is not primarily correlated with sociality but with feeding and foraging strategies, increased visual and multimodal input, and long-term memory36,63,64. Particularly, insect species which repeatedly visit several different foraging sites per day appear to have enlarged MBs63. This behavior certainly involves route memories and likely also time-place memories. Thus, these species might also have evolved elaborate interactions between the circadian clock and MBs as well as the circadian clock and the central complex similar to those we have described for honey bees65.

A hundred years ago Karl von Frisch and his students1,6,62 demonstrated that honey bees are capable to memorize the time-of-day an artificial feeder provides food and can associate the different features of a food source, i.e. color, odor, and the route towards the food source with the time-of-day. Today, we have the benefit to combine advanced molecular techniques with their experimental paradigms to discover the brain processes involved, making honey bee foragers an excellent model to study principles and mechanisms of time-memory in animals.

Methods

Animals

Apis mellifera colonies consisting of a naturally mated queen and around 8000 workers (i.e. 8 frames with ~ 1000 workers per frame) were procured from a local beekeeper and maintained on the campus of the National Centre for Biological Sciences - Tata Institute of Fundamental Research (NCBS-TIFR), Bangalore, India. For all experiments colonies were kept in an outdoor flight cage (12 m x 4 m x 4 m), a procedure that allows to control food availability under semi-natural conditions and a natural light/dark cycle. The experiments were performed in September and October 2021, when sunrise and sunset started around 06:10 h and 18:05 h respectively (timeanddate.com). Thus, the bees experienced an approximate 12:12 LD cycle.

Time-training paradigm

Colonies were allowed to adjust to the flight cage for 2–3 days prior to the start of the experiment. During this pre-training period, 1 M sucrose solution and pollen were provided ad libitum on two unscented but differently colored feeders (09:00 to 17:00). Then during the time-training only the sucrose feeder (1 M sucrose solution) was presented for 2 h from 09:00 to 11:00. Foragers were trained for 10 consecutive days to improve temporal accuracy of feeder visits66 and increase the total number of time-trained foragers, as a huge number of individuals were needed for the molecular study. Each day at the end of the training period the feeder plate was cleaned and put back at its location16. On the 7th day all foragers visiting the feeder, were marked at the feeder during the training period. From day 8 to day 10, only those marked foragers were remarked with different colors on each day when visiting the feeder plate during the 2 h training period. On the collection day, we caught consistent foragers66 that had visited the feeder on three of the four previous days, i.e. showing at least three different color marks. Further, we chose a training time period from 09:00 to 11:00 because the anticipatory activity phase of time-trained foragers is shorter in the morning41 which might further help to increase the level of physiological synchronization among the experimental group of foragers.

Collection of time-trained foragers

On the 11th day, food was not provided at the feeder plate. Marked consistent foragers were collected at 7 different time-points at regular 4 h intervals i.e., at 06:00, 10:00, 14:00, 18:00, 22:00, 02:00, and 06:00 for a complete circadian day (24 h) starting at 06:00 (11th day) to 06:00 (12th day). The collection time-points were set to include 10:00, i.e., 60 min into the training time to detect gene expression that has been synchronized with the training time16,40. All foragers collected at 06:00 were in “immobile-active state”, 10:00 were visiting the feeder plate, 14:00 were “active”67, 18:00 were mostly immobile but more active than those at 06:00, 22:00 and 02:00 which were in “second” or “third” sleep stages67. All collections apart from that at 10:00 (done from the unrewarded feeder plate) were done from the hive by temporarily taking out the comb frames from the hive. All night collections were done using dim red light. Collected foragers were immediately flash-frozen in liquid nitrogen and stored at -80 °C until further processing.

Cryo-sections of brain tissue for RNA-seq

Brains of flash-frozen bees were dissected using pre-chilled RNase-free cleaned dissection instruments. To avoid tissue thawing the embedding mold filled with tissue-freezing media (OCT; Leica Microsystems 3P, 14020108926) was slowly cooled on dry ice and dissected brains were placed into the mold once the lower half of the OCT was firmly frozen. After that the tissue block was kept on dry ice to complete the freezing process.

Before sectioning cryomolds containing brains were equilibrated to the cryostat (SLEE MEV) chamber temperature of -28 °C for at least 30 min. Each brain was sectioned at 10 μm thickness. Serial cryosections were mounted on PEN membrane slides disinfected by a 45 min UV treatment (ThermoFisher Scientific Catalogue no. LCM0522). Within 8 min from the mounting of the first section the slides were transferred into a desiccant beads containing slide box kept inside a dry ice container. None of the sections were allowed to dry at RT, to preserve RNA integrity.

Tissue staining and laser capture microdissection (LCM)

Tissue dehydration and staining was performed according to manufacturer’s protocol for Arcturus Histogene Frozen Section Staining Kit (KIT0401) with some alterations in timing of processing steps68. LCM was performed with Zeiss PALM MicroBeam Laser Microdissection System. Marked regions of interest, ROIs (sKCs) on each section were cut and captured using system’s pulsed UV laser (337 nm), with optimum energy and focus, set for the given sample. No more than 20–25 min were spent for each slide.

Sample pooling, RNA isolation, RNA quality check (QC)

The sKC-enriched tissue samples from one brain captured by LCM (total of 314 brains) were collected into adhesive caps of microcentrifuge tubes (Zeiss AdhesiveCap 200 clear). Then tissues were immediately lysed and homogenized using Qiagen RNeasy Plus Micro Kit components (Catalogue no. 74034). Manufacturer’s protocol was followed and microtubes with tissue lysates were transferred to dry ice and subsequently stored at – 80 °C till further processing to isolate RNA. Frozen tissue lysates from 11 to 12 individual brains per replicate of a time-point were thawed at 37 °C for 5 min. Then, samples were pooled and further steps of RNA isolation were done according to manufacturer’s instructions. A total of 27 out of 28 samples (7 time-points x 2 pooled replicates x 2 colonies/behavior experiments) were submitted for mRNA library preparation and RNA-seq at NCBS-TIFR Next Gen Sequencing facility. The final number of samples was 27 as one replicate was lost during capturing of tissue samples due to instrumental malfunction.

RNA-seq experimental details and data analysis

mRNA was isolated from extracted whole RNA using NEBNext® Poly(A) mRNA Magnetic Isolation Module (Catalogue no. E7490L). Stranded mRNA libraries were prepared with NEBNext® Ultra™ II Directional RNA Library Prep with Sample Purification Beads (Catalogue no. E7765L). mRNA library quality was evaluated for all samples with High Sensitivity D1000 Screen Tape ®, Agilent Technologies. Paired end RNA-seq data were obtained from Illumina Hiseq2500 platform using 2 × 100 bp sequencing read length. A total number of paired-end reads generated in a run was 851 and total number of million reads per sample ranged from 27.02 to 37.44.

Raw RNA-seq reads were trimmed to remove adaptors using Trimmomatic69. FastQC70 was done for each sample pre- and post-trimming. Reads were mapped to Apis mellifera genome (NCBI Assembly Amel_HAv3.1) using STAR RNA-seq aligner71. The number of reads aligning to each gene was counted using featureCounts72. To determine the number of detected genes, we set a filtering-criteria = minimum count value of 5 in one or more replicates per time-point, and with that we detected 9229 genes. The DESeq2 package within R73 was used for differential expression analyses, identifying 995 significantly differentiated genes (DEGs) through pairwise comparisons across all time-points with an adjusted p-value (p.adj) threshold of less than 0.05. These DEGs showed significant change (p.adj < 0.05) in their expression levels between at least two time-points. Additionally, DESeq2 was utilized to obtain normalized count values for these selected DEGs, and subsequently, these normalized counts were scaled to generate heatmap visualizations. The MetaCycle package within R was used to evaluate circadian oscillation among the detected genes74. P-values from multiple rhythmicity detection methods were combined using Fisher’s method, and Benjamini-Hochberg adjusted Q-values (BH.Q) were calculated to control the false discovery rate (FDR) across multiple tests. Genes with BH.Q < 0.05 were considered to have significant rhythmicity (FDR < 5%), while a more relaxed threshold of BH.Q < 0.3 was used in exploratory analyses to include additional potentially rhythmic genes (FDR < 30%). To generate heatmap visualizations for MetaCycle genes, normalized count values were scaled. The GO analysis was done using ShinyGO75 with parameters as mentioned in the legends corresponding to the respective figures. For identifying significantly enriched gene sets, we applied a False Discovery Rate (FDR) threshold of < 0.05. This threshold controls for multiple testing by limiting the expected proportion of false positives among enriched GO terms to 5%.

RNA in-situ hybridization with RNAscope technology

In total, with this technique we observed 10 brains for the spatiotemporal expression dynamics of Egr1 and pdfr and 4 brains for monitoring clock gene expression. Brain samples for the RNAscope study (an advanced type of smFISH76,77) were collected from time-trained foragers from honey bee colonies other than those used for the RNA-seq study. The feeder training procedure in both studies were the same. In the RNAscope experiment foragers were marked on the 2nd day they visited the feeder with a day-specific paint mark. From day 7 to day 10, all foragers already marked on day 2, were remarked with different colors on each day when visiting the feeder plate during the 2 h training period. On the collection day (Day 11), consistent foragers66 were caught that had visited the feeder on day 2 and three of the four previous days, i.e. showing at least four different color marks. On the collection day (Day 11) the feeder was rewarded to visualize target gene expression patterns in the brain during continuous food-rewarded foraging78. Additionally, one collection was done on day 2 at 09:00, first visit at the rewarded feeder. For the RNAscope study, we analyzed the expression pattern of brain serial sections of a single forager collected at 09:00 on day 2, and two brains of foragers collected at three different time-points: 08:00, 09:00 and 10:00 on day 11. All foragers originated from same colony. To check for colony effect, we also analyzed serial sections of a second group of brains from foragers of a different colony collected on day 11, one forager each at the 08:00, 09:00 and 10:00.

Collected brains were serial sectioned at 12 μm thickness using the same parameters as used for the LCM sectioning and mounted on Fisherbrand; Superfrost Plus microscope slides (Catalogue no. 1255015). The smFISH with RNAscope technology using the Multiplex Fluorescent Reagent Kit v2 (Catalogue no. 323100) was performed according to manufacturer’s protocol for fresh-frozen samples from Advanced Cell Diagnostics (ACD), a Bio-Techne Brand with minor alteration in the duration spent for fixing sections with 4% paraformaldehyde. The manufacturer (ACD) designed 20 ZZ RNAscope probes target all NCBI annotated transcript variants, and the targeted nucleotide sequences for the respective genes, but only their coding regions (exons; see table below). The details of fluorophores used are summarized in the table below. Slides were coverslipped after counterstaining with H-VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratories, Catalogue no. H-1200-10).

RNAscope probe details

Target ID

Target nucleotide sequence

cry2 (NCBI Gene ID 410197, NM_001083630.1)

70-1199

pdfr (NCBI Gene ID 412439, XM_016913344.2)

1277–2362

Egr1 (NCBI Gene ID 726302, XM_026445312.1)

2444–3914

per (NCBI Gene ID 406112, NM_001011596.1)

856–1824

Fluorophores

Specification

Catalogue No.

TSA CYANINE 5 SYSTEM

NEL705A001KT (50–150 SLIDES, Akoya Biosciences)

TSA CYANINE 3 SYSTEM

NEL704A001KT (50–150 SLIDES, Akoya Biosciences)

TSA Vivid Fluorophore 520

323,271 (ACD, Bio-Techne Brand)

Confocal microscopy and image analysis

All samples were imaged using an Olympus FluoView 3000 confocal microscope. All imaging parameters like dwell time, laser power, and gain were kept constant across samples. Representative confocal images were adjusted for brightness and contrast (evenly across images) in Fiji software (ImageJ79). All quantifications were done in Fiji. Area analysis was done by selecting the ROI, in only DAPI channel and then the same ROI was projected to all the other channels. For area analysis the raw image was median filtered and then thresholded with dark background following which area and area fraction were measured. The plots were generated in GraphPad Prism 10. In GraphPad Prism 10 the data were checked for normality (alpha = 0.05), using four complementary tests—D’Agostino & Pearson, Shapiro-Wilk, Anderson-Darling, and Kolmogorov-Smirnov. For all time points, the sKCs data met normality assumptions while the other KCs showed minor deviations at 09:00 in two tests but passed others, supported by Q-Q plots (Supplementary Figures S3, I and J). Based on these results, parametric analysis using one-way ANOVA (test for linear trend) was used to determine whether the area means for each time-point increases gradually from one time-point to the others, P value < 0.05. For quantification of correlation, JACoP plugin was used and Pearson’s coefficient (r), and Manders’ coefficients (M1 and M2) were determined from raw images.

Quantification and statistical analysis

All statistical details are described briefly in the figure legends and in depth in the method details respectively in the sections for RNA-seq and RNAscope labeling.