Dietary switch reveals fast coordinated gene expression changes in Drosophila melanogaster

Mortality rapidly acquires a restricted diet trajectory upon switch in diet

To identify the potential mechanism most proximally responsible for reduced mortality conferred by DR we applied an experimental protocol where adults are acutely switched from a full diet (control food; CF) to a restricted diet (restricted food; RF). Control food consists of sucrose, yeast and agar, while restricted food contains the same ingredients but with only 1/3 of the sucrose and yeast.

In previous reports, switching adult Drosophila from a full diet to a diet that is diluted for both yeast and sugar causes the mortality rate of the switched cohort to adopt the mortality level and trajectory of a continuously restricted cohort within 2 days [3, 13]. Here, we determined the mortality dynamics of a cohort that is switched between CF and RF. Adult Canton-S flies were collected within 24 hours upon eclosion window and distributed into vials with 25 males and 25 females. 240 vials were assigned to control food (CF cohort). 120 vials were placed on restricted food (RF cohort). Each day, flies were passed onto new food, and dead flies were removed and counted. Mortality rates were estimated daily to determine the time at which there was a measurable separation between the mortality trajectories of the CF and RF cohorts [15]. This appeared at day 40 the number of deaths in both cohorts was sufficiently large to reveal a consistent 1.5-fold difference in age-specific instantaneous mortality rate between cohorts. Accordingly, at day 40, 120 vials were switched from control food to restricted food (Switch Food (SF) cohort). Age-specific mortality of the SF cohort dropped to the level and trajectory of the RF cohort within 3 days (Figure 1A and S1). Our diet switch dynamics based recapitulates the rapid mortality switch reported previously [3, 13].

Transcriptional profiling of diet switch

Transcriptional changes that contribute to reduced mortality during DR must occur within the time-frame when a switched cohort adjusts its mortality to the level of the continuous RF cohort, that is within the three days following the acute diet switch. Hence we profiled gene expression in the CF, RF and SF cohorts at high resolution within the first 72 hours following the diet switch. Beginning at age 40 days we sampled 25 females and 25 males from each of at least 3 vials from each cohort (recording these decrements as censored observations in the life table) at 2, 4, 6, 8, 12, 18, 24, 32, 40, 48, 56, 72 hours (Figure 1B). mRNA was extracted from head and thorax of female flies for transcriptional profiling based on Affymetrix Drosophila 2.0 microarrays.

Transcript differences among continuously control-fed and restricted-fed cohorts

853 genes were differentially expressed between the CF and RF cohorts as measured across all time points between ages 40 to 43 days. These genes represent all the cumulative and secondary differences produced by 40+ days upon different diets as well as any differences that were required to produce or maintain mortality differences among these cohorts. The magnitude of these changes ranged between 0.04-fold for Lsp2 to 28.79-fold for CG8147, with most (62.6%) cases representing reduced expression. Fifteen KEGG pathways (see Methods) are significantly enriched among these differentially expressed genes [16] (Table 1). Down-regulated genes were particularly associated with pathways of carbohydrate and fatty acid metabolism. Pathways associated with up-regulated genes included folate biosynthesis, ubiquinone/terpenoid- quinone biosynthesis, and oxidative phosphorylation. Gene Ontology (GO) analysis of biological processes provides a broader platform for classification that recapitulates the KEGG outcomes and identifies up-regulated genes in the categories of proteolysis, cellular respiration and electron transport chain, and down-regulated genes in categories such as lipid modification, multicellular organismal aging and triglyceride homeostasis (Table S1).

Patterns of transcript dynamics upon diet switch

From the 853 genes that differentiate the CF and RF cohorts, genes likely to cause reduced mortality in DR flies will show temporal dynamics in the SF cohort. These candidate DR affecter genes are expected to depart from the expression level of the CF cohort and acquire the expression level observed in the RF cohort. In a statistical test these Category I switching genes show no significant difference between SF and RF but a significant difference between SF and CF (based on FDR threshold of 0.05, see Methods). 144 genes produced Category I switching dynamics, illustrated by the case of CG8147 in Figure 2. Category II and III contain genes whose expression level is determined by the nutritional history of the cohort. Category II genes (198 cases, Table 2) retained the expression level of the CF cohort after the time of the diet switch (at all time points: SF = CF and SF ≠ RF), exemplified by CG31446 in Figure 2. Category III genes (7 cases, Table 2) diverged from the CF cohort expression level after the switch but did not acquire the expression level of the RF cohort within the mortality switch timeframe; exemplified by CG8785 (Figure 2). Category IV genes were identified from those that did not differ between RF and CF (not in the set of 853); among these, 12 genes in the SF cohort diverged from the level measured in CF but did not adopt the level seen in RF, exemplified by the CG11594 gene (Figure 2). These deviating genes respond to the diet switch but not to chronic DR.

Of the original 853 differences between CF and RF, only 349 genes met the statistical criteria to be assigned to Categories I, II or III. The remaining genes displayed statically indeterminate behavior. Some of these cases, for instance, switched over the course of the entire 72 hours (Figures S2) but too slowly, with too little amplitude or with too much variability to meet statistical significance. These genes are assigned to Categories VI, VII and VIII according to the criteria described in Table S2. All remaining genes that were not significantly different between any of the diet cohorts were assigned to Category V (Table S2).

Pathway enrichment analysis

Categories I, II, III and IV are associated with distinctively enriched pathways, based on the same methodology applied to chronic DR genes. Fifty-five genes of Category I were present in annotated KEGG pathways while 110 were associated with at least one gene ontology. In both classification systems enrichment was most prominent for pathways of carbohydrate and fatty acid metabolism. In particular, down-regulated pathways include several observed for chronic DR, including pentose phosphate pathway, glycolysis/gluconeogenesis, and fatty acid metabolism (Table 3). In contrast, genes of Category I were up-regulated in only a single KEGG pathway, folate biosynthesis. GO analysis recapitulated most of the KEGG findings and further identified oxidation reduction process, oxoacid metabolic process, NADPH regeneration, and NADP metabolic process as enriched for down-regulated genes. Switching genes in the folate biosynthesis pathway include CG10592, CG3290, CG5150, CG8147, which function together to convert 2-amino-4-hydroxy-6-(erythro-1,2,3-trihydroxypropyl)-dihydropteridine triphosphate to dihydroneopterin. Interestingly, all proteins produced by these genes are alkaline phosphatases and show some of the largest levels of up-regulation in both chronic DR and in response to the diet switch (Figure S4B-E). No other alkaline phosphatases were detected as having a similar behavior (Figure S4A).

Forty-eight genes of Category II belonged to at least one KEGG pathways, and 123 fell within to at least one gene ontology. Down-regulated genes again occurred in pathways of carbohydrate and lipid metabolism, such as Mal-A1, Amyrel and CG6543. These are notably distinct from the pathways enriched for switching genes. And again, only one pathway was observed to contain up-regulated genes, in this case, oxidative phosphorylation, which was also found to be enriched for up-regulated genes in chronic DR. KEGG lists 28 genes within the Drosophila oxidative phosphorylation pathway; 11 of these were up-regulated Category II genes while the remaining 17 fell in uninformative Categories VI and did not demonstrate any switching behavior (Supplemental Data File 2). The oxidative phosphorylation pathway is enriched for genes that are up-regulated in chronic DR, but does not contain any switching genes. GO analysis recapitulated all the inferences from the KEGG pathways but further identified proteolysis as a category enriched for up-regulated refractory genes (Table S3).

Categories III and IV in general contain too few genes to provide statistical power for pathway enrichment analysis. It is therefore notable that enrichment with down-regulated genes with the organic cation transport pathway (KEGG) and trans-membrane transport (GO) was detected for Category IV.

Reduced insulin/IGF signaling extends life span in many species including Drosophila [17]. The Drosophila genome encodes 8 insulin-like peptides (dilp1-8). Ablation of DILP producing cells within adults extends lifespan, as does mutation of dilp2. Although single mutations of dilp2 or dilp5 do not affect the ability of DR to extend lifespan, the combined mutation of dilp2, 3, and 5 extends lifespan and appears to block the impact of DR upon aging [18]. Probes for dilp2, dilp5 and dilp6 were present on the Drosophila 2.0 microarrays. The level of mRNA for dilp2 was constant and similar amongst all the cohorts (Category VI, Figure 4C). dilp5 presented as a switching gene (Category I, Figure 4A). Although dilp6 was elevated in chronic DR, the variance of its expression level was too large to gain statistical significance as a Category I gene even though expression level in the SF cohort appears to match that of the RF cohort at late time points (Figure 4B).

The TOR pathway mediates at least part of the longevity response of DR [19-22]. 23 of the 26 genes in the TOR pathway (KEGGID 04150) were present on the microarray, and some of these were represented by multiple probesets for a total of 27 probesets. All probesets in the TOR pathway belonged to Category V (Supplemental Data File 2), i.e. they were not significantly different between any of the diet cohorts. We conclude that in our experiment effects of DR on the TOR pathway genes must occur post-transcrip-tionally and could not be detected by microarrays.

Regulatory elements in the promoters of switching genes

Many Category I genes presented remarkably quick dynamics, switching either up (45) or down (99 cases) within 4-6 hours of the change in diet. Several mechanisms could contribute to these responses, including induction or repression at promoters by transcription factor complexes, transcription initiation of stalled promoter complexes, and mRNA degradation by microRNA.

To identify potential regulatory elements and transcription factors that might coordinate this rapid switching response, we performed an in silico promoter analysis for conserved motifs. Over-expressed and under-expressed switching genes were curated into groups, conditioned on criteria of complete switch with 6-8 hours. This criterion produced a set of 18 genes with increased expression and 10 with reduced expression (Table S4). The FASTA repeat-masked sequences for the promoters of these genes was restricted to a region from −2000 to −1 from the most 3’ TSS to reduce overlap with neighboring genes. Gibbs analysis [23] revealed significant enrichment for WWTTTAATTR and VCWGCTGATY among the up-regulated genes (Figure S3A and S3B), and GTNATTNGTTTG and ANANACATNTTTNW among the down-regulated genes (Figure S3C and S3D).

Based on the Transfac and Jasper databases for regulatory factors [24, 25], the WWTTTAATTR motif corresponds to the binding site of the protein Lmx1b (q = 0.01), a LIM homeodomain transcription factor involved in the development of the brain and spinal cord. The other three motifs did not match any transcription factor in the two databases.

Promoter sequences were further analyzed using MatInspector [26], which uses a large library of position weight matrices for transcription factor binding sites to locate matches in DNA sequences. Five factor matrix matches were found in promoters of switching genes with increased expression with DR – Spalt-like transcription factor, PREB core-binding element, Autoimmune regulatory element binding factor, MEF3 consensus sequence binding factor, PR domain zinc finger protein 14 – and two for the down-regulated genes – Cyclin D binding myb-like transcription factor, and Zinc finger transcription factor RU49.

Contrasting effects between fasting and diet switching upon gene expression

An acute diet shift is likely to produce some rapid transcriptional changes required for metabolic adaptations independent of those responsible for reduced age-specific mortality. Although we cannot distinguish these classes within our current data, some insight is gained by comparing genes from our categories to transcriptional responses reported in cases for Drosophila adults acutely subjected to fasting, which typically decreases their lifespan. We would expect genes that mediate the longevity response to change in opposite directions in response to DR and starvation (Figure 5B).

Fujikawa et al. examined the gene expression changes in flies starved for 24 hours and observed 65 genes whose expression increased under starvation, and 48 genes whose expression decreased [27]. In both our chronic DR and Fujikawa et al. experiments the number of up-regulated genes exceeded the number of down-regulated genes although in slightly greater proportion in chronic DR (172% in DR versus 135% in starvation). Of the 853 genes differentially expressed in our study in response to chronic DR, 45 were also detected as differentially expressed in response to starvation (Figure 5A). Only 22% (i.e. 10 genes) changed in the same direction in both experiments, and of these, 4 genes were up-regulated in both experiments (CG1662, CG18135, and klu in category VI and CG16986 in category 2), and 6 were down-regulated in both experiments (CG14688, CG32425, CG943, fit, and Lsp2 in category I and sug in category VII) (Figure 5C). Most genes differentially expressed in both experiments, 78% (i.e. 35 genes), changed in opposite directions. Of these, a minority (8 genes) were up-regulated in DR but down-regulated in starvation and belonged primarily to category VI. The remaining 27 genes were up-regulated in DR and down-regulated in response to starvation, and more than half of them (14 or 52%) were categorized as switching genes. Pathway analysis performed on this group of 14 switching genes showed that the pentose phosphate pathway and pyruvate metabolism were significantly enriched (q value < 0.05). In particular a closer inspection of the pentose phosphate pathway revealed that a group of four key enzymes regulating several steps along most of the pathway – lactonase, 6-phosphogluconate dehydrogenase, transketolase, and transaldolase – were significantly down-regulated in response to both adult onset and chronic dietary restriction, but significantly up-regulated in response to starvation (Figure 5D-E). Fasting will increase the flux of glucose 6-phosphate to ribose 5-phosphate, and the production of NADPH+H and reduce glutathione, while DR will retard this flow.

Fruit fly husbandry and conditions

Canton-S flies were obtained from the Bloomington Drosophila Stockcenter at Indiana University (Bloomington, IN) and kept at 25°C in a temperature-controlled incubator at 50% humidity with a 12-hour on/off light cycle. The parental generation of flies was raised on food containing 30 g/L autolysed yeast, 120 g/L sucrose, 50 g/L cornmeal, 10 g/L agar with added dry live yeast. Experimental flies were collected in a 24-hour window, anesthetized under light CO2, sorted into vials at a density of 25 males and 25 females per vial, and randomly divided into treatment groups. The flies were then passed every day on either Control Food (CF, 150 g/L sucrose, 150 g/L autolysed yeast, and 20 g/L agar, all w/v) or Restricted Food (RF, 50 g/L sucrose, 50 g/L autolysed yeast, and 20 g/L agar, all w/v) and the number of dead flies recorded. On day 40, half of the flies on Control Food were switched to the Restricted Food. During the course of the experiment, age-specific instantaneous mortality rate was analyzed and the separation of mortality rate between the food conditions was verified before the switch. Flies were sorted under light CO2 and collected at fixed time intervals 2, 4, 6, 8, 12, 18, 24, 32, 40, 48, 56, and 72 hours after the switch time point via snap freezing in liquid nitrogen and were stored at −80°C. Heads and thorax of female flies were collected for microarray experiments.

Demographies

Demographies were performed on all experimental flies (total number of flies in demography: CF = 2892, RF = 2987, SF = 2278) and flies collected for microarray analysis were censored from the study on the day of their collection (total number of flies censored: CF = 1650, RF = 1825, SF = 1126). Log rank tests were performed using the survival package in R to obtain p-values. Maximum life span was calculated as the mean life span of the longest surviving 10% of the population. Instantaneous hazard and survivorship was computed via the “survival” R package (http://cran.r-project.org/web/packages/survival/).

Gene expression studies

Total head and thorax RNA was isolated from at least 75 females using Trizol (Invitrogen) and further purified using RNeasy columns (QIAGEN). 5 μg total RNA was used with Affymetrix One Cycle DNA conversion Kit (Cat # 900431) and all steps were carried out according to the Affymetrix manual. Briefly, first RNA was converted to double stranded cDNA followed by a clean-up step using spin columns. The double stranded cDNA was amplified in an in-vitro transcription reaction overnight at 37 °C using Affymetrix IVT labeling kit (cat # 900449), resulting in biotin labeled cRNA. After clean-up of the labeled cRNA with spin columns, 15 μg of cRNA were fragmented using metal induced hydrolysis. 10 μg of the fragmented RNA were hybridized to Drosophila 2.0 arrays overnight at 45 °C, 60 rpm. The array was stained using Affymetrix Hybridization-Wash-Stain kit and Fluidics Script FS450_0002 on the Affymetrix 450 fluidics station and finally, the arrays were scanned using an Affymetrix 3000 G7 scanner. Probes were mapped to CG numbers using the drosophila2.db annotation package from Bioconductor. The data was PVAC filtered, quantile normalized and summarized using Robust Multichip Analysis (RMA) to obtain expression scores in the log2 scale. Data have been deposited in the Gene Expression Omnibus with accession number GSE47631. qPCR was used to validate a subset of genes and time-points and we observed a consistent correlation between qPCR and microarray trends (Figure S4).

Data analysis

Paired t-test, two sided, was used to test for differential expression between RF and CF, SF and CF and SF and RF, using all time points. Points in each condition were paired for this test by their time of collection from the diet switch. A multiple testing correction was applied to the pooled p values from all tests and comparisons [42]. The False Discovery Rate (FDR) cutoff was set to 0.05 and genes were selected into different categories depending on q values between cohorts.

Heat map visualizations were conducted with GENE-E (http://www.broadinstitute.org/cancer/software/GENE-E/). Functional enrichment and pathway analysis were performed using Flymine [43], KEGG database [16], and the gene ontology (GO) term [44].