Gene expression Patterns
The CA1 hippocampal subregion (Methods) was used for gene expression microarray analysis. Data were filtered () to remove probe sets that were poorly annotated, redundant, or had weak expression levels. One-way ANOVA identified nearly 700 genes significantly influenced by SD. Approximately 8.4% of those are likely to be present due to the error of multiple testing as estimated by the False Discovery Rate (FDR) procedure. We consider this a fairly strong result, particularly compared to studies on normal aging, where FDRs are typically in the 20-25% range. However, the present work also has a higher n per treatment group, likely increasing statistical power.
The combination of significance and direction of change for significant genes constitutes the gene expression pattern for a single gene. Here, we used template matching 
to parse ANOVA-significant genes into expression patterns. Four experimenter-defined expression patterns (templates) were constructed to examine SD effects: sustained, transient, delayed, and linear ( A, B, C, D). Treatment group mean gene expression for every significant gene was correlated with each of these four templates, and each gene was assigned to the template with which it most strongly correlated. Positive correlations were interpreted as “increased” with sleep deprivation, and negative correlations as “decreased”. The complete list of all significant genes, along with their template assignments, mean expression values and 1-way ANOVA p-values, are provided in Table S1
Expression patterns for significant genes are shown.
Heat maps (, center) show color-coded standardized expression values for 10 representative genes from each pattern for each subject. On the right, the averaged expression values for all significant genes assigned to each template are shown, along with the total number of genes assigned to that template. Interestingly, most significant genes were assigned to either the sustained or transient templates. Further, the 72SD group tended to show a lesser (albeit non-significant) tendency to recover back to HC values.
To determine whether the number of genes categorized into each expression pattern was significant, we ran a Monte Carlo simulation. For the simulation, the data table used the same number of chips (columns), and same number of genes (rows), and the same ANOVA and post-hoc template matching strategy as the original microarray data set. However, all signal intensity values were replaced with randomly generated numbers. Then the number of significant “genes” that were categorized in each template was counted. This process was repeated 1000 times, with new randomly generated numbers each time. Out of these 1000 iterations, on average, the random data identified an average of 14 ‘genes’ in the sustained and delayed templates, 19 ‘genes’ in the transient template, and 9 ‘genes’ in the linear template. These results indicate that different templates have different random probabilities of having genes assigned to them, and this should be taken into account during this type of analysis. However, the actual data for all templates (, right) far exceed these Monte Carlo estimated values (p≤1.3×10−8 for all templates, binomial test). Thus, all template-identified genes are occurring at a significantly greater frequency than expected by chance.
Contrasting Sleep Deprivation and Stress Effects
The SD protocol used here has been shown to elicit a stress response 
. To help define sleep vs. stress-related changes, we performed a third experiment in which subjects were exposed to novel environment stress (NES) for 24 or 72 hours. This study was done in parallel with, and shares the same home cage (HC) controls as, cohort 1 of the SD study. The microarray analysis process was highly similar to that used for SD analysis (not shown). Significant results (405 genes, p≤0.05, FDR
0.18, 1-way ANOVA) were contrasted with results from the SD analysis (). The number of genes commonly regulated by the two experimental conditions was highly unlikely to have occurred by chance.
Novel environment stress (NES) elicited a transcriptional response highly similar to that seen with SD.
NES elicited a smaller transcriptional response than SD. However, the NES analysis had fewer arrays (N
21) than SD (N
53), likely resulting in reduced statistical discovery power. To determine whether this difference in N could explain the reduction in number of significant genes, we performed a resampling analysis in which the SD study was artificially restricted to only 21 observations. Among the 53 SD chips, a subset of 21 was tested, and the number of significant genes (1-way ANOVA; p≤0.05) noted. This process was repeated 1000 times with different randomly selected subsets of 21 chips, generating 1000 estimates of the number of significant genes expected if the SD study had only 21 chips. These estimates were plotted as a frequency histogram () and were well fit by a Gaussian distribution. The peak of this distribution (~476) represents the average number of significant genes we predict would be significant if N
21 in the SD study. Superimposing the number of genes actually found significant in the NES study (405 genes- gray area, ) revealed that the ‘transcriptional magnitude of effect’, between NES and SD was not significantly different. That is, although 405 NES genes (observed) is less than 476 (predicted) SD genes, it is not significantly less. Therefore, the high degree of overlap among genes changed in NES and SD, the agreement in direction of that change, and, when compensating for statistical power differences, the relative similarity in the transcriptional strength of these experimental manipulations all suggest that the functional processes associated with SD and NES should be highly similar.
To test this, each region of the Venn diagram () was separated by direction of change and analyzed by functional overrepresentation. We tested the prediction that the two treatments (SD and NES) would show relatively similar functional grouping results and that, if anything, SD would show a more refined dissection of functional groups identified in both SD and NES due to their similarity at the overlap level. Surprisingly, only the upregulated immune and apoptosis pathways supported this hypothesis. SD showed significant and opposite influences on a separate cohort of immune-related genes, and exclusive upregulation of machinery involved in mRNA translation to protein, and downregulation of neuronal components (illustrated in ).
Comparison to other SD Studies
The overlap analysis approach used for the following comparisons has a strong tendency to increase confidence in true positive findings at the expense of false negatives 
. It is also important to note that differences across laboratories, SD techniques, brain regions studied and species used likely account for disagreements among studies. Thus, genes identified in multiple studies should represent robust SD responses that are less likely to be species and/or brain region-dependent.
Complete microarray datasets are available for two mouse brain studies of sleep deprivation: GSE 6514 
; GSE 23628 
. Downloaded and compared to our data set, 495 total genes were present in all three studies. Significant genes: our study
148, and Thompson
63. By chance, 1 gene should be found significantly changed in the same direction by all three studies, while 4 were observed to change (p
0.024, binomial test). Downregulated in all three studies: Cirbp (cold-inducible binding protein), S100a1 (S100 calcium binding protein A1) and Ednrb (endothelin receptor type 1). Interestingly, the three downregulated genes were also robustly downregulated in our NES group.
Three studies examined SD-related transcriptional changes in rat hippocampus 
, rat cerebral cortex 
, or mouse forebrain 
. However, lists of selected significant genes, rather than complete data sets, are available for these studies. Thus, we restricted reports to genes found within our filtered data set, and used a statistical approach similar to that employed in DAVID analysis (679/2167
35% of our data set significant; selected genes then, 17.5% chance of significance and directional agreement across studies). Conti et al report 15 significant genes, 7 of which were significant and changed in the same direction in our study. Mongrain et al report 141 significant results (exclusively among control vs. sleep deprived intact animals). Fifty of these are significant and agree in direction in the present work. Cirelli et al report 28 genes, 12 of which were significant in our study. All of these comparisons were significant (p≤0.05; binomial test) and is annotated with these findings. Interestingly, Homer1 was significantly upregulated in all five prior microarray sleep deprivation studies, as well as in our own study, and was not significantly influenced by NES.
Comparison to Hippocampal Aging
We tested the prediction that SD would cause an aging-like shift in hippocampal gene expression by assembling a composite set of aging genes in rat hippocampus based on three studies 
. The simplified ‘aging union’ results (up- or downregulated in at least one of the prior aging studies- see Methods) are annotated in Table S1
. 617 genes changed with aging and 670 changed with SD. Functional overrepresentation analysis (see Methods) was possible () for genes uniquely regulated by SD (456 genes) or Aging (403 genes). Aging showed a selective upregulation of pathways related to transcription, lipid metabolism, and Ca2+
signaling. Further, downregulated categories were almost exclusively related to neuronal function. Interestingly, sleep deprivation appeared to also downregulate a strong, but different, cohort of neuronal genes ( lower right), suggesting that SD and aging may have additive influences on brain function. Upregulated exclusive SD genes included processes related to protein handling and degradation, as well as well-characterized upregulation of stress hormone stimulus pathways.
SD and aging influence a similar set of genes.
A significant subset (214 genes, p
0.028, binomial test; ) are changed by both SD and aging. Therefore, as with the comparison between SD and NES (), aging and SD appear to exert influence over a common set of genes. However, unlike the very strong tendency for SD and NES gene expression to agree in direction (), this could not be said of aging and SD (, lower), where significant genes were split- 128/214 agreed in direction. We partitioned the overlapping 214 genes into four quadrants based on direction of change in aging and SD. Because the resulting profiles were too small for reliable DAVID functional grouping statistical analysis, genes were manually categorized (- and see below). Immune/inflammatory, neuronal, and macromolecular synthesis categories showed the largest effects. Division/differentiation/apoptosis, energy metabolism, protection/antioxidant, and transcription factor changes appeared similar in aging and SD. Glial changes were upregulated in aging, and a subset of those upregulated changes moved in the opposite direction with SD. Finally, glucocorticoid sensitivity and circadian/sleep were both downregulated with age and upregulated by SD. Individual genes listed in each condition or category, along with notes regarding putative function, are included below.
1) Upregulated in both aging and SD (60 genes)
Division/differentiation/apoptosis: Ap1b1, Cd24, Cdc5l, Msh2, Vcp
Energy/metabolism: Aldoc, Ech1, Gatm, Sds, Slc2a1
Glial: Adk, Cryab, Ocln, Mt1a, Oplah, Pmp22
(although exclusively expressed in peripheral nervous system, mRNA is detected in CNS, see Allen Brain Atlas- 
Immune/inflammatory: C1qb, Cyp4f4, Fn1*, Hmgb2, Ndrg2, Ptgs1* (* inflammatory)
Macromolecular synthesis: Csda*, Eif4ebp1*, Eif5, Rpl10, Rpl21, Rpl23, Rps15, Rps21, Rps5, Rps7 (* suppresses synthesis)
Neuronal: Adra1d, Fxyd1, Gng5, Grip2, Ptpn1, Ralgds, Slc1a3, Sult1a1
Protection/antioxidant: Cited2, Gpx1, Prdx6, Sepp1, Sod3, *Xdh (*pro-oxidation)
Transcription factor: Dbp, Neurod1, Nfia
2) Downregulated in both aging and SD (64 genes)
Division/differentiation/apoptosis: Egr1, Igf2, Lxn, Met, Ngfrap1, Sc65
Energy/metabolism: Acat1, Atp9a, Amacr, Atp1a3, Dnm1l, Fez1, Got2, Ivd, Tst
Immune/inflammatory: Add1, Ap3m2, Csrp2, Cx3cl1, Dcn, Dpp3, Ide, Lamc1, Lrpap1, Muc1, Tubb5 (primarily fibril formation and inflammation-related genes)
Macromolecular synthesis: Cyp51, Faah, Fapb3, Hmgcr*, Sqle* (* rate-limiting enzymes in the sterol biosynthesis pathway)
Neuronal: Accn1, Adcy5, Agrn, Ap2b1, App, Calb1, Calb2, Capns1, Crmp1, Fez1, Gabbr1, Gda, Htr2c, Kcns3, Nnat, Pctk1, Pkia, Por, Ppp2r2c, Prkar1a, Ptpra, Ptprn2, Rab12, Slc1a1, Snca, Syt4
3) Upregulated in aging and downregulated by SD (46 genes)
Glial: Fgfr2, Gfap, Gjb1, Rab13, Tip2
Immune/inflammatory: Dbnl, Grn*, Lamp1*, Litaf*, Plat*, Psme1 (* inflammatory)
Macromolecular synthesis: Gpam, Insig1, Magat1, Pgr, Phgdh, Ppap2c, Rpl29, Srebf1
Neuronal: Akr7a2, Bin1, Cpd, Cplx2, Klk6, Nln, Pgr, Pla2g2c, Syngr1, Vamp1
4) Downregulated in aging and upregulated by SD (44 genes)
Neuronal: Apbb3, Begain, Calr, Celsr3, Cgref1, Chga, Hspa4, Ifrd1, Marcks, Mark3, Npy, Rgs14, Tac1, Vgf (Note: Tac1, Ifrd1, and Npy may also be related to immune or inflammatory signaling)
Immune/inflammatory: Arg2, Col11a1, Serpinh1, Tpm1 (generally opposes other inflammatory signals)
Glucocorticoid sensitivity: Chrbp1, Fkbp4 (expression opposes glucocorticoid action)
Circadian/sleep: Homer1, Per2