The flow of microarray data analysis is presented in Fig 1. In brief, data from 61,818 microarray probes were matched to 20,771 genes. Of these, 3,007 genes met the criteria of P ≤ 0.01 and a fold change ≥ 1.4, the latter chosen for comparability with previously published chicken liver transcriptomics (16). The number of up- and down-regulated genes were roughly equal in all comparisons, ranging from 124 up- and 157 down-regulated genes (comparing AL and F2), to 798 up- and 786 down-regulated (comparing F2 and SK) (Table 2).. In total, 1,675 genes showed upregulation and 1,630 showed downregulation in at least one treatment comparison with some genes showing both patterns depending on which groups were being compared (Additional figure 1)..
Cluster and pathway analyses
The gene set underwent a cluster analysis, sorting genes into four clusters (Fig 2, heatmap rows). The clustering analysis successfully differentiated between the four feeding treatments, grouping the two fully fed states (AL and F2) together and the underfed states of chronic restriction and fasting (CR and SK) together. The gene clusters were also split into two cluster pairs, with clusters 1 and 2 (634 and 912 genes respectively) showing general down-regulation in the fully fed states. Cluster 1 was upregulated in CR but not in fasting, with cluster 2 showing a near-opposite pattern. Clusters 3 and 4 (501 and 960 genes respectively) instead showed general up-regulation in the fully fed states and were mostly down-regulated in CR. Cluster 4 was also down-regulated in SK, but cluster 3 was not. (Fig 2).. Correlational analyses were performed to compare the mean expression of each cluster to known parameters of liver physiology and ArcN expression of appetite-related genes (Fig 3, Additional figure 2).. Clusters 1 and 3 were both significantly correlated with relative liver mass (expressed as % body weight), and the ArcN expression of the appetitive AgRP and the anorexic POMC (Fig 3).. The correlational patterns were opposite with higher Cluster 1 expression in individuals with low relative liver mass, low POMC expression and high AgRP expression and the reversed situation for Cluster 3. Clusters 2 and 4 were instead correlated with hepatic glycogen reserves and ArcN expression of the consummatory orexigenic signal NPY. Again, these two clusters showed opposite patterns with Cluster 2 expression showing negative correlation with glycogen concentration and positive correlation with NPY expression, and Cluster 4 the reverse (Fig 3)..
Each cluster was also subjected to a KEGG analysis. Approximately 40% of the cluster genes were covered by the KEGG database and about half of those were included in the pathway analysis (Fig 1).. Clusters 1, 2 and 4 were heavily loaded with metabolic pathways, with the general “Metabolic pathways” category making up 36.1%, 27.0% and 27.0% of the transcripts respectively (Fig 4).. Of these, 40.6%, 28.3% and 30.7% respectively were exclusive to that category. Cluster 1 mainly consists of genes related to amino acid metabolism, cluster 2 is heavily loaded with lipid and fatty acid metabolism genes and cluster 4 is home to most transcripts related to the metabolism of sugars and starch. Cluster 3, which was by far the smallest cluster in the pathway analysis, was instead made up of genes related to DNA replication and cell cycle progression (Fig 4).. While many of the main terms highlighted in the pathway analysis are ambiguous and largely overlapping, some particularly interesting pathway terms worth noting include cell cycle (30 genes), PPAR signaling (17 genes) and insulin resistance (11 genes). Less relevant pathway terms include “biosynthesis of antibiotics”, a set of 75 genes that entirely overlap with other pathways and are largely made up of genes involved in basic metabolic processes and “Herpes simplex infection” which contains a mostly unique collection of 15 genes involved in apoptosis (FADD, CASP8), inflammatory JAK/STAT signaling (SOCS3, JAK1, IFNGR2), transcriptional activation (TBPL1, ALYREF), splicing (SRSF2, SRSF3, LOC100859609) and protein synthesis (EIF2S1) as well as ERK signaling (FOS, JUN) an S-phase protein (SKP2) and a circadian regulator (CLOCK).
Expression pattern profiling
In addition to the cluster-based analysis we performed a pattern analysis looking for genes with expression profiles of particular interest. The same 3,007 genes that were subjected to the cluster and KEGG pathway analyses were also examined for expression patterns that could be of certain interest in the context of metabolic switching. For this reason we looked for two different expression patterns: either a pattern directly resembling the metabolic switch with F2 and SK being significantly different from each other or a chronic change in expression (F2 and SK both significantly different from AL and changed in the same direction) which would be expected for mechanisms involved in long-term health effects of metabolic switching.
A total of 967 genes met the switching criteria with the expression of at least one of the IF days being significantly different from AL, although this includes some duplicates due to retired annotations (Fig 1).. For further analysis, we chose to restrict ourselves to the 27 genes that were significantly different from AL under both F2 and SK conditions. Of these, seven entries also met the criteria for chronic change and will be considered as part of that dataset. A total of 13 genes were found to be unique, identifiable and have mammalian orthologs that have been characterized; one more was identifiable but uncharacterized (TCP11L2) and four were unidentifiable. The expression profiles of these genes are presented in Fig 5. Six out of the 13 characterized genes show a switching pattern where expression is significantly lower in F2 conditions but elevated during fasting (Fig 5a),, of these five are mainly related to energy metabolism (TMEM234, HAO2, CMBL, MMADHC, GYS2) and one is mainly related to immune function (F11). These genes were typically unaffected by the CR treatment. The remaining seven genes (Fig 5b-c) showed the opposite pattern of elevated expression during feeding but reduced expression in fasting, of these two are mainly implicated in cell proliferation (ODC1, DUSP14), one is immune-related (IRF9; two transcripts) and one mainly involved in energy metabolism (ITPR3). The remaining three are involved in telomere elongation (SMG6), membrane transport (XKRX), and tubulin formation (TUBB2B). Most of these genes showed some level of reduced expression in CR compared to AL. A brief overview of the function of these genes is given in Table 2.
The chronic change criteria were met by a total of 70 genes, of which 49 were unique, identifiable and characterized (Fig 1).. Six of these correspond to the seven entries (i.e. one gene was represented by two transcripts) also showing a switching pattern, and the remaining 43 genes were non-switching. Five of the six genes showing both chronic change and a switching pattern were chronically upregulated compared to AL (Fig 6a),, four of these are mainly related to energy metabolism (HMGCL, SUCNR1, NR1D2 (two transcripts) and AGPAT9) and one mainly to cell proliferation (PLZF). All of these showed a switching pattern with higher expression in fasting conditions, while the only gene that was down-regulated compared to AL (MTHFR, Fig 6b) was also further down-regulated in SK. The function of these genes is briefly summarized in Table 3. The remaining 43 genes were fairly evenly split into 21 characterized (Fig 7a-d) and five uncharacterized genes (Fig 7e) that were upregulated and 22 characterized (Fig 7f-i) and six uncharacterized genes (Fig 7i-j) that were down-regulated compared to AL controls. The upregulated gene set was made up of 13 cell proliferation genes (HTATIP2, PAICS, HISTH1, BIRC5, CDHR2, KIF20A, NHEJ1, FANCI, BRCA2, HIST1H3H (two transcripts), HISTH110, CIP2A, CA9), two energy metabolism genes (ACAT2, CHPT1), one circadian gene (PER3), one immune-related gene (AvBD13; two transcripts), one monoamine oxidase (MAOA), one adhesion molecule (CD99L2), one potential cytoskeletal effector (PYROXD2) and one negative growth factor (CG–16). The downregulated gene set was made up of five energy metabolism genes (SLCO1B3, MYLIP, KCNT2, LPL, HRASLS, UGT1A1), three cell proliferation genes (JARID2, SDC1, RALGPS1), three immune-related genes (CFH, C4A, C4BPA), two circadian genes (DDX5, TIPARP), two transport proteins (SLC40A1, SLC10A7), one Golgi protein (FAM198B), one cell motility gene (SPATA4), two deaminating proteins (ABHD12B, ACCS), one mitochondrial metabolism gene (YME1L1) and one antioxidant response transcription factor (NFE2L2, two transcripts). The functions of these genes are briefly described in Table 4.
Of the 49 genes whose expression levels were chronically affected by IF, ten were chronically changed compared to both AL and CR (Fig 7a-e, i).. The expression of each of these genes were correlated against the same traits of liver physiology and appetite regulation expression in the ArcN as in the cluster analysis (Fig 8).. The expression of HTATIP2, CDHR2, KIF20A, HISTH3H, CIP2A, CG–16, NFE2L2 and BIRC5 were significantly correlated with the relative mass of the liver (expressed as %BW) with only the last two mentioned showing negative correlations. The expression of CIP2A was also significantly and positively correlated with the hepatic total lipid concentration. No other significant correlations were found.