Gene expression profiling of puberty-associated genes reveals abundant tissue and sex-specific changes across postnatal development

Abstract

The timing of human puberty is highly variable, sexually dimorphic, and associated with adverse health outcomes. Over 20 genes carrying rare mutations have been identified in known pubertal disorders, many of which encode critical components of the hypothalamic-pituitary-gonadal (HPG) axis. Recent genome-wide association studies (GWAS) have identified more than 100 candidate genes at loci associated with age at menarche or voice breaking in males. We know little about the spatial, temporal or postnatal expression patterns of the majority of these puberty-associated genes. Using a high-throughput and sensitive microfluidic quantitative PCR strategy, we profiled the gene expression patterns of the mouse orthologs of 178 puberty-associated genes in male and female mouse HPG axis tissues, the pineal gland, and the liver at five postnatal ages spanning the pubertal transition. The most dynamic gene expression changes were observed prior to puberty in all tissues. We detected known and novel tissue-enhanced gene expression patterns, with the hypothalamus expressing the largest number of the puberty-associated genes. Notably, over 40 puberty-associated genes in the pituitary gland showed sex-biased gene expression, most of which occurred peri-puberty. These sex-biased genes included the orthologs of candidate genes at GWAS loci that show sex-discordant effects on pubertal timing. Our findings provide new insight into the expression of puberty-associated genes and support the possibility that the pituitary plays a role in determining sex differences in the timing of puberty.

Introduction

Puberty is a developmental process required across species for reproductive competency. The timing of puberty varies greatly in the general population, differs between boys and girls, and is associated with multiple adverse health outcomes in adult life (reviewed in Golub et al. and Cousminer et al. (1,2)). Puberty begins with an increased release of gonadotropin-releasing hormone (GnRH) from the hypothalamus, activating a cascade of events leading to pituitary-gonadal maturation. Its timing is affected by environmental and genetic factors, with approximately 50–80% of the variation being attributed to genetic background (3–7). Genetic studies of pubertal diseases, such as isolated hypogonadotropic hypogonadism (IHH), Kallmann syndrome (KS) and precocious puberty, have revealed fundamental neuroendocrine regulators of the hypothalamic-pituitary-gonadal (HPG) axis (8). However, the underlying mechanisms responsible for the normal variation observed in pubertal timing in the general population are largely unknown.

In addition to puberty-related disease genes, over 100 genomic loci that are significantly associated with variation in pubertal timing have been discovered by several genome-wide association studies (GWAS) for age at menarche (AAM) in women, or recalled age at voice breaking in men (9–12). As with many GWAS studies, the majority of single nucleotide polymorphisms (SNPs) fall in non-coding regions of the genome. Only a few GWAS loci are found near genes mutated in rare pubertal disorders including MKRN3, LEPR, TACR3 and PCSK1. The functional consequences of genetic variation at the vast majority of the other associated loci have yet to be placed in the context of pubertal development (10).

To gain further insight into the functional significance of genes relevant to pubertal timing, we first explored the gene expression profiles of ‘puberty-associated’ genes in human tissues using publicly available data sets. Informed by these results, we then profiled candidate genes from human GWAS loci as well as genes with mutations known to cause pubertal disorders such as precocious puberty, Kallmann syndrome and hypogonadotropic hypogonadism in mouse tissues. Focusing on HPG axis tissues, as well as the pineal gland (which we found enhanced expression of puberty-associated genes using the human data sets), we report the spatiotemporal and sex-specific patterns of puberty-related gene expression at five postnatal ages spanning the pubertal transition.

Results

Establishing tissue specific expression of human puberty-associated genes

To create a list of puberty-associated genes we first curated a list of genes associated with rare pubertal disorders, which we refer to as ‘Disease-genes’ in this study. We then collected candidate puberty-associated genes implicated by two recent GWAS studies (9,10). In these studies, a list of plausible candidate genes near AAM-associated loci were proposed by the authors using a combination of criteria that considered (i) the nearest gene, (ii) genes in linkage disequilibrium, (iii) expression quantitative trait loci (eQTLs), and (iv) neighbouring biological candidate genes (9,10). We refer to these candidate genes as ‘GWAS-genes’. Combining the two classes of genes, we compiled a list of 169 puberty-associated genes acknowledging that the ‘Disease-genes’ have more evidence for their involvement in pubertal timing than the ‘GWAS genes’, and that the causal GWAS genes may have been missed at certain loci (see Materials and Methods for details of gene selection; see Table 1 and Supplementary Material, Table S1 for the full list of genes). LEPR, MKRN3, PCSK1 and TACR3 were common to both disease and GWAS candidate lists. We then analyzed pre-processed RNA microarray data of 77 normal human tissues obtained from BioGPS database (13) by applying a gene expression specificity measurement based on Shannon entropy as described in Schug et al. (14) (see Materials and Methods for details). This allowed us to assess whether the expression of our list of candidate genes was more enhanced in any of the examined tissues.

Table 1.

Genes used in the human tissue specificity analysis. Gene symbols that are made bold are common to the list of ‘GWAS-genes’ (see second column) but are analyzed as ‘Disease-genes’ in this study

‘Disease-genes’ (n =  24)

‘GWAS-genes’ (n =  145)

CHD7, FGF8, FGFR1, GNRH1, GNRHR, HESX1, HS6ST1, KISS1, KISS1R, LEP, LEPR, MKRN3, NDN, NR0B1, NSMF, PCSK1, PROK2, PROKR2, SEMA3A, SOX10, SOX2, TAC3, TACR3, WDR11

ACAD11, ADARB2, ALOX15B, ARNTL, ASCC3, BCDIN3D, BCL11A, BDNF, BEGAIN, BEND6, BRD8, BRWD1, BSX, BYSL, C22orf34, CA10, CADM1, CADPS2, CBX7, CCDC85A, CCNL1, CENPW, COG4, COG6, CRTC1, CSMD1, CTBP2, DET1, DLGAP1, DLK1, DPYD, DST, EEFSEC, EIF4G1, ESR1, ETV5, EVI5L, FAM83B, FRS3, FSHB, FTO, FUT8, GAB2, GALNT10, GHR, GNPDA2, GPR45, GPRC5B, GTF2I, HCRTR2, HLA-A, HTR1F, IGF2BP2, IGSF11, IL20RB, IMPG1, INHBA, IRX3, JADE2, KCNK9, KCTD13, KDM3B, KDM4A, KDM4C, KLF12, KLHDC8B, LEKR1, LEPR, LGR4, LIN28B, LRP1B, MAGEL2, MAP2K5, MCHR2, MKL2, MKRN3, NCOA7, NEGR1, NFAT5, NPBWR1, NPHP3, NPTXR, NR4A2, NR5A2, NTRK2, NUCKS1, OLFM2, OLFM3, OR2K2, PARP10, PCDH7, PCSK1, PCSK2, PEX2, PGR, PHF21A, PLCL1, POU1F1, PRDM13, PTH, PTPRD, PTPRF, PTPRK, RAB29, RBM6, RDH8, RETN, RMI1, RORA, RXRG, SATB2, SCRIB, SEC16B, SEC23IP, SIM1, SIRT3, SIX6, SKOR2, SLIT3, SMARCAD1, STARD4, STXBP4, TACR3, TBX6, TCF7, TENM2, TEX29, THRB, THRSP, TMEM108, TMEM18, TMEM245, TMEM38B, TNNI3K, TRA2B, TRIM66, TRMT11, TRPC6, TYW3, UBA7, VDR, VGLL3, WDR25, WDR6, WSCD1, WWP2, ZNF131, ZNF483, ZNF654

In total we identified 10 tissues in which puberty-associated genes showed enhanced expression compared to genes randomly selected from the genome (adjusted empirical P value < 0.05, Fig. 1A;Supplementary Materials, Fig. S1, Table S2). We detected the highest degree of enrichment in the pituitary gland, followed by ‘pineal day’, adrenal cortex, ‘pineal night’, prefrontal cortex, parietal lobe, amygdala, hypothalamus, occipital lobe and cingulate cortex (Fig. 1A, see Materials and Methods for pineal gland tissue descriptions). As expected, the pituitary-specific enrichments included previously established pituitary-expressed genes such as GNRHR and FSHB (15,16) (Supplementary Material, Fig. S1). However, GNRH1 was not among the genes with hypothalamus-enhanced expression. Further inspection of the BioGPS microarray data confirmed that GNRH1 expression in the hypothalamus was indistinguishable from other brain regions, likely due to the low abundance of GNRH1-expressing neurons in the hypothalamus.

Figure 1.

Tissue specificity of the expression of human puberty-associated genes in human tissues. (A) X axis indicates the sum of specificity scores (presented as scaled and centered z-scores, normalizing to the scores of randomly selected genes; the lower the score, the higher the tissue specificity). Y axis indicates tissue types. The tissue specificity score of each tissue is indicated by gray diamonds. The normalized distributions of the sum of specificity scores of randomly selected genes (randomly selected from the genome, 1000 iterations) are shown as boxplots (middle bar represents median; box signifies upper and lower quartile; upper and lower whiskers represent upper quartile + 1.5 interquartile range (IQR) and lower quartile – 1.5 IQR respectively, outliers shown as black circles). Only tissues in which puberty-associated genes show statistically significant higher specificity than randomly selected genes are shown (10 out of 77). (B) Heatmap shows pineal-enhanced genes (rows, ranked by specificity score) with their log-transformed average expression levels (represented by color intensity) in the significant tissues. Blocks with borders and an ‘X’ represent genes which were selected as tissue-enhanced in corresponding tissues based on several criteria. For example, a gene ‘g’ will be marked with an ‘X’ in tissue ‘t’ if: 1) its Q value < 5% of all Q values; 2) expression value of gene ‘g’ in tissue ‘t’ > mean + 2SD of expression of gene ‘g’ in all tissues; 3) ranked as top 50% expressed genes in tissue ‘t’.

In addition, we found that 20 of the puberty-associated genes were enhanced in the pineal gland, two of which were in our list of ‘Disease-genes’ (WDR11 and PCSK1; Fig. 1B) (17), the rest being ‘GWAS-genes’. Six of the ‘GWAS-genes’ (TMEM38B, RXRG, PEX2, PCSK2, CA10, NFAT5) were associated with loci reproducibly found in both Elks et al. and Perry et al. studies (9,10). Interestingly, the pineal gland is involved in pubertal timing in seasonal breeders (18), and it has also been linked with puberty phenotypes in humans (19,20) and rats (21,22). Furthermore, genetic variation near TMEM38B has been reproducibly identified in multiple GWAS studies, and it is the second most statistically significant GWAS hit after the LIN28B locus (9,10,23).

Establishing a system to profile the expression of the mouse orthologs of puberty-associated genes

To examine how the expression of puberty-associated genes change during postnatal development, we used the C57BL/6J mouse as a model system (24). We profiled the expression of 183 genes in selected mouse tissues across the pubertal transition in both male and female mice using the Fluidigm BioMark qPCR array (see Materials and Methods for details). The 183 genes we profiled include mouse orthologs of human puberty-associated genes (‘Disease-genes’: n = 24; ‘GWAS-genes’: n = 140), genes with reported functions in HPG axis development (‘Literature-based genes’: n = 7), additional genes near GWAS loci whose expression was enhanced in hypothalamus, pituitary or pineal gland in our human expression dataset analysis (‘Analysis genes’: n = 8) and control genes (n = 4) (see Materials and Methods for details of gene selection, Table 2 and Supplementary Material, Table S1 for a complete list). We profiled the expression of these genes in HPG axis mouse tissues (hypothalamus, pituitary, ovary and testis) and added the pineal gland because of the enhanced expression of puberty-associated genes in this tissue (Fig. 1). We also profiled the liver as a relatively homogenous, non-HPG axis tissue that undergoes extensive metabolic changes and sex-specific gene expression during postnatal development (25). Based on data collected from C57BL/6J mice born in our facility, pubertal onset age ranges from postnatal day 25 to day 31 for male mice, and day 28 to day 36 for female mice. We collected samples at five postnatal days corresponding to: early development (day 12), pre-pubertal (day 22), pubertal (days 27 and 32), and post-pubertal (day 37) stages in both male and female mice.

Table 2.

Genes profiled in mouse tissues using microfluidic qPCR. Gene symbols that are made bold are common to the list of ‘GWAS-genes’ (see second column) but are analyzed as ‘Disease-genes’ in this study. In the ‘GWAS-gene’ category, MCHR2 and C22orf34 were excluded because they do not have an established mouse ortholog. PTH, SKOR2 and TBX6 were excluded due to failure to design suitable primers

‘Disease-genes’ (n =  24)	‘GWAS-genes’ (n =  140)	‘Literature-based’ genes (n =  7)	‘novel analysis genes’ (n = 8)	Reference genes (n =  4)
Chd7, Fgf8, Fgfr1, Gnrh1, Gnrhr, Hesx1, Hs6st1, Kiss1, Kiss1r, Lep, Lepr, Mkrn3, Ndn, Nr0b1, Nsmf, Pcsk1, Prok2, Prokr2, Sema3a, Sox10, Sox2, Tac2, Tacr3, Wdr11	Acad11, Adarb2, Alox8, Arntl, Ascc3, Bcdin3d, Bcl11a, Bdnf, Begain, Bend6, Brd8, Brwd1, Bsx, Bysl, Cadm1, Cadps2, Car10, Cbx7, Ccdc85a, Ccnl1, Cenpw, Cog4, Cog6, Crtc1, Csmd1, Ctbp2, Det1, Dlgap1, Dlk1, Dpyd, Dst, Eefsec, Eif4g1, Esr1, Etv5, Evi5l, Fam83b, Frs3, Fshb, Fto, Fut8, Gab2, Galnt10, Ghr, Gnpda2, Gpr45, Gprc5b, Gtf2i, H2Q10, Hcrtr2, Htr1f, Igf2bp2, Igsf11, Il20rb, Impg1, Inhba, Irx3, Jade2, Kcnk9, Kctd13, Kdm3b, Kdm4a, Kdm4c, Klf12, Klhdc8b, Lekr1, Lepr, Lgr4, Lin28b, Lrp1b, Magel2, Map2k5, Mkl2, Mkrn3, Ncoa7, Negr1, Nfat5, Npbwr1, Nphp3, Nptxr, Nr4a2, Nr5a2, Ntrk2, Nucks1, Olfm2, Olfm3, Olfr267, Parp10, Pcdh7, Pcsk1, Pcsk2, Pex2, Pgr, Phf21a, Plcl1, Pou1f1, Prdm13, Ptprd, Ptprf, Ptprk, Rab7l1, Rbm6, Rdh8, Retn, Rmi1, Rora, Rxrg, Satb2, Scrib, Sec16b, Sec23ip, Sim1, Sirt3, Six6, Slit3, Smarcad1, Stard4, Stxbp4, Tacr3, Tcf7, Tenm2, Tex29, Thrb, Thrsp, Tmem108, Tmem18, Tmem245, Tmem38b, Tnni3k, Tra2b, Trim66, Trmt11, Trpc6, Tyw3, Uba7, Vdr, Vgll3, Wdr25, Wdr6, Wscd1, Wwp2, Zfp131, Zfp654, Zkscan16	Eed, Hnf4a, Lhx3, Lin28a, Otx1, Otx2, Yy1	Gtf2ird1, Hace1, Hdhd2, Odf3, Pagr1a, Reep2, Sez6l2, Trappc9	Actb, Gapdh, Adprh, Ddx18

In total, we performed 53,676 qPCR reactions on six microfluidic chips. Excluding control samples, dilution series and reactions beyond detection limits, we effectively measured 46,450 gene expression levels (Supplementary Material, Fig. S2A). To determine the consistency of our experiment, we assessed the correlation between biological replicates (n = 6 for hypothalamus, pituitary and pineal samples, n = 4 for gonads and liver samples). Only one pituitary sample on one microfluidic chip showed an average Pearson correlation coefficient < 0.8 with its biological replicates and was excluded from further analysis (Supplementary Material, Fig. S2B) (see Materials and Methods for details). We successfully obtained data for 3 or more biological replicates in 93% of all the assays (an assay refers to the profiling of one gene in one male or female tissue at a certain age) (Supplementary Material, Fig. S2C).

The expression of puberty-associated genes is most prominent in the hypothalamus

As expected, genes with known tissue-enhanced expression patterns were found, such as Gnrhr, Fshb and Pou1f1 in the pituitary (15,16,26), Tex29 and Zfp654 in the testis (27) and Hnf4a in the liver (28) (Supplementary Material, Fig. S3, S4A). In addition, our qPCR method was sensitive enough to detect rare transcripts. For example, we detected hypothalamic-specific expression of Gnrh1 and Prdm13, which are expressed only by a limited number of GnRH neurons and the dorsomedial hypothalamus (DMH) respectively (Fig. 2A) (29,30).

Figure 2.

Puberty-associated genes show tissue-specific gene expression. (A) Expression levels of Gnrh1 and Prdm13 in hypothalamus and pituitary across development. The x-axis shows age, and the y-axis shows normalized expression level (ΔCt). The red color represents female gene expression, and blue represents male gene expression. Circles (female samples) and triangles (male samples) represent median expression levels of biological replicates, with error bars showing IQR and the smaller colored dots represent the expression levels of each biological replicate. These examples demonstrate our ability to detect rare transcripts in the hypothalamus. (B) 2D t-SINE map of hypothalamus, pituitary, pineal and gonad samples. Each data point represents a sample, with tissues shown by shapes, age by different sizes, and sex by colors. Neighbouring data points were circled and the tissues they were collected from were highlighted on the plot. (C) Heatmap showing the expression levels of HPG tissue and pineal-enhanced genes. Color intensity corresponds to the median expression level (ΔCt) of a gene in a tissue, considering all samples (including samples of all ages and both sexes). We used this median value to represent the expression level of a gene in a tissue. Each column represents a tissue (left to right columns: hypothalamus, pituitary, pineal gland, testis and ovary) and each row represents a gene.

The expression of mouse orthologs of human puberty-associated genes showed distinct tissue features in our selected tissues (Supplementary Material, Fig. S3). We visualized the similarity of our samples using a 2D t-Distributed Stochastic Neighbor Embedding (t-SNE) map (31), based on the expression of 120 genes which were detected in all samples used (Fig. 2B). Samples were separated to distinct clusters featuring different tissues. In all tissues, day 12 samples were relatively more separated from samples collected at other ages. As expected, gonad samples were clearly grouped based on sex. Separation of samples based on sex was also observed in pituitary, especially in samples collected at day 27 onwards.

We classified gene expression levels into five categories based on their relative expression levels (ΔCt values): not detected, low, weak, medium and strong (see Materials and Methods for details). Using these categories we found that our set of puberty-associated genes were most highly expressed in the hypothalamus (86% of reactions in medium or strong categories), followed by the pituitary (81%) and the pineal gland (77%) (Supplementary Material, Fig. S4B).

We identified 42 genes whose expressions are enhanced in only one of the HPG tissues or the pineal gland based on several criteria (see Materials and Methods for details). We found the largest number of such tissue-enhanced genes in the hypothalamus (n = 23) followed by the pituitary gland (n = 5), testis (n = 5), ovary (n = 5) and the pineal gland (n = 4) (P value < 0.05, Wilcoxon rank sum test, differences in median ΔCt value > 2) (Fig. 2C, Supplementary Materials, Fig. S3, S4A). We examined their biological relevance using functional enrichment analysis with the mouse genome as background. We identified that the mouse hypothalamus-enhanced genes are most significantly enriched for terms related to obesity based on Sim1, Tac2, Tacr3, Sox10, Ndn, Prokr2, Ntrk2, and Magel2 (‘obesity’ (P value = 9.79e-8); ‘abnormal eating behaviour’ (P value = 2.92e-5)), which is in concordance with the complex relationship between nutritional status and age at puberty (12,32). Ovary-enhanced genes are enriched for several terms including ‘gonad development’ based on Retn, Fgf8, Inhba, and Lep (P value = 5.15e-5) and testes-enhanced genes were enriched for ‘miR-346’ based on Prok2, Tex29, and Zfp654 (P value = 0.0379). The ‘GnRH signaling pathway’ was enriched in pituitary based on Fshb, Gnrhr genes (P value = 0.0379; see Supplementary Material, Table S3 for a complete list of enriched terms).

Overall, the tissue-specific expression of our mouse genes was largely in concordance with their human orthologs. For example the majority of genes we found enhanced in the human hypothalamus or pituitary were also strongly expressed in the mouse hypothalamus or pituitary (Supplementary Material, Table S2). Eight out of 20 orthologs of human pineal-enhanced genes were also strongly expressed in the mouse pineal gland; while 10, including Tmem38b were expressed at medium level in mouse.

Most temporal changes in gene expression of puberty-associated genes occur before puberty

We examined the temporal expression patterns of the puberty-associated genes in each tissue. 1,310 patterns were assessed and classified into 4 clusters (see Materials and Methods for details, Supplementary Material, Fig. S5). Clusters 1 and 4 contained genes whose overall expression increases or decreases around puberty respectively (day 22 to day 32). Cluster 2 contained genes whose expression drastically increases from day 12 to day 22. Expression patterns of puberty genes Gnrh1, Kiss1 and Tac2 in male hypothalamus fit into Cluster 2. In contrast, expression of Gnrh1 and Kiss1 in the female hypothalamus was classified into Cluster 1. This finding is particularly interesting because in our mouse colony, males undergo puberty at an earlier age than females (as measured by preputial separation (PS) and vaginal opening (VO)), consistent with the earlier rise in the expression of these genes among male compared with female mice. Finally, Cluster 3 was characterized by expression that decreases early and generally stays lower after day 12. Bcl11a and Mkrn3, two genes with enhanced expression in the hypothalamus, fit into Cluster 3 with Bcl11a showing the most similar temporal pattern to Mkrn3 in our time series analysis.

We then asked which genes show temporal expression changes when comparing pre-, peri-, and post-puberty development in mouse. We separately assessed the qPCR gene expression data of male and female mice, and identified 173 genes whose expression in at least one age group is different from other ages in at least one tissue (One-way ANOVA, P value < 0.05, Supplementary Material, Fig. S6A). Only five genes, Lep, Prdm13, Otx2, Six6 and Lhx3, did not show significant temporal changes in any tissues. When considering only pairs of adjacent developmental time points, most changes occurred between day 12 and day 22 (P value < 0.05, Fig. 3, Supplementary Material, S6B, S7A).

Figure 3.

The largest number of temporal gene expression changes is observed pre-puberty between days 12 and 22. Numbers of genes significantly changed (One-way ANOVA and Tukey’s post hoc test, adjusted P value of interaction effect < 0.05, see Materials and Methods for details) between each neighbouring developmental time points are shown for both female (top panels) and male (bottom panels) samples in all tissues (left to right, hypothalamus, pituitary, pineal gland and gonads. Height of bars in positive numbers shows the number of genes with positive expression changes between the labeled ages (day 22 > day 12, day 27 > day 22, day 32 > day 27, day 37 > day 32). Height of the bars in negative numbers represents number of genes with negative expression changes between the labeled ages (day 22 < day 12, day 27 < day 22, day 32 < day 27, day 37 < day 32). Colors represent genes of different categories.

We found that the largest number of temporally dynamic genes (n = 154) are expressed in testis, most of which decrease in expression from day 12 to day 22 (Fig. 3, Supplementary Material, S6A, S6B, Table S4). This finding is consistent with global gene expression trends during postnatal testis development (27,33). In contrast, it is interesting that the subset of our ‘puberty-associated genes’ that are specifically enhanced in the testis (Odf3, Prok2, Tex29, Lekr1 and Zfp654) are all significantly up-regulated during postnatal development and remain highly expressed after puberty. Expression of the puberty-associated genes in the ovary was less dynamic compared to that in the testis, which is in line with studies comparing postnatal gene expression profiles in male and female gonads (34,35).

In the hypothalamus, 17 puberty-associated genes were significantly up-regulated while 21 were significantly down-regulated from day 12 to day 22 in both males and females (Supplementary Material, Table S4). All of the up-regulated genes were expressed at medium or strong levels in hypothalamus, including three ‘Disease-genes’: Pcsk1, Tac2 and Tacr3. The top five genes that showed the greatest increase in gene expression are Dlk1, Wscd1, Cbx7, Jade2 and Tac2. The gene showing the most significant decrease in gene expression between day 12 and day 22 is Mkrn3. Notably, Mkrn3 is present in both of our ‘Disease-gene’ and ‘GWAS-gene’ lists, has known mutations causing precocious puberty, and is known to be down-regulated prior to pubertal onset in hypothalamus (36,37). The remaining down-regulated genes are all ‘GWAS-genes’, with the top five being Bcl11a, Sec16b, Fam83b, Igf2bp2 and Olfm2. However, Sec16b and Fam83b were weakly expressed in hypothalamus in all ages and both sexes (Supplementary Material, Table S4).

The temporal gene expression changes in the pituitary showed more sex-bias than other tissues profiled (Fig. 3, Supplementary Material, S4B, Table S4). Specifically, between day 12 and day 22, more genes showed significant temporal changes in male pituitary compared to female pituitary, and between day 22 and day 27 significant expression changes were only detected in male pituitaries. The pineal gland showed the lowest number of significant temporal changes, all of which occur between day 12 and day 22. After day 22, the expression of puberty-associated genes in the pineal gland was relatively stable.

Identifying sex-biased gene expression patterns in puberty-associated genes

We sought to gain insight into differences in male and female pubertal timing by determining sex-differences in the expression patterns of puberty-associated genes. Overall, we identified 48 genes that show sex differences in the pituitary gland, followed by 14 in the liver, and 11 in the pineal gland (P value < 0.05, student’s t-test) (Fig. 4A, Supplementary Material, Fig. S7B). These correspond to seven ‘Disease-genes’, fifty-three ‘GWAS-genes’, two ‘Literature-based’ genes and two ‘Analysis genes’. Hypothalamic tissue showed no significant sex-biases in gene expression which is concordant with results from previous studies in rat, mouse and rock pigeon (38–40) (Fig. 4A). All the genes with sex-biased expression in the pineal gland were ‘GWAS-genes’ and the differences were observed at day 37. Wscd1 was enhanced in the pineal gland and was more highly expressed in female mice; the other 10 genes were more highly expressed in male mice. As positive controls, we captured known sex differences such as Fshb expression in pituitary (16,39), and Ntrk2 expression in liver (41) (Supplementary Material, Fig. S9).

Figure 4.

The pituitary gland exhibits the greatest level of sex dimorphism. (A) The distribution of differences in expression between sexes for each gene, at each time point in hypothalamus, pituitary and pineal gland are plotted. Colored diamonds (jittered to avoid overplotting) indicate the genes with statistically significant sex differences (Student’s t-test, adjusted P value < 0.05, see Materials and Methods for details). ΔExpression: female expression levels (measured by median ΔCt of biological replicates) minus male expression levels (measured by median ΔCt of biological replicates). (B) Expression profiles of genes showing significant sex-biased change from d22 to d27. (C) Heatmap showing the sex differences of selected genes in the pituitary. Color scale represents the sex differences in gene expression (measured the same way as in panel A; red: higher in female; blue: higher in male). Hierarchical clustering was performed based on the difference between expression levels in females and males. Genes showing significant sex-biased expression at certain ages are labeled with a star (*) at the corresponding columns.

A critical time window for pubertal transition in both male and female mice is the period between day 22 and day 27. We asked if any genes showed sex-biased expression changes from day 22 to day 27. We identified seven genes (Cadps2, Irx3, Olfm3, Pgr, Retn, Tenm2 and Tyw3) in the pituitary with sex specific expression differences (two-way ANOVA, age and sex as factors, age: gender interaction effect P value < 0.05, Tukey’s HSD test) (Fig. 4B). The expression of Irx3 and Retn in males increased from day 22 to day 27, while expression in females decreased. In contrast, Cadps2, Olfm3, Pgr, Tenm2 and Tyw3 showed the opposite trend, with decreased expression in males and increased expression in females.

The pituitary gland displayed the largest number of sex differences (Fig. 4A, Supplementary Material, Table S5), and the notable aspects of these sex differences are: 1) Seven ‘Disease-genes’ (Kiss1, Fgf8, Pcsk1, Nsmf, Hesx1, Sema3a, and Tacr3) are identified as sex-specific, all in pituitary. Among these, only Pcsk1 is previously known to exhibit sex-biased expression in the pituitary. 2) 37 ‘GWAS-genes’ show sex differences in the pituitary gland, and two (Dlk1 and Fshb) display enhanced expression in this tissue compared to other tissues. Lin28a, which plays a functional role in pubertal onset in mice (42), is also significantly sex-biased in the pituitary gland at day 32 and day 37. 3) At day 12, three genes (Fshb, Hesx1 and Npbwr1) show significant sex differences in the pituitary. Both Fshb and Hesx1 are ‘Disease-genes’ and show similar expression pattern changes between sexes (higher in females at day 12 but higher in males at later ages). 4) Many sex differences in the pituitary gland were observed after puberty.

Only six out of 123 GWAS-associated loci associated with age at menarche have been demonstrated to exhibit sex-discordant effects on pubertal timing (2,12). The genes associated with these loci are OLFM2, RXRG, CSMD1, SIX6, SLIT3 and SIM1. It is notable that in the mouse tissues we profiled, Csmd1, Olfm2 and Rxrg showed significant sex differences in expression in the mouse pituitary gland at several ages, and Slit3 showed sex-biased expression in the pineal gland at day 37 (Supplementary Material, Table S5). The functional significance of these sex-specific gene expression patterns associated with pubertal timing phenotypes remains to be elucidated.

Because the pituitary showed the greatest sex-differences in puberty gene expression, we explored pituitary sex-biased genes in more detail. In the pituitary gland, similar numbers of male-biased (expressed higher in male samples) and female-biased (expressed higher in female samples) genes were identified. However, female-biased genes showed stronger gene expression differences compared to male-biased genes, and the differences occur earlier (Fig. 4C). We calculated the temporal sex differences between genes by deducting median expression of male samples from the median expression of female samples at each age. We then clustered the genes based on their sex-differences in expression (Fig. 4C). Globally, male-biased and female-biased genes formed separate clusters. Noticeably, the temporal sex differences of Fshb were most similar to those of Retn (Supplementary Material, Fig. S8A) while the temporal sex differences of Impg1, Kiss1 and Tnni3k were similar (Supplementary Material, Fig. S8B), forming two outermost groups in the clustering. Impg1, Kiss1 and Tnni3k were all weakly expressed in the pituitary gland and showed a prominent increase of expression around pubertal onset (from day 22 to day 27) in female pituitary tissue.

Discussion

Genome wide association studies are revealing an increasing number of loci where genetic variation can explain variation in the timing of puberty. As an initial step towards identifying gene regulatory networks that regulate the timing of puberty, we leveraged a human gene expression compendium to identify tissue enhanced expression patterns of candidate GWAS genes. To gain insight into the spatial, temporal and sex-biased gene expression patterns of puberty associated genes we then used a microfluidic qPCR strategy to profile the expression of 178 mouse orthologs of human puberty-associated genes. Our study assessed multiple puberty-related tissues simultaneously, addressed the temporal pattern of gene expression across the pubertal transition, and looked at sex-specificity of gene expression, which is often overlooked in gene expression profiling studies. Our approach provides new insight into the regulation of pubertal timing. For example, our data raise the possibility that the pituitary, not the hypothalamus, may be the seat of much of the sex-specific regulation of pubertal timing by these genes.

Our microfluidic qPCR approach had the sensitivity to detect the expression of genes, which are either overall lowly expressed or expressed only in a small number of cells within a heterogeneous tissue. As such, we detected the expression of the majority of puberty-associated genes in the hypothalamus. This result supports the central role of the hypothalamus in modulating pubertal onset, although we acknowledge that many of these genes are broadly expressed in the central nervous system. In all tissues, we observed the most dynamic gene expression changes between days 12 and 22. Additional studies are needed to assess the degree to which this dynamic expression is specific to puberty-associated genes. Understanding the role that factors such as diet have on puberty-associated genes during this postnatal developmental stage will also be important. For example, little is known regarding how weaning, which involves a sudden switch from a lipid rich diet to a carbohydrate rich diet, influences puberty-associated gene expression.

Many of the puberty-associated genes, especially the ‘Disease-genes’ have established functions in the hypothalamus. In fact, only three ‘Disease-genes’ (Fgf8, Hesx1 and Lep) were not detected in the hypothalamus. We also detected the expression of some ‘Disease-genes’ outside of the hypothalamus. For example, we detected Kiss1/Kiss1r expression in the pituitary. This is consistent with several previous studies, which showed that kisspeptins can directly regulate gonadotropin release at the pituitary level (43–46). Another example is the expression of Prok2/Prokr2 in the testis, which has also been described previously (47). In addition, the expression of Fgf8 and Lep is enhanced in the ovary while the expression of Hs6st1 is enhanced in the liver. However, the exact function of these genes in these tissues and whether the non-hypothalamic expression of the mutated genes contributes to disease phenotypes are not clear.

While the increased sensitivity in our qPCR method permitted observation of relevant changes in key pubertal regulators known to be expressed in a limited number of neurons (Gnrh1, for example), our approach did not allow us to distinguish cell-type-specific expression. Future work such as purification of specific cell-types from within the hypothalamus represents a way to examine more closely the cell-type specific expression of these genes.

Finding an enrichment of pineal-expressed genes among the human puberty-associated genes is intriguing as there have been conflicting findings with regards to the potential role of the pineal gland and melatonin in regulating human puberty (19,20,48). It is possible that some of the AAM-associated genes have undetermined roles in the pineal gland and contribute to the modulation of puberty timing. However, not all human pineal-enhanced genes showed strong expression in mouse pineal gland in our experiment. Interestingly, all of the mouse orthologs of human-pineal enhanced genes were strongly expressed in the mouse hypothalamus, and most were also strongly expressed in mouse pituitary. It is therefore possible that these genes are enhanced in the brain broadly and were more sensitively detected in the pineal gland in the human microarray dataset.

Some genes with an established function in other tissues showed enhanced mouse pineal expression. For example, Kiss1r is highly expressed in the mouse pineal gland in our study and evidence has shown that kisspeptin signalling can be affected by melatonin (51). Lhx3 is known to be important for pituitary gland development and is also known to be expressed in prenatal mouse pineal gland (52). Our study showed that it is also strongly expressed in the pineal gland during postnatal development. An important consideration for our pineal gland expression patterns is the known differences in pineal gland physiology between and within species. For example, the pineal gland in C57BL6/J mice expresses very little melatonin (49). It will be interesting to compare expression patterns of puberty related genes in melatonin expressing mouse strains (e.g. C3H, (50)). Our findings regarding the pineal expression of puberty-associated genes represent an area of important future research. To further explore the significance of this observation, it will be important to determine, for example, if these genes are parts of pathways linking the pineal gland with the hypothalamus and pituitary.

In this study, we focused on gene expression profiling in HPG axis tissues and the pineal gland. Future work is warranted in the other human tissues with enhanced expression of puberty-associated genes. For example, the adrenal cortex, together with the hypothalamus and pituitary, constitutes the hypothalamic-pituitary-adrenal (HPA) axis, a neuroendocrine system that controls reactions to stress and could influence pubertal timing. The amygdala is involved in the regulation of the HPA axis, and furthermore, has been shown to modulate puberty timing through multiple neuroendocrine signalling pathways in rodents (53,54). Several studies have provided evidence of these associations in humans too. For example, stressful early life events (e.g. adoption, single-parenting) have been associated with an earlier onset of puberty (55–58). Thus, the expression of genes in HPA axis tissues suggests a possible role for these genes in the regulation of the HPA axis and modulating the timing of puberty.

Except for the hypothalamus, which showed roughly an equal number of genes with increased or decreased expression from day 12 to day 22, at least 66% of genes with significant changes over time showed decreased expression. The onset of puberty is modulated by genes that stimulate the HPG axis, such as Kiss1, whose expression might be expected to increase prior to puberty, but also by genes that inhibit puberty, whose expression would decrease prior to the onset of puberty. MKRN3, which has identified to be near GWAS loci and also known to cause precocious puberty, is an example of a gene whose expression decreases prior to puberty and the function of the gene is predicted through its expression pattern. We sought to explore if there are other genes whose expression patterns resemble that of Mkrn3 in the mouse, and we identified Bcl11a as such a gene. To our knowledge, no protein-protein or genetic interactions have been identified between Mkrn3 and Bcl11a. Bcl11a is a transcriptional repressor essential for postnatal development and lymphopoiesis, and is known to be expressed in the hypothalamus (59,60). Its expression pattern raises the question of whether it too is an important inhibitor of the HPG axis. Relevant to pubertal timing, the over-expression of LIN28B (the gene near the most significant GWAS associated locus for pubertal timing in males and females) in erythroblasts down-regulates BCL11A (61). This additional information raises the possibility that LIN28B and BCL11A expression is mechanistically linked to pubertal timing.

Pubertal timing varies between sexes and there are sex-differences in the occurrence of pubertal disorders. In total, we observed that 7 out of 24 ‘Disease genes’ showed sex-biased expression at various ages, all of which occurred in the pituitary. Additionally, 53 out of 140 ‘GWAS-genes’ we profiled showed sex-biased expression. Although a high proportion of AAM loci are also associated with pubertal timing in males, some loci show larger effect size in one sex than in the other. There are six loci associated with discordant effects on pubertal timing in boys and girls and the candidate genes in these loci include OLFM2, RXRG, CSMD1, SIX6, SLIT3 and SIM1 (2,11,12). In our study, we examined the expression of the mouse orthologs of these candidate genes. We discovered that four out of the six genes (Olfm2, Rxrg, Csmd1 and Slit3) showed sex-biased expression in the mouse pituitary, pineal gland or liver at different ages. All of these sex-biased expression patterns were observed post-puberty (day 27). Our results support the use of mouse models to further explore these candidate GWAS genes, and suggest that fine mapping the genetic variants at these loci should give new insight into gene regulatory networks involved in sex-biased gene expression.

Using mouse as a model, our study also showed that the expression of puberty-associated genes is more sex-dimorphic in the pituitary gland, compared to the hypothalamus during postnatal development. The sex-biased gene expression was not only detected at individual time points, but also across pubertal maturation. This sex specificity is intriguing and could provide insight into sex-specific regulation of the HPG axis and into the differences in pubertal timing among males and females. The effect of estrogen on pituitary gene expression in mouse has been previously characterized by Kim et al. (62). Impg1, Kiss1, and Pgr were shown to be up-regulated by estrogen in the pituitary gland and thus represent examples of sex hormone-sensitive genes (62,63). Another group of interesting genes was Dlgap1, Csmd1, Sema3a, Bdnf, Cadps2 and Thrb. These genes were all expressed in pituitary at medium level and showed significant female-biased expression starting sometime between days 22 and 27 and continuing thereafter. Cadps2 encodes a protein CAPS2, which has been shown to modulate the release of BDNF from granule cells (64,65). It is also interesting that physical interactions between proteins encoded by Rxrg and Thrb have been shown (66). The biological significance of the sex-coordinated expression between these genes remains to be elucidated.

Additional interesting pituitary sex-biased gene is Retn, which encodes Resistin, an adipocytokine that is involved in insulin resistance (67). We observed the previously reported sex-specific developmental regulation of Retn in pituitary and also found that expression of Retn in the ovary peaks pre-puberty (68,69), potentially highlighting the importance of ovary-expressed adipocytokines.

One limitation of our study design is that, without the support of more functional data, causal genes for each GWAS loci might still be missing from our analysis. During the final revision of this manuscript an even larger scale GWAS analysis of ∼370,000 women identified 389 independent signals (70). The authors incorporated Hi-C data and suggested that a majority of these GWAS loci locate in topologically associating domains (TADs), with an average of 5 genes in the same TADs. Exploring expression patterns of these additional candidate genes using the framework we set up in this study would be an important step towards understanding the gene regulatory networks underlying puberty onset.

In conclusion, we report the first large-scale postnatal expression profiling of the mouse orthologs of known puberty-associated genes in the HPG axis, the pineal gland and the liver. We observed a high degree of sex-specific gene expression of puberty genes in the pituitary gland, setting the stage for future studies regarding the role that the pituitary plays in differences in the timing of puberty between males and females, including, for example, how these genes change in response to environmental stimuli such as sex hormones and diet. Overall, our study provides a foundation for further functional exploration of the biology of pubertal onset.

Materials and Methods

Selection of puberty-associated genes

We first assembled a list of genes referred to as ‘puberty-associated genes’ (Table 1, Supplementary Material, Table S1), and used this list for the human expression analysis:

1) ‘Disease-genes’ (n = 24): genes with known mutations detected in patients diagnosed with Kallmann syndrome (KS), normosmic hypogonatropic hypogonadism (HH) or precocious puberty; 2) ‘GWAS-genes’ (n = 145): candidate genes near genetic variants identified in large-scale GWAS studies for age at menarche in women (9,10) (Table 1, Supplementary Material, Table S1).

We then selected mouse orthologs of these human ‘puberty-associated genes’ and also included additional genes (‘Literature-based’, ‘Analysis’ and ‘Reference’ genes, see below) for expression profiling in mouse tissues. MCHR2 and C22orf34 were excluded from groups 1 and 2, as they do not have an established mouse ortholog. PTH, SKOR2 and TBX6 were excluded, because optimal primers could not be designed.

The additional genes we profiled fell into three categories: 3) ‘Literature-based’ genes (n = 7) which were selected based on reported relevance to puberty regulation. Lhx3, Otx1 and Otx2 are important for normal physiological functions of the pituitary (71–74); Eed, and Yy1 are known as epigenetic regulators of puberty (75); Lin28a is a functional homologue of the GWAS gene LIN28B and is known to result in a pubertal phenotype when over-expressed (42); Hnf4a is enriched in liver and was included as a control (28). 4) ‘Analysis genes’ (n = 8): we examined all protein-coding genes close to (<100kb) AAM-associated GWAS loci and identified 10 additional genes (not included in candidate genes described in Elks et al. and Perry et al. (9,10)) that show enhanced expression (expression level in tissue of interest larger than mean + 2SD of its expression level in all tissues) in human hypothalamus, pituitary or pineal gland. We excluded two genes (MVP and ERICH3) because they were only enhanced in one of the pineal samples (‘Pineal_day’ but not ‘Pineal_night’). 5) ‘Reference genes’ (n = 4): to identify appropriate reference genes for our qPCR analysis, the RefGenes tool by Genevestigator was used (76). 13 candidate reference genes were further validated in silico using the FANTOM5 mouse promoterome dataset (data not shown) (77). Two most stably expressed genes; Ddx18 and Adprh, together with Actb and Gapdh were selected as reference genes for the qPCR experiment. A complete list of genes used in both studies and their sources of selection can be found in Supplementary Material, Table S1.

Analysis of tissue specificity expression of puberty associated genes

Transcriptome data of human and mouse tissues were downloaded from BioGPS (http://biogps.org/downloads/) (13). Microarray probes were mapped to human and mouse genes respectively using both the BioMart database and probe annotation file provided by BioGPS. Data were log transformed and expression values for multiple probes that map to the same gene were averaged. Data from neoplastic tissues were excluded. Tissues named ‘Pineal_day’ refer to pineal glands collected from donors who died during day time while ‘Pineal_night’ refer to pineal glands collected from donors who died during night time. Next, for each gene in each tissue, a specificity score (Q) was calculated as described in Schug etal. (14). Specifically, for a gene g in a tissue t, its relative expression level is defined as P_g,t=W_g,t/Σ_1≤t≤nW_g,t, where W represents the actual expression level and n represents total number of tissues. The entropy of the gene expression distribution is defined as H_g=Σ_1≤t≤nP_g,tlog₂(P_g,t). To measure tissue specificity, a score Q is calculated: Q_g,t=H_g-log₂(P_g,t). The lower the Q score, the more specifically a particular gene is expressed in the corresponding tissue. We used a sum of the specificity scores, Q_sum,t= Σ_1≤t≤nQ_g,t of all the puberty-associated genes in a tissue to represent the level of specificity of puberty-associated gene expression in the particular tissue and to allow us to compare puberty-associated gene expression in one tissue versus another. To test the significance of the observed scores, we randomly sampled the same number of genes (n = 169) from the whole genome 1000 times and plotted the distribution of these Q_sum scores as background. We then calculated the Z score of Q_sum relative to the background distribution (Z= (Q_sum − μ)/σ where μ is the mean and σ is the standard deviation of Q_sum of randomly selected genes). An empirical P value indicating the significance of enrichment was calculated for each tissue and corrected for multiple testing. Genes marked with an ‘X’ in Figure 1B and Supplementary Material, Fig. S1 were selected based on: 1) Q value < 5% of all Q values; 2) expression value in tissue t > mean + 2SD of its expression in all tissues; 3) ranked as top 50% expressed in tissue t.

Animals. All studies and procedures were approved by the Toronto Centre for Phenogenomics (TCP) Animal Care Committee (AUP 09-08-0097) in accordance with recommendations of the Canadian Council on Animal Care, the requirements under Animals for Research Act, RSO 1980, and the TCP Committee Policies and Guidelines. Experimental mice were maintained under controlled conditions (25 °C, 10/14-h light/dark circle) at the TCP in sterile, individually ventilated cages, and were provided irradiated chow (standard chow from Harlan (Harlan, Teklad Global 18% Protein Rodent Diet, 2018, Madison, WI)) and sterile water ad libitum via an automated watering system. In all experiments, one female and one male C57BL/6J mouse were mated for one week, after which the male was transferred to an individual cage. The date of birth of pups was monitored daily from 9:00 am to 12:00 pm. Pups were weaned on postnatal day 21. Male and female pups were then housed separately (3–5 pups per cage).

Tissue collection. Six male and six female mice were sacrificed and dissected on postnatal days 12, 22, 27, 32 and 37 (total n = 30 males and n = 30 females). These postnatal ages were chosen based on data collected at our facility, corresponding to early development (day 12), pre-pubertal (day 22), pubertal (days 27 and 32), and post-pubertal (day 37) stages in both male and female mice as identified by physical manifestations of puberty in mice (VO and PS) (24). On the day of dissection, mice were weighed and, for days 22–37, phenotyped for puberty state by assessing VO and PS. In our cohort, all male mice on day 27 had undergone PS whereas half of the females had not yet exhibited VO. By day 32, all mice had entered puberty as evidenced by achieving PS or VO (see Supplementary Material, Table S6 for mouse breeding information).

Mice were euthanized using the automated Euthanex Carbon Dioxide (CO₂) lids followed by decapitation. All samples were collected between 9:30 am and 11:30 am. Hypothalamus, pituitary gland, testis, ovaries, liver and pineal gland tissues were dissected. The hypothalamus was dissected in block according to established coordinates and landmarks (78,79). Ovaries were transferred to a petri dish containing RNAlater and excess fat, fallopian tubes, and uterine tissue were removed under a microscope following initial macro-dissection. The pineal gland was micro-dissected from the skull under a microscope following removal of the skull from the mouse brain, flash frozen on dry ice and stored at −80 °C until RNA extraction. All other tissues were directly moved to RNAlater (containing 10% w/v sodium citrate tribasic dihydrate and 60% w/v ammonium sulphate following dissection, and stored at −20 °C until RNA extraction.

RNA extraction and cDNA synthesis. Total RNA was extracted from mouse tissues with the NucleoSpin® miRNA kit (Macherey Nagel) in combination with TRIzol lysis (Invitrogen) following manufacturers' protocols. Prior to RNA extraction, tissue samples were placed into bead mill tubes containing six 1.4 mm ceramic beads (MoBio Laboratories), and homogenized for 30 s at 6.5 m/s at 4 °C using Omni Bead Ruptor 24 bead mill. RNA quantity was determined using Nanodrop and the quality was assessed by Agilent 2100 Bioanalyzer. First strand cDNA was synthesized from 200 ng of RNA using SuperScript® VILO™ MasterMix (Invitrogen) and following manufacturer’s instructions.

Primer design and validation. Primers were designed using either the Universal ProbeLibrary Assay Design Center by Roche (based on Primer3 software), Primer-BLAST tool by the National Center for Biotechnology Information (NCBI) or the Realtime PCR Tool by Integrated DNA Technologies. Parameters were given to yield ideal product sizes of around 100bp. Primer pairs were then verified using the Primer-BLAST tool by NCBI (80) to ensure that the primer pair amplified all variants of the gene of interest and were also specific to only that gene. The insilico PCR tool by UCSC Genome Bioinformatics was also used to ensure primer pairs were amplifying a product that spanned at least one intron. Candidate primer pairs were first wet-tested using cDNA produced from pooled RNA isolated from mouse pituitary, pineal, hypothalamus, testes, ovaries and liver. For each gene, we designed 3 pairs of primers and selected the most specific one based on visual inspection of melting curves and Ct values. Three genes (Pth, Skor2, Tbx6) were excluded as no specific primers could be designed. A complete list of the 183 primer pairs used in the microfluidic qPCR experiments can be found in Supplementary Material, Table S1.

Microfluidic qPCR. Expression of the selected 183 genes was measured in total of 260 mouse cDNA samples (5 tissues, 5 ages, 2 sexes, 6 biological replicates for hypothalamus, pituitary and pineal gland, 4 biological replicates for gonads and liver) using 96.96 dynamic integrated fluidic circuit (IFC) arrays (Fluidigm, San Francisco, CA). First, specific target pre-amplification of each cDNA (12.5 ng) was performed using pooled primer assays (final concentration 50 nM for each) and PreAmp Mastermix (100–5580, Fluidigm, San Francisco, CA) according to manufacturer’s protocol. In parallel, mouse reference standard genomic DNA (2.5 ng) was pre-amplified, and ValidPrime™ assay (81) for mouse was used to correct for RT-qPCR signals derived from gDNA (TATAA Biocenter, Gothenburg, Sweden). A dilution series (50 ng/µl, 10 ng/µl, 2.5 ng/µl, 0.5 ng/µl) of a pool of selected cDNA samples from all tissues was included on each sample plate to examine primer efficiency. Next, the unincorporated primers were removed from the pre-amplified cDNA samples with Exonuclease I (NEB, Ipswich, MA) digestion (30 min at 37 °C, 15 min at 80 °C). Resulting samples were diluted 1:5 with 1× TE buffer (Sigma, St Louis, MO).

To load the dynamic array IFC, 5 µl of sample mixture was prepared for each sample containing 1× SsoFast EvaGreen Supermix with Low ROX (Bio-Rad, 172-5210), 1× DNA Binding Dye Sample Loading Reagent (Fluidigm, 100-3738) and each of the diluted pre-amplified cDNA. As well, 5 µl of primer assay was prepared using 5 µM of each primer assay (or 0.5 µM ValidPrime assay) and 1× Assay Loading Reagent (Fluidigm, 85000736). Next, 96 × 96 chip was primed in the IFC Controller HX, followed by loading the samples and primers into the microfluidic chip. Chips were run following either the GE 96 × 96 PCR + Melt v1 or GE 96 × 96 Standard v1 protocol in the Biomark using the Data Collection Software (Fluidigm).

Fluidigm BioMark qPCR analysis

Ct values were obtained with the Fluidigm Real-time PCR Analysis software. Ct values of assays marked as ‘Fail’ (Ct = 999) or beyond BioMark detection limit (Ct > 25) were set to NA. The impact of gDNA was determined and corrected using the ValidPrime method (81). Ct values of assays with gDNA contributing to >60% of their signals were set to NA (see Supplementary Material, Table S7 for raw Ct values). Dilution curves were used to estimate primer efficiencies on each plate.

The detected expression levels of conventional reference genes Actb and Gapdh were higher than 99% of all other genes while Adprh and Ddx18 were expressed at medium to high levels compared to other genes. Thus, for each assay, delta (Δ) Ct values were calculated using Adprh and Ddx18 as reference genes with: ΔCt (geneX) = Ct (mean of Adprh and Ddx18) – Ct (gene X). ΔCt values were used in further data analysis (see Supplementary Material, Table S3 for ΔCt values). We classified gene expression levels into five categories based on their ΔCt values: not detected; low (ΔCt < −10); weak (−10 ≤ ΔCt < −5); medium (−5 ≤ ΔCt < 0) and strong (ΔCt >0). Six samples were replicated twice: Hypothalamus day 12 female sample No.6 and Testis day 32 sample No.2 were repeated due to sample loading failure and the failed reactions were discarded. Hypothalamus day 12 male sample No.4, Pineal day 32 female sample No.4, Pineal day 27 male sample No.3 and Pineal day 37 male sample No.5 were repeated due to low correlation with other biological replicates and the mean Ct values of primers for the replicated samples was used. Quality of biological replicates was assessed by calculating the average Pearson correlation coefficient (PCC) between a certain sample and all of its other biological replicates. Primer efficiencies were first evaluated using dilution curves on sample plate1. Primers with an efficiency < 0.7 or > 1.3 were tested again in a separate regular qPCR experiment using dilution series of hypothalamus or pituitary cDNA. These primers all showed efficiencies 0.8 - 1.2 in regular qPCR.

Statistical analysis of qPCR data

All data analysis and plotting were performed in R (version 3.2.3). All plots were generated using R package ‘ggplot2’ (82).

Visualization of sample clustering using t-SNE

For the t-SNE analysis, liver samples and pituitary day 27 sample No. 3 were excluded from the analysis. Only genes detected in all samples were used. R package ‘Rtsne’ was used to perform the t-SNE analysis with perplexity set to 30 and maximum iterations to 5000 (31,83).

Identify tissue-enhanced genes in qPCR experiments

For each gene, if its overall expression in one tissue (considering all ages, sexes and replicates) is significantly higher than all other tissues used in the analysis (Wilcoxon rank sum test, adjusted P value < 0.05; log2 fold change (ΔΔCt of median value) > 2), it will be defined as enhanced for that tissue.

Functional enrichment analysis with genes

All functional enrichment analysis in the paper were performed using g:Profiler (accessed Jan. 2017) (84) with default settings. The background gene sets were chosen as described in the text.

Clustering of temporal patterns

For each combination of gene, tissue, age and sex, a median of the Ct values of replicates was calculated and used to represent expression for that gene in that condition during subsequent analyses. For each of these combinations, a corresponding temporal pattern (TS_T,S,X, temporal pattern of gene X in sex S, tissue T) was generated by deducting the median Ct value of day 12 samples from the median Ct values of subsequent time points separately for male and female samples. The variation of each temporal pattern (TS_T,S,X) was assessed by its SD (TS_T,S,X)) across ages. Temporal patterns with the lower quartile of variations (SD < 1.4) were excluded from clustering. Clustering of temporal patterns was performed using the Partitioning Around Medoids (PAM) algorithm (n = 4, determined based on the silhouette information, R package ‘cluster’) (85) with distance measurement using CORT coefficient (k = 4, R package ‘TSclust’) (86). Patterns from all tissues and both sexes were clustered together then plotted separately.

Differential gene expression analysis between ages

To identify genes showing significant temporal expression changes and to examine expression differences between ages, each gene was analyzed in each tissue for male and female separately at each age. One-way analysis of variance (ANOVA) tests with age as the factor were applied. Comparisons for which data of more than 2 ages were missing were excluded. For genes showing a statistically significant main effect of age (P values corrected for number of genes tested adjusted P value < 0.05), pair-wise comparisons between time points were further investigated using Tukey’s honest significant difference (HSD) post hoc test. P values of comparisons between neighbouring ages were further corrected for the number of genes tested and were used to identify significant temporal changes (adjusted P value < 0.1). All P values reported in this work are corrected for multiple testing unless specified (with p.adjust(method=‘BH’) in R).

Differential gene expression analysis between male and female samples

A student’s t-test was used to assess the sex differences for each gene in each tissue at each age. We chose to use t tests based on the assumption that the expression of one gene is normally distributed on the log scale within the population. Only comparisons with ≥ 3 replicates for both groups were performed. P values were corrected for the number of genes tested within each tissue separately. A significance cut-off of adjusted P value < 0.05 was used.

Supplementary Material

Supplementary Material is available at HMG online.