All human brains were obtained from the Brain Donor Program at the University of California—Irvine, and the studies were approved by both the University of California—Irvine and the University of Michigan Institutional Review Boards. Written informed consent was obtained from the next-of-kin of the deceased. Information regarding diagnosis, treatment and other clinically relevant variables were obtained from medical records, the coroner's investigation and family interviews. Table 1 shows a list of subject demographics for controls and individuals with MDD used in this study. Six subjects per group were used for microarray analyses, and an additional four subjects per group were added for the qRT-PCR validation study. No significant differences in age, gender, post-mortem interval or brain pH were detected between groups. All subjects also had an agonal factor score (Gustafsson et al., 2003) of 0. Previous studies found that AFS and brain pH are the strongest factors influencing gene expression (Li et al., 2004). Therefore, all brains used in this study had a pH greater than 6.6. The brains were removed at the time of autopsy, cooled to 4°C and then sliced into 0.75 cm coronal slabs which were then immediately stored at −80°C to preserve integrity (Jones et al., 1992). The slabs were then manually dissected while kept frozen at −80°C by trained staff into various brain region blocks, including the hippocampus which was used for this study.

Previously dissected blocks from the post-mortem human hippocampus were sectioned (−20°C) at 10 μm and stored at −80°C. One section was placed onto a SuperFrostPlus slide (Thermo FisherScientific, Waltham, MA) and stored at −80°C until processing. Every 50th section was selected for mRNA in situ hybridization throughout the hippocampus, as previously described (Lopez-Figueroa et al., 2004). Briefly, slides were fixed in 4% paraformaldehyde, rinsed in 2X SSC, incubated in an acetic anhydride/triethanolamine solution and dehydrated. Sections were then hybridized with ³⁵S-UTP and ³⁵S-CTP cRNA probes overnight at 55°C. The next day, slides were washed, rinsed in increasing stringency of SSC, dehydrated and exposed to Kodak Biomax MR film (Eastman Kodak, Rochester, NY). Exposure time was empirically determined using test slides to visualize the region of interest. The cRNA probes were synthesized from human cDNA cloned in-house. The GAD67 (NM_000817, 738-964) cRNA probe was used to determine the anatomical level of the human hippocampus, and the transthyretin (TTR; NM_000371, 477-701) cRNA probe was used to visualize the CP. In situ hybridization autoradiograms were then digitized and scanned using the ScanMaker 1000XL Pro Flatbed Scanner (Microtek, Carson, CA) and SilverFast Ai Imaging software (LaserSoft Imaging, Sarasota, FL).

The hippocampus was identified using GAD67 mRNA in situ hybridization, see Figure 1. The hippocampus was identified based on the C-like shape of the dentate gyrus. Transthyretin has been shown from mouse studies to be highly abundant in the CP and can be used as a reliable mRNA marker of this tissue (Marques et al., 2011). Sections containing the CP were processed through the following dehydration protocol: room temperature for 30 s, 75% ethanol for 30 s, dH₂0 for 30 s, 75% alcohol for 30 s, two 95% alcohol washes for 30 s, 100% alcohol for 30 s, two xylene washes for 5 min each, and left to air dry for 20 min. Laser capture microscopy of the CP was performed at the level of the caudal hippocampus (levels 4–6, based on Amaral and Insuasti, 1990). Two field-of-views of the CP from one section was captured using the AutoPix LCM system (Arcturus, Mountain View, CA) onto CapSure Macro LCM caps (Arcturus, Mountain View, CA). Laser settings were set to 70 mW and 1.5 ms. Figure 2 shows a low magnification image (4X) of the CP before and after capture (area circled in red). The large blood vessels were not captured resulting in a sample of primarily epithelial cells.

Total RNA was extracted from adherent cells (on LCM caps) using the PicoPure RNA isolation kit (Arcturus, Mountain View, CA). Cell extracts were stored at −80°C between RNA extraction and final isolation. An optional DNAse treatment was used in the isolation process. RNA was eluted in a final volume of 11 μl. Quality and quantity of cRNA was determined on the Agilent 2100 Bioanalyzer using Picochip kits (Agilent, Palo Alto, CA). The quality of 18S and 28S peaks were determined as previously described (Bernard et al., 2011). For an average area of 2.7 mm², we obtained an average concentration of 1 ng/μl and RNA quality was determined as previously described (Kerman et al., 2012). This process yielded an average of 11 ng of total RNA. The total isolated RNA was then amplified by the RiboAmp Plus 1.5-round RNA Amplification kit (Molecular Devices, Sunnyvale, CA). The final biotin-labeled cRNA was generated with the Bioarray High Yield RNA transcription kit (ENZO life sciences, Plymouth Meeting, PA). The quantity was determined on the Nanodrop ND-1000 spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA). After amplification, the total biotin-labeled cRNA was an average of 5 μg per sample.

Equal amounts (750 ng) of amplified total RNA sample from each human sample was hybridized to HumanHT-12v4.0 Expression BeadChips and scanned on the BeadStation system (Illumina Inc., San Diego, CA) following the manufacturer's instructions (Turner et al., 2011). Each Beadchip provides coverage of more than 47,000 probes. The microarray data was quantile normalized in BeadStudio using Illumina's error model (Illumina Inc., San Diego, CA). Fold-changes and p-values were calculated from the average array signal. Data were also analyzed by Ingenuity Pathways Analysis Software v8.0 (Ingenuity©Systems, Redwood City, CA). The Functional Analysis tool identified biological functions and/or diseases that were most significant to the data set. Molecules from the dataset that had a fold-change cutoff of 1.1, a p-value less than 0.05 and were associated with biological functions and/or diseases in Ingenuity's® Knowledge Base were considered for the analysis. To follow-up on the top pathway identified by Ingenuity Pathway Analysis, we evaluated gene ontology using Cytoscape v2.8.2 with a BinGO plug-in.

Table 2 shows a list of primers used for qRT-PCR validation. Amplified cRNA (1 μ g) was reverse transcribed using the iScript cDNA synthesis kit (Bio-Rad Laboratories, Hercules, CA) in a total reaction volume of 20 μ l. cDNA (1 μl) was used as the template for real-time PCR assays with a MyiQ real-time PCR system (Bio-Rad Laboratories). The quantitative PCR was conducted in duplicate using iQ SYBR Green Supermix, according to the manufacturer's instructions (Bio-Rad Laboratories). Relative expression of the gene of interest was normalized to β-actin expression in each sample. It is important to note that β-actin did not differ between controls and individuals with MDD. The expression level of the gene of interest was evaluated using the 2^−(ΔΔCt) method and values for each gene were expressed as fold-changes (Livak and Schmittgen, 2001). The PCR product quality was monitored using post-PCR melt-curve analysis at the end of the amplification cycles.

For qRT-PCR, a Student's t-test was performed using SPSS (IBM, Armonk, NY). For pathway analysis, a right-tailed Fisher's exact test was used to calculate a p-value determining the probability that each biological function and/or disease assigned to that data set is due to chance alone. For gene ontology, Benjamini & Hochberg false discovery rate (FDR) correction was applied to the dataset.

The most highly expressed known transcript in the human post-mortem CP was transthyretin (TTR). The primary role of TTR is to transport thyroxine (T4) and retinol in the brain (Fleming et al., 2009). However, TTR also has proteolytic activity (e.g., neuropeptide Y and β-amyloid). TTR protein levels had previously been associated with depression, such that low levels of TTR have been correlated with high suicidal ideation and low 5-HIAA (Sullivan et al., 2006). A previous study also found TTR to be downregulated in the CSF of individuals with MDD (Ditzen et al., 2011). However, TTR gene expression was not altered in the CP in MDD subjects in our study.

Supplementary Table 1 lists the 11,506 transcripts that were significantly detected across all samples in the control CP. Notably, half of the 14 mostly highly expressed transcripts were transcripts that encode ribosomal proteins. However, none of these transcripts were altered in subjects with MDD.

In general, there were 169 transcripts differentially expressed between MDD subjects and controls, see Supplementary Table 2 for the complete list. The majority of these transcripts (75%) were downregulated with only a small percentage of transcripts upregulated (25%). The fact that the majority of the transcripts were downregulated is similar to what we have observed in the hippocampus in subjects with MDD (unpublished observations).

When Ingenuity pathway analysis was performed on the dataset, the top five significant functions were connective tissue disorder, dermatological diseases, developmental disorder, genetic disorder, and metabolic disease. The top two significant pathways were hepatic fibrosis and calcium signaling. The top network focused on transforming growth factor-beta (TGFβ). As shown in Figure 3, all of the transcripts associated with the TGFβ network were downregulated in MDD subjects compared to controls. Gene ontology analysis confirmed that growth factor binding (corrected p-value: 0.00885) was the most significant function in this dataset. Indeed, seven members of the TGFβ network validated by this analysis. TGFβ signaling is known to play a role in cytoskeletal dynamics and actin reorganization (Baghdassarian et al., 1993; Gagelin et al., 1995). Since TGFβ is associated with various components of the extracellular matrix, we decided to interrogate transcripts with high fold-changes related to structural support and integrity, some of which are part of the TGFβ pathway.

We selected eight transcripts for real-time PCR validation, see Table 3. Since we can predict the direction of change based on the microarray results, we performed one-tail t-tests for qRT-PCR. Three transcripts were confirmed by qRT-PCR and two transcripts exhibited non-significant trends in the same direction as the microarray results. We will describe these five transcripts in the next few paragraphs. Fibulin 2 (FBLN2) was significantly decreased in subjects with MDD compared to controls. FBLN2 is a secreted glycoprotein that is produced by epithelial cells and can bind calcium. In terms of function, FBLN2 interacts with many other proteins, including the laminins and integrins, to stabilize the extracellular matrix (Zhang et al., 1996; de Vega et al., 2009). It is possible that downregulation of this transcript may lead to destabilization of the extracellular matrix.

A myosin heavy chain transcript and a calcium-activated potassium channel were also altered in individuals with MDD. Myosin heavy chain 11 (MYH11) was significantly decreased in MDD subjects compared to controls. This myosin motor can bind actin and play a role in ATP hydrolysis (Renard et al., 2011; Armstrong et al., 2012). Thus, the actin cytoskeleton might be altered in subjects with MDD. The calcium-activated and voltage-dependent potassium channel β1 subunit (KCNMβ1) was also significantly decreased in individuals with MDD. Although not much is known about the role of this large conductance channel in the brain, it has been linked to cardiovascular disease and hypertension in the periphery (Grimm et al., 2009). However, it is plausible that this channel also may play a role in filtration, as total gene expression in the kidney is similar to that of the CP suggesting that the two tissues may share similar functions (Sathyanesan et al., 2012). This particular subunit of the channel is also the regulatory subunit of the channel and may alter surface expression of the receptor (Toro et al., 2006).

There were also non-significant trends for mitogen-activated protein kinase 1 (MAPK1) and the purinergic receptor, P2YR5, to be altered in MDD subjects. MAPK is part of the TGFβ network (see Figure 2) and is known to play a role in tyrosine kinase receptor signaling. MAPK is also known to bind to and phosphorylate cytoskeletal proteins (Veeranna et al., 2000). Moreover, P2YR5 (aka LPAR6) plays a role in endothelial cell morphology and may be involved in actin reorganization, as it is a receptor for lysophosphatidic acid (Ishii et al., 2009). It is also possible that this receptor may regulate vascular permeability in the CP. In summary, all five of the transcripts from the qRT-PCR experiments were downregulated in subjects with MDD. Taken together, these results suggest that there may be an altered extracellular matrix or cytoskeleton in MDD subjects.