High Throughput Gene Expression Analysis Identifies Reliable Expression Markers of Human Corneal Endothelial Cells

Considerable interest has been generated for the development of suitable corneal endothelial graft alternatives through cell-tissue engineering, which can potentially alleviate the shortage of corneal transplant material. The advent of less invasive suture-less key-hole surgery options such as Descemet’s Stripping Endothelial Keratoplasty (DSEK) and Descemet’s Membrane Endothelial Keratoplasty (DMEK), which involve transplantation of solely the endothelial layer instead of full thickness cornea, provide further impetus for the development of alternative endothelial grafts for clinical applications. A major challenge for this endeavor is the lack of specific markers for this cell type. To identify genes that reliably mark corneal endothelial cells (CECs) in vivo and in vitro, we performed RNA-sequencing on freshly isolated human CECs (from both young and old donors), CEC cultures, and corneal stroma. Gene expression of these corneal cell types was also compared to that of other human tissue types. Based on high throughput comparative gene expression analysis, we identified a panel of markers that are: i) highly expressed in CECs from both young donors and old donors; ii) expressed in CECs in vivo and in vitro; and iii) not expressed in corneal stroma keratocytes and the activated corneal stroma fibroblasts. These were SLC4A11, COL8A2 and CYYR1. The use of this panel of genes in combination reliably ascertains the identity of the CEC cell type.


Introduction
The human corneal endothelium (CE), comprised of a monolayer of hexagonal corneal endothelial cells (CECs) attached to a basement Descemet's membrane (DM) composed of collagen, forms a selective barrier between the anterior chamber of the eye and the corneal stroma. CECs are the most metabolically active cells in the cornea expressing fluid pumps that actively move fluid from the stroma back into the anterior chamber of the eye. The dynamic balance between the ''leaky'' barrier and constant pump activity maintains corneal deturgescence, thereby keeping the cornea transparent. Corneal blindness is often due to endothelial dysfunction and is the second leading cause of visual impairment [1].
Human CECs have limited proliferative capacity within the eye [2,3]. Hence, in order to replace dead or damaged CECs, existing cells spread out to maintain the functional integrity of the CE [4,5]. In cases of severe cell loss due to genetic corneal endothelial dystrophy or trauma, decompensation of the CE may occur, reducing its capacity to pump fluid out of the stroma, resulting in stromal and epithelial edema, loss of corneal clarity and visual acuity, and eventually bullous keratopathy [1]. The conventional solution to restore vision is through cornea transplantation. However, there is a global shortage of transplant-grade donor cornea. This shortage is expected to worsen as the demand for corneal transplantation increases with an aging global population. Therefore, considerable interest has been generated for the development of suitable endothelial grafts through tissue engineering, which can potentially alleviate the shortage of corneal transplant material.
With the introduction of surgical procedures such as Descemet's Stripping Endothelial Keratoplasty (DSEK) and Descemet's Membrane Endothelial Keratoplasty (DMEK), less invasive, keyhole surgery options for the selective replacement of the CE layer are now possible [6,7,8]. The advent of such surgical procedures provides further impetus for the development of alternative endothelial grafts -either by means of ex vivo expansion of human CECs, or by de novo generation of CECs from pluripotent human embryonic stem cells, induced pluripotent stem cells or multipotent adult stem cells, which are unlimited sources of cells. However, a major obstacle to such endeavors is the lack of specific markers for CECs, resulting in an inability to definitively identify such putative stem cell-derived CECs.
Currently, the most commonly used markers in the characterization of cultivated CECs include ZO-1 [9,10], a tight junction protein involved in signal transduction at cell-cell junctions, and Na + /K + -ATPase [11], an essential enzyme involved in the active transport of ions across the CE. Although the co-expression of both proteins indicates the presence of key components of CE fluid transport function, it is not a definite indication of the identity of CECs because both ZO-1 and Na + /K + -ATPase are ubiquitously expressed in many other cell types [12,13,14,15]. This paper presents a thorough gene expression analysis of CECs and proposes a panel of markers that reliably identifies CECs in vivo and in vitro. We used RNA-sequencing (RNA-seq) [16] and High Throughput Quantitative Polymerase Chain Reaction (HT-PCR), together with various Bioinformatics platforms such as BioGPS (http://biogps.org), DAVID Functional Annotation Clustering Tool [17] and PANTHER Classification System [18], to select a panel of markers based on the following criteria: 1) Highly expressed in CECs from young and old donors: 2) Expressed in CECs in vivo, as well as in ex vivo CEC cultures; 3) Ideally not expressed in other cell types; 4) Not expressed in corneal stroma keratocytes or activated corneal stroma fibroblasts.

Results
To identify markers for CECs, global gene expression analysis of CECs stripped from donor cornea along with the Descemet's membrane (CEC-DM) was carried out using RNA-seq [19]. Gene lists generated were analyzed using DAVID Functional Annotation Clustering Tool [17] and PANTHER Classification System [18] to identify over-represented ontology groups and molecular pathways.

Identification of Genes most Highly Expressed in CECs
Gene expressions of the following samples were analyzed using RNA-sequencing: 1) CEC-DM pooled from 5 young donors; 2) CEC-DM pooled from 5 old donors; 3) CEC cultures and 4) corneal stroma pooled from 5 young donors ( Figure 1). Description of the isolation process can be found in Materials and Methods. Full dataset for the RNA-sequencing performed is found in Table S1.
The sequence depth ranged from 1.8 million to 4.6 million reads ( Figure 2A). Hierarchical clustering showed that the young and old CEC-DMs cluster closest to each other, followed by CEC culture, and lastly the corneal stroma ( Figure 2B). This indicates that the two CEC-DM samples are closer in gene expression than they are to the CEC culture, and that the corneal stroma sample has the greatest gene expression difference compared to the other 3 samples.
We first sought to identify the genes most highly expressed in the young CECs. The top 20 genes expressed in CECs include those that play a role in cellular metabolism (ENO1, GAPDH, CA3, LDHA, ALDOA, ATP5B, ATP5A1), and genes important for trans-membrane transport (SLC2A1, ATP5B, ATP1A1, ATP5A1) ( Table 1). Some of these genes have also been previously identified to be highly expressed in CECs. CA3, PTGDS, LDHA, MGP, and C4orf49 have been identified by Sakai et al. [20], while ENO1, GAPDH, and PTGDS have been noted by Gottsh et al. [21] to be among the top 50 most highly expressed genes in CECs.
Furthermore, analysis using the DAVID Functional Annotation Clustering Tool showed that the most over-represented ontological groups for the top 200 genes in young CECs are cellular metabolism, regulation of cell death, and membrane transport ( Table 2). Taken together, these results describe a cell type that is metabolically active, and have a function in ion and water transport, corroborating with previous descriptions of CECs [1,22].

Analysis of Genes that have been Previously used as CEC Markers
We next examined the expression levels of commonly used CEC markers in our samples. Table 3 shows that most of the markers previously used by other groups are either not expressed in culture (reads per million (RPM) ,10), do not differentiate between CECs and stroma (fold change of young/stroma ,5), or lack specificity as catalogued in BioGPS. Of the genes listed, COL8A2 and SLC4A11 emerged from our study as the best candidates as markers for CECs, but these too are not ideal. Expression of COL8A2 is high in CECs and low in stroma, but is also present in the retina. SLC4A11 is highly expressed in CECs, not ubiquitously expressed in many cell types (data not shown), but is 6 times lower in expression in CEC cultures than in CEC-DM (Table 3). Therefore, we sought to probe for other candidates that could be used to identify CECs.

Identification of Genes Important for CEC Function and Physiology
To identify genes important for CEC physiology, genes that are highly expressed in CEC-DM and CEC cultures were selected and analyzed. The list was filtered for genes with RPM.10 in young CEC-DM, old CEC-DM and CEC culture, at least two times higher in young CECs than in corneal stroma, and less than 2-fold difference in expression between young and old CEC. The top 50 genes ranked in descending order of expression in young CEC-DM are shown in Table 4. Each gene was analyzed in BioGPS to obtain information on the expression of each gene across different tissues (data not shown). Genes that are not ubiquitously expressed were further analyzed using high-throughput QPCR to compare the expression of these selected genes in various human tissue types.
QPCR results show that ATP5B is expressed in all the tissue types tested. SLC2A1, ENO1, and COL8A2 are highly expressed in CEC cultures and CEC-DM but are also expressed in many other tissue types, albeit at lower levels. MYOC and SLC4A11 on the other hand are highly expressed in CEC cultures and CEC-DM but not in other tissue types ( Figure 3).
Interestingly, the expression of CA3, PTGDS, IER3 and SFRP1 were high in both young and old CEC-DM, but were lost in CEC cultures, indicating that the expression of these genes are not necessary for CEC survival in vitro. While they could be useful for identifying CECs in vivo, they should not be used as CEC markers in experiments involving CEC cultures. On the other hand, MGP is expressed much higher in CEC cultures than in CEC-DM and other tissue types ( Figure 4).
Taken together, the results above suggest that SLC4A11 and COL8A2 are suitable markers for CECs based on their high expression in CEC-DM and CEC-cultures compared to other cell types. MYOC is not an ideal choice because it is expressed in the trabecular meshwork where it participates in the obstruction of fluid outflow.

Identifying Genes that are Expressed in CECs but not in Corneal Stroma
There was a trace amount of SLC4A11 and COL8A2 in corneal stroma keratocytes and/or its activated stromal fibroblasts ( Figure 3). Therefore, we sought to identify other genes that were not expressed or lowly expressed in corneal stroma. We selected genes that have RPM value of less than 1 in corneal stroma but more than 10 in the other 3 samples (Table 5). Each gene was analyzed in BioGPS (data not shown). Genes that are not ubiquitously expressed were chosen for further analyses. High throughput QPCR was first performed using Fluidigm Biomark system to compare expression of these genes across different human tissue types. Results show that of all the genes analyzed, CYYR1 was the only gene that was expressed in CEC-DM and CEC cultures, but not expressed in corneal stroma and stroma fibroblast cultures ( Figure 5).
Confirming the Stability of Expression of SLC4A11, COL8A2 and CYYR1 across Multiple Biological Samples QPCR analysis was performed using a 7500 Fast Real Time PCR machine to compare the expression of SLC4A11, COL8A2, and CYYR1 in CECs and corneal stroma from 3 individual donors. Expression of CYYR1 was similar in all three donors (average 6 standard deviation: 495654), while that of SLC4A11 and COL8A2 varied slightly (average 6 standard deviation: 3186181 and 126657 respectively). On average, expression of SLC4A11 was 84-fold higher in CECs than in stroma (p,0.05), while expression of COL8A2 and CYYR1 was 16-fold higher in CECs than in corneal stroma (p,0.05 and p,0.01 respectively). Expression of all three genes was negligible in stromal fibroblast cultures obtained from 2 individual donors ( Figure 6). Taken together, the combination of SLC4A11, COL8A2 and CYYR1 present a useful panel of genes to differentiate between CECs and corneal stroma/stroma fibroblast.

Gaining Insight into CEC Physiology from Gene Expression
The transparency of the cornea is predominantly attributed to the ion and fluid transport capacity of CECs. We were therefore interested to examine the expression of transporters in CECs (Table S2). Genes that are classified as transporters under the PANTHER classification system were extracted. The gene list was filtered for RPM.10 in young CEC-DM, old CEC-DM and CEC culture, at least two times higher in young CECs than in corneal stroma, and less than 2-fold difference in expression between young and old CEC. Table 6 shows the list of transporters that fit these criteria, ranked in descending order of expression in young CECs.
The high throughput gene expression analyses performed can also be used to further understand the signaling pathways that govern CEC physiology. To achieve this aim, we looked at the expression of cytokine receptors (Table S3) and transcription factors (Table S4) in CECs. Genes that are classified as cytokine receptors or transcription factors under the PANTHER classification system were extracted. The gene lists were filtered for RPM.10 in young CEC-DM, old CEC-DM and CEC culture, at least two times higher in young CECs than in corneal stroma, and less than 2-fold difference in expression between young and old CEC. Only 3 cytokine receptors pass the selection criteria (Table 7). Table 8 shows a similar list of transcription factors expressed in CECs.
We next performed pathway analysis of the transcription factors using DAVID Functional Annotation. The signaling pathways represented by the expressed transcription factors include TGFb and Wnt signaling pathways ( Table 9). The expression of components of these signaling pathways were further examined (Tables S5 and S6). Interestingly, we found that CECs do not express all SMADs (intracellular mediators of TGFb signaling). The most highly expressed SMADs (SMAD2, SMAD3 and SMAD4) in CECs are those involved in the Activin/Nodal branch of TGFb signaling (Table 10). Such information provides further clues for future work to delineate the signaling pathways controlling CEC physiology and gene expression.

Expression Markers for CECs
The challenges and important Clinical Quality Assurance (CQA) points that should be taken into consideration for CEC therapy include identity, potency, purity, and safety. To date, no one specific gene has been identified as a marker of CECs. While generic markers such as ZO-1 and Na + /K + -ATPase indicate the presence of functional fluid transport, they do not definitively identify CECs. Indeed, co-expression of ZO-1 and Na + /K + -ATPase has been observed in mammalian blastocyst [23,24], and in various adult tissues including intestinal epithelium [25] and retinal pigmented epithelium [26]. The use of our proposed panel of genes (SLC4A11, COL8A2 and CYYR1) together ascertain the identity of the CEC cell type, thereby fulfilling the first CQA point that renders the cell suitable for clinical use.
Both SLC4A11 and COL8A2 are known to play important functions in the cornea. SLC4A11 is a multi-pass membrane protein important for sodium-mediated fluid transport. It prevents morphological abnormalities of the cornea caused by increased sodium chloride concentrations in the stroma. Mutations in SLC4A11 have been known to cause Harboyan Syndrome (congenital corneal endothelial dystrophy with progressive perceptive deafness) [27], Corneal Endothelial Dystrophy Type 2 (bilateral corneal dystrophy characterized by corneal opacification and nystagmus) [28], and Fuchs Endothelial Dystrophy (ocular disorder characterized by focal wart-like guttata that arise from DM and develop in the central cornea, epithelial blisters, reduced vision and pain) [29].    COL8A2 encodes the alpha 2 chain of type VIII collagen. Type VIII collagen is a major component of the DM [30]. Mutations in COL8A2 have been shown to cause Fuchs Endothelial Corneal Dystrophy and Posterior Polymorphous Corneal Dystrophy Type 2 [31,32,33,34,35,36,37].
Little is known about CYYR1, also called cysteine and tyrosinerich protein 1. It is a recently identified gene located on human chromosome 21 whose product has no similarity to any known protein and is of unknown function [38]. Analysis of expressed sequence tags revealed high human CYYR1 expression in cells belonging to the diffuse neuroendocrine system, which may be the origin of neuroendocrine tumors [39].

Differentiating between Corneal Endothelial Cells and Corneal Stroma Cells
The inclusion of CYYR1 as part of our panel to identify CECs is crucial, because it helps differentiate CECs from corneal stroma keratocytes and the serum-activated keratocytes known as corneal stroma fibroblasts. This is especially important for researchers studying CEC explants and CEC cultures as slow-growing CECs can be contaminated with, and over-taken by the highly proliferative corneal stroma fibroblasts that may be co-isolated during the isolation of CEC for culture [40]. In addition, CECs are the closest to corneal stroma keratocytes in terms of embryonic origin. During development, both cell types are derived from cranial neural crest cells, which migrate and settle in the head mesenchyme. In chick and human, the first wave of migration of mesenchymal cells forms the endothelial layer on the inner surface of the cornea, adjacent to the lens. Shortly thereafter, a second wave of mesenchymal cells invades the primary stroma where the mesenchymal cells differentiate into keratocytes [41,42]. Therefore, a reliable panel of genes to identify de novo CECs generated from pluripotent stem cells must have considerably lower expression in stromal keratocytes/fibroblasts compared to CECs.

Signaling Pathways Controlling CECs Physiology
The high throughput gene expression analysis performed is not only useful for identifying markers for CECs, but also provides insight into the signaling pathways controlling CEC physiology. Examination of the cytokine receptors and transcription factors expressed revealed that TGFb (in particular the Activin/Nodal branch) and Wnt signaling pathways are active in CECs.
Indeed, exogenous TGFb in the aqueous humor has been shown to suppress S-phase entry in CEC cultures, suggesting that TGFb could play a role in the inhibition of proliferation of CECs [43]. In the developing eye, TGFb from the lens has been shown to control the development of the neural crest-derived eye structures (corneal stroma, corneal endothelium, anterior iris and trabecular meshwork) [44]. The migration of periocular mesenchyme into the region of the prospective corneal stroma and endothelium is attributed to a response to TGFb signals from the lens. Furthermore, TGFb2 released from the lens is required for the expression of transcription factors PITX2 and FOXC1 [44], whose expressions in periocular mesenchyme are necessary for the proper formation of the anterior segment of the eye [45].
Wnt signalling has also been implicated in the development of CECs. Canonical Wnt signalling is required for maintenance of PITX2 expression in ocular neural crest cells during cornea development [46]. Interestingly, the Wnt antagonist DKK2 is an essential downstream target of the PITX2 homeodomain transcription factor in neural crest during eye development, to provide a mechanism to locally suppress canonical Wnt signaling activity during eye development [47].
Therefore, manipulation of TGFb and Wnt signaling pathways in CECs could help improve conditions for ex vivo culture of CECs, and direct differentiation of pluripotent stem cells or neural crest cells into CECs.
In conclusion, the high throughput expression analysis of CECs presented in this paper revealed a panel of genes (SLC4A11, COL8A2 and CYYR1) that could be used to identify CECs and proposed pathways (TGFb and Wnt) that could be explored for expansion of CEC in vitro or generation of CECs de novo.

Research-grade Human Corneoscleral Tissues
Research-grade corneoscleral tissues from cadaver human donors considered unsuitable for transplantation were procured from Lions Eye Institute for Transplant and Research Inc. (Tampa, FL, USA). Overall general health of the donor before death was also considered which included previous history or medical treatment that might damage or affect the growth of the corneal endothelium. Research corneas were preserved and transported in Optisol-GS at 4uC, and were used within 13 days from preservation. The ages of donors ranged from 19-76 years (Table 11).

Isolation of Human Corneal Endothelial Cells
Research corneas were incubated in three washes of antibiotic/ antimycotic solution in PBS, 15 minutes each. Corneoscleral rims were placed endothelial-side-up on a disposable cornea vacuum punch (Ripon, England), and mildly stabilized by the vacuum suction created. A brief 30 seconds treatment with Trypan Blue solution (0.1%) was used to delineate Schwalbe's line. The CEC-DM layer was carefully stripped off, approximately 1 mm anterior to the Schwalbe's line (away from the trabecular meshwork) from the posterior stroma under the dissecting microscope (Nikon, Japan).

Culture of Human Corneal Endothelial Cells
Primary cultures of CECs were established as described in [48], and propagated with some modifications. Briefly, the CEC-DM layer was digested enzymatically in collagenase A (2 mg/ml) for at least 2 hours and up to 6 hours. This allowed full detachment of the CECs from the DM, which tended to conglomerate into tightly-packed CEC clusters. The CEC clusters were rinsed once in PBS and further dissociated in TrypLE Express (TE) for 5 minutes. Cell pellets collected after a mild centrifugation (800 g for 5 minutes) were plated in culture-ware coated with FNC coating mixture. Isolated cells were left to adhere overnight in a stabilization medium made up of Human Endothelial-SFM supplemented with 5% FBS and 1x anti-biotic/anti-mycotic. Adhered hCECs were then cultured in F99 medium containing Ham's F12 and M199, mixed in a 1:1 ratio, supplemented with 5% FBS, 20 mg/ml ascorbic acid, 1x Insulin-Transferrin-Selenium, 1x anti-biotic/anti-mycotic and 10 ng/ml bFGF. When the cultured cells reached 80-90% confluence, they were re-exposed to the stabilization medium for at least one week before passaging [49]. Confluent hCECs were passaged using TE, and seeded onto FNC-coated culture ware at a plating density of approximately 10,000 cells per cm 2 for subsequent passage. All incubation and cultivation of hCECs were carried out in a humidified incubator at 37uC containing 5% CO 2 unless otherwise stated. Fresh media were replenished every two days.

RNA Extraction
For direct RNA extraction of donor tissues, isolated CEC-DM and corresponding corneal stroma button, punched out from the  donor cornea tissue using an 8.0-mm diameter trephine, were rinsed once in PBS and placed directly into 1 ml Trizol reagent. It should be noted that although most of the corneal epithelium spontaneously sloughed off the cornea surface during the transport and process of the cornea, remnant cells of the basal corneal epithelium might still be present on the corneal stroma button used for RNA extraction. Tissues were homogenized using a handheld homogenizer before addition of 200 ul of chloroform. After vigorous shaking, samples were spun at 13000 rpm for 15 min at 4uC. The upper phase, containing total RNA, was transferred to a new tube containing an equal volume of 70% ethanol. The resulting solution was transferred into a QIAGEN RNeasy column and procedures were performed as per manufacturer's protocol with a DNAse digestion step incorporated.

RNA Sequencing
Reverse transcription and amplification. RNA-seq libraries were prepared using an AB Demonstrated Protocol similar to the one reported in Tang et al (2009). For each sample, 100 pg of total RNA was reverse transcribed by SuperScript III reverse transcriptase (Cat no: 11752-250), using a poly(T) primer with a UP1 anchor sequence (50uC for 30 min followed by an enzyme inactivation step of 70uC for 10 min). The remaining free primers were removed by incubating with Exonuclease I (37uC 30 min, 80uC 25 min). Subsequently, a poly (A) tail was added to the 39 end of the first strand cDNA by terminal deoxynucleotidyl transferase (37uC for 15 min, and 70uC for 10 min).  The resulting product was split into 4 tubes and used for the synthesis of the second strand cDNA, using a poly(T) primer with another anchor sequence (UP2) and TaKaRa Ex Taq HS, in a volume of 22 ml each (1 cycle of 95uC for 3 min, 50uC for 2 min and 72uC for 6 min). The product of each tube was then separately amplified in a volume of 41 ml, using primers UP1 and UP2 and TaKaRa Ex Taq HS (18 cycles of 95uC for 30 sec, 67uC for 1 min and 72uC for 6 min +6 sec per cycle). The amplified cDNA was combined and purified using a QIAquick PCR purification kit and eluted with 50 ml EB buffer.
The purified cDNA was re-split into 4 tubes and re-amplified with a volume of 90 ml, using 1.2 ml of template for each tube (95uC for 3 min followed by 12 cycles of 95uC for 30 sec, 67uC for 1 min and 72uC 6 min +6 sec per cycle). For this amplification NH2-UP1 and NH2-UP2 primers were used. The amplified cDNA for each sample was then combined, purified and sizeselected (500 bp to 3 kb) by gel purification.
Library preparation. The ABI standard library preparation protocol (Chapter 2 of AB SOLiD 4 system library preparation guide -PN 4445673) was used for non-barcoded library preparation. The SOLiD fragment library barcoding kit module protocol (AB PN 4443045) was used for barcoded library preparation. The libraries were generated by using the SOLiD fragment library construction kit (PN 4443471) and the SOLiD fragment library barcoding kit (AB PN 4444836).
In brief 1 mg of amplified cDNA was sheared to approximately 165 bp, end repaired and P1, P2 adapters ligated. The product was then nick translated and amplified for 5 cycles before being size selected to be between ,240 bp to 270 bp. The libraries were further processed using the AB SOLiD 4 system templated bead preparation guide (PN 4448378) and sequenced on 1/8th of a SOLiD slide.
Alignment and counting. The resulting RNA-seq reads were processed using the ABI Bioscope pipeline. The reads were aligned to the hg19 human reference genome. Gene expression was measured by counting the number of reads mapping uniquely to both strands of each gene footprint and normalized to the total uniquely mapped reads to the entire genome. The value was adjusted to show reads per million (RPM).

Gene Expression Analysis
Reverse transcription. Reverse transcription was carried out using Superscript III (Cat no: 11752-250) according to manufacturer's protocol.

High-throughput
Quantitative Polymerase Chain Reaction (QPCR). Fluidigm BioMark TM System was used for high throughput QPCR and procedures were performed as per manufacturer's protocol. Briefly, 5.6 ng of cDNA and 5x Taqman Gene Expression Assays (Applied Biosystems) were used per reaction and were loaded on a 48648 Dynamic Array. Data was analyzed using BioMark TM HD System Software Suite. Expression from each gene was normalized to house-keeping genes (GAPDH, ATP6V0E1 and H2AFY) Error bars represent standard deviation from three replicates. The list of primers used can be found in Table 12.
QPCR. 7500 Fast Real-Time PCR System (Life Technologies) was used for QPCR and procedures were performed as per manufacturer's protocol. Briefly, 12.5 ng of cDNA and 2x Taqman Gene Expression Assays (Applied Biosystems) were used per reaction and loaded on a 96-well plate (for Life Technologies PCR system). Expression of each gene was normalized to GAPDH in the same run. Expression levels across samples were normalised to that in CEC culture (100 in CEC culture). Error bars represent standard deviation from three replicates. The list of primers used can be found in Table 12.