Widespread GLI expression but limited canonical hedgehog signaling restricted to the ductular reaction in human chronic liver disease

Canonical Hedgehog (Hh) signaling in vertebrate cells occurs following Smoothened activation/translocation into the primary cilia (Pc), followed by a GLI transcriptional response. Nonetheless, GLI activation can occur independently of the canonical Hh pathway. Using a murine model of liver injury, we previously identified the importance of canonical Hh signaling within the Pc+ liver progenitor cell (LPC) population and noted that SMO-independent, GLI-mediated signals were important in multiple Pc-ve GLI2+ intrahepatic populations. This study extends these observations to human liver tissue, and analyses the effect of GLI inhibition on LPC viability/gene expression. Human donor and cirrhotic liver tissue specimens were evaluated for SHH, GLI2 and Pc expression using immunofluorescence and qRT-PCR. Changes to viability and gene expression in LPCs in vitro were assessed following GLI inhibition. Identification of Pc (as a marker of canonical Hh signaling) in human cirrhosis was predominantly confined to the ductular reaction and LPCs. In contrast, GLI2 was expressed in multiple cell populations including Pc-ve endothelium, hepatocytes, and leukocytes. HSCs/myofibroblasts (>99%) expressed GLI2, with only 1.92% displaying Pc. In vitro GLI signals maintained proliferation/viability within LPCs and GLI inhibition affected the expression of genes related to stemness, hepatocyte/biliary differentiation and Hh/Wnt signaling. At least two mechanisms of GLI signaling (Pc/SMO-dependent and Pc/SMO-independent) mediate chronic liver disease pathogenesis. This may have significant ramifications for the choice of Hh inhibitor (anti-SMO or anti-GLI) suitable for clinical trials. We also postulate GLI delivers a pro-survival signal to LPCs whilst maintaining stemness.


Introduction
Cirrhosis as a result of chronic liver disease (CLD) is a major cause of worldwide morbidity and mortality [1]. Liver transplantation remains the only curative treatment option. Further, a lack of suitable donors renders this option implausible for the majority of patients. Therefore, novel therapeutic approaches for the successful treatment of CLD are in high demand.
To develop such approaches, it is essential to elucidate the cellular biology that underpins the regulation of pathogenic liver microenvironmental niches. A key feature of chronic human and mouse liver disease pathology [2,3] is the ductular reaction, thought to include expansion of putative bipotential liver progenitor cells (LPCs) [4][5][6]. We were the first to describe that primary cilia (Pc) are present on LPCs in mouse liver tissue [7]. Pc are single, non-motile, membrane-bound cellular organelles, essentially an 'antennae' (reviewed by [8]) that receive extracellular environmental cues (e.g. ligand) and translate these cues into internal cellular outputs (e.g. transcriptional response).
In vertebrate cells, canonical Hh signals are transduced through the Pc [9][10][11]. This occurs following Hh ligand binding (Sonic Hh, SHH; Indian Hh, IHH; or Desert Hh, DHH) to its receptor PATCHED1. This action de-represses SMOOTHENED (SMO), facilitating the activation/translocation of SMO into the Pc. Consequently, the GLI family (GLI1, GLI2, GLI3) of transcription factors are stabilized into their transcriptionally active forms, driving GLI target gene expression [12]. This is known as canonical Hh/Pc/SMO/GLI-mediated signaling. However, GLI signals can also be driven independently of Pc/SMO. This is referred to as Pc/SMOindependent GLI-mediated signaling, recently reviewed [12].
A thorough, accurate assessment of the intrahepatic cell populations contributing to Pc/ SMO-dependent GLI-mediated signaling and Pc/SMO-independent GLI-mediated signaling in CLD is imperative, due to the therapeutic potential of SMO small molecule inhibitors in the treatment of CLD-currently in stage I clinical trials [13]. A major limitation of former studies is the assumption that GLI2 + cells are 'actively' signaling via a canonical Hh pathway [14][15][16][17][18]. As we have previously asserted, this is a misleading assumption [7,19,20].
In the experimental thioacetamide (TAA) model of cirrhosis, we recently demonstrated that canonical Hh/Pc/SMO/GLI-mediated signaling amplifies the LPC response [7]. Further, our results suggest that GLI signaling is mediated by at least two mechanisms within intrahepatic populations in mouse liver tissue-Pc/SMO-dependent and Pc/SMO-independent GLImediated signaling. However, it is plausible our observations apply specifically to a single model of experimental liver disease. Importantly, cell populations signaling via a Pc/SMOindependent GLI-mediated mechanism would not be directly targeted using anti-SMO therapy; these populations may instead benefit from anti-GLI therapy.
To address the limitations of former human studies, and potentially reconcile these with our own observations, in this paper we specifically addressed: 1) Whether canonical Hh/Pc/ SMO/GLI-mediated signaling participates in regulation of the human LPC niche in vivo; 2) Clarification of whether GLI-mediated signaling is driven through Pc/SMO in other intrahepatic cell populations in human CLD; 3) The contribution of GLI inhibition to LPC viability, proliferation and gene expression in culture. Our results have direct therapeutic implications for the design of human clinical CLD trials involving Hh inhibitors.

Patient samples
Ethics for human liver tissue collection were firstly submitted to, and approved by, the Human Research Ethics Committee, Sydney Local Health District, RPAH Zone. Human liver tissue was collected from consenting patients according to methods outlined in approved X10-0072 protocol. Participants provided written informed consent. Samples included representative donor (n = 5) and cirrhotic samples from various disease etiologies (alcoholic liver disease (ALD) n = 6; non-alcoholic steatohepatitis (NASH) n = 3; primary biliary cirrhosis (PBC) n = 1). Patient age, sex, and fibrosis staging (Scheuer score [21]) are provided in Table 1.

Imaging
Screening of human liver samples for SHH and GLI2 was initially performed using a Leica DM6000B. Identical exposure conditions were used for imaging the protein of interest and matching IgG control (S1 Fig). Confocal imaging was performed using a Leica SP5. For Pc screening, n = 8-10 fields of view (FOV) were imaged through the fibrotic septae where Pc + cells were located (n = 3 ALD samples, Pc/LPC marker (EpCAM or pan-CK), Pc/Vimentin, Pc/CD31; n = 1 ALD sample Pc/CD45). All Pc images were obtained using the following imaging parameters: 63x objective, 1.5X digital zoom, 6x 0.7μm steps, 1024x1024 pixels. All Pc images represent a maximum intensity projection (MIP); quantification was performed on the MIP. For Pc quantitation, only fully formed Pc (axoneme-comprising α-acetylated tubulinassembled from the basal body comprising γ-tubulin) were counted. To determine the % of LPCs expressing Pc, a total of 566 Pc and 1259 LPC cells (either EpCAM + or pan-CK + ) were counted. To determine the % of vimentin + hepatic stellate cells (HSCs) expressing Pc, a total of 510 Pc and 3375 vimentin + cells were counted. A triple stain for EpCAM, vimentin and GLI2 was performed to quantitate the % of GLI2 + LPCs and GLI2 + HSCs within the same FOV. Images (ALD n = 3, n = 8 FOV/sample) were obtained through the fibrotic septae, using identical imaging parameters. A total of 1129 EpCAM + cells and 1902 Vimentin + cells were counted. Cells were considered GLI2 + if they expressed nuclear GLI2.

Electron microscopy
Human liver tissue (~1 mm 3 ) was washed in PBS prior to placing in fixative (2% (v/v) glutaraldehyde 1% (w/v) sucrose PBS, pH 7.4) O/N 4˚C. BMOL1.2 and BMOL-TAT were grown on thermanox coverslips and fixed for 1 h RT in 4% Glutaraldehyde-EM-grade in PBS pH 7.4 to equal amounts of medium and mixed gently. For scanning electron microscopy (SEM) cells were osmicated (1% OsO 4 /0.1 mol/l sodium cacodylate buffer), dehydrated in an ethanol gradient to 100%, and incubated for 2 min in hexamethyl-disilazane. Coverslips were then mounted on stubs, sputter coated with platinum, and examined using a JEOL 6380 SEM. For transmission electron microscopy (TEM), coverslips and tissue samples were embedded in Spurr's resin following the post fixation and dehydration steps. Ultrathin sections were obtained and viewed on a Phillips CM10 TEM.

Quantitative PCR
Total RNA was extracted from liver tissues with TRIzol (Life Technologies). cDNA was generated from total RNA with high capacity reverse transcription kits (Life Technologies). The abundance of transcripts was assessed by quantitative real-time PCR (qPCR) on a Corbett Rotor-Gene 3000 with SsoFast EvaGreen (Bio-Rad). The level of each transcript was determined by a standard curve. Expression of SHH, GLI1 and GLI3 was normalised for the efficiency of amplification with both PPIA and ACTB. The sequences of primer pairs used in qPCR were PPIA sense GGCAAATGCTGGACCCAACACAand PPIA antisense CTAGGCATGG GAGGGAACAAGG; ACTB sense ATTGGCAATGAGCGGTTC and ACTB antisense GGATGCCA CAGGACTCCAT; SHH sense AGTTTCACTCCTGGCCACTG and SHH antisense GATGAAGAA AACACCGGAGC; GLI1 sense TAGCTACTGATTGGTGGTGGGand GLI1 antisense ACTCCAGC CCTGGACCG; GLI3 sense GGCTGCATAGTGATTGCGT and GLI3 antisense CGAACAGATGT GAGCGAG AA.

Western blot
Protein from donor (n = 3 samples) and ALD (n = 3 samples) frozen liver was extracted using RIPA buffer containing protease and phosphatase inhibitors (Roche). Proteins were resolved on a 4-12% NuPAGE Bis-Tris gel, transferred to polyvinylidene difluoride (PVDF) membranes (Millipore, Billerica, MA), and then blocked with 10% (w/v) skim milk powder in PBS for 1 h at RT. The GLI1 rabbit polyclonal (1:2000, sc-20687, Santa Cruz Biotechnology) and GAPDH monoclonal (1:10,000, #ab8245, Abcam) antibodies were incubated with the membranes for 16 h at 4˚C. After being washed three times, the membranes were incubated with horseradish peroxidase conjugated secondary antibodies for 1 h at RT. After washing three times with PBS, the membrane was incubated with Pierce ECL Western blotting substrate (Thermo Fisher Scientific). Blots were quantified with NIH Image J and normalised by GAPDH.
Cell culture BMOL1.2 and BMOL-TAT cells were obtained from Dr. Tirnitz-Parker [22] and cultured as previously described [7]. Experiments were performed between passages 2-6 after revival from N 2 (l). Cell lines were confirmed negative for mycoplasma using the MycoAlert Detection Kit (Lonza).

RT 2 PCR profiling arrays
Cells (5x10 5 ) were plated into 10cm dishes, and medium replaced with 0.2% FBS culture medium for 16 h prior to treatments. Cells were treated with GANT61 (10 μM) or DMSO equivalent in 0.2% FBS for 8 h. Cells were then washed with 1xPBS, and RNA isolated on ice by scraping and RNeasy kit (Qiagen). The experiment was repeated in triplicate per line. cDNA was generated from RNA (1 μg) using the RT First Strand Synthesis kit (Qiagen).
Custom 384-well murine gene arrays (CAPM12471E; S1 Table) were loaded using an EpMotion robot and qPCR carried out using an ABI-7900HT (Applied Biosystems) at the ACRF Facility, Garvan Institute. Controls included PPC, RTC, and MGDC wells, four constitutive gene expression controls (gapdh, hsp90ab1, b2m, rn18s), and the cycle threshold set to ensure a consistent PPC value. Analysis using online SABioscience PCR Array software was performed for each individual experiment, cell line (n = 3) and combination of LPC lines (n = 6). Data were standardised to b2m and gapdh and a 1.5 fold-change set for gene transcript comparison. Consistent gene changes were generated from n = 3 analyses, and included only genes that modulated in the same direction in !2/3 experiments. Data were converted to log 2 scale and displayed in heat-map format using GENE-E.

Hh-producing cells in human donor and cirrhotic tissue
We firstly established the cellular source of Hh ligand in donor and cirrhotic human liver tissue specimens. In line with previous studies [23], we observed minimal SHH ligand expression (C-terminal) in donor liver samples ( Table 1 [23]. Varying patterns of SHH expression were seen, outlined in Table 1. Importantly, the majority of Hh ligand production in human liver disease tissue appeared to be derived from hepatocytes, located at the epithelial-mesenchymal interface or within the regenerative nodule. High levels of SHH in ALD (vs. donor) were also confirmed at the mRNA level ( Fig 1B).
Interestingly, we noticed rare SHH + cells around the portal tracts in donor human liver. These were confirmed to be EpCAM + and were present in all donor samples (n = 5) screened (Table 1, Fig 1C) but not observed in the mouse model [7]. Notably, the EpCAM + structures were observed as pseudo-ductular strings [24], indicative of the cellular component of the ductular reaction-a reactive process that is induced at the portal/parenchymal interface in chronically injured livers and that encompasses proliferation of bipotential LPCs amongst inflammatory, neural, vascular and extracellular matrix changes [25]. Further, EpCAM + SHH + ductular reactions were also noted at the epithelial-mesenchymal interface in ALD ( Fig 1D).

GLI2 expressing cell populations in human donor and cirrhotic tissue
Next, we sought to identify potential Hh-responsive cells using GLI2, the marker used in all prior studies of Hh signaling within a liver disease context [14,16]. Upon screening for GLI2 in our cohort of human donor and cirrhotic liver samples, we found GLI2 to be widely expressed within the lobule during liver homeostasis (Fig 2A, n = 5 donor; representative image). The number of GLI2 + cells was found to increase during human cirrhotic injury (ALD n = 6, NASH n = 3, PBC n = 1) (Fig 2A; representative images shown). In a way akin to the TAA model, GLI1 and GLI3 transcripts were found~10-fold increased in ALD compared to donor  difference between means (One-sided student t-test, **p<0.005). Western blot for full-length GLI1 protein (>150 kDa) in donor (Don) or ALD patient liver tissue (Fig 2B). GLI1 protein was also significantly increased in ALD vs. donor liver tissue (p = 0.0093), albeit to a lesser degree (~0.3-fold increase) than transcript ( Fig 2B). Importantly, GLI2 was readily expressed within the EpCAM + ductular/LPC population in donor, ALD, PBC and NASH liver specimens (Fig 2C, representative images).

Identification of Pc on the human ductular reaction/LPCs
In order to define canonical Hh signaling, we examined human cirrhotic liver tissue for Pc structures. Pc were readily identified at the epithelial-mesenchymal interface in human cirrhotic liver disease (Fig 3A). It is well known that mature bile ducts express Pc [29], however Pc structures were also identified on immature EpCAM + cells comprising the ductular reaction and believed to contain LPCs, in diverse human liver tissues including donor, ALD, NASH and PBC human samples (Fig 3A). A secondary axoneme marker (Arl13b), was also shown to co-localise with α-acetylated tubulin, corroborating the correct identification of the structure (Fig 3B). Arl13b staining was readily identifiable on EpCAM + ductular reactions/ LPCs in a similar pattern to α-acetylated tubulin (Fig 3B). We duly note that LPCs are a heterogeneous cell population, and that no current gold-standard marker in humans is capable of singularly identifying LPCs without co-staining either biliary or hepatocyte epithelium. Nevertheless, the obvious anatomical differences outlined earlier between a mature human bile duct and LPCs/the ductular reaction allowed us to confirm the presence of Pc on cells morphologically representing LPCs/ductular reaction. Furthermore, a second gold standard LPC marker (pan-CK) was used to confirm the presence of Pc on ductular reactions/LPCs (S2B Fig). Pc structures were also confirmed on non-biliary, non-HSC cells located within the fibrotic septae in human ALD tissue using TEM (Fig 3B, arrow).
To demonstrate the importance of these structures within the LPC/ductular population, we quantitated the presence of Pc on LPCs/ductular reaction (EpCAM + or pan-CK + ). Single, fully formed Pc structures were found on 58.6% LPCs in donor liver (162 Pc, 279 EpCAM+ cells, n = 3 samples) and 42.11% LPCs in ALD liver (566 Pc, 1259 EpCAM + or pan-CK cells, n = 3 samples), respectively (Fig 3C). Further, α-acetylated tubulin staining was found consistently associated with the ductular reaction (90% of LPCs; data not shown), even though single Pc structures were not resolved on every cell.
Intuitively, a cell reabsorbs its Pc in order to enter the cell cycle. Interestingly, we were able to show that Pc + LPCs/ductular reaction (white dashed arrows) are adjacent to  Only 1.92% of vimentin + HSCs/myofibroblasts in chronic liver injury expressed Pc, although 99% of vimentin + cells were GLI2 + (Figs 2E, 3C, 4A and 4B). Axoneme staining for Arl13b suggested the presence of rare vimentin + cells that express a fully assembled Pc (Fig 4A,  arrow). When screening for Pc, rare Pc + vimentin + cells were detected in ALD chronic liver injury (Fig 4A). Pc were not detected on CD31 + ECs in the ALD samples screened (n = 3; Fig 4C, representative image). Further, absence of Pc on leukocytes (CD45 + ) (n = 1 sample screened; data not shown) and hepatocytes assisted to validate the specificity of our staining, with these cell types previously established to be Pc -ve [30].
This suggests that in human liver, the ductular/LPC niche is supported either by hepatocyte-derived (paracrine) or ductular/LPC-derived (autocrine) canonical Hh signals, with Hh ligand capable of binding to the EpCAM + Pc + GLI2 + Hh-responsive ductular/LPC population. This appears to occur frequently within liver homeostasis and in a chronic injury context.

Effect of GLI inhibition on LPC function and gene expression
Our mouse and human in vivo data provide robust evidence for active GLI signaling within many intrahepatic cell populations including LPCs [7]. We have previously ascertained that GLI signaling can be driven either canonically (Pc/SMO-dependent) or non-canonically (Pc/SMO-independent) in the mouse LPC line BMOL1.2 [7]. Further, SMO inhibition (or hGLI3-R overexpression) significantly decreases LPC proliferation/viability, whilst hGLI1 overexpression significantly increases proliferation/viability [7]. To further explore the nature of GLI signaling within the LPC and to determine the effect of GLI inhibition on LPC proliferation and gene expression, we used the two well-characterized LPC lines BMOL1.2 and BMOL-TAT [22]. Characterisation of LPC-specific marker OPN [24], Pc and GLI expression within these cell lines is provided in Fig 5. Both BMOL1.2 and BMOL-TAT cells showed a significant reduction in viability following addition of 10 μM of specific GLI inhibitor GANT61 (GLI1, GLI2) compared to matched DMSO controls (p<0.001) in three separate experiments (Fig 6A). A difference between the two LPC lines was observed following addition of low concentration GANT61 (1 μM), which significantly increased cell viability in BMOL-TAT cells (p<0.001) compared to controls. Interestingly, medium supplementation with active concentrations of Notch, c-Met and Wnt inhibitors had no effect on BMOL1.2 or BMOL-TAT cell viability, within a 72 h time frame (Fig 6C-6H).

Discussion
The key observations in this paper include the finding that canonical Hh (Pc/SMO-dependent, GLI-mediated signaling) is predominately restricted to the EpCAM + ductular compartment in human liver tissue. In contrast, GLI-mediated signals appear to be Pc/SMO-independent within multiple Pc -ve GLI2 + human intrahepatic compartments (namely, HSC, hepatocyte, leukocyte and endothelial). Other novel aspects of this study include the observation that GLI delivers a pro-survival signal to LPCs in culture. An obvious caveat to our work is our key observations are mostly descriptive and based on a small sample size of human liver specimens. Nevertheless, our results provide the first methodical and comprehensive assessment of Pc/GLI expression in human liver disease specimens, and identify that GLI signals are mediated at least via two mechanisms in CLD. This has immediate translational significance: anti-GLI therapies (possibly in combination with anti-SMO) may be a better Hh target for novel CLD therapies, currently in stage I trials.
The impact of this study is several-fold. Firstly, we have made the novel discovery that Pc are expressed by LPCs (EpCAM + , pan-CK + ) in human liver tissue-both in donor and cirrhotic liver of several etiologies (ALD, NASH, PBC). This paper provides the first evidence that Pc-mediated signaling pathways e.g. Hh (and possibly others yet to be identified) are of importance/biologically significant in human liver tissue pertaining to regulation of the LPC niche. These data have not been reported previously, with our group being the first to identify Pc on LPCs in mouse liver tissue in 2014 [7]. The presence of Pc on normal intrahepatic bile ducts has long been appreciated. Yet Pc-regulated signaling has not previously been  studied within the context of the human LPC niche, although many stem cell niches in other organ systems e.g. brain [31], muscle [32], lung [33], and also embryonic stem cells [34], are known to be regulated (at least in part) via Pc-mediated signals. Future studies are required to understand the contribution of Pc-mediated signaling pathways to the evolution of the human LPC niche from homeostasis to chronic injury, as gaining these insights will most likely have important ramifications to gaining understanding in to how liver cancer progresses.
Secondly, we have identified gene changes in BMOL1.2 and BMOL-TAT LPC lines that occur with GLI inhibition (discussed below). Thirdly, we have validated in human tissue samples that GLI-mediated signals are mediated by at least two mechanisms in chronic liver injury; this has important ramifications for current clinical trials, and is discussed in greater detail below. Fourthly, we have provided additional verification to prior studies pertaining to characterising Hh-producing and Hh-responding cell populations in human chronic liver disease, which is required to strengthen the field. Our in vitro culture studies provide further support to our prior studies [7] and human data, and strongly suggest the GLI transcriptome is a critical pathway in LPC biology. Inhibition of GLI saw a similar downregulation of a defined gene set within the two LPC lines. In general, three main gene groups appear to be GLI-dependent: stem cell genes, hepatocyte/biliary differentiation genes, and Hh/Wnt pathway genes.
Stem cell genes inhibited by GANT61 include Lgr5, Hes1, Hey1, Ly6a, Wnt2, Wnt4, and Wnt5b transcripts. Lgr5 is a well-known marker of proliferating Wnt-driven stem cells [35] and marks proliferating murine LPCs in vivo [35]. Lgr5 was downregulated by up to 100-fold. Likewise, Wnt2 was highly downregulated (35-to 150-fold). Furthermore, two of the four chief pluripotency genes (Klf4, myc) were upregulated in BMOL1.2 cells following GANT61 treatment, concurrent with an increase in multiple Hh-related genes (Fgf9, Foxe1, Npc1, Shh). This potentially demonstrates a compensation event, if GLI signals are indeed maintaining LPC stemness and survival. Overall, our results imply that GLI is involved in maintaining stemness in the LPC lines. We hypothesise from our data that GLI delivers a pro-survival   signal to LPCs whilst maintaining their stemness, subsequently allowing for appropriate hepatocyte (Wnt-driven) and biliary (Notch-driven) differentiation responses [36].
The major findings of this study confirm our prior observations in the TAA model [7] that GLI signaling in CLD is mediated by at least two mechanisms: Pc/SMO-dependent and Pc/ SMO-independent GLI-mediated signaling. The interpretation of these data has important ramifications, given anti-Hh therapy is currently being used as a novel target for hepatocellular carcinoma and cirrhosis treatment (presently in stage I clinical trials [13]). Importantly, current trials have selected an anti-SMO inhibitor (LDE225) for cirrhosis treatment; however, our results suggest that GLI (either with or without combinatorial targeting of SMO) may be a more appropriate signaling node within the Hh pathway to target. In support of our observations, mouse fibrosis intervention studies in the lung [37] and kidney [38] have indicated that anti-GLI rather than anti-SMO inhibitors are effective at decreasing fibrosis-and indicate that GLI signaling is (at least partially) controlled via SMO-independent mechanisms in these models. Prior studies in mouse liver injury models have indicated that deletion of SMO in HSCs decreases the degree of fibrosis [39]. However, in light of our observations (both in the TAA mouse model and now confirmed in human liver tissue samples), combinatorial targeting of SMO/GLI should be tested in liver injury models, and anti-GLI inhibitors considered for the design of clinical trials involving anti-Hh therapies for the treatment of CLD.
To conclude, this work provides the first thorough assessment of Pc/GLI expression in human liver disease, and confirms (following our initial observations in the mouse) that GLI signals are likely to be mediated via at least two mechanisms in CLD. These results have immediate translational significance and should be considered prior to initiating stage II clinical trials using SMO inhibitors. Human ALD liver tissue samples were examined for primary cilium expression (α-acetylated tubulin, green; γ-tubulin, red) on LPCs (pan-CK) + (grey) indicated by white dashed arrows. Yellow solid arrows indicate intermediate hepatocytes found adjacent to these Pc + LPCs. DAPI, blue. (B) LPCs (green; EpCAM or pan-CK) were co-stained with the proliferation marker ki67 (red). Intermediate hepatocytes adjacent to LPCs were often ki67 + (yellow arrows). All images obtained using confocal microscopy, 63x objective. Ã non-specific staining in ki67 channel. (TIF) S1 Table. Custom 384-well murine gene array (CAPM12471E) layout. (XLSX)