In order to discover global gene expression patterns characterizing subgroups of colon cancer, microarrays were hybridized to labeled RNAs obtained from seventeen colonic specimens (nine carcinomas and eight normal samples). Using a hierarchical agglomerative method, the samples grouped naturally into two major clusters, in perfect concordance with pathological reports (colon cancer versus normal colon). Using a variant of the unpaired t-test, selected genes were ordered according to an index of importance. In order to confirm microarray data, we performed quantitative, real-time reverse transcriptase–polymerase chain reaction (TaqMan RT–PCR) on RNAs from 13 colorectal tumors and 13 normal tissues (seven of which were matched normal-tumor pairs). RT–PCR was performed on the gro1, B-factor, adlican, and endothelin converting enzyme-1 genes and confirmed microarray findings. Two hundred and fifty genes were identified, some of which were previously reported as being involved in colon cancer. We conclude that cDNA microarraying, combined with bioinformatics tools, can accurately classify colon specimens according to current histopathological taxonomy. Moreover, this technology holds promise of providing invaluable insight into specific gene roles in the development and progression of colon cancer. Our data suggests that a large-scale approach may be undertaken with the purpose of identifying biomarkers relevant to cancer progression.
Colorectal cancer is the second most common cause of cancer-related deaths in the United States, with an estimated 130 200 new cases and 56 300 deaths occurring in the year 2000 (American Cancer Society, 2000). Early detection of colon cancer in its premalignant stages prevents progression to invasive cancer (American Cancer Society, 2000). Therefore, it is crucial to discover lesions early in their neoplastic evolution. Furthermore, biomarkers capable of predicting progression of polyps to cancer could identify subgroups of patients most likely to develop recurrent polyps or new cancers.
cDNA microarray technologies have now made it possible to characterize tissue samples by monitoring the expression of thousands of genes simultaneously. Previous studies have been successfully performed in acute leukemia, lymphoma, breast cancer, and others (Chu et al., 1998; Spellman et al., 1998; Eisen and Brown, 1999; Iyer et al., 1999; Perou et al., 1999, 2000; Alizadeh et al., 2000). These breakthrough studies have begun to create a modern molecular taxonomy, which capitalizes on the statistical power of large gene expression datasets (Bassett et al., 1999). This comprehensive approach permits the classification of diseases or disease lesions based on comprehensive characterizations of their phenotypes, rather than on limited histologic or other clinical features (Perou et al., 1999, 2000; Alizadeh et al., 2000; Selaru et al., 2002a,b).
While cDNA microarray data comprises precise ratios of expression levels between each sample and a reference probe, it is difficult to interpret the enormous data flow that results from this process. In order to efficaciously extract information from this data, mathematical and bioinformatics tools have been developed. For example, the bioinformatics program Cluster (Eisen et al., 1998) generates hierarchical and K-means clusters from tab-delimited text files containing expression ratios comparing two RNA species' expression levels. In addition, the program Treeview (Eisen et al., 1998) generates two-dimensional graphic displays of gene expression in red and green colors (Chen et al., 1998; Spellman et al., 1998; Eisen and Brown, 1999; Iyer et al., 1999; Perou et al., 1999; 2000; Alizadeh et al., 2000). Significance Analysis of Microarrays (SAM), a software program developed at Stanford University (Tusher et al., 2001), was used for finding the most significantly differentially expressed genes in a cancer relative to matching normal specimens. The output consists of lists of genes ordered in decreasing order of an index of significance.
We applied SAM to identify genes whose expression was most relevant to distinguishing between normal and cancerous colonic epithelium. Our results suggest that bioinformatics analyses of microarray data can precisely classify colorectal pathological states, and that candidate genes responsible for this classification can be identified in this process.
Hybridizations were successfully performed on all specimens. Cluster software was then applied, and a dendrogram view was generated using TreeView (Figure 1).
The degree of relatedness of specimens, as well as genes, was quantitated by the closeness of branches in the dendrogram. Cluster grouped the seventeen colonic specimens into two major clusters. The first cluster contained all of the normal colon samples, while the second contained all of the colon cancer samples. This classification was derived with no previous pathological data, using only gene profiling data. Thus, normal colon tissue was differentiated from cancerous colon tissue based solely on global patterns of gene expression.
SAM was then used to identify 250 genes that were most influential in the distinction between normal and cancerous colon, with a false discovery rate (FDR) of 0.57 (Web Table, www.microarray.umaryland.edu/manuscripts/11415/). This Table contains links to further information regarding each gene (q.v.). Among these 250 genes, 150 were expressed at higher levels in normal than in cancerous colon; the remaining 100 were expressed at higher levels in cancer than in normal tissue.
As a proof-of-principle, we verified the relative expression levels of four of these 250 genes by real-time (TaqMan) RT–PCR. The four genes tested were adlican, gro-1 oncogene, properdin/B factor, and endothelin converting enzyme 1. All four genes were confirmed by TaqMan RT–PCR as showing the same differential expression patterns found in the primary cDNA microarray data. The first three of these genes were overexpressed in colon cancers, while the last was overexpressed in normal colon (Figure 2).
This Figure displays results of real-time PCR products performed on 13 colorectal carcinomas and 13 normal tissues, including seven paired normal and cancerous colorectal samples. Nine of the tumors and seven of the normals were analysed by both microarray and TaqMan; the remaining four cancers and six normals were analysed only by TaqMan, but were in agreement with the differential expression pattern seen in the original cDNA microarray data.
In order to obtain additional insight into global gene expression patterns and to identify individual genes distinguishing among colon cancer subgroups, we performed significance analysis of microarrays (SAM) (Tusher et al., 2001) based on right versus left colon lesions and on lymph node positive (Dukes' C) versus lymph node negative (Dukes' B) tumors. Genes selected by SAM were then entered into Cluster (Eisen et al., 1998; Eisen and Brown, 1999). Results of these two cluster analyses are displayed in Figure 3.
As shown in Figure 3a, clustering performed on 203 genes selected by SAM was 100% successful in correctly classifying colon cancers as arising in the right versus left colon. To our knowledge, this is the first demonstration of global gene expression differences between the right and left colon. In Figure 3b are displayed the results of clustering performed based on 224 genes selected by SAM as significantly differentially expressed between lymph node-negative versus lymph node-positive tumors. Clustering was still fairly successful in this comparison, misplacing only one LN-positive specimen in the LN-negative group.
The above data suggest that gene filtering methods based on genomic data are capable of distinguishing normal from cancerous human colon. Furthermore, these data show that gene comparison algorithms (such as SAM) can be used to identify genes whose differential expression is most relevant to distinguishing between two diagnoses, such as normal and cancerous colon. In addition, these results suggest the combined approaches used herein can discriminate among subgroups of colon cancer.
In our study, genes identified by SAM as being significantly differentially expressed included many with known or suspected relevance to cancer. For example, among the genes whose expression was verified by quantitative RT–PCR, the Gro1 oncogene (also known as the cytokine growth-related oncogene or melanoma growth-stimulating activity alpha) has been localized to region 4q21. This chromosomal region is frequently involved in clonal aberrations found in primary melanomas (Grammatico et al., 1995). Moreover, gro-1 has been implicated in central nervous system tumors (Robinson et al., 2001). Similarly, properdin/B factor has been implicated in the etiopathogenesis of breast cancers (Perou et al., 1999); and the retinoic acid receptors RAR alpha, RAR beta, and RAR gamma mRNA are overexpressed in the colon cancer cell line, LoVo (Stewart and Thomas, 1997). Conversely, many genes overexpressed in normal colon were also relevant to cancer. For example, endothelin converting enzyme 1 expression is altered in breast and lung cancers (Ahmed et al., 2000; Patel and Schrey, 1995).
Additional cancer-related genes identified by our SAM-based analysis of cDNA microarray data included members of the protein phosphatase 1 gene family, which show genetic alterations and abnormal expression in human cancers (Takakura et al., 2001; Sogawa et al., 1997); gastrin-releasing peptide (GRP) and its receptor (GRP-R), which are frequently expressed by cancers of the gastrointestinal tract, breast, lung, and prostate (Jensen et al., 2001); members of the retinoic acid receptor gene family, which have been implicated in tumor growth inhibition (Nicke et al., 1999); and certain mitogen-activated protein kinases, which have actually been reported as downregulated in colon cancers (Wang et al., 2000).
The differences in gene-filtered expression patterns that we observed between right and left colon are consistent with published observations on biological differences between the two sides of the colon (Ikeda et al., 2001; Meltzer et al., 1989). Moreover, the near perfect classification of LN-positive versus LN-negative tumors is not surprising, considering that this clinical parameter forms part of the basis for the traditional Dukes' classification.
The majority of published cDNA microarray studies use this technology primarily as a global gene profiling tool (Duggan et al., 1999; Khan et al., 1999; Alizadeh et al., 2000; Elek et al., 2000). Data presented in the current manuscript suggest that cDNA microarrays are also a good screening tool to identify individual genes relevant to a particular disease process, such as human cancer. Our TaqMan RT–PCR results suggest that microarrays are accurate in gauging differential expression. Moreover, our findings suggest that gene comparison algorithms, such as SAM, are highly accurate in identifying genes most influential in making distinctions between two types of study materials, such as normal and cancerous colon. Thus, they support the broad use of cDNA microarrays and bioinformatics in large-scale gene discovery studies.
Materials and methods
Tissues and patients
Patients were enrolled sequentially during a three-year period from the University of Maryland or Baltimore VA Hospitals. Patients with a presumed diagnosis of colonic carcinoma were enrolled in the study, with histologic confirmation after enrollment. Informed consent was obtained from all patients prior to enrollment under a protocol approved by the University of Maryland/Baltimore VA Hospital Institutional Review Board (IRB). All tissues were immediately frozen on dry ice and stored under liquid nitrogen at −180°C until further use. All presumptive diagnoses were confirmed by histopathologic examination: parallel samples from each anatomic location were obtained for histological analysis simultaneously with each research specimen. Clinical parameters of patients analysed in our study are included in Table 1.
RNA extraction and amplification
Total RNA was extracted from freshly frozen tissues and cell lines with the Rneasy Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Samples were treated with Rnase-free DNase (Qiagen) on the RNeasy columns. RNA was dissolved in Rnase-free water and amplified using T7 RNA polymerase using a linear, nondistorting protocol (Van Gelder et al., 1990; Luo et al., 1999). Amplified RNA (aRNA) was generated from 10 μg of total RNA using a cDNA Synthesis System (GIBCO–BRL) and the Ampliscribe T7 Transcription Kit (Epicentre Technology). Using this approach, approximately 100-fold amplification of mRNA from total RNA was obtained.
Reference (process control) probe
A reference probe was produced from a mixture containing equimolar aliquots of RNA from the cell lines Hct116, HT29, CaCo-2, Hct15, HTB114, MCF-7, HeLa and AGS. HTB114 was derived from a patient with leukemia, HeLa from a cervical cancer, MCF-7 from a breast cancer, and AGS from a gastric cancer. The remaining four cell lines (Hct116, HT29, CaCo-2, and Hct15) were derived from colorectal cancers. These cell lines were chosen to represent a variety of cell types in order to garner a baseline (i.e., green or Cy3) expression level 1) for comparison by ratio with red (Cy5) lesion-derived signal and 2) as a process control to ensure that each clone was generating detectable signal on the array (Alizadeh et al., 1998; Iyer et al., 1999; Perou et al., 1999, 2000).
Labeling of the aRNA probe
For each two-way comparison, 3–6 μg of aRNA prepared from the reference cells or esophageal lesion were labeled by incorporating Cy3- or Cy5-labeled dCTP using random primers and Superscript reverse transcriptase. The resulting probes were purified with a Microcon microcentrifuge filter device and recovered in a volume of 25 μl. These probes were used for each hybridization under a 40×22 mm coverslip at 65°C in 2.24X SSC, 0.25% SDS in a final volume of 35 μl.
Preparation of micro-assay clones
We followed protocols obtained from the National Cancer Institute–Advanced Technology Center (NCI–ATC; Lance Miller, director; David Petersen, personal communication). The 95% non-redundant, sequence-verified, periodically annotation-updated cDNA library prepared by the Lawrence Livermore Laboratories was used as a source of clones (Research Genetics, Huntsville, AL, USA). One μl of each bacterial stock was amplified in a 108 μl reaction containing 2.5 U Taq polymerase (Life Sciences, Gaithersburg, MD, USA). The master mix for each plate contained 9.5 ml of double-distilled water, 1 ml of 10× buffer, 10 μl each of 1000 μM M13 primers (forward and reverse), 20 μl of each of the four dNTPs (100 mM), and 50 μl of Taq polymerase (5 U/μl). PCR conditions consisted of an initial denaturation step at 96°C for 30 s followed by 30 cycles of 45 s at 94°C, 45 s at 55°C and 2 min 30 s at 72°C, then concluded by a final extension of 5 min at 72°C. The amplified inserts were then purified using a Qiagen PCR purification kit (Cat. No. 963141, QiAGEN) on a Qiagen BioRobot 9600 liquid handling robotic workstation. After purification, PCR products were dessicated in 96-well plates using a large Speed-vac apparatus, then reconstituted in 30 μl of distilled water.
Glass slide coating protocol
We prepared lysine-coated slides derived from the NCI–ATC and Stanford University protocols (http://www.microarrays.org/protocols.htm). Slides were cleaned using a solution containing 3.5N sodium hydroxide and 68% ethanol, then coated with 1% poly-L-lysine in 1% PBS. Before printing, slides were aged a minimum of 2 weeks, but not longer than 6 weeks.
The 8000 clones were printed using eight pins in a 32-pin print head (Majer Precision Engineering, Tempe, AZ, USA) on a GeneMachines Omnigrid Arrayer (GeneMachines, Oxnard, CA, USA). The printed slides were UV-crosslinked, post-treated with succinic anhydride to reduce backgound, and subjected to hybridization.
Microarray hybridization and washing
Each slide was incubated in 35 μl of hybridization solution containing Cy3- and Cy5-labeled target, 1 μl of 50× Denhardt's blocking solution (Sigma, St. Louis, MO, USA), 20 μg of Human COT 1-DNA (Roche Diagnostics Corporation, Indianapolis, IA, USA), 10 μg of yeast tRNA (Roche), 8–10 μg of Poly-A (Roche), in 2.24× SSC/0.25% SDS at 65°C overnight. Hybridization was performed under a 22×40 mm cover slip. The slide was then placed in a sealed hybridization chamber (Teleckem, Sunnyvale, CA, USA) containing two side wells with a total of 50 μl of water for humidification at 65°C overnight. On the next day, the slide was washed in 500 ml 2× SSC, 0.1% SDS at room temperature, during which time the coverslip fell off and washing continued for 2 min. The slide was placed in 1× SSC for 2 min at RT. it was then washed once with 0.2× SSC at RT and once with 0.05× SSC for 2 min and air-dried.
The hybridized slides were scanned using a GenePix 4000A dual-laser slide scanning system (Axon) at wavelengths corresponding to each probe's unique fluorescence. The resulting GenePix report was then reformatted for importance into the Cluster software program.
Data imported from GenePix were manipulated and clustered, using established algorithms implemented in the software program Cluster (http://rana.lbl.gov/) (Eisen et al., 1998; Eisen and Brown, 1999). Prior to average linkage clustering, data was log transformed, then median centered on genes and on arrays and, finally, normalized for genes and arrays, as described (Eisen et al., 1998). Average linkage clustering with centered correlation was used. TreeView software (ibid.) generated visual representations of the clusters.
SAM was used for identifying differentially expressed genes between normal colon specimens and colon cancer specimens (Tusher et al., 2001). SAM, a computer program specifically designed for manipulating microarray data, reports the most statistically significant differentially expressed genes between two groups of samples. In addition SAM reports an estimate of the Median False Discovery Rate (FDR), which is the percentage of genes falsely reported as showing statistically significant differential expression. SAM uses an algorithm based on the Student's t-test and also performs data permutations in order to determine the FDR.
SAM compared results for tumor to those for normal colon. SAM makes this comparison by comparing tumor/reference probe ratios to normal colon/reference probe ratios (i.e., it makes direct comparisons of the primary data, which are themselves ratios of tissue RNA to reference probe RNA). Genes listed in the web link table as being overexpressed in tumor are overexpressed in tumor relative to normal colon; likewise, genes listed as overexpressed in normal colon are overexpressed in normal colon relative to tumor.
Samples used for verification of gene expression levels
Total RNAs from the same samples studied in cDNA microarray analysis were used for verification of gene expression levels observed on cDNA microarrays. In addition, we extracted total RNAs from seven normal and seven tumor paired colon tissues.
Design of quantitative RT–PCR primers and probes
Quantitative RT–PCR was performed on a TaqMan real-time PCR machine (ABI 7700, Applied Biosystems, Foster City, CA, USA). For quantitative RT–PCR, the amplicon spanned an exon–exon boundary in order to exclude genomic DNA contamination. We obtained exon–exon boundary information on genes of interest from the Ensembl Genome Server of the Sanger Centre (http://www.ensembl.org/). RNA sequence and exon–exon boundary information was placed into quantitative PCR primer design software (PrimerExpress version 1.5, Applied Biosystems). Primer and probes are summarized in Table 2.
Probes were labeled with the reporter dye 6-carboxyfluorescein (6′-FAM) at the 5′ end and with the reporter dye 6-carboxytetramethylrodamine (TAMRA) at the 3′ end.
Acquisition and normalization of data
In each TaqMan run, serial dilutions of a single standard cDNA (derived from one colon cancer) were amplified to create a standard curve, and values of unknown samples were estimated relative to this standard curve. PCR reactions of each sample were run in duplicate; the mean value of two reactions was defined as representative of the sample. TaqMan Ribosomal RNA Primer and Probe (VIC Dye labeled: Applied Biosystems) were used to normalize quantitative data. The formula for normalization was: ratio of sample to reference cDNA =Gene(s)/Gene(r)/(Ribo(s)/Ribo(r)), where Gene(s) and Gene(r) were expression levels of each gene in the sample and reference cDNA, respectively, and Ribo(s) and Ribo(r) were ribosomal RNA expression levels in the sample and reference.
Real-time RT–PCR methods
For reverse transcription we used Superscript II (GIBCO–BRL). In a 20 μl reaction volume, 1 μg of total RNA was used with the supplied buffer, 40 U of Rnase Out recombinant ribonuclease inhibitor, 100 μM random hexamers, 1 mM each deoxynucleotide triphosphate, and 200 U of Superscript II. Reaction conditions were as follows: preheating of RNA sample for 5 min at 65°C, addition of the reaction mixture on ice, and incubation for 10 min at 25°C, then for 50 min at 42°C, followed by heating at 70°C for 15 min for enzyme denaturation, then rapid cooling to 4°C.
PCR reaction mix contained 12.5 μl of Taqman Universal MasterMix (Applied Biosystems), 0.25 μl of each forward and reverse primers (10 μM), 2.0 μl of dual-labeled probe (2.5 μM), cDNA from 50 ng total RNA, and H2O up to 25 μl total volume. PCR reaction conditions were as follows: 50°C for 2 min, 95°C 10 min, and 40 cycles of 95°C for 15 s and 60°C for 1 min.
Ahmed SI, Thompson J, Coulson JM, Woll PJ . 2000 Am. J. Respir. Cell. Mol. Biol. 22: 422–431
Alizadeh A, Eisen M, Botstein D, Brown PO, Staudt LM . 1998 J. Clin. Oncol. Immunol. 18: 373–379
Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, Powell JI, Yang L, Marti GE, Moore T, Hudson Jr J, Lewis DB, Tibshirani R, Sherlock G, Chan WC, Greiner TC, Weisenburger DD, Armitage JO, Warnke R, Staudt LM et al . 2000 Nature 403: 503–511
American Cancer Society. 2000 Cancer Facts and Figures"NPGsj5_1.dtd" (www.cancer.org)
Bassett Jr DE, Eisen MB, Boguski MS . 1999 Nat. Genet. 21: 51–55
Chu S, DeRisi J, Eisen M, Mulholland J, Botstein D, Brown PO, Herskowitz I . 1998 Science 282: 699–705
Duggan DJ, Bittner M, Chen Y, Meltzer P, Trent JM . 1999 Nat. Genet. 21: 10–14
Eisen MB, Brown PO . 1999 Methods Enzymol. 303: 179–205
Eisen MB, Spellman PT, Brown PO, Botstein D . 1998 Proc. Natl. Acad. Sci. USA 95: 14863–14868
Elek J, Park KH, Narayanan R . 2000 In Vivo 14: 173–182
Grammatico P, Roccella M, Catricala C, Roccella F, Bucher S, Mordenti C, Amantea A, Di Rosa C, Del Porto G . 1995 World J. Surg. 19: 350–351
Ikeda Y, Oda S, Abe T, Ohno S, Maehara Y, Sugimachi K . 2001 Oncology 61: 168–174
Iyer VR, Eisen MB, Ross DT, Schuler G, Moore T, Lee JCF, Trent JM, Staudt LM, Hudson Jr J, Boguski MS, Lashkari D, Shalon D, Botstein D, Brown PO . 1999 Science 283: 83–87
Jensen JA, Carroll RE, Benya RV . 2001 Peptides 22: 689–699
Khan J, Saal LH, Bittner ML, Chen Y, Trent JM, Meltzer PS . 1999 Electrophoresis 20: 223–229
Luo L, Salunga RC, Guo H, Bittner A, Joy KC, Galindo JE, Xiao H, Rogers KE, Wan JS, Jackson MR, Erlander MG . 1999 Nat. Med. 5: 117–122
Meltzer SJ, Zhou D, Weinstein WM . 1989 Exp. Mol. Pathol. 51: 264–274
Nicke B, Riecken EO, Rosewicz S . 1999 Gut 45: 51–57
Patel KV, Schrey MP . 1995 Br. J. Cancer 71: 442–447
Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT, Pergamenschikov A, Williams CF, Zhu SX, Lee JC, Lashkari D, Shalon D, Brown PO, Botstein D . 1999 Proc. Natl. Acad. Sci. USA 96: 9212–9217
Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lonning PE, Borresen-Dale AL, Brown PO, Botstein D . 2000 Nature 406: 747–752
Robinson S, Cohen M, Prayson R, Ransohoff RM, Tabrizi N, Miller RH . 2001 Neurosurgery 48: 864–873
Selaru FM, Zou T, Shustova V, Xu Y, Yin J, Mori Y, Sato F, Wang S, Olaru A, Shibata D, Greenwald BD, Krasna MJ, Abraham JM, Meltzer SJ . 2002a Oncogene 21: 475–478
Selaru FM, Xu Y, Yin J, Zou T, Liu TC, Mori Y, Abraham JM, Sato F, Wang S, Twigg C, Olaru A, Shustova V, Leytin A, Shibata D, Harpaz N, Melzter SJ . 2002b Gastroenterology 122: 606–613
Sogawa K, Masaki T, Miyauchi A, Sugita A, Kito K, Ueda N, Miyamoto K, Okazaki K, Okutani K, Matsumoto K . 1997 Cancer Lett. 112: 263–268
Spellman PT, Sherlock G, Zhang MQ, Iyer VR, Anders K, Eisen MB, Brown PO, Botstein D, Futcher B . 1998 Mol. Biol. Cell. 9: 3273–3297
Stewart LV, Thomas ML . 1997 Exp. Cell Res. 233: 321–329
Takakura S, Kohno T, Manda R, Okamoto A, Tanaka T, Yokota J . 2001 Int. J. Oncol. 18: 817–824
Tusher VG, Tibshirani R, Chu G . 2001 Proc. Natl. Acad. Sci. USA 98: 5116–5121
Van Gelder RN, von Zastrow ME, Yool A, Dement WC, Barchas JD, Eberwine JH . 1990 Proc. Natl. Acad. Sci. USA 87: 1663–1667
Wang Q, Ding Q, Dong Z, Ehlers RA, Evers BM . 2000 Anticancer Res. 20: 75–83
NIH grants CA85069, DK47717, CA95323, CA77057, and the Medical Research Office, Department of Veterans Affairs.
About this article
MALAT1 gene expression in colorectal cancer and its clinical significance: Data mining based on multiple gene expression databases
World Chinese Journal of Digestology (2019)
Epigenetic identification of ADCY4 as a biomarker for breast cancer: an integrated analysis of adenylate cyclases
4-Hydroxy-7-oxo-5-heptenoic Acid Lactone Is a Potent Inducer of the Complement Pathway in Human Retinal Pigmented Epithelial Cells
Chemical Research in Toxicology (2018)
Pathology - Research and Practice (2018)
Journal of Cellular and Molecular Medicine (2017)