Abstract
Since circulating leukocytes, mainly B and T cells, continuously maintain vigilant and comprehensive immune surveillance, these cells could be used as reporters for signs of infection or other pathologies, including cancer. Activated lymphocyte clones trigger a sensitive transcriptional response, which could be identified by gene expression profiling. To assess this hypothesis, we conducted microarray analysis of the gene expression profile of lymphocytes isolated from immunocompetent BALB/c mice subcutaneously injected with different numbers of tumorigenic B61 fibrosarcoma cells. Flow cytometry demonstrated that the number of circulating T (CD3+CD4+ or CD3+CD8+) or B (CD19+) cells did not change. However, the lymphocytes isolated from tumor cell–injected animals expressed a unique transcriptional profile that was identifiable before the development of a palpable tumor mass. This finding demonstrates that the transcriptional response appears before alterations in the main lymphocyte subsets and that the gene expression profile of peripheral lymphocytes can serve as a sensitive and accurate method for the early detection of cancer.
Introduction
The early detection of tumor cells in vivo could significantly enhance the success of cancer treatments. Peripheral blood lymphocytes (B and T cells), which have a cell surface receptor–based recognition system (surface immunoglobulins [Ig] and the T-cell receptors [TCR], respectively), represent a system naturally engaged in this detection through immunologic surveillance. The recognition of infectious agents or cancer cells by the immune system may trigger different patterns of gene expression that are associated with the nature and site of the insult. This possibility has been referred to as a “pathognomonic gene expression signature” that could, therefore, be used to diagnose hidden or nascent diseases.
B and T cells have sensitive signal transduction systems that are triggered by surface receptors that recognize free antigen (surface Ig on B cells) or major histocompatibility complex (MHC)-antigenic peptide complexes (the TCR on T cells) and allow lymphocytes to respond to injury at any site in the body (1–3). The recognition of cancer cells by lymphocytes may result in a different pattern of gene expression that can be associated with the nature and site of an early tumor (4). Our group has pursued the use of blood lymphocytes as diagnostic reporters of disease, including autoimmune diseases in man and experimental cancer in mice, through gene expression profiling (5–7).
Microarray technology is the ideal method for the determination of gene expression profiles because of its sensitivity and robustness, and it allows for the measurement of thousands of genes in a single array. Moreover, the bioinformatic analyses of microarrays that are currently available allow for the accurate hierarchical organization of experimental samples and genes.
Our aim in this study was to evaluate the gene expression profiles in peripheral lymphocytes in response to a small number of fibrosarcoma cells in immunocompetent BALB/c mice before the development of a visible tumor mass. We used an in vivo model system in which BALB/c mice received subcutaneously injections of transformed B61 cells (3T3 fibroblasts transfected with a mutant p21 Ha-ras oncogene) (8, 9), which, over time, develop into fibrosarcoma at the site of injection (5).
For the gene expression profiling, we used a glass slide cDNA microarray containing 4500 sequences that represent the main biological functions of the murine immune system, including interleukins and their receptors; genes involved in cell cycle control, metabolism, signal transduction, DNA repair/recombination and apoptosis; and expressed sequence tags (ESTs) of unknown function.
Using a significance analysis of microarrays (SAM) algorithm (10) and ANOVA (11), we were able to statistically validate the microarray data and to identify genes that were induced or repressed. Hierarchical clustering was useful in comparing the different hybridization signatures observed between the mice subjected to bacterial infection or inflammation and the mice injected with B61 fibrosarcoma cells.
By comparing the flow cytometric analysis of peripheral blood lymphocytes with the microarray data, we demonstrate that specific gene expression signatures could be associated with each treatment group, despite the absence of changes in the number of T (CD3+CD4+ or CD3+CD8+) and B (CD19+) cells. These data demonstrate that the blood lymphocyte transcriptional response, as determined by the microarray assay, is a sensitive and specific method for the early detection of tumor cells in vivo.
Materials and Methods
Ha-ras-1–Transformed B61 Cell Line.
Mouse BALB/3T3 fibroblasts were transfected with a plasmid expressing a mutated human c-Ha-ras-1 oncogene (pEJ, which expresses the p21 ras protein with a G→V substitution at position 12). This transformed cell line, designated B61, contains 50–100 copies of the active mutant c-Ha-ras-1 oncogene and was generated by Kovary and colleagues (8, 9). The B61 adherent cells were cultured in DEMF10 medium supplemented with 10% fetal bovine serum plus antibiotics in an incubator with 5% CO2 at 37°C for 1 week. The cultures were trypsinized and supplemented with fresh medium at 3-day intervals.
Injection of B61 Fibrosarcoma Cells into Mice, Peripheral Lymphocyte Separation, and Extraction of Total RNA.
We injected 0.1 ml of sterile nonpyrogenic saline containing 103 or 105 B61 cells subcutaneously (sc) in the scapular region of 4- to 6-week-old female immunocompetent BALB/c mice.
Three days after injection, before the emergence of any tumor mass at the site of injection, peripheral blood was collected (approximately 0.9 ml per animal) and pooled for a total of 10 ml. The mononuclear cells were separated by centrifugation on a Ficoll-Hypaque (GE Healthcare, Uppsala, Sweden) cushion, resuspended in DEM medium, and incubated at 37°C in a tissue culture plate (Corning, Monterey, Mexico). The nonadherent cells, consisting mainly of lymphocytes, were collected, and total RNA was isolated by using the Trizol reagent (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. All RNA samples were extracted 3 days after tumor inoculation. The integrity of the RNA samples was evaluated by conventional agarose gel electrophoresis, and only those preparations free of contaminating DNA, proteins, or phenol were used (data not shown). The gene expression experiments were independently repeated three times, as required by the SAM program (10), thereby reducing the false-discovery rate. All manipulations of mice were approved by the Animal Research Ethical Committee of the University of São Paulo, Campus of Ribeirão Preto, Brazil.
Bacterial Infection, Inflammation, and Control Groups.
Total RNA was also extracted from the peripheral lymphocytes isolated from groups of BALB/c mice subjected to bacterial infection or inflammation. Mice in the bacterial infection group received an injection given sc of a culture of total bacteria from mouse feces (107 colony forming units [cfu] in 0.1 ml sterile apyrogenic saline per animal), and mice in the inflammation group received an injection of Zymozan (Sigma, St. Louis, MO) (500 μg in 0.1 ml sterile apyrogenic saline per animal) given sc. Mice that received an. injection of 0.1 ml sterile apyrogenic saline given sc served as controls. As described for animals injected with tumors cells, all RNA samples were extracted 3 days after injection.
Flow Cytometry Analysis.
Peripheral blood mono-nuclear cells (1 × 107 cells per milliliter) from mice in each group were incubated for 40 mins at 4°C with Fc block (1 μg per 106 cells; Pharmingen, San Diego, CA). Cells were incubated with the appropriate monoclonal antibody (0.75 μg per 106 cells) for 30 mins at 4°C in the dark. The following anti–mouse cell surface marker antibodies were used: (i) PE-conjugated anti-CD8 (clone 53–6.7), anti-CD19 (clone 1D3), and anti-IgG2a (G155–178); and (ii) FITC-labeled anti-CD3 (clone 145–2C11), anti-CD4 (RM4–5), and anti-IgG2a (clone R35–95). All antibodies were purchased from Pharmingen and used according to the manufacturer’s instructions. The cells were analyzed with a FACsVantage flow cytometer (Becton Dickinson, Foster City, CA) using the FACSort software (Becton Dickinson). The cells were defined according to size (forward scatter), granularity (side scatter), and fluorescence intensity. The number of CD3+, CD4+, CD8+, and CD19+ cells was then determined. The flow cytometry experiments were independently repeated 10 times.
cDNA Microarray Method.
The gene expression of lymphocytes was assessed by using glass slide cDNA microarrays according to standard protocols (12). The arrays were prepared on silane-coated UltraGAPS slides (#40015, Corning, New York, NY) containing a total of 4500 target cDNA sequences. These sequences were from the Soares thymus 2NbMT normalized library, represented EST cDNA clones prepared from the thymus of a 4-week-old male C57BL/6J mouse, and are available at the IMAGE Consortium (http://image.llnl.gov/image/html/iresources.shtml). The cDNA inserts were homogeneous in size (near 1 kb), cloned into three vectors (pT7T3D, pBluescript, and Lafmid), and amplified in 384- or 96-well plates by using vector-PCR amplification with the following primers that are specific for the three vectors: LBP 1S GTGGAATTGTGAGCGGATACC forward and LBP 1AS GCAAGGCGATTAAGTTGG reverse.
The microarrays were prepared by using PCR products from the cDNA clones and a Generation III Array Spotter (Amersham Molecular Dynamics, Sunnyvale, CA).
Complex cDNA Probe Preparation and Hybridization.
The cDNA complex probes derived from the total RNA isolated from lymphocytes of each group were prepared by reverse transcription using 10 μg of total RNA and were labeled with Cy3 fluorochrome using the CyScribe postlabeling kit (GE Healthcare). The hybridization took place for 15 hrs and was followed by washing with an automatic slide processor system (ASP, Amersham Biosciences). The washed microarrays were then scanned with a Generation III laser scanner (Amersham Biosciences). As a reference for the hybridization procedure, we used equimolar quantities of cDNA generated from unrelated total RNA (mouse thymus total RNA). This approach allowed us to estimate the amount of cDNA target in each microarray spot.
A complete file providing all genes and ESTs present in the microarrays used in this study is available online (www.rge.fmrp.usp.br/passos/mmu_array).
cDNA Microarray Data Analysis.
Microarray image quantification was performed by using Spotfinder software (http://www.tm4.org/spotfinder.html). The normalization process was carried out by using the R platform (http://www.r-project.org), and the statistical data were analyzed by the Multiexperiment Viewer (MeV) software (version 3.1; available online at http://www.tm4.org/mev.html) (13).
To analyze the gene expression patterns, we used an unsupervised hierarchical clustering method that grouped genes on the vertical axis and samples on the horizontal axis on the basis of similarity in their expression profiles. The similarities and dissimilarities in gene expression are presented as dendrograms (in which the pattern and length of the branches reflect the relatedness of the samples or genes) and as heat maps that were generated by Cluster version 3.0 and Java Tree View (http://rana.lbl.gov/EisenSoftware.htm).
Oligonucleotide Primer Design and Quantitative Real-Time Polymerase Chain Reaction (PCR) for Microarray Data Confirmation.
Two genes that were induced and two that were repressed in the mice given injections of 105 B61 cells were selected on the basis of the hierarchical clustering–based expression pattern. The cDNA sequences of the selected genes were retrieved from the National Center for Biotechnology Information GenBank database (http://www.ncbi.nlm.nih.gov) by using the following accession numbers: Ada, NM_007398.3; Cdk4, NM_009870; and Parp3, NM_145619.2. The Primer3 web tool (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) was used to select pairs of oligonucleotide primers that spanned an intron and exon junction and had an optimal melting temperature of 60°C. The following pairs of primers were used: Ada (3′ CCTATGAGGGCGCAGTAAAG 5′ and 3′ TAACCATGTCCCACCCTTC 5′), Cdk4 (3′ CAATGTTGTACGGCTGATGG 5′ and 3′ GGAGGTGCTTTGTCCAGGTA 5′), and Parp3 (3′ CCTGCTGATAATCGGGTCAT 5′ and 3′ TTGTTGTTGCCGATGTT 5′). The cDNA samples were prepared by using the SuperScript II enzyme as recommended by the manufacturer (Invitrogen). The expression of the genes mentioned above was quantified by a 7500 Real-Time PCR System (Applied Biosystems). The expression was normalized to the expression of the housekeeping gene Gapdh (NM_008084; 3′ GGGTGTGAACCACGAGAAAT 5′ and 3′ CCTTCCACAATGCCAAAGTT 5′). The real-time PCR experiments were independently repeated 10 times.
Data Mining.
Data mining, including the gene ontology, of the differentially expressed genes (induced or repressed) that characterized each group of mice was performed by using the SOURCE (http://source.stanford.edu/cgi-bin/source/sourceSearch) and DAVID (http://apps1.niaid.nih.gov/david) databases.
Results
Peripheral Lymphocyte Subsets.
As shown in Figure 1, there were no significant changes in the T (CD3+CD4+ or CD3+CD8+) and B (CD19+) cell subsets among mice given sc injections of Zymozan (inflammation), the bacterial suspension (infection), or B61 fibrosarcoma cells (103 or 105 cells per animal) and those in the control group, which was given injections of only sterile saline. All analyses were carried out 3 days after the respective injections. This time point was long enough to elicit local inflammation in the Zymozan-injected mice and peritoneal infection in the infection mice, but no palpable or biopsy-identifiable tumor was found at the site of injection in the B61 cell–injected mice.
Peripheral Lymphocyte Transcriptional Response.
From the set of the 4500 cDNA sequences present in the microarray used to determine the differential gene expression profile, we selected only those sequences that were present in 80% of the microarray hybridizations (n = 3600) for statistical analysis. To determine the variation in the gene expression in the full data set and to identify genes that were differentially expressed by the different groups of mice, we used two different statistical methods: a SAM algorithm (10) and a one-way ANOVA (11).
Using the multiclass SAM algorithm with a 5% false-discovery rate, we identified 425 genes that were significantly and differentially expressed (Fig. 2A). The resulting data were further analyzed by using an unsupervised hierarchical clustering method, which discriminated the groups of mice on the basis of their respective expression signatures. The first SAM sample cluster included the control group (saline-injected mice) and the group of mice given injections of 103 B61 tumor cells. The second SAM sample cluster included the infection and inflammation groups, which were positioned together because of their similar expression signatures. Finally, the third SAM sample cluster was the only exclusive cluster and contained the group of mice given injections of 105 B61 tumor cells.
Using the ANOVA approach (bootstrapping with 5000 permutations), we identified 711 informative genes that exhibited P < 0.05 (Fig. 2B). On the basis of this analysis, we again used the unsupervised hierarchical clustering method to analyze the data.
Similar to the results obtained with the SAM data, we obtained three sample clusters. The first ANOVA sample cluster included the control group (saline-injected mice) and the group given injections of 103 B61 tumor cells, which clustered together. The second ANOVA sample cluster included the infection and inflammation groups. Although these two groups displayed disparate expression signatures, the dissimilarities were not discriminatory, and the groups still clustered together. Finally, the group given injections of 105 B61 tumor cells comprised the third ANOVA sample cluster. The animals from this group featured an expression signature significantly different from those of all of the other groups, forming a distinct cluster.
Overall, both hierarchical clustering analyses positioned the groups of mice into three large clusters (Fig. 2A and B), and both statistical approaches demonstrated that gene expression of peripheral lymphocytes was different according to the treatment. Although the number of genes was distinct for each statistical approach, both analyses yielded similar results, further emphasizing the relevance of the data.
A one-way ANOVA was applied to evaluate the variation in the gene expression in the full data set. Bootstrapping from 1000 permutations and a significance level of P < 0.05 were used to select 711 genes. The data from these 711 genes were then analyzed by using unsupervised clustering. Using the same full data set and the SAM algorithm (false-discovery rate < 0.05 and 1000 bootstrapped permutations), we found 425 significantly and differentially expressed genes. Interestingly, 246 significantly and differentially expressed genes as determined by the SAM algorithm were included among the 711 informative genes as determined by the ANOVA analysis. This result statistically validates the microarray data. A file providing the entire quantitative normalized microarray data set obtained in this study is available online (http://rge.fmrp.usp.br/passos/earlyB61/dataset).
Gene Expression Assessed by Real-Time PCR.
The Ada, Cdk4, and Parp3 genes were selected to confirm the results obtained by the cDNA microarray method. These genes were chosen on the basis of their statistical significance determined by the SAM algorithm. The real-time PCR analysis confirmed that these genes were induced (Ada and Cdk4) or repressed (Parp3) in the group of mice in which 105 B61 tumor cells were injected (Fig. 3).
Two major conclusions may be drawn from these results. First, the different treatments produced specific expression signatures, as detected by two distinct hierarchical clustering analyses. Second, at least 246 differentially expressed genes were detected and validated by two distinct statistical approaches. Some of these validated genes, which could be informative for the presence of the B61 cells, were chosen for discussion (Table 1).
Discussion
There is great interest in uncovering new methods for accurate detection of early cancer in asymptomatic patients. To illustrate this interest, Sharma et al. (14) considered breast cancer, where mammographic screening is the most consistent method to detect tumors. Although effective, it has important limitations. For example, in absence of microcalcification, mammography often fails to detect tumors that are less than 5 mm in size. Therefore, these authors developed a method for breast cancer detection based on microarray gene profiling of peripheral blood cells in asymptomatic individuals.
Most published results of cancer diagnosis/prognosis based on gene expression have involved clinical samples (15). We agree with the authors who believe that obtaining samples for clinical purposes requires a prior knowledge of both their presence and their location in the body. Therefore, a gene expression-based assay not requiring samples from the diseased area remains a focus of research.
We previously demonstrated that a primary immune system organ such as the thymus exhibits a differential transcriptional response in the presence of fibrosarcoma developing at a distant site in the body (5). This observation led us to search for an assay based on the gene expression profile of immune system cells for the detection of cancer.
Because B61 tumor cells require at least 6 days to form a palpable tumor mass at the site of injection (5), we chose 3 days after the injection of tumor cells as a single time point to analyze the gene expression profile. This time point assured the presence of tumor cells in the animals in absence of a tumor mass.
Circulating blood cells, mainly lymphocytes, represent a logical choice of cells to investigate because of their vigilant and comprehensive surveillance for signs of infection or other threats, including cancer (16). In addition, blood is easily obtained in a noninvasive manner.
Interestingly, we have observed differential gene expression profiles as early as 3 days after injection of tumor cells. T-cell responses often take 7 days to develop, whereas innate immune responses develop faster but involve other cells. Therefore, the expression profile observed may be implicated in the molecular mechanisms of such cells before the cellular response.
A recent study using a proteomic approach based on plasma protein fractionation demonstrated that an in-depth proteomic analysis may provide a useful strategy for early cancer detection in a genetically engineered mouse model and in patients with pancreatic cancer (17). In the present study, we demonstrate a simpler strategy based on microarray transcriptome profiling, which may be of more practical use. Our aim was to define specific hybridization signatures that characterized mice injected with fibrosarcoma tumor cells. The infection and inflammation groups were also assayed to determine the specificity of the observed hybridization signature. Because different tumors produce particular antigens, the lymphocyte gene expression profile in response to different tumor cell lines, although not included in this study, needs be investigated.
Another interest was to investigate whether at an early time point there were alterations in the frequency of the main sets of circulating lymphocytes, which could possibly explain the alterations in gene expression. Figure 1 demonstrates that the number of T (CD3+CD4+ or CD3+CD8+) and B (CD19+) cells did not significantly change between the control, inflammation, infection, and tumor cell-injected groups of mice.
The unsupervised analysis of the differentially expressed genes allowed for the detection of specific hybridization signatures and, consequently, hierarchical clustering of these groups. These findings strongly suggest that the differential hybridization signatures observed is a result of the early modulation of gene expression in peripheral lymphocytes.
The SAM algorithm (10), which is based in the classic t test, was specifically adapted for the high-throughput analysis of microarray data and is widely used for the identification of differentially expressed genes. ANOVA (11) was used in parallel as a second method to validate the results obtained by the SAM algorithm (Fig. 2).
Although the microarray used in this study is not widely used by researchers because it was prepared in our laboratory, it identified genes that were expressed at significantly different levels. Furthermore, the data from this array were statistically and experimentally validated by traditional methods such as the SAM algorithm (10) and quantitative real-time PCR.
Figure 2 demonstrates that the group of mice into which 103 B61 tumor cells were injected clustered together with the control group, which indicates a similar hybridization signature. This result suggests that 103 B61 tumor cells did not elicit a specific gene response in peripheral lymphocytes.
The infection and inflammation groups of mice were positioned in the same cluster and exhibited similar hybridization signatures. Although these two groups were tested to better define the specificity of the hybridization signatures, a general conclusion may be drawn from this result: host responses to bacterial infection or inflammatory stimuli share common mechanisms and/or effector immune cells that modulate a common set of genes. Table 1 lists the genes identified by both statistical approaches for each group of mice.
Finally, we were able to identify a set of genes that were exclusively modulated in the lymphocytes of mice injected with 105 B61 tumor cells, which were characterized by their specific hybridization signature (Fig. 2A and B). On the basis of the statistical significance of their expression profile, some genes were selected for discussion.
Among the induced genes, we have highlighted Rps11 (ribosomal protein S11, accession number NM_013725), which encodes a protein constituent of the ribosome, and Ppp2r5c (protein phosphatase 2, regulatory subunit B [B56], gamma isoform, accession number NM_012023), which participates in the signal transduction cascade. These two genes participate in the regulation of the cell cycle. We also identified Cdk4 (cyclin-dependent kinase 4, accession number NM_009870), another gene involved in the regulation of the cell cycle by cyclin-dependent protein kinase activity. Finally we detected the induction of the Ada gene (adenosine deaminase, accession number NM_007398.3). The protein encoded by this gene is important in the development and function of the immune system in humans and mice (18).
Among the repressed genes we have highlighted Mre11a (meiotic recombination 11 homolog A, accession number NM_018736.2) and Parp3 (poly ADP-ribose polymerase family, member 3, accession number NM_145619.2), which encode proteins involved in the response to DNA damage. The Phtf1 transcription factor (accession number NM_013629), which encodes an enzyme involved in protein amino acid phosphorylation, was also repressed in lymphocytes following tumor cell injection.
These induced and repressed genes represent candidate genes for the detection of cancer by a gene profiling assay. These genes may be useful in further investigations using an in vivo carcinogenesis model (e.g., colon carcinoma induced by azoxymethane and dextran sodium sulfate treatment of mice) (19). Although an orthotopic model of cancer could be more physiologically relevant, it features intrinsic experimental limitations that would have made the analysis presented in this paper difficult. Specifically, the identification of specific animals bearing tumor cells before the appearance of a tumor mass is difficult in an orthotopic model. As this study aimed to examine the expression profile of lymphocytes before a tumor could be detected, we used a system in which the number of tumor cells could be precisely controlled.
The statistically significant shift in the gene expression profile of peripheral lymphocytes shortly after the injection of cancer cells validates the development of an assay that uses microarray-based gene expression profiling of peripheral blood immune cells as a method of early detection of cancer. Moreover, these results serve as a first step in the development of a study to examine the peripheral lymphocytes of patients with cancer and the development of a data bank with the expression signatures corresponding to the different cancers. This could be useful as a reference data bank of lymphocyte gene expression signatures in response to cancer.
Genes Differentially and Significantly Expressed in Peripheral Blood Lymphocytes of BALB/c Mice at an Early Stage of Fibrosarcoma, as Detected by SAM or ANOVA Statistical Tests (P ≤ 0.05)
Flow cytometric analysis to determine the number of CD3+, CD4+, CD8+, and CD19+cells in the blood of mice given sc injections of different numbers of B61 tumor cells and the control groups. No significant changes were detected (ANOVA, P ≤ 0.05).
Dendrograms of samples constructed on the basis of the ANOVA (A) or the SAM algorithm (B) confirming the separation of the different groups of mice given injections of different substances (B61 tumor cells, Zymozan [inflammation] or bacterial suspension [infection]) and control mice on the basis of gene expression signatures.
Quantitative real-time PCR was used to confirm the induction of Ada and Cdk4 and the repression of Parp3 in the mice into which 105 B61 cells were injected. The expression was compared with that in the B61 103, inflammation, infection, and control groups of mice. The expression levels were normalized to Gapdh expression (n = 0; mean ± standard error of the mean; one-way ANOVA, * P < 0.001).
Footnotes
MMCM was recipient of a FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo, Brazil) fellowship. This project was funded by FAPESP and CNPq (Conselho Nacional de Desenvolvimento Cientìfico e Tecnológico, Brazil).
Acknowledgements
The flow cytometry and real-time PCR were performed in the laboratories of Dr. Célio Lopes Silva and Dr. Zilá Luz Paulino Simões, respectively, at the University of São Paulo campus of Ribeirão Preto, Brazil. The cDNA clones used for the microarray preparation were kindly provided by Dr. Catherine Nguyen from the INSERM (Institut National de la Santé et de la Recherche Médicale) U928, Marseille, France. This work is a part of the Ph.D. thesis of MMCM, which was awarded in 2008 by the CAPES (Coordenação de Aperfeiçoamento Pessoal de Nível Superior)/Ministry of Education, Brazil.