SEARCH

SEARCH BY CITATION

Keywords:

  • needle biopsy;
  • comparison;
  • microarray;
  • transcriptional profiles

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

BACKGROUND

Gene expression profiling should be applicable to needle biopsy samples if microarray technology is to become practically useful for clinical research or management of breast carcinoma. This study compared gene expression profiles derived from fine-needle aspiration biopsy (FNAB) and from core needle biopsy (CBX).

METHODS

Total RNA was extracted from single FNAB and CBX samples. Corresponding pairs of FNAB and CBX were analyzed for similarity of gene expression profiles using cDNA microarrays that contain 30,721 human sequences. A subset of genes that distinguished CBX samples from FNAB samples was evaluated in a larger group of needle biopsy samples and in a published genomic database derived from 78 sporadic breast carcinomas with known clinical outcome.

RESULTS

Sixty-eight patients with newly diagnosed breast carcinoma were included in the current study. Sixty-five patients underwent FNAB (17 had both FNAB and CBX) and 3 underwent CBX only. Extracted RNA was of suitable quality for hybridization in 46 (71%) FNABs and 15 (75%) CBXs. Total RNA yield in those samples was similar for single-pass FNAB (mean = 3.6 μg and median = 2.2 μg; n = 46) and CBX (mean = 2.8 μg and median = 2.0 μg; n = 15), with 1 μg or more of total RNA in all cases. Transcriptional profiling was performed successfully in all cases when it was attempted, in a total of 50 samples (38 FNABs and 12 CBXs), including matched FNAB and CBX samples from 10 patients. There were differences in gene expression profiles in 10 matched FNAB and CBX sample pairs. Genes that were expressed differently in CBX samples, compared with FNAB samples, were recognized as being predominantly from the endothelium, fibroblasts, myofibroblasts or smooth muscle, and histiocytes. Corresponding microscopic cell counts from FNABs demonstrated means of 80% tumor cells, 15% lymphocytes, and 5% stromal cells, whereas CBXs contained 50% tumor cells, 20% lymphocytes, and 30% stromal cells. Considering that CBXs are approximately six-fold richer in nonlymphoid stromal cells than FNABs and that CBXs differentially express a set of recognized stromal genes, the authors used these biopsies to define a transcriptional profile of breast carcinoma stroma. A set of 120 genes differentially expressed in CBXs was assessed independently in a published breast carcinoma genomic database to classify breast carcinomas based on stromal gene expression. Subgroups of tumors with low or high stromal signal were identified, but there was no correlation with the development of systemic metastases within 5 years.

CONCLUSIONS

Both FNAB and CBX yield a similar quality and quantity of total RNA and are suitable for cDNA microarray analyses in approximately 70–75% of single-pass samples. Transcriptional profiles from FNAB and CBX of the same tumor generally are similar and are driven by the tumor cell population. The authors concluded that each technique has relative advantages. The FNABs provide transcriptional profiles that are a purer representation of the tumor cell population, whereas transcriptional profiles from CBXs include more repre sentation from nonlymphoid stromal elements. Selection of the preferred needle biopsy sampling technique for genomic studies of breast carcinomas should depend on whether variable stromal gene expression is desirable in the samples. Cancer 2003;97:2960–71. © 2003 American Cancer Society.

DOI 10.1002/cncr.11435

Microarray transcriptional profiling is a powerful, high-throughput technology that can generate a substantial amount of genomic information from a single cancer sample. Comprehensive gene expression analyses promise to influence our understanding of classification, prognosis, and prediction of response to treatments for breast carcinoma.1–5 This technology has been applied mostly to stored macroscopic samples of breast carcinoma from surgical resection specimens.1, 2, 4, 6, 7 However, future clinical or research applications of this genomic technology will likely be ancillary to routine diagnostic studies and may be limited to an additional pass from a fine-needle aspiration biopsy (FNAB) or from a core-needle biopsy (CBX). Pilot studies indicate that it is possible to obtain enough RNA for transcriptional profiling from FNAB and CBX samples of some breast carcinomas.5, 8, 9 At least 1 μg of total RNA is now acceptable for some microarray protocols. Ellis et al.10 reported a median of 1.34 μg total RNA yield from single-passage CBX. In a different report, only 15% of FNABs were estimated to have sufficient RNA for transcriptional profiling.8 Preliminary steps to assess the RNA yield from small biopsy specimens can pose a challenge to the use of those samples for transcriptional profiling experiments. The amount of RNA required to be loaded for electrophoresis (to determine RNA quality) can leave insufficient remaining RNA from small biopsy samples for microarray hybridization.5, 8, 9 However, new microcapillary bioanalysis technology does allow for accurate measurement of RNA yield and quality from small biopsy samples, while leaving most of the RNA in the sample available for hybridization experiments.

We investigated how reliably a single FNAB or CBX sample obtained in a clinical biopsy setting can yield sufficient RNA for transcriptional profiling with a cDNA microarray. These baseline data of RNA yield will be a useful guide to investigators planning studies that utilize clinical sample procurement for transcriptional profiling. We compared the cellular composition and transcriptional profiles derived from FNAB and CBX samples of breast carcinoma and described important differences in the expression of some genes. Differentially expressed genes then were evaluated in a separate published database from breast carcinoma tumor tissue samples with annotated clinical outcome, to look for heterogeneity of expression and possible prognostic significance.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Patients

Needle biopsy samples of breast carcinoma were collected for transcriptional profiling as part of a clinical trial at The University of Texas M. D. Anderson Cancer Center (MDACC). These samples were from tumors of women with newly diagnosed, AJCC Stage I–III, palpable breast carcinoma before they underwent any systemic therapy. Clinical characteristics are summarized in Table 1. This study was approved by the MDACC institutional review board and all patients signed a specific informed consent form.

Table 1. Clinical Characteristics of Patients and Tumors
CharacteristicsNo.
  1. +: positive; −: negative; IHC: immunohistochemistry; FISH: fluorescent in situ hybridization.

Age (yrs) 
 Median50
 Range29–77
Race 
 Caucasian46
 African American8
 Hispanic7
 Asian7
Histology 
 Invasive ductal62
 Invasive lobular6
Clinical stage before therapy 
 Tumor size 
 T18
 T238
 T314
 T48
 Lymph node positive39
 Lymph node negative29
Estrogen receptor status 
 ER +38
 ER −30
HER-2/neu status 
 IHC score 0–1+42
 IHC score 2+11
 IHC score 3+7
 FISH not amplified39
 FISH amplified12
 FISH not performed17

FNAB Samples

The FNAB was performed by an experienced cytopathologist using a 23-gauge or 25-gauge aspiration needle (the applied local anesthesia was an ethyl chloride spray). Samples were collected for this and other planned studies. Generally four FNAB samples were obtained. Two samples were each placed into two vials containing RNAlater solution (Ambion, Austin, TX) to preserve RNA, 1 sample was pass snap frozen in a plain vial and stored at −80 °C, and the sample obtained from first pass was prepared as 10–20 cytologic smears on glass slides. The samples in RNAlater solution were kept at room temperature for 20–30 minutes, then snap frozen and stored at −80 °C. The adequacy and cellularity of the first FNAB pass was assessed by the cytopathologist, who evaluated two Diff-Quik–stained (Baxter Scientific, McGraw Park, IL) cytologic smears. The cellular composition of the FNAB samples was determined by cell counts of the Diff-Quik–stained slides using light microscopy at high magnification (×400). In a representative area of the slide, the first 300 cells were classified as tumor cells, lymphocytes, or stromal cells. Those results were recorded as percentages.

CBX Samples

The CBX tissue samples were obtained for transcriptional profiling in some patients who also required a histologic diagnosis of their breast carcinoma. Six tissue core samples were obtained by an experienced cytopathologist using a 14-gauge core needle device (Tenmo, Bauer Medical International, Santa Domingo, Dominican Republic) through a small skin incision in a sterile field (the applied local anesthesia was an injection of xylocaine). The adequacy of these core samples was assessed at the time the biopsy was performed by the cytopathologist using touch imprint cytology stained with Diff-Quik. Three tissue core samples were processed for routine histolologic diagnosis. They were fixed in formaldehyde, embedded in paraffin, 4-μm sections, and stained in hematoxylin and eosin (H & E). The cellular composition was determined by cell counts from representative cancerous areas in the H & E-stained slides. The remaining three core samples were utilized for transcriptional profiling. Two core samples were placed in two separate vials containing RNAlater solution and one core sample was snap frozen in a plain vial and stored at −80 °C. The samples in RNAlater solution were kept at room temperature for 20–30 minutes then snap frozen and stored at −80 °C. When both FNAB and CBX were performed, the FNAB was performed immediately before the CBX.

RNA Isolation and cDNA Microarray Hybridization

RNA was extracted from one of the single-pass FNAB or CBX samples from each patient, using the RNAeasy Kit (Qiagen, Valencia, CA). Samples stored in RNAlater were preferentially used within 6 months of the biopsy. The amount and quality of RNA were assessed with an Agilent 2100 Bioanalyzer RNA 6000 LabChip kit (Agilent Technologies, Palo Alto, CA). First-strand cDNA synthesis was performed with Superscript II (Invitrogen, Carlsbad, CA) in the presence of [100 mCi/mL] 33P-dCTP (Amersham, Little Chalfont, UK) from 1–2 μg of total RNA. The generated cDNA probes were hybridized (without further amplification) to high-density cDNA microarray membranes proprietary to Millennium Pharmaceuticals (Cambridge, MA). The membranes contained 30,721 human sequence clones, including 10,890 expressed sequence tags (EST) obtained from Unigene (National Center for Biotechnology Information, Bethesda, MD) and verified by direct sequencing.

Data Analyses

The cDNA microarray hybridization method generated expression data from a single RNA sample and did not use control RNA for comparison. Raw gene expression values were log transformed and mathematically normalized to a median value of 1.0 for each membrane, to enable comparison of relative expression of genes in breast carcinoma samples. Sequences with a low absolute expression in an individual sample may produce unreliable results due to interference from the background noise of experiments and because small absolute differences in expression can produce artificially high relative differences in expression. Therefore, sequences with normalized expression values below 0.3 were assigned arbitrarily an expression value of 0.3. Consequently, small absolute differences in minimally expressed sequences were no longer detectable and did not skew the results. Genes and ESTs with known “alu” repeats were removed because of likely cross-hybridization between those sequences. Alu repeats are transposon-derived short interspersed repetitive sequences that preferentially insert in regions with a high GC content and range from 100 to 400 base pairs. Relative expression data were restricted further to only retain sequences that had variable relative expression in different breast carcinoma samples. Sequences with a normalized expression value ≥ 2.0 in at least 2 samples (from 20 matched FNAB and CBX samples) were retained for analysis. These 2 filters (removing alu repeats and sequences without significant expression) resulted in 10,993 sequences (approximately 36% of the 30,721 genes analyzed) for hierarchical cluster analysis and calculation of Pearson correlation values. To identify genes that distinguish CBX from FNAB samples, we were less stringent. Genes and ESTs with alu repeats were removed, but sequences with expression values ≥ 1.0 in at least 2 samples were retained. These two filters (which removed alu repeats and sequences without significant expression) resulted in 20,023 sequences (approximately 65% of the 30,721 genes analyzed) for analysis.

We used hierarchical cluster and treeview programs (Stanford University Office of Technology Licensing, Palo Alto, CA) to compare the expression profiles from different samples. Essentially, hierarchical cluster analysis compares each sample with all the other samples. We also calculated the Pearson correlation statistic to compare the gene expression profiles of 10 matched pairs of FNAB and CBX samples from the same tumor. To identify sequences that distinguish CBX samples from matched FNAB samples, we calculated the signal-to-noise ratio (SNR) value for each sequence ([μCBX − μFNA]/[σCBX + σFNA]) and expressed SNR as an absolute value. For this calculation, μ and σ represent the mean and standard deviation of expression from the 10 samples in the CBX class and the 10 matched samples in the FNAB class. Using this method, sequences that are expressed differentially in CBX and FNAB samples have higher SNR values. All 20,023 sequences then were ranked in descending order of SNR value. The marker gene lists of 51 and 200 genes were derived from those rankings of SNR values.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Total RNA Yield from Needle Biopsy Samples

Sixty-eight patients with newly diagnosed breast carcinoma were sampled by needle biopsy for this study. Sixty-five patients underwent FNAB (including 17 who had both FNAB and CBX) and 3 other patients underwent CBX only. Of these, two FNAs and one CBX were not submitted for analysis of RNA because the tumor sampling was considered to be inadequate when microscopic study of the corresponding slides did not demonstrate adequate cancer cellularity. In addition, the initial sample of 11 FNABs and 2 CBXs in RNAlater did not contain satisfactory quality RNA (i.e., indistinct 18S and 28S subunits), so a second (back-up) sample was analyzed. However, in 9 of 11 FNAB sample and 1 of 2 CBX samples, the RNA yield from that second sample also was unsatisfactory. RNA was extracted and found to be of suitable quality for hybridization in 46 of 65 (71%) FNAB samples and in 15 of 20 (75%) CBX samples. The total RNA yield in those samples was similar for single-pass FNABs (mean = 3.6 μg and median = 2.2 μg; n = 46) and CBXs (mean = 2.8 μg and median = 2.0 μg; n = 15) and they all contained at least 1 μg of total RNA (a sufficient sample for our cDNA microarray). Frequency histograms (Fig. 1) demonstrate the distribution of total RNA yields for each needle biopsy technique.

thumbnail image

Figure 1. Frequency histograms demonstrate similar total RNA yields from 46 of 65 (71%) fine-needle aspiration biopsy samples (A) (mean = 3.6 μg and median = 2.2 μg total RNA) and from 15 of 20 (75%) core-needle biopsy samples (B) (mean = 2.8 μg, median = 2.0 μg). Count (Y-axis) indicates the number of samples with a particular total RNA yield (X-axis).

Download figure to PowerPoint

There were 65 FNAB samples from 65 patients, 46 of which had sufficient RNA. Of these 46 samples, 38 were selected for transcriptional profiling because these patients participated in the associated clinical trial. There were 20 CBX samples from 20 patients, 15 of which had sufficient RNA. Of these 15 samples, 12 were selected for transcriptional profiling. Seventeen patients underwent both FNAB and CBX (matched pairs). Thirteen FNAB samples had sufficient RNA. Ten matched CBX samples also had sufficient RNA. These 10 matched FNAB and CBX samples were selected for transcriptional profiling.

Differences in Cellular Components

Microscopic cell counts were performed on the available slides from the 38 FNAB and 12 CBX samples that had been transcriptionally profiled. The counts demonstrate that there are more stromal cells (fibroblasts, endothelial cells, histiocytes, adipocytes) present in CBX samples (Fig. 2). The mean percentage of each cell type from each sample technique is as follows: 80% tumor cells in FNAB samples versus 50% tumor cells in CBX samples; 15% lymphocytes in FNAB samples versus 20% lymphocytes in CBX samples; and 5% stromal cells in FNAB samples versus 30% stromal cells in CBX samples. These results demonstrate that FNAB samples are enriched in tumor cells and contain fewer nonlymphoid stromal elements than the CBX samples. This difference in the nonlymphoid stromal cell component is illustrated in photomicrographs from two selected FNAB and CBX pairs (Fig. 3).

thumbnail image

Figure 2. Cellular composition (% cells) from microscopic counts performed on slides from fine-needle aspiration biopsies (A) compared with core-needle biopsies (B). Cells were categorized into three groups: tumor cells (red), lymphocytes (blue), and other stromal cells (yellow).

Download figure to PowerPoint

thumbnail image

Figure 3. Photomicrographs demonstrate the differences between fine-needle aspiration biopsy (FNAB; Diff-Quik stain) and core-needle biopsy (CBX; H & E stain) sample types. (A, B) Sample from Patient 120 whose FNAB and CBX samples had a Pearson correlation value of 0.34 and clustered separately (see Fig. 4). (A) An FNAB smear with large adenocarcinoma cells and scattered small lymphocytes. (B) A CBX section with invasive nests of adenocarcinoma cells surrounded by desmoplastic stromal tissues. (C, D) Sample from Patient 177 whose FNAB and CBX samples had a Pearson correlation value of 0.75 and clustered together (see Fig. 4). (C) An FNAB smear with adenocarcinoma cells and scattered small lymphocytes (similar to A). (D) The CBX section with nests of adenocarcinoma cells and associated desmoplastic stromal tissues (similar to B) (Original magnification, ×400).

Download figure to PowerPoint

thumbnail image

Figure 4. Unsupervised hierarchical cluster analysis of 10 matched fine-needle aspiration biopsy (FNAB) and core-needle biopsy sample pairs using 10,993 genes demonstrates that 7 of 10 sample pairs have closely related transcriptional profiles. The other three CBX samples (that did not cluster with their respective FNAB samples) actually clustered together (Samples 157, 120, and 117).

Download figure to PowerPoint

Gene Expression in Paired FNAB and CBX Samples

Transcriptional profiling has been performed successfully in all 50 samples (38 FNAB and 12 CBX samples). Overall transcriptional profiling of the 38 FNAB samples yielded reliable results, and those are reported separately.11 The current study focuses on the comparison of FNAB and CBX samples. Transcriptional profiling was performed on paired FNAB and CBX samples from 10 patients. An unsupervised complete linkage analysis of 10,993 cDNA clones (sequences) demonstrated clustering of 7 of 10 FNAB and CBX pairs from the same breast tumor (Fig. 4). Each of the three CBX samples that did not cluster with their respective FNAB samples (Samples 157, 120, and 117) had a histiologic component of stroma that was similar to the other CBX samples (Fig. 2). Individual FNAB and CBX sample pairs also were compared using Pearson correlation statistics, with the following correlations: Sample 115 was 0.90, Sample 120 was 0.34, Sample 136 was 0.75, Sample 150 was 0.66, Sample 123 was 0.79, Sample 135 was 0.67, Sample 157 was 0.83, Sample 117 was 0.73, Sample 177 was 0.75, and Sample 182 was 0.51 (Fig. 4). These paired correlation statistics demonstrated good, but incomplete correlation of genes expressed in these FNAB and CBX paired samples. Only Sample 120 did not cluster by hierarchical analysis and also showed a weak correlation (Pearson statistic) between the FNAB and CBX samples. The histopathologic appearance of a CBX sample did not appear to predict whether the gene expression profile clustered or correlated with the matched FNAB sample (Fig. 3). Sequences that most distinguished CBX samples from matched FNAB samples (highest 51 absolute SNR values) are presented in Table 2 and are expressed more strongly in CBX samples than in FNAB samples (Fig. 5). Those 51 genes are expressed predominantly in nonlymphoid stromal cells, such as endothelial cells, fibroblasts, myofibroblasts or smooth muscle, histiocytes, and adipocytes (Table 2). They are associated with angiogenesis, fibrosis, tissue remodeling, myocontractility, and phagocytosis (Table 2). The presence of more nonlymphoid stromal cells in CBX samples and the stronger expression of a set of recognized stromal genes in CBX samples supports our interpretation that this gene set is a transcriptional profile from nonlymphoid breast carcinoma stroma.

Table 2. The 51 cDNA Clones that Statistically Most Strongly Distinguish between FNA and CBX Sample Types, Ranked in Decreasing Order of SNR Value
RankSNRGeneGene descriptionRelevant function
  1. SNR: signal-to-noise ratio; EST: expressed sequence tags; ECM: extracellular matrix; BM: basement membrane; VEGF: vascular endothelial growth factor; DAG: diacylglycerol.

11.53LOXLysyl oxidaseFibrosis: crosslinking elastin and collagen14, 15
21.52IGFBP7Insulin-like growth factor binding protein 7Angiogenesis: capillary vessel formation16, 17
31.52LMOD1Leiomodin 1 (smooth muscle)Smooth muscle and myofibroblast contractility18, 19
41.51 ESTs//(Hs.93102;)∧462603∧AA704965?
51.49AGTRL1Angiotensin receptor-like 1? Angiogenesis: related to adrenomedullin receptor20, 21
61.49C1QR1Complement component C1q receptorPhagocytosis and adhesion to endothelium22
71.43AQP1Aquaporin 1 (channel-forming integral protein)Angiogenesis: endothelial permeability23, 24
81.41RGS5Regulator of G-protein signalling 5Angiogenesis (see ref. 14) and myofibril contractility25, 26
91.35VWFvon Willebrand factorAngiogenesis: secreted by endothelium27
101.35 Polymerase I and transcript release factorFibrosis: transcription of collagen Type I28
111.33LAMA4Laminin, alpha 4Angiogenesis: endothelial attachment to BM29
121.31MMP2Matrix metalloproteinase 2 (Type IV collagenase)Angiogenesis and invasion: proteolysis of ECM30, 31
131.27DLL4Delta-like 4 homolog (Drosophila)Angiogenesis: capillary vessel formation32
141.27RGS5Regulator of G-protein signaling 5Angiogenesis: inhibits endothelial growth signaling25, 26
151.27COL6A2Collagen, Type VI, alpha 2Fibrosis: activated fibroblasts and myofibroblasts33
161.27 ESTs//(Hs.50382;)∧296568∧W00794::N73843?
171.27CDH5Cadherin 5, Type 2, VE-cadherin (vascular)Angiogenesis: endothelial adhesion and permeability34–36
181.26SMPD1Sphingomyelin phosphodiesterase 1, acid lysosomalUbiquitous: cell membrane integrity, ceramide production (proliferation, apoptosis, differentiation)37
191.25CDH5Cadherin 5, Type 2, VE-cadherin (vascular)Angiogenesis: endothelial adhesion and permeability34–36
201.23CALD1Caldesmon 1Smooth muscle and myofibroblasts38
211.21SEI1CDK4-binding protein p34SEI1Unknown relevance: cell proliferation39
221.2CD34CD34 antigenAngiogenesis: endothelial surface molecule40
231.19 KIAA0758 protein?
241.19PCOLCEProcollagen C-endopeptidase enhancerFibrosis: enhances collagen deposition and crosslinking41
251.19SPARCL1SPARC-like 1 (mast9, hevin)Angiogenesis26
261.19 Polymerase I and transcript release factorFibrosis: transcription of collagen Type I28
271.19 HUEL (C4orf1)-interacting proteinUnknown relevance: possible transcriptional regulation42
281.18 Homo sapiens HSPC285 mRNA, partial coding sequence?
291.17IGFBP3Insulin-like growth factor binding protein 3Ubiquitous: binds IGF1, expressed in fibrous tissue43, 44
301.16CD34CD34 antigenAngiogenesis: endothelial surface molecule40
311.14LOC51705Endomucin-2Angiogenesis: endothelial surface molecule45
321.13TCEB3Transcription elongation factor B (SIII), (elongin A)Unknown relevance: regulator of transcription46
331.13COL4A1Collagen, Type IV, alpha 1Angiogenesis and fibroblasts: BM protein
341.12COL1A2Collagen, Type I, alpha 2Fibrosis
351.12PDGFRBPlatelet-derived growth factor receptor β polypeptideFibrosis: fibroblast cell membrane receptor47
361.11CYR61Cysteine-rich, angiogenic inducer, 61Angiogenesis and tissue remodeling: secreted ECM protein involved in vessel branching and integrin binding48–50
371.1FBLN1Fibulin 1Angiogenesis and cell motility: adhesion and motility51
381.1FLT1VEGF receptorAngiogenesis: endothelial surface receptor for VEGF52
391.09 ESTs//(Hs.30089;)∧754449∧AA410298::AA410480?
401.09A2MAlpha-2-macroglobulinAngiogenesis and fibrosis: secreted by endothelium and fibroblasts into ECM to scavenge cytokines53
411.09PPAP2BPhosphatidic acid phosphatase Type 2BUbiquitous: cell membrane integrity and DAG signal54, 55
421.08RARRES2Retinoic acid receptor responder 2Unknown relevance: retinoid-induced differentiation56
431.06 Homo sapiens mRNA; cDNA DKFZp434C2016?
441.06COL15A1Collagen, Type XV, alpha 1Angiogenesis and tissue remodeling: BM protein surrounding vessels, adipocytes, fibroblasts, smooth muscle cells67
451.06LOC83468GycosyltransferaseUbiquitous: biosynthesis of glycoproteins and glycolipids
461.05COL6A1Collagen, Type VI, alpha 1Fibrosis: activated fibroblasts and myofibroblasts33
471.05COL3A1Collagen, Type III, alpha 1Fibrosis
481.05CYR61Cysteine-rich, angiogenic inducer, 61Angiogenesis and tissue remodeling: secreted ECM protein involved in vessel branching and integrin binding48–50
491.05 Homo sapiens cDNA: clone HRC08686?
501.04 ESTs, highly similar to T42626 (Mus musculus)?
511.04IGF1Insulin-like growth factor 1 (somatomedin C)Tissue remodeling and paracrine cell stimulation: expressed in breast carcinoma stromal cells43, 58
thumbnail image

Figure 5. The top 51 cDNA clones that were identified by the signal-to-noise ratio analysis of the 10 matched sample pairs also showed differential expression in the entire study population of 38 fine-needle aspiration biopsy and 12 Core-needle biopsy (CBX) samples. Strong expression (red), and weak expression (green) are shown. These predominantly stromal genes (Table 2) were expressed more strongly in the CBX samples.

Download figure to PowerPoint

Expression of Nonlymphoid Stromal Genes in Breast Carcinoma Samples

The 51 sequences that most distinguish CBX samples from their matched FNAB samples (Table 2) were used in a supervised cluster analysis of all our 50 samples (12 CBX and 38 FNAB samples). Those genes were expressed more strongly in CBX samples and consistently distinguished between the two sample types (Fig. 5). Core biopsy samples with the lowest proportion of stromal cells (Samples 136, 120, 123, and 117) had relative expression of the 51 stromal genes similar to the CBX samples with the highest proportion of stromal cells (Samples 115, 135, 150, and 182) from histopathologic counts (Figs. 2B, 5). Therefore, we found no apparent relation between the percentage of nonlymphoid stromal cells in tissue sections and the relative strength of expression of recognized nonlymphoid stromal genes in corresponding tissue samples. It is likely that the histologic appearance of breast carcinoma stroma does not indicate the transcriptional activity in those cells.

The stromal gene list then was examined in a different genomic database from breast carcinoma tissue samples to investigate heterogeneity of expression and possible clinical value. Currently, all publicly available expression profile databases for breast carcinoma are derived from surgical resection tissue specimens, but we know that the tissue composition of breast carcinoma in surgical biopsies is usually similar to that of core biopsies. We used the published microarray gene expression data from van't Veer et al.4 (Available from URL: http://www.rii.com/publications/ default/htm) because this was the only publicly available data annotated with clinical outcome. Normalized gene expression data were available from 78 surgical resection specimens of primary invasive breast carcinoma from patients who either did (34 patients) or did not (44 patients) develop systemic metastasis within 5 years of the initial diagnosis.4 Because all the cDNA sequences from our array would not be represented on other array platforms, we expanded our stromal gene set to include the top 200 cDNA sequences. We identified 120 genes from our 200 stromal gene set in that published database.4 Unsupervised cluster analysis separated the 78 breast carcinoma tissue specimens into two main categories (likely to be stroma rich and stroma poor) with a possible smaller intermediate category (Fig. 6). Patients who developed systemic metastasis within 5 years of diagnosis were evenly represented in these groups, suggesting no relation between the extent of stromal gene expression and prognosis (Fig. 6). Supervised cluster analysis (according to metastatic status within 5 years) was also performed on the published database using our 120 stromal genes.4 This analysis did not demonstrate any recognizable pattern of gene expression, even when the 78 tissue specimens were ranked in order of aggressiveness according to the original publication.4

thumbnail image

Figure 6. Unsupervised cluster analysis of the stromal gene set (120 sequences) in 78 independent breast tumor tissue samples4 demonstrates 2 main clusters that are related to higher (red) or lower overall stromal gene expression levels. A less-defined intermediate cluster may exist (middle). The development of metastases within 5 years of diagnosis occurs with similar frequency in the two main clusters and appears to be unrelated to stromal gene expression.

Download figure to PowerPoint

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Transcriptional (gene expression) microarray profiling can be performed with samples from single-pass FNAB or CBX. We report that FNAB and CBX techniques yield similar amounts of total RNA. Approximately 75% of single-pass biopsies yield at least 1 μg of quality total RNA. The mean RNA yield was 3.6 μg for FNABs and 2.8 μg for CBXs. These results suggest that RNA retrieval from single-pass FNAB and CBX samples is equivalent and usually suitable for microarray hybridization studies. Our success rate for FNABs (approximately 75%) is higher than that previously published in a smaller feasibility study (15%).8 We believe that our rate is due partly to commitment of the entire sample into RNAlater for optimal RNA preservation and stability. For optimal clinical utility, we would have prefered a success rate > 90%. To achieve this, two or three FNAB passes should be considered for microarray analysis. In a recent study, Dunmire et al.12 determined (from ex vivo FNAB samples of resected breast tumor tissue samples) that approximately half of the total RNA yield can be retrieved from the supernatant phase during extraction of RNA from FNAB samples collected and stored in RNAlater. The RNA in the supernatant generates nearly identical gene expression data as the RNA from the cell pellet.12 Capturing the RNA in solution should significantly improve the total RNA yield from FNAB samples.

The cellular composition of FNAB and CBX samples differs, according to microscopic cell counts. The FNAB samples contain more tumor cells (80% vs. 50%) and fewer nonlymphoid stromal cells (5% vs. 30%) than the CBX samples. Our cell counts yielded similar results to those reported in a recent study of five paired FNAB and surgical biopsy samples, in which FNABs contained 66–93% (mean, 83%) tumor cells and tissue sections contained 37–78% (mean, 62%) tumor cells.13 The proportion of lymphocytes is similar in both FNAB (15%) and CBX samples (20%). The fundamental difference in cellular composition between CBX and FNAB samples is the proportion of nonlymphoid stromal cells. The proportion of nonlymphoid stromal cells is approximately six-fold higher in CBX compared with FNAB samples (30% vs. 5%). The smaller yield (5%) of nonlymphoid stromal cells in FNAB samples might contribute less total RNA in the sample because these cells are presumably less transcriptionally active than tumor cells. These reasons most likely explain the stromal genes that we identified in the CBX samples, not in the FNA samples (Table 2) (Fig. 5).

When transcriptional profiles of matched FNAB and CBX samples from the same tumor are evaluated for overall similarity by unsupervised hierarchical clustering methods that compare each sample to all the others, profiles from FNAB and CBX samples from the same patient's tumor often cluster together (7 of 10 pairs). Comparing individual pairs of matched FNAB and CBX samples, the Pearson correlation statistic was > 0.50 for 9 of 10 pairs. Therefore, the transcriptional profile obtained from needle biopsies of breast tumor tissue predominantly arises from the cancer cell population. The malignant cells most likely are more transcriptionally active than other cells, and that tumor cell profile is present in both samples. The matched FNAB and CBX pairs were then analyzed for differentially expressed genes. We have identified a set of 51 genes from a larger pool of 200 differentially expressed sequences. These are recognized genes expressed by nonlymphoid stromal cells, such as endothelium, fibroblasts, myofibroblasts or smooth muscle, histiocytes, or adipocytes (Table 2). Stromal genes have been identified in samples from resected breast tumor tissue and are variably expressed.1 A supervised cluster analysis was performed using that test set of 51 genes and clearly separated the total sample population of 12 CBX and 38 FNAB samples into two groups, according to biopsy type (Fig. 5). There was markedly higher expression of these genes in the CBX samples and they strongly characterize stromal elements (Table 2). However, the relative expression of those genes did not relate to the proportion of nonlymphoid stromal cells in the corresponding tissue sections from the CBX samples (Figs. 2, 3). The likely explanation is that the histopathologic appearance of the tumor stroma does not accurately reflect stromal transcriptional activity, although intersample variability also may contribute to our findings. Overall, the results of the current study suggest that FNAB is a preferred sampling method for genomic studies of breast carcinoma cell populations, whereas CBX is a preferred sampling method for genomic studies of combined breast carcinoma cell and stromal populations.

The histopathologic extent and appearance of breast carcinoma stroma is variable in different tumor tissue samples. Recognized stromal genes are present in transcriptional profiles from CBX samples of breast tumor tissue and are variably expressed. However, because the histopathologic appearance of the tumor stroma is not informative of the transcriptional activity of the stromal component, to our knowledge there is not currently a way to reconcile the amount of stromal cell contribution to the overall gene expression profile from a CBX or resected breast tumor tissue sample. Having identified a nonlymphoid stromal gene profile using matched FNAB and CBX samples, it might be possible to develop an informatics tool that could mathematically subtract stromal genes from the data weighted to their relative expression levels in that sample. Another approach for future study would be to use laser capture microdissection of frozen tumor samples to separate stromal and cancer cell populations. At a molecular level, the interactions between breast carcinoma cells and host stromal cells are complex. It is not known whether transcriptional profiles of breast tumor stroma are independently clinically meaningful, whether stromal gene expression in tumor tissue samples creates a level of background signal, or even whether biologic interactions between stroma and cancer cells are represented indirectly in the transcriptional profile of the tumor cells. Stromal interactions with breast tumor cells typically occur through extracellular receptor or adhesion molecules, many of which interface with signal transduction pathways that influence DNA transcription.

If stromal genetic information is obtained from CBX samples but not FNAB samples, it becomes relevant to know whether the expression of those genes in breast tumor tissues is variable and if the stomal genetic information is clinically significant. To test our stromal gene set against an independent set of breast tumor tissue samples, we analyzed the expression of these genes in what to our knowledge was the only publicly available, clinically annotated, profiling database.4 There are inherent difficulties in applying information from a microarray experiment to a different dataset that was obtained under different experimental conditions. Hybridization conditions, array imaging, and data analysis methods would be different. There also would be differences in the methods of sample procurement and purification of RNA. The potential error is increased further if the other dataset was obtained using a different microarray platform. There are several important differences between the data generated by the van't Veer et al. study4 and the current study. van't Veer et al. used oligonucleotide glass arrays with fluorescent labeling to compare each sample with an experimental control of pooled samples from the 78 tumors. We used nylon membrane-based cDNA arrays that utilize 33P-isotope labeling of individual samples. Predictably, the two proprietary platforms also contained different sets of sequences with likely differences in the binding affinities of individual probes. We identified 120 genes from our 200 stromal gene set in the published database by van't Veer et al.4 Despite these limitations, our analysis demonstrates that some degree of cross-validation is possible with gene sets discovered in one platform and applied to an independent database generated by another laboratory. The observed separation of these tumors using our stromal gene set in an unsupervised cluster analysis suggests that stroma-rich tumor tissue samples can be distinguished from stroma-depleted tumor tissue samples (Fig. 6).4 We do not know if there were histopathologic differences in the stromal component of these two groups of tumor tissue samples, but we also should not assume that all the different cell types observed in a tumor mass would have equivalent transcriptional activity. van't Veer et al.4 reported that tumors with at least 50% tumor cells by histopathologic assessment were used, but they did not explain whether that represents microscopic area or cell counts. These possible differences in the tumors might not be relevant. Our results suggest that observed differences in the proportion of stromal cells do not explain differences in stromal gene expression. It is noteworthy that the breast tumor tissue samples in the referenced study4 were of earlier stage (T1, T2) disease than the tumors we sampled. Nevertheless, this dataset was selected because the transcriptional profile data are linked to an established clinical outcome measure. The 78 primary invasive breast tumor tissue samples were selected from patients who either did (34 patients) or did not (44 patients) develop systemic metastasis within 5 years of initial diagnosis.4 We performed supervised cluster analysis (according to metastatic status within 5 years) using our 120 stromal genes. We observed no correlation of the stromal gene expression profile with disease recurrence, even when the cancers were ranked in order of aggressiveness according to the original publication.4 An extended series of prospectively collected tissue samples with clinical annotation is required to explore fully the clinical potential of classifying breast carcinoma by the transcriptional profile of the stroma.

The transcriptional profiles derived from FNAB or CBX of breast tumor tissue are similar overall, but consistent differences in the expression of some genes were found according to sample type. We attribute these differentially expressed genes to the nonlymphoid stromal elements in CBX samples. Cell counts suggest that the relative absence of nonlymphoid stromal cells in FNAB samples provides a purer representation of the transcriptional profile from tumor cells, although both sample types have an equivalent lymphoid cell component. The histopathologic appearance of the stroma may not indicate its transcriptional activity. The component of the transcriptional profile provided by nonlymphoid stromal cells is quite variable in breast carcinoma tissue samples and does not relate to prognosis. However, the selection of either FNAB or CBX as a sampling technique for genomic study of breast carcinoma should be influenced by whether stromal gene transcription is desired in the particular experimental design.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES