Gene expression profiling in CD34 cells to identify differences between aplastic anemia patients and healthy volunteers

Acquired aplastic anemia (AA) is a bone marrow (BM)–failure syndrome that is characterized by low blood cell counts and bone marrow hypocellularity.¹  On the basis of clinical observations of high response rates to combined immunosuppressive therapy, immune-mediated suppression of hematopoiesis has been considered to play an important role in most cases of AA.^2-5  Laboratory findings, including inhibition of hematopoietic cell growth by patient lymphocytes and their overproduction of myelosuppressive cytokines, such as interferon-gamma (IFN-γ) and tumor necrosis factor (TNF), have supported this hypothesis.^6-9  Similarly to other autoimmune diseases, antigen-specific T cells in the BM of AA patients are expanded; these lymphocytes are likely to mediate organ-specific cytotoxicity for bone marrow hematopoietic cells.^10-14  To date, only limited information has been available concerning the characteristics of stem cells in AA. The precise antigenic targets of cytotoxic T cells are unknown, and the effects of T-cell attack on hematopoietic target cells are poorly characterized. Although the expression levels of a few genes, such as FMS-related tyrosine kinase3 ligand (FLT3L) and GATA2, appear to be different in AA patients and healthy donors,^15-17  a more general transcriptome pattern of CD34 cells in AA patients has not been described.

Oligonucleotide microarrays allow quantitation of expression levels of a large number of genes in a cell, and thus provide a powerful tool to study the molecular mechanisms of disease at the messenger RNA level. Recently, the gene expression pattern in healthy human CD34 stem/progenitor cells has been reported.¹⁸  Using microarray technology, Steidl et al¹⁹  successfully compared the gene expression profile in CD34 cells derived from bone marrow or granulocyte colony-stimulating factor (G-CSF)–mobilized peripheral blood cells. Microarray has also provided an image of gene expression in autoimmune disease, such as multiple sclerosis lesions.²⁰  Here we apply DNA chip technology to measure the gene expression profile in CD34 cells from the bone marrow of patients with newly diagnosed AA.

Patients were evaluated at the Hematology Branch of the Clinical Center of the National Institutes of Health. The diagnosis of AA was established by bone marrow biopsy and peripheral blood counts as recommended by the International Study of Aplastic Anemia and Agranulocytosis²¹ ; severity was classified by the criteria of Camitta et al.²²  Thirty-six patients with newly diagnosed moderate or severe AA were selected for our experiments (Table 1). Controls were 12 healthy volunteers whose sex and age were approximately matched. To obtain marrow, informed consent was obtained according to protocols approved by the Institutional Review Board of the National Heart, Lung, and Blood Institute.

BM mononuclear cells (BMMNCs) were obtained by aspiration of the iliac crest of patients and healthy donors and prepared with the use of lymphocyte separation medium (Cappel, Aurora, OH). CD34 and CD4 cells were positively selected by means of the mini-MACS immunomagnetic separation system (Miltenyi Biotec, Auburn, CA), according to the manufacturer's instructions. In brief, to obtain normal CD34 cells, 10⁸ or fewer BMMNCs were washed twice and then suspended in 300 μL sorting buffer composed of 1 × phosphate-buffered saline (PBS), 2 mM EDTA (ethylenediaminetetraacetic acid), and 0.5% bovine serum albumin. Cells were incubated with 100 μL human immunoglobulin–Fc receptor (FcR) blocking antibody and 100 μL monoclonal hapten-conjugated CD34 antibody (clone QBEND/10; Miltenyi Biotec) for 15 minutes at 4°C. After washing, cells were resuspended in 400 μL sorting buffer, and 100 μL paramagnetic microbeads conjugated to antihapten antibody were added, followed by incubation for 15 minutes at 4°C. After washing, cells were resuspended in sorting buffer, passed through a 30-μm nylon mesh, and separated in a column exposed to the magnetic field of the MACS device. The column was washed twice with sorting buffer and removed from the separator. Retained cells were eluted with sorting buffer by means of a plunger and subjected to a second separation. Purity of CD34 cells was 90% to 97% by flow cytometry analysis. After washing, 10⁷ or fewer of CD34^– cells were resuspended in 80 μL sorting buffer; 20 μL CD4 microbeads was added and incubated for 15 minutes at 4°C. Washed cells were resuspended and passed through the column, and the subsequent steps were performed as described.

Total cellular RNA was extracted from CD34 cells by means of TRIzol reagent (Invitrogen, Carlsbad, CA) or the High Purity RNA Isolation Kit (Roche Diagnostics, Indianapolis, IN), according to the manufacturers' protocols. To provide sufficient total RNA for processing, samples were pooled. An RNA pool from 24 AA patients (equal amounts of RNA from each individual) was named pool-AA₁, and pool-AA₂ was obtained from another cohort of 6 AA patients. For controls, pool-N₁ was prepared from 8 healthy individuals and pool-N₂ from an additional 4 healthy individuals. In the initial oligonucleotide array experiments, triplicate technical RNA aliquots from pool-AA₁ or pool-N₁ were prepared separately and subjected to subsequent cDNA synthesis, labeling, hybridization, and analysis. For subsequent oligonucleotide array analyses, biologic duplicates, termed pool-AA₂ and pool-N₂, were prepared from different patients and healthy volunteers, respectively. In addition, pool-AA₃ was prepared from a further 6 AA patients for real-time polymerase chain reaction (PCR) assay (TaqMan; PE Applied Biosystems, Foster City, CA).

The GeneChip Eukaryotic 2 Cycles Small Sample Target Labeling protocol developed by Affymetrix (Santa Clara, CA) was employed to produce biotinylated cRNA from small amounts of total RNA. This protocol uses 2 cycles of cDNA synthesis combined with in vitro transcription (IVT). In the first cycle, first-strand cDNA is synthesized from total cellular RNA, which in turn becomes a template to generate second-strand cDNA, resulting in double-strand (ds) cDNA. As a final step in the first cycle, unlabeled cRNA is created from the ds-cDNA. In the second cycle, the unlabeled cRNA is converted into ds-cDNA through first-strand and then second-strand cDNA syntheses, followed by synthesis of biotinylated cRNA. In our study, 500 ng pooled total RNA was used as a template to generate first-strand cDNA with the SuperScript Choice reagents (Invitrogen) in combination with an oligo-deoxythymidine (oligo-dT) primer containing the T7 RNA polymerase binding site (5′-GCCAGTGAATTGTAATACGACTCACTATAGGGAGGCGG-(dT)24-3′) (Genset, La Jolla, CA), according to the manufacturer's instructions. After generation of ds-cDNA from the first-strand cDNA, unlabeled cRNA was synthesized by in vitro transcription with the use of the Ambion MEGAscript T7 Kit (Ambion, Austin, TX) in the provided protocol. In the second cycle, first-strand cDNA was synthesized with the use of the unlabeled cRNA as a template and random primers (Invitrogen), and subsequently converted into ds-cDNA. For probing on Affymetrix arrays, biotinylated cRNA was generated with the Enzo BioArray High Yield Transcript Labeling Kit (Enzo Diagnostics, Farmingdale, NY). The biotinylated cRNA was purified with the RNeasy Kit (Qiagen, Valencia, CA), followed by fragmentation of an aliquot (15 μg) of the biotinylated cRNA. Samples were frozen at –20°C until use.

Hybridization, washing, staining, and scanning of Affymetrix probe arrays were performed as described in the standard Affymetrix protocol (P/N 700 222 rev 4) for Human Genome U95A version 2 Arrays (HG-U95AV2) with the use of 15 μg fragmented RNA.

Gene expression levels were determined by means of Affymetrix's Microarray Suite 5.0 (MAS 5.0); this software's algorithms allow quantitative estimation of a gene expression and a P value to establish a confidence level that the mRNA of interest is accurately measured. To correct for technical variation between chips, the mean expression of the 50th percentile of each chip was scaled to a common value of 1000. Scaled expression levels and P values were exported for individual GeneChips for subsequent analysis with the use of Silicon Genetics's GeneSpring software (version 5.1) (Silicon Genetics, Redwood City, CA). Once imported into GeneSpring, each gene was normalized by using the median of its measurements in all samples. The mRNA expression levels for patients and controls were determined in 2 steps: means of gene expressions among the 3 technical replicates were used as the best estimate of expression levels for pool-AA₁ and pool-N₁, and these means were then averaged with the biologic replicates, pool-AA₂, and pool-N₂, respectively. The averaged expression level of the 2 biologic samples was used in subsequent analysis by GeneSpring software.

Genes differentially expressed in the patients were identified by normalizing the expression levels of pooled AA by those of normal pools. Lists of genes for further study were created by filtering genes with at least a 2.0-fold change. As only 2 biologic replicates were possible for each group, a rigorous t test with a multiple testing correction produced no significant genes. For exploratory analysis of the data, the most reliable measurements were identified with an uncorrected t test on individual genes, and genes with P values less than .05 were retained. An additional filter, based on the P < .05 according to MAS, was added to eliminate genes that were not accurately measured in at least one of the samples used.

For some functional gene assignments, we also used the Cancer Molecular Analysis Project of the National Cancer Institute Web site (http://cmap.nci.nih.gov/. Accessed October 1, 2003).

TaqMan real-time reverse transcription–PCR (RT-PCR) was performed to confirm expression levels of RNA transcripts with sequence-specific oligonucleotide primers and methylglyoxal bis(guanylhydrazone) (MGB) probes (Table 2), according to the manufacturer's instructions (PE Applied Biosystems). For relative quantification, beta-actin mRNA served as an external control. In brief, first-strand cDNA was synthesized from total cellular RNA with an oligo-dT 12-18 primer (Pharmacia, Piscataway, NJ) with the use of the SuperScript Choice reagents. The obtained cDNA was amplified in a final volume of 20 μL with 300 nM of each primer; 200 nM probe; 3.5 mM MgCl₂; 1 × TaqMan Buffer A; 200 μM deoxyadenosine triphosphate (dATP), deoxycytidine triphosphate (dCTP), and deoxyguanosine triphosphate (dGTP); 400 μM deoxyuridine triphosphate (dUTP); 0.2 U AmpErase uracil N-glycosylase (UNG); and 0.5 U AmpliTaq DNA polymerase. All PCR consumables were purchased from PE Applied Biosystems. Primers and probes were designed with the use of Primer Express (PE Applied Biosystems) and synthesized by PE Applied Biosystems. The thermal cycling included 2 minutes at 50°C and 10 minutes at 95°C, then proceeded with 40 cycles at 95°C for 15 seconds and 60°C for 1 minute. All reactions were performed in the Model 7700 sequence detector (PE Applied Biosystems). Each target (pool-AA₁, pool-AA₃, or pool-N₁) was measured in the same plate for the same gene, and every sample was examined in duplicate. The threshold cycle (Ct) was used to quantify mRNA levels of samples with beta-actin normalization. The following equation was used for relative mRNA calculation²³ : Relative mRNA = 2^–ΔΔC_T. (ΔΔC_T = ΔC_T,X – ΔC_T,R; X indicates the difference in threshold cycles for target; R, housekeeping gene).

We analyzed the gene expression profile of bone marrow CD34 cells from patients with newly diagnosed AA using Affymetrix oligoarrays containing sequences of 12 627 genes. Highly enriched CD34 cells (purity, 90% to 97%) were isolated from AA patients and healthy volunteers. In AA patients, the numbers of bone marrow CD34 cells are extremely low, and it is impossible to obtain sufficient mRNA from CD34 cells of a single patient for individual testing. To account for differences among individuals and to obtain adequate quantities of RNA for the analysis, we pooled equal amounts of CD34-cell RNA from patients (pool-AA₁ or pool-AA₂) or healthy controls (pool-N₁ or pool-N₂). Technical replicates were subsequently created from pool-AA₁ and pool-N₁ to examine the reproducibility of the Small Sample Protocol. The standard sample preparation Affymetrix GeneChip protocol requires at least 5 μg total RNA as a starting material for each target preparation reaction. Owing to the extremely limited numbers of CD34 cells in AA patients, we used the Small Sample Protocol developed by Affymetrix, which provides for 2 cycles of standard cDNA synthesis, followed by IVT for GeneChip target amplification. The principle of this method is that the first cycle provides initial amplification of total RNA, which results in unlabeled cRNA. In the second cycle, during IVT synthesis, biotin-ribonucleotides are incorporated to produce labeled antisense cRNA target. To evaluate this method for microarray expression analysis, we used several parameters, including the yield of labeled cRNA, expression levels of transcripts used as positive controls, and reproducibility of expression levels among technical replicates. The cRNA yield was compared in the Small Sample and the standard protocols, with the use of 500 ng or 5 μg total RNA of CD4 cells from healthy donors, respectively (Table 3). The quantities of cRNA obtained from 500 ng or 5 μg RNA in 2 replicate experiments were 55.5 and 53.2 μg, or 54.5 and 52.2 μg, respectively, indicating similar yields. The 500 ng RNA samples resulted in 45.6% “present” calls, comparable to 45% obtained with 5 μg starting RNA labeled by the standard protocol. The correlation of expression levels showed 91% reproducibility. The Small Sample Protocol gave rise to a higher 3′-to-5′ ratio of individual genes, including control genes such as GAPDH, presumably owing to the generation of shorter products toward the 3′ end of mRNA in the second cycle of amplification. In this study, the ratio was 1.5 to 3.27 for the Small Sample Protocol and below 2 for the standard protocol. Our method therefore met the quality control metrics provided by Affymetrix for the Small Sample Protocol. All these parameters were comparable in the Small Sample and standard protocols, suggesting that results using the Small Sample Protocol would be reliable.

To identify major sources of experimental variability, 3 technical replicates were prepared with the use of 500 ng RNA samples of CD34 cells from AA patients (pool-AA₁-1, pool-AA₁-2, and pool-AA₁-3) or healthy volunteers (pool-N₁-1, pool-N₁-2, and pool-N₁-3), respectively. Each RNA sample was converted to ds-cDNA, followed by synthesis of the first-cycle cRNA. With the use of 3 μg cRNA as a template for the second cycle, ds-cDNA and then biotinylated cRNA target were generated (Table 3). The “present” calls of the 8 pools were between 41.9% and 48.5%. The technical replicates showed that the Small Sample Protocol was highly reproducible: the correlation coefficients between replicates from pool-AA₁ were 0.987, 0.990, and 0.994, and for replicates of pool-N₁, 0.991, 0.991, and 0.996. There was modestly more variation between biologic replicates: the correlation coefficient was 0.919 between pool-AA₁ and pool-AA₂, and 0.904 between samples pool-N₁ and pool-N₂.

A comparison of pool-AA₁ with pool-AA₂ showed 5542 genes were present in all 3 replicates from pool-AA₁, and 6116 genes were present in the single pool-AA₂. There were 5169 genes present in both pool-AA₁ and pool-AA₂, which represented 93.3% of the genes present in pool-AA₁ and 84.5% of those in pool-AA₂. For the normal pools, 5291 or 5868 genes were present in pool-N₁ or pool-N₂, respectively. Venn diagram analysis revealed that 4854 genes were present in both N₁ and N₂ pools, of which 91.7% of genes were judged present in pool-N₁ and 82.7% in pool-N₂. Genes identified as absent were not well correlated, indicating that the reported hybridization data of genes with low expression levels and/or absent calls were unreliable. In contrast, a present call indicates low experimental variability and high reproducibility.²⁴ 

Genes expressed differentially were identified by comparing the average of the biologic pools. Overall, about 8% of the total genes were differentially expressed in patient samples, and most were up-regulated compared with controls: 805 genes were increased in expression compared with 238 genes decreased in expression. An overview of the gene expression profile in AA patients compared with healthy donors is shown in Figure 1.

The 805 genes up-regulated at least 2.0-fold in AA patients belonged mainly in the functional categories of defense/immune response, cell death and apoptosis, cell cycle/cell proliferation, cytokine/chemokine, signal transducer, metabolism, transport, stress response, transcription factor, and cell adhesion. The 238 genes showing at least 2.0-fold down-regulation in AA patients were grouped into cell cycle/cell proliferation, growth factor, cell growth and maintenance, antiapoptosis, nucleic acid binding, cell adhesion, oncogenes/transcription factor, signal transduction, enzyme/enzyme inhibitor, metabolism, immune response, and genes of unknown function categories. (Figures 1 and 2)

The most striking results were obtained for the gene categories related to immunity and cell death. A large number of immune/defense response genes were highly expressed in CD34 cells from AA patients. In Affymetrix HG-U95AV2 arrays, 150 of the 290 genes (56%) related to the immune response were at least 2.0-fold changed in their expression in AA; almost all (141) were upregulated: 20 genes for cytokines and cytokine receptors, 21 genes for chemokines and chemokine receptors, 36 signal transduction-mediation genes, and 64 other immune response genes (antibodies, enzymes, complement/component receptors, IGFBP4, and toll-like receptors). In contrast, lower expression in AA was observed for a small number (9) of immune response genes, including FCE1A, pro-platelet basic protein, PF4, and PPBP.

Apoptosis genes also were differentially expressed in patients' samples at a much higher rate than in the global pattern of the transcriptome. Sixty-seven out of 356 (19%) apoptosis genes, including 9 death receptor pathway genes, 3 caspase-related genes (CASPER, CASP1, and CASP8), 5 granzyme and perforin pathway genes, 21 other signal transduction-related pathway genes (JUN, JUNB, KBF1, TNFSF2, and MAP4K4), and 26 genes otherwise involved in other apoptosis pathways (serine/threonine kinase 17a and 17b, and TOSO), were up-regulated. In contrast, 3 genes including TIAF1, which has been implicated in antiapoptotic regulation, were down-regulated in AA. In the death pathway, 5 death receptors and 4 death ligands showed enhanced expression in AA.

Cell cycle and cell proliferation genes (54 out of 348; 16%) also showed differences between AA patients and healthy volunteers. Eleven signal transduction–related genes, including STAT1 and IGF1; 17 cell proliferation-negative control genes; and 6 other cell cycle-related genes were up-regulated. Of these genes, most are believed to exert negative effects on cell proliferation and to inhibit entry into cell cycle. In contrast, several genes that exert positive effects on cell cycle progress and cell proliferation control were down-regulated: 2 members of the cyclin-dependent kinase (CDK) family; 3 of the cell division cycle (CDC) family; and 15 signal transduction or other cell cycle control genes, including M-phase phosphoprotein 9, MYC, and BUB1.

Genes encoding proteins that bind to DNA were also differentially regulated in AA patients compared with controls. In patients, 25 DNA-binding protein genes, including members of the zinc finger protein family, and RNA-binding genes, were down-regulated. Conversely, 53 genes of these types were up-regulated, including RNA polymerase II, which is overexpressed in cells undergoing apoptosis. Genes for several cell adhesion molecules and cell adhesion receptors were up-regulated in AA, including VCAM1 and ICAM1, expression of which is increased following T-cell engagement. Two genes related to platelet differentiation, CD62P and CD42b, were down-regulated in patients. Growth factor and cytokine genes, such as FLT3, GATA2, and PF4, were down-regulated in AA patients, as well as several oncogenes including c-myb. A large number of other genes involved in signal transduction pathways, such as transcription factors, membrane proteins, and enzymes, also showed differential expression in AA.

For quantitative analysis using TaqMan Quantitative PCR, we selected 9 genes from the initial GeneChip analysis: 5 genes appeared to be up-regulated and 4 were down-regulated, over a range of 2.7- to 77.4-fold. Three pools were assayed: the original samples prepared for the GeneChip analysis (pool-AA₁ and pool-N₁) as well as RNA from a new group of patients (pool-AA₃). TNFR2 and IL-8 showed 3.2- and 77.4-fold increases, respectively, in chip analysis of pool-AA₁; with the use of real-time PCR, these genes were increased 1.8- and 13-fold in pool-AA₁, and 9.6- and 12-fold in pool-AA₃. Similarly, CD34, c-myc, GATA2, and FLT3, which were all decreased by GeneChip analysis of AA CD34 cells, were down-regulated in real-time PCR analysis (Figure 3).

In spite of the extremely limited numbers of CD34 cells present in the bone marrow of patients with AA, we were able to analyze the transcriptome pattern in these cells by combining the use of pooled RNA samples and a Small Sample amplification technique. Because of the small numbers of cells, the use of pooled samples, and the Small Sample amplification method, there was a strong possibility of error and of generating misleading data. However, we showed, first, the high reproducibility of results among replicate samples from the same pool of RNA of either AA patients or healthy individuals. Second, we found a high correlation in gene up- and down-regulation in patient samples as compared with healthy individuals when separate patient and control pools were compared. Third, the ratio of representation of the 3′ and 5′ ends of the genes assessed, a measure of the adequacy of RNA synthesis, was within the parameters specified for this technique and close to that obtained with standard GeneChip analyses. Finally, we selected individual genes for comparison using real-time PCR amplification. While a minority of genes could not be confirmed to be dysregulated in AA with the use of this more rigorous methodology, the majority of the genes that we identified by chip analysis were similarly up- or down-regulated in a third pool of AA patient samples. Therefore, we believe that our method is an adequate screening technique for the scant numbers of CD34 cells in bone marrow failure patients and should be capable of providing data for hypothesis generation, with the understanding that initial results should be confirmed by gene amplification or other methods.

We have proposed that the pathophysiology of AA can be simplified to T-cell–mediated, organ-specific attack of cytotoxic lymphocytes on CD34 hematopoietic stem and progenitor cells.²⁵  Most obviously in the current analysis, CD34 cells from AA patients showed ample evidence of the expression of genes involved in the signal transduction pathways for apoptosis and terminal cytolytic enzyme generation. Conversely, antiapoptotic genes appeared to be expressed at lower levels in patients' CD34 cells as compared with healthy voluteers. Among the up-regulated genes involved in the death receptor pathway were several receptors and ligands, such as the death receptors Fas, DR3, and DR5, TNFRII, and TRAIL. High expression of TNFR2 has been associated with the pathogenesis of other immune-mediated diseases.^26,27  Other apoptosis-related genes were increased in patients: stress- and cytokine-inducible GADD45 B family proteins, which function as specific activators of mitogen-activated protein three kinase 1 (MTK1) (a mitogen-activated protein kinase kinase kinase [MAPKKK] upstream in the p38 pathway that can induce apoptosis),^28,29  and nuclear factor kappa-B (NFKB) inhibitory protein NFKBIA (nuclear factor of kappa light polypeptide gene enhancer in B cells inhibitor alpha), which could influence the function of NFKB and enhance apoptosis.³⁰ 

Direct evidence of immune system attack was also inferred from increased expression of a large number of defense and immune response genes in patient samples. Anticipated to be increased in expression were a number of interferon-response genes, stress-related genes, and chaperone protein genes, such as HSP40. However, a number of cytokine, chemokine, and T-cell effector protein genes also were apparently active in patients, including IFN-γ, TNF-α, perforin, and granzyme protein genes. These results are consistent with some reported data suggesting that CD34 cells are capable of cytokine production and release,^19,31  but they also could be explained by contamination of even our relatively purified CD34 populations, especially from scanty cell samples of marrow failure patients, with effector lymphocytes themselves, the presumed source of these inhibitory or cytotoxic cytokines and perforin family members. IL-1β, IL-6, and IL-8 also showed up-regulation in patient samples. The receptor for IL-10 was increased in expression consistent with an IFN-γ effect; IL-10 inhibits in vitro hematopoietic suppression as well as production of IFN-γ and TNF-α by peripheral blood MNCs (PBMNCs) from patients with AA.³²  IL-10 is also thought to play a role in limiting immune-mediated pathology during the host response to pathogens.³³  We observed up-regulation of several chemokine genes including CXC (IL-8 and SDF1) and CC (MCP-2 and MCP-1), increased expression of which occurs in other autoimmune diseases.^34,35  Finally, a large number of genes involved in signal transduction following immune activation were increased in patient samples. In total, the expression pattern of immune response genes in our chip analysis was supportive of the hypothesis of immune-mediated marrow destruction in AA.

Thirty-four of 54 genes in the class of cell proliferation and cell cycle were up-regulated in AA CD34 cells; 17 of these genes were assigned a negative regulatory function in the software and publicly available databases that we employed for annotation (only 1 up-regulated gene was characterized as a positive proliferation regulator, and the remainder were of mixed or indeterminate function). Conversely, of the 20 genes in this class that were down-regulated in AA, 14 were identifed as positive promoters of cell proliferation and cycling (with the remainder of mixed or indeterminate function [Figure 2]). These data imply suppression of proliferation of CD34 cells as well as direct induction of cell death by T-cell attack. Of some interest, genes for several constitutive centromere proteins that are essential for spindle-pole body duplication showed markedly decreased expression in AA, a suggestive finding given the propensity of patients to develop aneuploidy over time. Cell cycle control genes that were down-regulated included, for example, CDK6, which plays an essential role in controlling the G₁/S transition, and cell cycle regulators like cyclins E and A.^36,37 CDK2, important in the initiation of both centrosome duplication and DNA synthesis, was down-regulated. In summary, the pattern of involvement of multiple genes that control cell cycle progression might explain the inability of remaining stem and progenitor cells to competently replicate and ultimately compensate for destruction within the hematopoietic cell compartment, despite the abundance of hematopoietic growth factors and even after seemingly successful immunosuppression has removed extrinsic inhibitory factors. Down-regulation of several cell cycle “checkpoint” genes, such as FANCG, c-myb, and c-myc, would also be consistent with the ultimate development of premalignant or aneuploid cells in survival patients, who are susceptible to conversion to myelodysplasia or frank leukemic transformation. Conversely, transforming growth factor–β1 (TGF-β1) was up-regulated; the gene product inhibits G₁ and G₂ cyclin-dependent kinesis.³⁶ CDK2, which is regulated by TGF-β1, was markedly decreased in AA. Cell cycle progression through the G₁ phase into S is a major checkpoint for proliferating cells and is under multiple levels of control by p21.³⁸  Of the growth factor genes and their receptors, we confirmed previously described FLT3 and FLT3 ligand changes in AA,¹⁶  showing especially markedly elevated FLT3 ligand expression. Decreased FLT3 receptor expression suggests impairment of FLT ligand signaling in this disease. Also, a number of insulin growth factor genes and genes for their receptors were elevated in patient samples, implicating this important family of mitogens for the first time in marrow aplastic. We also confirmed down-regulation of GATA-2 in AA patients;¹⁷ C-myb also was down-regulated, and decreased expression of c-myb and GATA-2 probably affects the growth and differentiation of CD34 cells in marrow failure. Finally, a large number of genes that were apparently abnormally up- or down-regulated in patients have not been previously suspected as involved in AA. Examples include vascular cell adhesion molecules, such as VCAM-1, and intercellular adhesion molecule ICAM-1, both of which were greatly increased in patients' CD34 cells. Other adhesion molecules, some of which have been associated with platelet function (CD62P and PF4), were down-regulated. These aberrations in gene expressions need to be confirmed by appropriate studies, but they suggest further experimental approaches for both the understanding of the pathophysiology of AA and the improvement of therapy. For example, expressions of some adhesion molecules are altered by T-cell engagement, and interruption of this interaction may be generally beneficial in autoimmune diseases.³⁹