Expression analysis of multiple myeloma CD138 negative progenitor cells using single molecule microarray readout.

We present a highly sensitive bioanalytical microarray assay that enables the analysis of small genomic sample material. By combining an optimized cDNA purification step with single molecule cDNA detection on the microarray, the platform has improved sensitivity compared to conventional systems, allowing amplification-free determination of expression profiles with as little as 600ng total RNA. Total RNA from cells was reverse transcribed into fluorescently labeled cDNA and purified employing a precipitation method that minimizes loss of cDNA material. The microarray was scanned on a fluorescence chip-reader with single molecule sensitivity. Using the newly developed platform we were able to analyze the RNA expression profile of a subpopulation of rare multiple myeloma CD138 negative progenitor (MM CD138(neg)) cells. The high-sensitivity microarray approach led to the identification of a set of 20 genes differentially expressed in MM CD138(neg) cells. Our work demonstrates the applicability of a straight-forward single-molecule DNA array technology to current topics of molecular and cellular cancer research, which are otherwise difficult to address due to the limited amount of sample material.

Introduction

Microarrays have become the paramount tool for comparative global analysis of biological samples. Yet, current technology requires relatively large amounts of sample material, which hampers the specific analysis of complex and heterogeneous biological tissues. These technological limitations are converse to the emerging interest in gene regulation of rare cellular subpopulations. Most researchers therefore revert to polymerase chain reaction (PCR) or RNA polymerase-based amplification to increase the signal on the microarray.

In recent years, a few amplification-based approaches were developed particularly for expression analysis of small RNA samples. In 2006 Kurimoto et al. established a method for RNA expression profiling that relies on a combination of robust, strongly controlled PCR amplification and microarray-based quantification. They were able to perform expression profiling even of a single cell (Kurimoto et al., 2006). However many controls were required to guarantee the linearity of the sample amplification rendering this method laborious and time consuming. Marcus et al. developed a microfluidic device for parallel RT-qPCR quantification of 72 RNA samples. Using few amplification steps they were able to detect low abundant mRNA templates making the assay useful for analysis of small populations (Marcus et al., 2006) and even single neurons (Citri et al., 2011). The required amplification steps, however, have the disadvantage that the native expression levels may be distorted (Bustin, 2000) (Bustin and Nolan, 2004), rendering unbiased expression profiling of small cell populations still difficult.

Also sequencing approaches were adapted for RNA-expression profiling (RNA-Seq) (Wang et al., 2009) (Brooks et al., 2011). Here, the RNA population is converted into cDNA fragments and analyzed using deep sequencing technology. The procedure is very robust and sensitive, with a requirement of only 20 ng of total RNA, though it is still prohibitively expensive for many laboratories.

As an alternative approach we have developed a single molecule readout platform for the quantification of small amounts of cDNA in a massively parallel format without the need for prior sample amplification (Hesse et al., 2006). The ultra-sensitive biochips allow for competitive analysis of over 2000 target genes simultaneously, requiring only 200 ng of total RNA, which is extremely demanding for parallel PCR approaches (Hesse et al., 2006; Marcus et al., 2006). The novel analysis method relies on direct single molecule counting, making it independent of labeling efficiency or label brightness. It allows for accurate background rejection and gives precise information on the number of molecules specifically hybridized to the microarray spots. As an additional advantage, the array costs and total RNA handling efforts are comparably low.

We demonstrated the potential of the single molecule microarray platform by recording the expression profile of multiple myeloma CD138 negative progenitor (MM CD138neg) cells. Multiple myeloma (MM) is a malignant and basically incurable disease characterized by uncontrolled proliferation of CD138pos clonal plasma cells. Previous studies have shown that MM comprises two populations of plasma cells with distinct malignant properties. The majority of MM cells is CD138pos and displays a differentiated phenotype, while a rare subpopulation of MM CD138neg cells shows features of cancer stem cells (CSCs) including self-renewal, clonogenic growth and tumor initiation upon transplantation into immunodeficient recipient mice (Matsui et al., 2004). Using the developed single molecule microarray platform we characterized the human MM cell line NCI-H929 cells on a molecular level and identified a set of genes differentially expressed in the rare CD138neg subpopulation.

Materials and methods

Cell culture and FACS analysis

NCI-H929 cells were grown as described previously (Matsui et al., 2004) in RPMI1640 media (PAA), 10% fetal bovine serum (FBS; PAA), penicillin (62.5 μg/ml) and streptomycin (100 μg/ml) at 37 °C in a humidified atmosphere of 5% CO2. NCI-H929 cells (ATCC) were initially depleted for necrotic cells by density centrifugation. CD138 cell surface staining was carried out in phosphate buffered saline (PBS) containing 0.5% FBS (PAA) plus 2 mM EDTA for 30 min at room temperature with phycoerythrin (PE) conjugated mouse anti human CD138 antibody (BD Pharmingen; Cat: 552026) in combination with 7AAD (BD Pharmingen) to exclude dead cells. Fluorescence activated cell sorting into CD138pos and CD138neg populations was done on a BD-FACS-vantage machine (Matsui et al., 2004). Specificity of CD138-PE negatively stained cells was counterchecked via usage of an isotype control whereas intensity of PE signal was checked by specific control staining of Jurkat (ATCC) cells via usage of anti human CD3-PE antibody (ImmunoTools; Cat: 21270034).

Doxycycline inducible GLI expressing HaCaT cells were grown in HEPES-buffered DMEM (pH 7.2), containing 10% FCS, 100 mg/l streptomycin and 62.5 mg/l penicillin at 37 °C. GLI expression was induced by adding to the culture media tetracycline (Invitrogen) to a final concentration of 1 μg/ml. (Regl et al., 2002). Total RNA was harvested after 48 h of tetracycline treatment.

RNA isolation

Total RNA isolation was carried out using the phenol-free RNAqueous® Technology (Ambion; AM1912). From the sorted 106 cells of the CD138neg population of the NCI-H929 cell line an amount of 2 μg of total RNA could be isolated. 106 cells of the CD138pos population obtained in the same FACS run were used to isolate 2 μg total RNA for comparative analysis. 600 ng total RNA of each population were used for screening for alterations in gene expression, 1.4 μg for validation of determined genes of interest.

cDNA preparation

In brief, 20 μg, 5 μg or 0.6 μg of total RNA were reverse transcribed with Superscript II (Invitrogen Life Technologies) in the presence of Alexa Fluor 555-aha-dUTP or Alexa Fluor 647-aha-dUTP (Molecular Probes, Invitrogen). When cDNA obtained from 1 μg RNA was purified using affinity columns according to the manufacturer's instructions, no microarray signal could be detected. Therefore, we used a precipitation approach for cDNA preparation. The method is based on ammonium acetate and linear polyacrylamide (LPA, GenElute) precipitation of the DNA, which is effective in dNTP depletion and recovery of larger DNA fragments. In the first step, the 60 μl (0.025 M NaCl) feedstock from the hydrolysis reaction (kit specific RNA degradation reaction) was incubated with 17.5 μl 10 M ammonium acetate (2–2.5 M in the hydrolysis solution). Second, 2.5 μl LPA was added to the hydrolysis solution (0.6 mg/μl of LPA). The solution was mixed in 160 μl of ethanol (96%, 2× total volume) and incubated on ice for 30 min. To purify the cDNA and separate free nucleotides and dye molecules, the sample was centrifuged with ∼14.000 × g and the supernatant was removed carefully. Finally, the sample was washed with 200 μl of 70% ethanol, centrifuged with ∼18.000 × g for 5 min and dried after discarding the supernatant. The remaining pellet was re-suspended in 15 μl H2O.

Hybridization

Labeled and purified cDNA adjusted to 3xSSC (3x Saline-Sodium Citrate: 45 mM sodium citrate, 450 mM NaCl, pH 7.4)/0.1% SDS (Sodium Dodecyl Sulfate) buffer was heated to 96 °C for 3 min and subsequently hybridized for 16 h at 50 °C under a LifterSlip (Erie Scientific Company). During hybridization the sample was permanently mixed in a hybridization station (SlideBooster, Advalytix AG, mix./pause ratio 3:7, mix. power: 27). Afterwards the microarrays were washed in three sequential steps in 1xSSC/0.1% SDS, 0.1xSSC/0.1% SDS and finally 0.1xSSC each for 5 min at RT. Prior to fluorescence readout, microarrays were covered with 0.1x SSC buffer and sealed using hybridization chambers (Sigma–Aldrich, Secure Seal, SA500) (Hesse et al., 2006).

Readout and microarray analysis

The biochip readout was performed on a single molecule sensitivity fluorescence scanner described in detail before (Hesse et al., 2006, 2004). Briefly, the set up is based on an epifluorescence microscope (Axiovert 200, Zeiss) which is equipped with Ar+- and Kr+-ion lasers (Innova, Coherent) for selective fluorescence excitation of Alexa Fluor 555 at 514 nm and Alexa Fluor 647 at 647 nm, respectively. The samples were illuminated in objective-type total internal reflection (TIR) configuration using a 100x oil immersion objective (NA = 1.45, α-Fluar, Zeiss). Fluorescence light is collected using the same objective and, after appropriate filtering using standard Cy3 and Cy5 filter sets (Chroma Technology Corp.), imaged onto a back-illuminated CCD camera (SPEC10:100B, Princeton Instruments; quantum efficiency = 90%, gain = 0.77 counts/e−). For large area readout the scanner was operated in time-delay and integration- (TDI-) mode and equipped with a focus hold system that maintains the focal position during imaging (Hesch et al., 2009).

For measuring the specific hybridization of labeled cDNA, the microarrays were scanned at 200 nm resolution with an excitation intensity of 0.12 kW/cm2 and an integration time per pixel of 116 ms. After readout of the specific hybridization signal, the arrays were stained with labeled random hexamer oligonucleotides (Supplementary), and the slides were imaged at low resolution using a hardware binning of 10. These low resolution images were used for obtaining the spot coordinates. For each spot sub-images were generated and analyzed using an à trous wavelet based peak counting approach (Muresan et al., 2010). More detailed analysis of peak characteristics (brightness and width) confirmed that these signals originated from individual hybridized cDNA molecules (Hesse et al., 2006). The majority of spots showed only low number of peaks corresponding to a weak hybridization signal.

qPCR analysis

The sequences of the primers used for amplification are listed in the supplemental material section (Supplementary table S1). Primers used that are not listed in table S1 were as referred (Schnidar et al., 2009). Primer design was done with Primer3 v. 0.4.0 online software via usage of standardized primer (length: 20–27 bp; Tm: 70–72 °C) and product size (100–200 bp) parameters. Comparative qPCR analysis was carried out on a Rotorgene3000 (Corbett Research) using SYBR-Green-Supermix (BioRad Laboratories). Large ribosomal protein P0 (RPLP0) was used as a reference for normalization (Martin et al., 2001).

Results

In order to optimize imaging conditions and preparation steps for the minute sample size of MM CD138neg cells, we performed test experiments using a Tetracycline (Tet)-inducible human keratinocyte cell line (HaCaT) expressing the GLI oncogene under Tet-control (Regl et al., 2004). Differences in gene expression between Tet-treated (GLI expressing) and untreated (GLI-negative) samples were analyzed by competitive two-color microarray hybridization experiments.

We previously developed a method, which enables expression profiling with tiny amounts of only 104 cells with a detection limit of 1.3 fM (∼39,000 target molecules/sample volume 50 μl) (Hesse et al., 2006). Here we first extended this platform to two color analysis. Alexa Fluor 647 and Alexa Fluor 555 labeled cDNA was synthesized from 5 μg total RNA isolated from Tet-treated and untreated control cells, respectively. 4% of the purified cDNA–equivalent to 200 ng total RNA – was used for hybridization to custom-made microarrays (Supplementary), which contained a set of 120 genes with >60 replicates taken from the Human Genome Oligo Set V3 (Operon). After hybridization over night, unbound sample was removed by a series of washing steps.

The microarrays were scanned sequentially in the red and the green channel, yielding images of 1.4 cm × 2.4 cm size at diffraction-limited resolution and single molecule sensitivity (Fig. 1). Single molecule counting was performed as described in (Muresan et al., 2010). Subsequently, we applied a second hybridization that stained all spots with 1 nM fluorescent random hexamer sequences (Supplement Array stain & Signal Amplification for Gridding), and rescanned the same area again. These images were used to localize all spots on the microarray.

Fig. 1

Fluorescence images of a 1.4 cm × 2.4 cm microarray (upper row) after hybridization of HaCaT cDNA labeled with Alexa Fluor 555 (left column) and Alexa Fluor 647 (right column); the whole array was scanned with 200 nm resolution. The columns in the magnified image regions (A and B) show replicate spots of a few different genes. The spots show homogenous signal with low variation in intensity and shape in both colors. The highest magnification (C and D) shows two spots with different densities of single molecule signals.

To account for potential influences of the fluorophore on the hybridization and labeling efficiency, a color-swap was performed. Three technical replicates were recorded for each sample/dye combination using cDNA from the same reverse transcription. For comparison a conventional comparative analysis was performed on a commercial system (Corning slides, Perkin Elmer reader) with cDNA identically prepared from hundred-fold higher amounts of total RNA. The profile obtained on the single molecule microarrays correlated well with the results obtained on the commercial platform. Moreover a color swap on the ultra-sensitive array did not have any influence on the correlation with the commercial platform (Pearson correlation of 0.85/0.81 respectively) (Fig. 2).

Fig. 2

Correlation plot of expression profiling ratios obtained from the ultrasensitive platform (200 ng total RNA) and a commercial microarray (20 μg total RNA). 29 gene ratios of Tetratcycline treated/untreated HaCaT cells detected at the single molecule level were compared. The diagram shows a high correlation of gene ratios with a Pearson correlation coefficient of 0.85. The black dots show ratios between data obtained using Alexa Fluor 647 with Tet-treated cells and Alexa Fluor 555 with untreated cells. The open rectangles represent the ratios determined after a color swap using the opposite sample-dye combination.

One major factor that limits the sensitivity in mRNA expression profiling is the loss of material during reverse transcription and purification of the resulting cDNA (Bustin and Nolan, 2004; Marcus et al., 2006). Therefore standard methods for RNA preparation needed to be adapted for handling of minute amounts of sample. To mimic the limited amount of sample material, we have reduced the cell number to 104 HaCaT cells for RNA extraction. Using the RNAqueous® kit, 600 ng total RNA could be obtained and was used for transcription into fluorescently labeled (Alexa Fluor 647 or Alexa Fluor 555) cDNA utilizing the Superscript II kit. To prevent loss of cDNA due to irreversible binding to chromatography purification columns, we set up a precipitation-based purification approach using linear polyacrylamide (LPA) as matrix. The high yield of this method allowed splitting the purified cDNA into three sub-samples and hybridizing each of them to separate microarrays. We used from here on a selection of 2086 genes from the Operon human signal transduction set (Supplementary Gene_list.xls). The signal intensities obtained from the technical replicates show excellent correlation with signals obtained using 5 μg total RNA purified classically and detected on the ultrasensitive platform (Pearson correlation coefficient ∼0.9, Fig. 3).

Fig. 3

Correlation of the gene expression levels between sample preparations using 5 μg and 600 ng total RNA as starting material. For both conditions, cDNA corresponding to 200 ng total RNA from the Alexa Fluor 647 labeled cDNA from Tet-treated HaCaT cells was used for hybridization. The determined expression levels show a Pearson correlation coefficient of 0.9.

We then used the new method for mRNA expression profiling in multiple myeloma (MM) cancer stem cells. MM CD138neg cells were isolated from the total population via FACS based on the lack in surface expression of CD138 (Huff and Matsui, 2008; Krivtsov et al., 2009; Matsui et al., 2004; Peacock et al., 2007). The CD138neg subpopulation resembled ∼2.5% of the total NCI-H929 population and only 2.500 cells could be collected per sorting run. To get enough material for one experiment multiple FACS runs were pooled prior to RNA extraction. For validation of FACS selection, the isolated cells were tested for differential expression of Hedgehog (HH) downstream signaling targets. The RNA obtained from CD138neg and CD138pos cells was labeled with Alexa Fluor 647 and with Alexa Fluor 555, respectively, and competitively hybridized on chips spotted with the human signal transduction set representing 2086 genes as described above (Fig. 4). Due to the low sample concentration, the average fluorescence signal measured on the microarray was low, with many spots having just a few molecules hybridized. Still, with the single molecule approach we could unambiguously identify the hybridized molecules, as exemplified on the zooms shown in Fig. 4.

Fig. 4

Array hybridized with cDNA reverse transcribed from 200 ng total RNA obtained from the NCI-H929 cell line. Most of the genes can only be detected at the single molecule level. The zoomed out two colors scanned spots show sharp, randomly distributed single molecule signals with different densities.

In the regime of very low binding probabilities a fair quantification of the detection limit needs to account for the stochasticity of the binding process (Hesse et al., 2006). To ensure statistically robust results, we set a threshold of 100 molecules bound per spot for ratiometric analysis. After applying the threshold more than 400 genes remained analyzable with the obtained expression levels and correlated well between the technical replicates. The expression ratios (CD138neg/CD138pos) of most of the genes were found to be smaller than 2 meaning that genes are expressed at similar levels in both cell populations. However, a strong alteration (>2 or <0.5) could be found for a set of 25 genes.

For validation of these results, a subset of 20 targets with strongly altered expression in CD138neg MM cells compared to CD138pos was chosen (Table 1). For this, the remaining 1.4 μg mRNA was analyzed using qPCR. Of the 20 genes, the induction of 11 genes could be validated (genes marked with * in Table 1); 4 genes gave diverging results and 5 genes could not be detected by qPCR. The positively validated genes show a good accordance with the microarray data.

Table 1

Ratios between differentially expressed genes.

Gene name (* – validated)	Array result (CD138neg/CD138pos)	p-Value of array results	qPCR result (CD138neg/CD138pos)
FGF18 *	17.43 ± 5.6	0.016	4.28
SMOH *	15.61 ± 0.53	0.002	25.99
DYRK3 *	9.16 ± 0.43	0.006	28.84
ABCB11	8.05 ± 3.61	0.013	Not detected
CYP19 *	6.82 ± 0.65	0.092	3.86
PRKAA1 *	6.21 ± 0.8	0.062	4.59
GNB2	5.68 ± 2.09	0.041	1.51
NFATC2 *	5.26 ± 0.32	0.007	2.92
PCDHγB4 *	4.88 ± 2.28	0.008	50.21
GRB10 *	4.87 ± 0.44	0.020	2.22
CSEN	4.69 ± 3.3	0.219	1.15
KCNIP1	4.26 ± 2.2	0.040	Not detected
BLR1	4.25 ± 4	0.013	Not detected
KCNQ2 *	4.02 ± 0.58	0.014	7.21
MYO1C	3.96 ± 0.02	0.015	1.07
OSMR *	2.73 ± 0.52	0.006	9.19
PCDHγB7 *	2.25 ± 0.75	0.046	2.14
CASQ1	0.45 ± 0.01	0.010	Not detected
CSNK1A1	0.33 ± 0.03	0.006	2
UBE2I	0.30 ± 0.02	0.004	Not detected

Discussion

We developed an inexpensive, robust and amplification-free ultrasensitive microarray assay for competitive hybridization, which was successfully applied to mRNA expression profiling of MM CD138neg cells, a very rare subpopulation of self-renewing tumor cells responsible for tumor growth, minimal residual disease and patients’ relapse (Matsui et al., 2004). The key challenges of the presented approach were of technical and biochemical nature: the platform required a two-color extension for the detection and a robust analysis system capable of correct alignment and fast analysis of large arrays at single molecule level (Muresan et al., 2010). Moreover, while RNA isolation using standard kits worked reliably, cDNA purification using a commercial kit resulted in substantial loss of sample material. Using the human keratinocyte cell line HaCaT, we implemented a precipitation-based purification, which, in combination with single molecule microarray analysis, allowed amplification-independent expression profiling with as little as 200 ng total RNA corresponding to about 10.000 cells as starting material. The ratios obtained via molecule counting on single molecule microarray data showed no dependency on the labeling as differential labeling efficiencies do not distort the quantification; here: a molecule is counted as long as it carries at least one dye label. This eliminates the commonly required color swap experiments further reducing the amount of biological material and time-consuming lab work.

The platform was used to analyze the gene expression profile of MM CD138neg cells found in the human cancer cell line NCI-H929. The low frequency at which CD138neg cells occur in tumor samples rendered standard analysis approaches inappropriate in terms of sensitivity. The single molecule microarray approach provided quantitative information on 400 genes represented in the human signal transduction set. This yielded a set of 20 genes differentially expressed between the CD138neg and CD138pos population. Using qPCR, the altered expression of 11 genes could be validated, 4 genes showed contradictory results and 5 genes were not detected in the qPCR analysis. This result demonstrates i) the high sensitivity of the single molecule platform and ii) the good correlation of qualitative expression changes. However, we determined quantitative differences in the differential mRNA expression between single molecule microarray and qPCR analysis. This might be e.g. due to unspecific hybridization on the microarray, variations in amplification efficiencies depending on the primer design as well as the unspecific background signal of the SYBR Green technology used for qPCR analysis (Morey et al., 2006).

Previous studies of have implicated the Hedgehog/GLI pathway in the control of MM CSC (Ghosh and Matsui, 2009; Peacock et al., 2007). Inhibition of the essential Hedgehog effector Smoothened (SMO), a seven transmembrane domain protein with similarities to G-proteins coupled receptors (Ayers and Therond, 2010), prevents self-renewal and disease propagation of CD138neg MM cells by promoting differentiation of CSC (Peacock et al., 2007). In this context it is noteworthy that we found high expression of SMO in the CD138neg population compared to CD138pos cells. Together with high-level expression of the Hedgehog-regulated transcription factor GLI1 and Hedgehog ligands (data not shown), this suggests that MM CSC may be preferentially sensitive to SMO antagonists, in analogy to the recent report showing that high level SMO expression has been correlated with sensitivity of chronic lymphocytic leukemia cells to SMO inhibitors (Decker et al., 2012). Further, increased expression of the dual-specificity tyrosine-phosphorylation regulated kinase 3 (DYRK3) points to an additional layer of regulating Hedgehog signal strength in MM CD138neg cells, as DYRK kinases are well known modifiers of Hedgehog signaling by affecting the transcriptional activity of GLI proteins (Mao et al., 2002) (Varjosalo et al., 2008). In summary, the gene expression profile of MM CD138neg cells determined with our single-molecule microarray platforms provides descriptive evidence supporting an involvement of Hedgehog signaling in MM CD138neg cells, an observation that warrants further experimental investigations.

On the technical side, the presented work clearly demonstrates that the combination of single molecule microarray analysis with an improved cDNA purification method provides a powerful and robust technology platform. It allows cost effective and sensitive analysis of known targets in very small biological samples by comparative hybridization, a field where other techniques like single molecule sequencing (Harris et al., 2008; Korlach et al., 2008) and single cell genomics (Kalisky and Quake, 2008) are currently too expensive but may become affordable in the future. Our platform marks a valuable technological advancement for the molecular analysis of small and complex heterogeneous samples such as rare subpopulations of highly malignant cancer cells.

Funding

This work has been supported by the Austrian Science Fund (FWF): L422-N20 and Y250-B03, by the GEN-AU program of the Austrian Federal Ministry of Education, Science and Culture, by the State of Upper Austria and by the European Fund for Regional Development (Hochsensitive Analyze-Methoden für die Untersuchung von biologischen Proben, Wi-219244/20–2010/Kr/Zs) and by the Focus Area “Biosciences and Health” of the University of Salzburg.