Position of the fluorescent label is a crucial factor determining signal intensity in microarray hybridizations.

A key issue in applications of short oligonucleotide-based microarrays is how to design specific probes with high sensitivity. Some details of the factors affecting microarray hybridization remain unclear, hampering a reliable quantification of target nucleic acids. We have evaluated the effect of the position of the fluorescent label [position of label (POL)] relative to the probe-target duplex on the signal output of oligonucleotide microarrays. End-labelled single-stranded DNA targets of different lengths were used for hybridization with perfect-match oligonucleotide probe sets targeting different positions of the same molecule. Hybridization results illustrated that probes targeting the labelled terminus of the target showed significantly higher signals than probes targeting other regions. This effect was independent of the target gene, the fluorophore and the slide surface chemistry. Comparison of microarray signal patterns of fluorescently end-labelled, fluorescently internally random-labelled and radioactively end-labelled target-DNAs with the same set of oligonucleotide probes identified POL as a critical factor affecting signal intensity rather than binding efficiency. Our observations define a novel determinant for large differences of signal intensities. Application of the POL effect may contribute to better probe design and data interpretation in microarray applications.

INTRODUCTION

Increasingly, high-density oligonucleotide microarrays have been exploited for all fields of biological research on a large scale (1,2). Therefore, it is compulsory to obtain high specificity of oligonucleotide probes and high sensitivity of microarray hybridizations (1), especially for the application in the field of single-nucleotide polymorphism (SNP) analysis and diagnostics. However, several key issues of this technology such as the design of optimal oligonucleotide probes and their actual versus their predicted hybridization behaviour still have to be resolved (3). All microarray-based techniques for the analysis of DNA variation rest on the hybridization of complementary strands of DNA (4,5). The interpretation of microarray-based experiments relies on the signal intensity derived from labelled samples hybridized to the array. However, many variables contribute to the level of the signal detected. It has been often observed that some surface-bound probes targeting certain regions show low or even false negative signals, and even carefully designed oligonucleotide probes often display two orders of magnitude difference in signal output (6). Attempts to explain this behaviour include e.g. signal-suppressing parameters such as secondary structure of the target molecules, or steric hindrance of duplex formation, selectively acting on different probe binding sites (4). Less attention appears to have been paid to the relative position of label (POL) with respect to the duplex formed.

In the course of developing a diagnostic microarray for nitrogenase gene diagnostics (nifH phylochip), we compared in this study hybridization patterns of multiple surface-bound, short oligonucleotide probes targeting different regions of the target-DNAs. The results suggest that POL is a crucial factor determining signal intensity. Our observations provide an important insight into factors affecting microarray performance and define a novel determinant for large differences of signal intensities. Application of the POL effect will contribute to an improved probe design and improved data interpretation in microarray applications.

MATERIALS AND METHODS

End-labelled targets

The target sequences were fragments of the nifH gene encoding the nitrogenase iron protein of the nitrogen-fixing endophytic bacterium Azoarcus sp. strain BH72 (AF200742) (7). When stated otherwise, nifH of Azoarcus indigens VB32T (8) or vnfH of Azotobacter vinelandii DSMZ 366 was used. End-labelled targets were obtained by PCR using three different primer pairs, generating three amplicons with different lengths (Z1 = 362 nt length, position 737–1098 nt on AF200742; Z2 = 322 nt length, position 737–1058; Z3 = 209 nt length, position 850–1058 nt). Chromosomal bacterial DNA used as a template was purified by standard procedures (9). Degenerate oligonucleotides typically used for amplification of divergent nitrogenase genes (nifH, anfH and vnfH) (10) were applied to amplify target Z1, the forward primer being labelled with Cy5, Cy3 or biotin, respectively, at the 5′ end [Z1-nifH-F: 5′-TG(CT) GA(CT) CC(AGCT) AA(AG) GC(AGCT) GA-3′], and the reverse primer being modified with biotin or Cy3, respectively, at the 5′ end [Z1-nifH-R: 5′-A(AGT)(ACGT) GCC ATC AT(CT) TC(ACGT) CC-3′]. For the shortened fragments Z2 and Z3, primers were designed to specifically amplify nifH of Azoarcus sp. strain BH72 [Z2-nifH-F-Cy5, 5′-TGC GAC CCG AAG GCT GA-3′, and Z2-nifH-R-Biotin, 5′-GGG CCT TGT TTT CGC G-3′; Z3-nifH-F-Cy5, 5′-CCT GTC GGT CGG CTT C-3′, and Z2-nifH-R-Biotin, 5′-GGG CCT TGT TTT CGC G-3′]. Amplification reactions were performed as described previously (11). Amplification of a nifD fragment (365 nt length, position 2109–2473 nt on AF200742) (7) was carried out with primer nifD-BH72-F labelled with Cy5 at the 5′ end [ATC GGC GAC GAC ATC GAA GCC] and biotinylated primer nifD-BH72-R [TAG TTC ATC GAG CGG TAG CAG T]. Amplification conditions were an initial denaturing for 5 min at 95°C, then 40 cycles of 95°C for 1 min, 60°C for 1 min, 72°C for 1 min, with a final extension at 72°C for 5 min. Successful amplification was confirmed by analysing PCR products on 2% agarose gels, and fragments were purified with the QIAquick PCR purification kit (Qiagen, Hilden, Germany).

For hybridization, fluorescence end-labelled single-stranded DNAs (ssDNAs) were isolated from the double-stranded DNA amplification products. Strands were separated with streptavidin-coated paramagnetic particles (Roche, Mannheim, Germany) as described previously (12), and the concentration of ssDNA was determined by UV spectrometry. Shortly before use, the ssDNA was routinely denatured at 95°C for 10 min and then snap-cooled on ice.

Alternatively, the nifH target was end-labelled with [γ-33P]ATP (25 TBq/mmol; Hartmann Analytics, Braunschweig, Germany) by using T4 polynucleotide kinase (Fermentas, St Leon-Roth, Germany) according to manufacturer's instructions; the probe was separated from unincorporated label by gelfiltration with P6 Biogel.

Random-labelled target

Unlabelled Z2 sense was generated by PCR using unlabelled primer Z2-nifH-F and primer Z2-nifH-R-Biotin. After strand separation, ssDNA was used for multiple random fluorescent labelling with Alexa Fluor 647 included in ULYSIS nucleic acid labelling kit (Molecular probes, Inc., Eugene, Oregon) following the manufacturer's instructions. Shortly before use, the fluorescent-labelled ssDNA was routinely denatured at 95°C for 10 min and then snap-cooled on ice.

Oligonucleotide probes

All oligonucleotides targeting the nifH gene of Azoarcus sp. strain BH72 were designed with GeneDoc [K. B. Nicholas and H. B. J. Nicholas (1997) ] according to the following criteria: their binding region had similar characteristics with respect to length (16–18 nt), %GC content, melting temperature (Tm) calculated with MELT 1.1.0 (J. P. Sanders), and ΔG taking into account the complementary strand. Their sequence is shown in Table 1 and their relative position with respect to the target fragment in Figure 1. They all carried polyadenosine triphosphate spacers (6 A), had either a 5′- or 3′-amino modification (Aminolink C6), and were synthesized by Thermo Electron GmbH (Ulm, Germany). Secondary structure predictions for the target strands were carried out using the Mfold program (13). The same criteria and methods were used to design oligonucleotides targeting a different gene, nifD of strain BH72, designated as nifD-A1 and so forth (Table 1). As examples of longer oligonucleotide probes, 50mers specific for the nifH gene of Azoarcus sp. strain BH72 (7) were designed: A1–50 (position 737–786 nt on AF200742), A19–68 (position 755–804 nt on AF200742) and A174–223 (position 910–959 nt on AF200742).

Table 1

Oligonucleotide probes used in this study

Probesa	Sequences (5′–3′)b	Tm (°C)c	Length (nt)	GC%	Positiond
A1	AAAAAA TCAGCCTTCGGGTCGCA	68	17	64	1–17
A4	AAAAAA GAGTCAGCCTTCGGGTC	68	17	64	4–20
A8	AAAAAA GGTGGAGTCAGCCTTCG	68	17	64	8–24
A14	AAAAAA CAGACGGGTGGAGTCAG	68	17	64	14–30
A20	AAAAAA GAGGATCAGACGGGTGG	68	17	64	20–36
A64	AAAAAA CTTCAGCGGCCAGGTG	68	16	68	64–79
A114	AAAAAA GAAGCCGACCGACAGG	68	16	68	114–129
A188	AAAAAA GGCGGTGATAACGCCAC	68	17	64	188–204
A235	AAAAAA GAAGTCGAGTTCGTCGTC	67	18	55	235–252
A307	AAAAAA GGGCCTTGTTTTCGCG	65	16	62	307–322
S20	AAAAAA CCACCCGTCTGATCCTC	68	17	64	327–343
S64	AAAAAA CACCTGGCCGCTGAAG	68	16	68	284–299
S114	AAAAAA CCTGTCGGTCGGCTTC	68	16	68	234–249
S188	AAAAAA GTGGCGTTATCACCGCC	68	17	64	160–176
S235	AAAAAA GACGACGAACTCGACTTC	67	18	55	113–130
S307	AAAAAA CGCGAAAACAAGGCCC	65	16	62	41–56
A20-3′	GAGGATCAGACGGGTGG AAAAAA	68	17	64	20–36
A307-3′	GGGCCTTGTTTTCGCG AAAAAA	65	16	62	307–322
nifD-A1	AAAAAA TTCGATGTCGTCGCCGAT	67	18	55	1–18
nifD-A8	AAAAAA AGACGGCTTCGATGTCGT	67	18	55	8–25
nifD-A77	AAAAAA ACACGCCACGGAAGCCTT	69	18	61	77–94
nifD-A185	AAAAAA CGATGATGGTGACGTCGT	67	18	55	185–202
nifD-A307	AAAAAA CTTCGGGGTGTTCTCCAT	67	18	55	307–324
VB-A34	AAAAAA TGTGCCTTGCTGTGCAG	66	17	58	34–50
VB-A164	AAAAAA GCAGCCGACGCCGGGTT	71	17	76	164–180
VB-A307	AAAAAA GGGCCTTGTTTTCGCG	65	16	62	307–322
vnf-A47	AAAAAA ATGACGGTGCCCTGGG	68	16	68	47–62
vnf-A110	AAAAAA GCCGATCTGCAGCACGT	68	17	64	110–126
vnf-A309	AAAAAA TCCTGGGCCTTGTTCTCG	69	18	61	309–326

aA and S represent antisense reverse or sense probes, respectively, the numbers indicating the corresponding regions in the target fragment of strain BH72: e.g. S20 corresponds to sequence 20–36 nt of the sense strand, and A20 is reverse complementary to S20; 5′ or 3′: the oligonucleotide was amino-modified on the 5′ or 3′ end, respectively.

bPolyadenosine triphosphate spacer (6A) on either 5′ or 3′ end shown with gap.

cTm values were calculated with MELT 1.1.0 (Jo P. Sanders) at a salt concentration of 600 mM.

dPositions of the probes were shown according to the binding regions on the complementary targets. Binding positions of sense probes on the antisense target were shown in italics.

Figure 1

Schematic representation of the probe positions relative to the targets. Azoarcus sp. BH72 nifH Z-fragments of different lengths amplified from genomic DNA of Azoarcus sp. BH72 were used as fluorescence-labelled targets. The regions harbouring primers for amplification by PCR are shown as dotted lines. Short solid lines represent the regions where the probes were selected; e.g. S20 was designed from the region between 20 and 36 nt of the sense strand, and A20 is reverse complementary to S20. The fluorescent label is always located at the 5′ end of the strand and shown as a star.

Fabrication of microarrays

The DNA microarray format used in this study was based on standard microscopic glass slides (Menzel, Braunschweig, Germany). Chemicals and solvents were from Fluka (Neu-Ulm, Germany). The substrates were cleaned, silylated and activated (isothiocyanate-functionalized surface) as described by Beier and Hoheisel (14). The activated surfaces were used directly for immobilization of either 5′- or 3′-amino-modified capture oligonucleotides via covalent binding. Probes were spotted onto the activated slide surface using the piezo-driven spotting device Robodrop (BIAS, Bremen, Germany) bearing a glass pin, or MicroGrid II Compact 400 (Genomic Solutions Ltd, Huntingdon, Cambridgeshire) with an ArrayIt™ split pin (TeleChem International, Inc., Sunnyvale, CA), respectively. The concentration of the amino-modified oligonucleotides in 1% glycerol was 10 µM. Volumes of deposited oligonucleotide solutions were ∼200 pl, resulting in spots with a diameter of ∼200 µm. To complete covalent binding, after being spotted with probe solutions, slides were incubated overnight at room temperature in a wet chamber to prevent evaporation of the spots. Blocking of the slides was performed in 6-amino-1-hexanol (50 mM) and diisopropylethylamine (150 mM) in dimethylformamide (15).

Five types of commercial slides with different chemically modified surfaces were used for comparison. Three types of slides were purchased from Genetix (Dornach, Germany) with aldehyde, amine or ‘aldehyde plus’ surface; one was purchased from PEQLAB Biotechnologie (Erlangen, Germany) with amino surface coating (QMT Amino Slides); one amine slide (Pan® Amine slide) was purchased from MWG Biotech (Ebersberg, Germany). Spotting and hybridization of microarrays were performed following manufacturer's instructions.

Microarray hybridization and evaluation

Hybridization and washing were done in a PersonalHyb oven (Stratagene). Hybridization time varied from 1 h to overnight, without significantly different results. Incubation was performed at 25°C. The final concentration of target ssDNA in hybridization buffer [4× SET (9), 10× Denhardt's solution and 50% formamide) was 10 nM. To guarantee a uniform moistening of the slide surface, the sample was covered with a cover slip. Washing was done routinely with 2× SET/0.1% SDS for 5 min and 1 × SET/0.1% SDS for 10 min at room temperature. Spin-dried slides were imaged at a resolution of 10 µm with a GenePix4000A microarray scanner (Axon, Union City, CA) at the same laser power and sensitivity level of the photomultiplier for each slide so that absolute signal intensities (arbitrary units) from independent experiments can be compared. One-way Analysis of Variance (ANOVA) was calculated with Graphpad instat 3.01 (GraphPad Software, Inc., San Diego, CA). Alternatively, microarrays with radioactively labelled targets were exposed for 1 h at room temperature to a phosphoimager screen and analysed by a Typhoon™ 8600 Variable Mode Imager (Molecular Dynamics Inc., Sunnyvale, CA) in combination with ImageQuant® v.5.1.

RESULTS

The design of the test system

During the development of a diagnostic microarray for environmental monitoring of nitrogenase gene diversity (nifH-phylochip), optimization experiments suggested that the relative position of the fluorescence label might have an impact on signal intensities. In order to test the effect of different oligonucleotide probe positions with respect to the fluorescence label of the target sequence, we focussed on the nifH gene from Azoarcus sp. strain BH72 encoding the iron protein of the nitrogenase complex of this bacterium. NifH fragments were either end-labelled by fluorescence-labelled primers used for amplification by PCR, or randomly labelled after amplification. Oligonucleotides targeting different sequence regions of these fragments were used as probes for microarray hybridizations with single-stranded targets, binding either the sense or the antisense strand. The binding positions with respect to the target sequence are shown in Figure 1. Antisense probes binding the sense strand are designated as A, sense probes binding the antisense stand as S, the numbers referring to the relative nucleotide position of the probe with respect to the sense strand of target Z1. All oligonucleotides represented perfect-match probes. Probe thermodynamics (length, Tm, %GC, ΔG and secondary structure) were carefully controlled to fall in a range as narrow as possible (Table 1).

Effect of the position of fluorescent label relative to the probe-target duplex for end-labelled ssDNA targets: impact of target and secondary structure

Six pairs of complementary probes targeting different regions on the sense strand of Z1 were first evaluated for variations of signal intensities. Cy3-labelled Z1 sense strand ssDNA was used as target for hybridization with antisense probes spotted on the slide. Surprisingly, A20 showed the highest signals despite an expected strong steric hindrance by the non-bound target fragment (Figure 2A). The signal intensity decreased with increasing distance of the labelled nucleotide from the duplex formed by hybridization, being 638-fold lower for oligonucleotide A188 in a middle position for which the lowest intensity was obtained (Figure 2A). The increase of signal intensity when oligonucleotides hybridize more closely to the end-labelled target is referred to as ‘POL effect’ or simply ‘position effect’.

Figure 2

Hybridization of Cy3-labelled Z1 sense strand (A) or antisense strand (B) ssDNAs with antisense reverse or sense probes, respectively, spotted on slides. Schematic representation of three pairs of hybrids on the surface is shown for three positions A/S 20, 114 or 307, respectively. The label located at the 5′ end of the target is shown as a star, poly(A) spacers of the probes as a dashed line. Four spots are shown for the probes targeting six positions on antisense or sense strand, which are in the same order as the probe names in the quantitative analysis below (arbitrary fluorescence units given with standard deviations). The hybridization images were adjusted for best viewing (quantitative conclusions drawn from the image may be misleading). Data are from 24 spots on three different slides.

To analyse the impact of putative secondary structure formation of the target on the observed effect, the microarray hybridization strategy was reversed (Figure 2B). The antisense strand was used as 5′ end-labelled target, and the reverse complement of the oligonucleotides was used as probes binding to the same regions of the target as in Figure 2A (e.g. sense probe S20 reverse complement of A20). Since the complementary DNA strands of the target DNA were expected to form similar secondary structures in the corresponding regions (not shown), hybridization results would be expected to be similar for sense and antisense probes in case the secondary structure was the major determinant of the expected effect. In contrast, again the probe hybridizing closest to the fluorescent end of the target (S307, Figure 2B) provided the strongest signal but not the distant probe S20, with the lowest signals occurring for probes in middle positions (for S188 in comparison with S307 112-fold lower). This strengthened the assumption that the relative position effect caused differences in signal intensities.

To analyse the impact of a shorter distance between the end-label of the target and the hybridized region, a different target Z2 was utilized that differed from Z1 by amplification with perfect-match primers to nifH of strain BH72 instead of degenerate primers. This allowed the design of perfect-match probes spanning also the region of the PCR primer (A1–A14 in Figure 3). As expected from our POL hypothesis, probe A1 showed the highest signal, with a successive decrease of signal intensities with distance increase down to Probe A188. The difference in signal intensities of probe A1 and A64 was ∼15-fold, that of probe A1 and A188 was >355-fold (Figure 3), the latter can be regarded as a false negative signal. Thereafter (A188–A307) signals increased slightly.

Figure 3

(A) Hybridization of Cy5-labelled Z2 or Z3 sense strand ssDNA, respectively, with antisense reverse probes targeting different regions. Four spots are shown for the probes targeting 10 positions on Z2 sense strand (upper panel) and 4 positions on Z3 sense strand (lower panel). (B) Hybridization of a Cy5-labelled nifD sense strand ssDNA with antisense reverse probes targeting different regions. The hybridization images were adjusted for best viewing (quantitative conclusions drawn from the image may be misleading). The quantitative analysis below (arbitrary fluorescence units given with standard deviations) was calculated from 24 spots on three different slides.

To exclude the possibility that the probes corresponding to the middle positions provided low signals because the decrease of signal intensities was dependent on the sequence, they were brought more closely towards the labelled end of the target by shortening the target. Target Z3 differed from Z1 by a truncation of 153 nt at the 5′ end that was also Cy5-labelled. Probe A114 that gave almost negligible hybridization results with full-length target showed high signal intensities when it became positioned at the labelled end of the truncated target (Figure 3, dashed line), demonstrating that it can effectively hybridize with the target. In concordance with the postulated position effect, more distantly hybridizing probes resulted in lower signals (5.2-, 2.5- or 1.8-fold lower for A188, A235 or A307, respectively, statistically significant at P ≤ 0.01 level).

The position effect was not specific for the nifH gene of Azoarcus sp. BH72, since similar results were obtained for the Cy3-labelled Z1 fragment of nifH amplified from another species, A.indigens VB32T with perfect-match probes numbered according to the same criteria (VB-A34, VB-A164 and VB-A307) (94% identity of DNA sequence). The probe hybridizing most closely to the labelled end showed the highest signal, 39.5 ± 5.2 or 13.8 ± 2.9-fold higher than VB-A164 or VB-A307, respectively. The differences were statistically significant (P < 0.001). Also for a phylogenetically more divergent gene encoding the vanadium nitrogenase vnfH of A.vinelandii (84% DNA identity), results showed the same trend: vnf-A47 showed the highest signal intensity, vnf-A110 13.8 ± 2.6-fold, or vnf-A309 4.4 ± 0.98-fold lower signals, respectively. The differences were statistically significant (P < 0.001). Moreover, we used an entirely different gene, nifD, encoding a component of the dinitrogenase that is phylogenetically, functionally and structurally different from nifH encoding the dinitrogenase reductase. A fragment of 365 nt length (position 2109–2473 nt on AF200742) was amplified, end-labelled with Cy5, and the ssDNA hybridized as in the previous experiments. Hybridization to the five oligonucleotide probes specifically binding to this fragment showed a sharp decline of fluorescence with increasing distance to the label (Figure 3B): probe nifD-A1 showed a 342-fold higher signal than probe nifD-A185 located in a middle position. Thus, a large POL effect was also observed for a non-homologous gene.

Impact of the fluorophore, slide surface and other factors on the position effect

In order to test whether the nature of the fluorescent label had an impact on the position effect, the target sense strand was labelled with Cy5 instead of Cy3 (Figure 3). For the six probes tested in previous experiments (A20–A307) and a target of similar length labelled with Cy5, similar results were obtained (Figure 3, Z2).

The impact of the slide surface of the microarray on the position effect was tested for five types of commercial slides with different chemically modified surfaces: aldehyde, amine (QMT Amino Slides, Pan® Amine slides) or ‘aldehyde plus’ surfaces. With respect to the immobilization step, the isothiocyanate-functionalized surface in our lab has the advantages of simplicity and homogeneous distributions over the other five types of slides, since it does not require fixation steps, such as baking at elevated temperature or irradiation with UV light. The formation of strong ionic interactions or a varying number of covalent bonds between the surface and the DNA oligonucleotides are avoided. A20–A307 and S20–S307 were used as probes; Cy3-labelled Z1 sense and antisense ssDNAs were used as targets. With respect to the signal intensities obtained from hybridization experiments, the slides with aldehyde, amine or ‘aldehyde plus’ surface from Genetix and the slides with amine surface from MWG Biotech were found to yield low hybridization signals. However, the isothiocyanate-functionalized surface and the QMT Amino Slides revealed ∼10-fold higher fluorescence signals and more homogenous distributions. With respect to the POL effect, all five kinds of commercial slides showed similar signal variation patterns as the self-made slides, indicating that the observed phenomenon could not be attributed to the differences between microarray surfaces.

To assess the effect of possible steric hindrance of the non-hybridized part of the target, the same target was used for hybridization; however, we reversed the orientation of the probes on the microarray by attaching them at their 3′ end (Figure 4). The position effect was also apparent, probe A20 or A20-3′ providing higher signals compared with A307, irrespective of the orientation of the bound target. The 3′-immobilized probe A307-3′, however, showed a false negative result (Figure 4), which is probably caused by a combination of position effect, steric hindrance of hybridization and quenching of fluorescence close to the slide surface.

Figure 4

Hybridization of Cy3-labelled Z1 sense strand ssDNA with two pairs of antisense reverse probes with different immobilizing orientations. Schematic representation of different hybrids on the surface is shown along with four spots of hybridization image. The label located at the 5′ end of the strand is shown as a star, poly(A) spacer of the probe as a dashed line. The hybridization images were adjusted for best viewing (quantitative conclusions drawn from the image may be misleading). The quantitative analysis below (arbitrary fluorescence units given with standard deviations) was calculated from 16 spots on two different slides.

Impact of internally, randomly labelled ssDNA or radioactively labelled ssDNA on signal intensity

The data suggested that the relative position of the fluorescent end-label of the target is crucial for the observed effect. Therefore randomly labelled target was tested, which should diminish the effect. After strand separation of the Z2 target amplified with non-fluorescent primers, the sense strand ssDNA was labelled internally and randomly with Alexa Fluor 647 included in the ULYSIS nucleic acid labelling kit, and then used as target to hybridize with the same set of 10 antisense reverse probes spotted on the slide as in Figure 4. In comparison with the microarray results of end-labelled targets, the variation of signal intensities of different probes was strongly reduced (Figure 5A). The ratio between the highest signal (A8) and the lowest signal (A188) was only 4.7-fold (P < 0.001, ANOVA, also for probes A1–A20 in comparison with probe A64 and higher), suggesting that the stronger effects observed for end-labelled probes were indeed related to the position of the fluorescence label but not to the differences in hybridization efficiencies.

Figure 5

Hybridization of alternatively labelled Z2 sense strand ssDNA with antisense reverse probes targeting different regions. (A) Random fluorescent labelling with Alexa Fluor 647. (B) Radioactive end-labelling with [33P]. Four spots are shown for the probes targeting 10 positions on Z2 sense strand. The hybridization images were adjusted for best viewing (quantitative conclusions drawn from the image may be misleading). The quantitative analysis below (arbitrary fluorescence units given with standard deviations) was calculated from 24 spots on three different slides (A) or 16 spots from two different slides (B).

If the effect was related to fluorescence, a strong POL effect would not be expected for a radioactively labelled target. Therefore, the same set of probes was hybridized to the same type of probe end-labelled with the isotope [33P] (Figure 5B). Here, no sharp decrease in signal intensity was observed. Except for probes A14 or A20, respectively, versus A307, the signals were not significantly different (P ≥ 0.05, ANOVA).

Impact of oligonucleotide probe length on the POL effect

In many applications, e.g. in microbial ecology, longer oligonucleotide probes in the range of 50–70 nt are immobilized to the slide surface. Therefore we analysed the effect of probe length (50mers) on the signal intensity in hybridizations with randomly labelled (Alexa Fluor 647) or 5′ end Cy5-labelled target fragment Z2 of nifH. The signals for the different probes close to the label (A1–50) or in the middle position (A174–50) were not significantly different for the randomly labelled target: A1–50, 29 750 ± 1073; A19–68, 27 811 ± 3060; A174–223, 30 982 ± 956. In contrast, there was an enhancement of signal intensity when the end-label was close to the probe binding site, however, the factor was only 3.8: A1–50, 18 918 ± 712; A19–68, 14 575 ± 366; A174–223, 4975 ± 313. This showed that in microarrays utilizing longer oligonucleotides, the POL effect is still occurring, albeit at a weaker level.

DISCUSSION

The results of our study indicated that the signal intensity in oligonucleotide-based microarray experiments is influenced by the relative position of the nucleic acid hybrid towards the fluorescence label of the target: when oligonucleotides hybridize to the region closer to the end that is labelled, the fluorescence signal is stronger. We term this effect as ‘position effect’ or ‘POL effect’. The following lines of evidence suggest that this is a significant, previously unknown factor in microarray evaluation.

The probes targeting different regions of our test targets shared similar characteristics; moreover, probes from six positions were also applied as reverse complementary probes for comparison of the impact of complementary strands (Figure 1). Theoretically the complimentary probe pairs should have formed the same hybrids with their corresponding targets and thus given similar results in microarrays hybridizations. However, it was found in repeated experiments that probes targeting the regions close to the labelled ends generated higher signals than other probes, regardless of the target sequences (Figures 1 and 2).

In most of the previous studies, signal variations were attributed to the effect of secondary structure and steric hindrance, co-acting with certain extent. Secondary structure contributes to the low binding efficiency more predominantly, mainly owing to the complex secondary structure of long targets used in those experiments. In our study, several aspects suggest that the effect of secondary structure was not the main cause of the large variation between signal intensities: (i) ssDNA was routinely denatured before hybridization, and the hybridization was performed at high stringency (50% formamide) preventing the ssDNA from forming complex structures; (ii) base pair forming in different target regions calculated in silico (13) could not explain the observed hybridization phenomenon, moreover reverse complementary hybridization experiments did not result in identical signal strengths (see above); (iii) ΔG and ΔS of all formed hybrids were estimated in the same range (16,17) (data not shown); and (iv) radioactively labelled target that is expected to have the same binding and secondary structure features did not show a significant POL effect (Figure 5B). The target used in these experiments was relatively short (maximum 362 nt), therefore the chance of formation of complex secondary structures was alleviated. Utilization of long target molecules could be a reason why the POL effect was not observed in previous oligonucleotide microarray applications. Long target sequences in comparison with short ones are more probably to fold on themselves as a result of intra-molecular Watson–Crick base pairing, which hides parts of the target from the oligonucleotide probes (18) and obscures quantitative results.

The large variation of signal intensities observed is also unlikely to be the effect of steric hindrances. It has been observed in previous studies that immobilized probes could form hybrids more easily when they are away from the support surface (4). Although it was shown that the optimal spacer length should be at least 40 atoms (19), several studies have illustrated that spacer with 5, 6 or 12 atoms in length could affect the signal intensity significantly as well, especially one using a similar test system as this study (18). The extension of spacers from A6 to A12 in our study did not have a significant impact on the position effect (data not shown). Also a steric hindrance of hybridization owing to long overhanging ends of non-hybridized target DNA close to the array surface cannot explain the POL effect: comparison of different hybrids formed by the target with the same oligonucleotide having different immobilization directions and thus different steric hindrance (Figure 4) resulted in equally high signals as long as the fluorescence label was located close to the hybrid. However, probes resulting in sterically favourable hybrids showed very low signal intensities when the label was distant to the hybrid (see e.g. Figure 2, S20 versus A20).

Both signal-suppressing parameters are more or less correlated with the binding efficiency. However, the hybridization results of randomly, internally labelled target and of radioactively labelled target clearly showed that there was only a weak or no difference of binding efficiency for the different probes used (Figure 5). These minor variations might indeed be attributed to secondary structure or steric effects. The obvious difference among hybrids with end-labelled targets is POL and fluorescence. Since the observed strong effect cannot be attributed to differences in hybridization efficiency, it is most probably caused by increased fluorescence signal intensity. The dramatic variation of signal output might be attributed to multiple factors, however the physical and chemical basis of the POL effect is not yet fully understood. The flexibility of the labelled strand might increase with the distance to the double helix, therefore strands might fray with each other, leading to a decreased fluorescence. However, a nucleotide–fluorophore quenching effect may occur. It has been shown that fluorescence resonance energy transfer and contact-mediated quenching exist between fluorophores and different nucleotides (20). Probe-target hybrids with distant label, having a higher flexibility of the strand, might have a higher chance of fluorophore–nucleotide interaction and thus quenching of the fluorescent signal. Likewise, fluorescence might increase in proximity to a DNA double helix.

We obtained similar results with other PCR products of the nifH gene (genes from other microorganisms, truncated fragments), and with a second gene (nifD) that is phylogenetically, functionally and structurally different. This indicates that the effect is sequence independent and applicable to a wide range of targets. Moreover, our data suggest that the effect is largely independent from the chemical nature of the fluorophore or the microarray slide surface. Therefore it should be applicable to a wide range of oligonucleotide microarrays using end-labelled targets. They are widely used techniques in the field of SNP analysis, diagnostics, as well as microbial ecology and diversity studies. Moreover, the effect is not only observed for short oligonucleotide probes, but also for longer probes (50mers), albeit at a weaker level.

There is always a combined set of factors influencing signal output in microarray experiments. However, the POL effect should be considered as a critical factor for microarray performance when probes of similar features are compared. The observed differences of probe efficiency towards the same target molecule indicate the importance of the POL effect for probe design. Oligonucleotide probe design taking into account the POL effect can minimize false negative results, and can improve the selection of probes for maximum signal intensities. The sensitivity of microarray applications could be increased if the probes were selected from the proper regions. Also data interpretation, i.e. the comparison of signal intensities obtained for a set of probes, is affected: consideration of the POL effect will improve quantitative evaluation of microarray results. In conclusion, our observations provide an important insight into factors affecting microarray performance and define a novel determinant for large differences of signal intensities.