Elimination of laboratory ozone leads to a dramatic improvement in the reproducibility of microarray gene expression measurements

To obtain valid results from microarray experiments, many factors affecting microarray technology must be recognized and controlled [1–3]. Microarray analysis of gene expression depends on the relative binding (hybridization) of cyanine dye-labeled cDNAs or cRNAs to DNA probes covalently attached to microscope slides. The quality of microarray data depends on many factors and among the most important are stable ratios of the bound Cy5 and Cy3 dyes. Thus, each dye must remain intact from the completion of the hybridization process through the duration of the scanning process. The cyanine dye Cy5 is subject to ozone (0₃) oxidation resulting in a decrease in fluorescence intensity [4]. Because two-color microarray experiments depend on the ratio of the Cy5 and Cy3 signal intensities for relative gene expression measurements, specific, rapid, and uncontrolled degradation of the Cy5 dye would result in inaccurate gene expression/repression (Cy5/Cy3) ratios and erroneous interpretation of microarray data. Oxidation occurs primarily after the hybridization washing procedures have been completed and the microarray becomes exposed to air containing environmental ozone.

Ozone in the lower atmosphere is present at all times of the year and is generally higher during the summer months. Ozone is a principal component of smog so ozone levels are higher in urban and/or industrialized areas compared with rural areas. The mixture of hydrocarbons and nitrogen oxide from automobile exhaust and factory emissions combined with exposure to sunlight in the presence of little air movement leads to the generation of ozone in the lower atmosphere [5]. Cy5 oxidation can occur at ozone levels as low as 5–10 ppb and possibly even lower. Average monthly ozone levels measured in central Arkansas between 1 January 2005 and 30 June 2005 ranged between 16 ppb and 43 ppb. However individual hourly readings reached as high as 103 ppb during this interval [6]. In addition to seasonal changes, a diurnal variation of ozone levels occurs because more ozone is produced as more atmospheric pollutants are generated during daylight hours.

To overcome this widespread problem, a simple laboratory engineering solution was developed that resulted in low and consistent ozone levels throughout the entire laboratory. The data presented here demonstrate the effects of both ozone and humidity on microarray fluorescence measurements for both in-house fabricated arrays and those manufactured by Agilent Technologies (Palo Alto, CA, USA), and the dramatic improvement in fluorescent dye stability under the newly implemented carbon filtration system. We also provide several other simple practices that can reduce the detrimental impact of atmospheric ozone on microarray experiments.

An extensive database for air quality is maintained by the Arkansas Department of Environmental Quality (ADEQ) and was made available for this project. Figure 1 shows the average ozone concentrations during the months of January 2005 and June 2005, and illustrates the substantial seasonal differences in ozone levels in our region of central Arkansas. These data also show the diurnal fluctuation of the ozone level with the lowest level during the early morning hours and a peak between the hours of approximately 0900 and 1800. The average daily ground ozone level, in June 2005 ranged from 16 ppb to 66 ppb with individual hourly readings often in the 70–90 ppb range; a high of 103 ppb ozone was reached on June 28. Ozone readings within the microarray laboratory were similar to the environmental levels.

To reduce the laboratory ozone levels, a High-Efficiency Gas Adsorber (HEGA) carbon filter was installed in the laboratory air supply system (see Methods section for details). The airflow was also adjusted to keep the laboratory air pressure positive with respect to the hallway, thus preventing non-filtered air from entering the laboratory. Ozone readings taken within the carbon-filtered lab during June 2005 ranged from 2.6 ppb to 4.4 ppb (Figure 1). While filtration reduced the ozone levels to well below environmental ozone levels outside the lab, a slight diurnal variation is also seen within the lab. Two additional changes were also made in our scanning protocol. First, microarray hybridization experiments were scheduled so that washing and scanning were conducted early in the morning hours (0700–0900) when ozone levels are the lowest. Second, the microarray slides were scanned immediately after the final wash by spinning the slide dry for 10 seconds at 6,000 rpm using a micro-centrifuge equipped with a microscope slide adaptor and placing the microarray directly into the scanner. Fare et al. [4] demonstrated that degradation of the Cy5 signal by ozone is not significant as long as the slides remain in liquid.

The effects of ozone on an in-house fabricated microarray are shown in Figure 2. In this experiment, pairs of hybridized microarrays were first scanned in the carbon filtered laboratory. Following this, one of the microarrays was moved to a loosely covered slide box in the adjacent hallway where ozone was present at 25 ppb, while the other was kept in the carbon-filtered laboratory environment. The slides were scanned alternately in the carbon-filtered lab. Thus, the only time the slide was exposed to ozone was during the time it was in the hallway, which was for approximately 57 minutes during the 114 minute experiment. The reduction in Cy5 (red) signal in individual features (Cy5 dye excited at 635 nm) was dramatic across this time period. Similar reductions in Cy5 intensities were also observed in Agilent microarrays (images not shown).

Several experiments were conducted to establish a time course for microarray degradation in a non-carbon-filtered atmosphere and to demonstrate the improvement due to the procedural and engineering modifications described above. Figure 3A shows median feature intensity data from microarrays printed in-house that were scanned immediately after washing in the carbon-filtered laboratory (ozone ~2–4 ppb). After scanning, one of the microarrays was moved to the adjacent hallway which does not have carbon-filtered supply air. The hallway was used as an ozone exposure environment to enable direct measurement of ozone in both locations and to move the microarray into the scanner as quickly as possible during the experiment. The microarrays were scanned alternately with approximately 12–14 minutes between individual slide scans. This was the minimum time to scan the microarray, save the image data to the computer, and change to the alternate microarray. One microarray slide never left the carbon-filtered laboratory environment, while the other slide was returned to the non-carbon-filtered hallway after each scan. During the first 13 minutes, the reduction of the Cy5 signal was approximately 44% in the non-carbon-filtered environment (ozone level was ~25 ppb) while the reduction was less than 1% in the carbon-filtered laboratory. By contrast there was no reduction in Cy3 intensity. After 39 minutes, Cy5 intensity was reduced by 6% in the microarray maintained in the carbon-filtered laboratory environment and by 75% in the microarray from the non-carbon-filtered environment. Cy3 intensity increased slightly in the carbon-filtered laboratory, but decreased by about 5% in the non-carbon-filtered hallway after 39 minutes. Such experiments were repeated two times with similar results. Thus, using these microarray scanning procedures and engineering modifications, ozone degradation of the Cy5 dye was minimized.

The question arose as to whether microarrays supplied by a commercial vendor would respond to ozone differently than those printed "in-house". To examine this, an ozone degradation experiment was conducted on two hybridized Agilent Technologies microarrays (Fig. 3B). These results show a rapid decline in the Cy5 signal that is very similar to that observed using microarrays printed in-house. Ozone levels in the carbon-filtered laboratory remained constant at about 2–4 ppb; ozone in the non-carbon-filtered hallway was approximately 10 ppb during the scanning duration.

The effect of high humidity alone on the stability of the cyanine dyes was also assessed. Twenty in-house fabricated microarrays containing 10,000 rat oligonucleotides were hybridized with Cy3- and Cy5-labeled rat cDNA. They were initially scanned in the carbon-filtered laboratory environment in which the relative humidity was maintained at 42%. Following this, 10 microarray slides were placed in a humidified chamber at a constant humidity of 98–99% and the remaining 10 slides were left in the carbon-filtered, 42% humidity laboratory environment. After 2 hours and 4 hours, 5 slides from the high humidity chamber and 5 slides kept under laboratory environment conditions were rescanned at the original PMT/power settings. The data indicated little change in the median feature intensities of either cyanine dye when the slides were maintained at 42% humidity for 2 hours [Cy5 mean % change ± SD = -0.2 ± 1.3%; Cy3 mean % change ± SD = 1.0 ± 2.1%] or four hours [Cy5 mean % change ± SD = -6.6 ± 3.3%; Cy3 mean % change ± SD = 2.2 ± 3.3%]. However, modest but significant increases in cyanine dye intensities were seen after exposure to high humidity. After a 2-hour exposure to ~98% humidity, Cy5 and Cy3 mean feature intensities were increased by 18.6 ± 2.9% (p < 0.00009) and 7.0 ± 2.5% (p < 0.015), respectively, over initial mean intensity levels. After a 4-hour exposure, Cy5 and Cy3 mean intensities were increased by 6.8 ± 2.2% (p < 0.002) and 5.6 ± 1.5% (p < 0.12), respectively, over initial mean intensity levels.

Data from microarray experiments are affected by many variables including microarray printing, RNA extraction and purification, cDNA production, dye incorporation, and hybridization conditions [7]. A breakdown in any of the critical elements of these procedures can lead to unreliable microarray data. The data presented here, and elsewhere [4], demonstrate that even at the very last step in the process of generating microarray data, i.e., the time interval between the final hybridization wash and the completion of microarray scanning, substantial errors in the Cy5/Cy3 ratios can be introduced.

A typical microarray experiment in our laboratory involves handling 5 or 6 microarrays in a batch. Considering that it takes 6 minutes to scan a 20,000 feature microarray slide, the microarrays may be exposed to ozone from 1 to 36 minutes. As illustrated in Figure 3, significant degradation of the Cy5 signal intensity occurs even at ~25 ppb ozone during this period of time, while the Cy3 signal intensity remains stable. Thus, the signal intensity ratio used for gene expression calculations changes with the amount of time the microarray is exposed to ozone. This, of course, would introduce technical variability within the microarray experiment and mask true experimental affects. In addition, there are several high throughput scanning devices that are capable of batch microarray scanning in which the microarray slides are exposed to laboratory ozone levels for prolonged and varying periods of time. These instruments can take as long as 4 hours to scan a batch of microarrays. Clearly, this would not be suitable in a laboratory without some means of dramatically reducing ozone levels.

LOWESS normalization is a method used to normalize a two-color array gene expression dataset to compensate for non-linear dye-bias [8] and we typically use this normalization method before further data analysis. To examine the consistency of the microarray data during the 8 scans shown in Figure 3A, correlations were calculated between the data obtained from each scan and the data from the initial scan, before and after LOWESS normalization. As shown in Table 1, this normalization method resulted in improvements in the correlation of the gene expression ratios in the non-carbon-filtered data set. However, this improvement does not result in the highly consistent data derived under the carbon-filtered conditions. The main reason for this is that the Cy5 feature intensities do not decay at a uniform rate, resulting in low correlations among scans done at different times (see Table 1). In other words, the data obtained from the initial scans will not be consistent with data collected at later times, and will lead to erroneous results. Because of the non-uniform reduction in Cy5 feature intensities, no standard systematic normalization can correct for data collected under high ozone conditions.

Dye swap experiments, in which a second array is hybridized with cDNAs labeled with the opposite orientation of cyanine dyes, are often used to correct for scanning- and/or labeling-related differences between Cy5 and Cy3 dyes. These sorts of experiments require both cyanine dyes to be stable. Since only the Cy3 dye is stable while the Cy5 dye is highly susceptible to ozone-induced degradation, dye swap would result in a second array with ratio measurements that also change with the amount of time spent in high ozone conditions. Thus, dye swap would provide no improvement in either the data or the conclusions drawn from the data.

The data from the humidity experiments indicate a slight increase in fluorescence signals as a result of exposure to high humidity. The extreme humidity levels were chosen to increase the chance of detecting any influence of humidity on the decay of the cyanine dyes. In a practical sense, microarray laboratories would rarely, if ever, experience humidity in the 98–99% range, so this would not adversely impact the Cy5/Cy3 ratios in normal experiments. Although extremely high humidity may modestly affect cyanine dye fluorescence intensities, the dominant environmental factor that must be controlled is ozone concentration. Uncontrolled ozone has the potential to be the largest variable in microarray experiments.

Simple installation of carbon filters in the laboratory air handling system, coupled with making the laboratory air pressure positive with respect to the adjacent area, resulted in low and consistent ozone levels. Without the laboratory modifications, the intensity of the Cy5 signal dropped dramatically over a short time period (44% in 14 minutes when outdoor ozone levels were approximately 25 ppb) while the Cy3 signal intensity remained relatively constant. Such specific, rapid, and uncontrolled degradation of the Cy5 dye results in inaccurate and highly variable gene expression measurements. LOWESS normalization could not negate the effects of the degradation of the Cy5 intensity caused by ozone in the non-carbon-filtered environment. The laboratory modification dramatically increased the stability of the Cy5 dye to match the stability of the Cy3 dye. This allowed the accurate determination of gene expression by microarray using Cy5 and Cy3 fluorescent dyes.