Increased mdr1 gene transcript levels in high-grade carcinoma of the bladder determined by quantitative PCR-based assay.

Overexpression of the multidrug resistance (mdr1) gene has been implicated in resistance to a number of the chemotherapeutic agents currently used in the treatment of bladder cancer (doxorubicin, vincristine and epirubicin). We report the development and validation of a quantitative assay for the determination of mdr1 gene transcript levels based on reverse transcription and the polymerase chain reaction (PCR), sensitive to less than 2-fold variations in transcript levels. Using these techniques, mdr1 mRNA levels were investigated in 32 primary untreated transitional cell carcinomas of the bladder. mdr1 mRNA was detected in all samples, with levels varying between individual tumours over a 63-fold range. These variations were not associated with the proliferative status of the tumour. mdr1 mRNA levels were significantly higher in poorly differentiated high-grade (G3) tumours than in well- and moderately differentiated low-grade (G1 and G2) tumours (P = 0.0057). The results suggest that this relationship may extend to mdr1 mRNA levels being an indicator of poor prognosis, as anticipated on the basis of the observed relationship to tumour stage and grade. No evidence was found to implicate mdr1 mRNA levels as a predictor of tumour recurrence or progression. Given that mdr1 mRNA levels are increased in a proportion of high-grade bladder tumours that are routinely subjected to chemotherapy, we discuss the possibility that mdr1 mRNA levels may be clinically significant as determinants of chemotherapeutic response and outcome in bladder cancer.

Summary Overexpression of the multidrug resistance (mdrl) gene has been implicated in resistance to a number of the chemotherapeutic agents currently used in the treatment of bladder cancer (doxorubicin, vincristine and epirubicin). We report the development and validation of a quantitative assay for the determination of mdrl gene transcript levels based on reverse transcription and the polymerase chain reaction (PCR), sensitive to less than 2-fold variations in transcript levels. Using these techniques, mdrl mRNA levels were investigated in 32 primary untreated transitional cell carcinomas of the bladder. mdrl mRNA was detected in all samples, with levels varying between individual tumours over a 63-fold range. These variations were not associated with the proliferative status of the tumour. mdrl mRNA levels were significantly higher in poorly differentiated high-grade (G3) tumours than in well-and moderately differentiated low-grade (GI and G2) tumours (P = 0.0057). The results suggest that this relationship may extend to mdrl mRNA levels being an indicator of poor prognosis, as anticipated on the basis of the observed relationship to tumour stage and grade. No evidence was found to implicate mdrl mRNA levels as a predictor of tumour recurrence or progression. Given that mdrl mRNA levels are increased in a proportion of high-grade bladder tumours that are routinely subjected to chemotherapy, we discuss the possibility that mdrl mRNA levels may be clinically significant as determinants of chemotherapeutic response and outcome in bladder cancer.
Cancer of the urinary bladder is the fifth most common malignancy in males of the western world, approximating to 16 new cases per 100,000 males per year in western populations (Davies, 1982). Transitional cell carcinomas (TCCs) of the urothelium constitute more than 90% of urothelial malignancies (Raghavan, 1988). Well-differentiated superficial tumours make up 60-65% of the bladder TCCs and may be treated by local resection with 5 year survival of 80% (Raghavan, 1988;Kiemeney et al., 1993). However, the prognosis for patients presenting with muscle-invasive, dedifferentiated tumours is poor, with typical survival rates of 40-50% following combination radiotherapy and radical cystectomy (Skinner & Lieskovsky, 1984;Hendry, 1988).
Systemic chemotherapy is increasingly being used in the treatment of invasive bladder cancer, with combination regimens proving the most successful. Using an MVAC (methotrexate, vinblastine, doxorubicin, cisplatin) regimen, Sternberg et al. (1988) achieved a 67% response rate (37% complete) with median remission of over a year. Similarly, Harker et al. (1985) produced a 56% total (28% complete) response rate with median survival of 8 months using a CMV regimen. Both studies showed responses at all sites of the disease. However, it is clear that invasive tumours of the bladder are not wholly responsive to chemotherapy, with the majority of patients still dying of their disease. The reasons for this remain unclearseveral cellular mechanisms have been described that may confer upon tumours resistance to many of the currently used chemotherapeutic drugs, of which the multidrug resistance (mdrl) gene has been the most widely studied to date.
Classical multidrug resistance is manifested by crossresistance to a number of functionally unrelated lipophilic drugs of little structural similarity, including the vinca alkaloids (vincristine, vinblastine), the anthracyclines (doxorubicin, epirubicin), antibiotics (actinomycin D, mitomycin C) and taxol, and is thus implicated in resistance to a number of the agents currently used in the treatment of invasive bladder cancer. Acquisition of a multidrug resistance phenotype has been causally associated with expression of the mdrl gene, which encodes P-glycoprotein (PGP), a 170 kDa plasma membrane protein that functions as an energy-dependent drug efflux pump resulting in decreased drug accumulation (Endicott & Ling, 1989;Van der Bliek & Borst, 1989). Cell line studies have shown good correlation between mdrl mRNA levels and the degree of multidrug resistance (Shen et al., 1986;Fojo et al., 1987;Chan et al., 1988;Noonan et al., 1990). mdrl expression has been widely observed in many different human tumour and tissue types, with increased mdrl levels frequently observed at relapse following chemotherapy (Fojo et al., 1987;Goldstein et al., 1989;Noonan et al., 1990). Intrinsic variation in mdrl expression levels may be an important determinant of tumour response, yet few studies have investigated variation of mdrl mRNA levels within a specific tumour type prior to treatment, or related this to factors such as tumour grade and survival. Undetectable or low levels of mdrl mRNA have been reported in both drugsensitive and drug-resistant tumours prior to chemotherapy, lying close to or below the detection limits of conventional methods (Goldstein et al., 1989;Noonan et al., 1990). In bladder neoplasia, detection rates in untreated tumours have been mixed; at the protein level, Naito et al. (1992) reported detectable PGP in 32% of tumours by immunohistochemistry, while Benson et al. (1991) reported PGP expression in 71% of tumours using flow cytometry methods. At the message level, Goldstein et al. (1989) using Northern blot analysis reported only one weakly mdrl mRNA-positive bladder tumour in six analysed. No previous studies have systematically investigated mdrl mRNA expression in bladder cancer.
Assay insensitivity coupled with the often limited sample material available from tumours has hampered the detection of what may be clinically significant levels of mdrl mRNA. To investigate whether such mdrl mRNA levels lie below these detection limits, we report the development and validation of an assay based on reverse transcription and the polymerase chain reaction (PCR) for use in the determination of mdrl mRNA levels in clinical samples. Using these methods we have determined the incidence of and variation in mdrl transcript levels in a series of untreated TCCs of the bladder, and investigated the relationship of these levels to tumour stage, grade and rate of proliferation, prognosis, survival, progression and recurrence.
Materials and methods mdrl transcript levels were measured relative to those of 18S ribosomal RNA as an internal reference. The MHC class II related protein, P2-microglobulin (02-M), has previously been used as an internal reference for the determination of mdrl transcript levels by polymerase chain reaction (PCR)-based methods (Kuwazuru et al., 1990;Noonan et al., 1990). We have also investigated the suitability of P2-M mRNA levels as an internal reference for the measurement of mdrl mRNA levels in bladder cancer.
Tumours and tissues TCC tumour samples obtained at resection or cystectomy were immediately snap frozen in liquid nitrogen and transferred to a -80°C freezer for storage. A portion of the sample was sent for histological assessment and the tumours were staged and graded according to UICC (1978) criteria. All samples described in this study had not received prior chemotherapy. Eleven out of 18 of the patients with invasive tumours (T2-T4) went on to undergo chemotherapy. Adrenal tissue was obtained from patients undergoing radical nephrectomy.
Care was taken to limit the proportion of the tumour sample contaminated by normal tissue. Because of their papillary growth pattern superficial tumours were readily removed without contamination from underlying normal tissue. Although the invasive tumours were more difficult to separate from the underlying lamina propria and muscle, contamination by normal tissue was minimised by only taking samples from the protruding mass of the tumour. Although it was difficult to ensure complete elimination of normal tissue from these samples, histological examination indicated an upper limit of 10-15% normal tissue.

Cell lines
Multidrug-resistant cell lines known to overexpress the mdrl gene and their parental controls were used for assay development and validation. All lines were tested and found to be mycoplasma negative.
The KK47 cell line was established from an untreated grade 1, stage Ta TCC of the bladder, from which the multidrug-resistant cell line KK47/ADM was derived by stepwise selection in increasing concentrations of doxorubicin (Kimiya et al., 1992). Both lines were grown as monolayer cultures in complete minimum essential medium (MEM) at 37°C in a 5% carbon dioxide atmosphere, and subcultured weekly by harvesting with a 0.25% trypsin -0.02% EDTA solution and reseedingy at 3 x 106cells per 175 cm2 flask.

RNA extraction
Total RNA was extracted from the remaining tumour tissue and cell lines using the RNAzol method (Chomzynski & Sacchi, 1987). Briefly, in the case of solid tumours, tissue was pulverised in liquid nitrogen and rapidly lysed in a phenolguanidinium thiocyanate-i-mercaptoethanol mixture. Cell lines were harvested at 70% confluence by scraping in icecold phosphate-buffered saline (PBS) (KK47 and KK47/ ADM) or by centrifugation followed by washing in PBS (H69 and H69/LX4). Both methods preceded direct lysis as described. RNA was isolated by chloroform extraction and precipitated by isopropanol addition. Following washing in ethanol and resuspension in sterile water, RNA yield and purity were checked by spectrophotometric determination at 260 nm and 280 nm and the integrity assessed by electrophoretic size separation on 1.4% agarose gels. cDNA synthesis First-strand cDNA was synthesised from total RNA by reverse transcription using the random primer extension method. A 10 fig aliquot of total RNA was heated to 100°C for 5 min, then added to a 200 fLI reaction (final volume) consisting of the following: pd(N)6 random hexamers (0.88 units; Pharmacia), dithiothreitol (0.01 M; Gibco, BRL), dATP, dCTP, dTTP and dGTP (1 mM each; Pharmacia), Moloney murine leukaemia virus reverse transcriptase (1200 units, Gibco BRL) and human placental RNAse inhibitor (35 units; Gibco BRL) made up in 1 x reverse transcriptase buffer (Gibco BRL). The reaction was allowed to proceed at 37°C for 1.5 h before termination by heating to IOOC for a further 5 min. Following synthesis, cDNA was purified by passage through a cDNA spun column containing Sephacryl S-300 (Pharmacia) to remove any remaining proteins, free nucleotides and random hexamers remaining from previous procedures. The second cDNA strand was synthesised by the extension of sequence-specific oligonucleotide primers in the first cycle of a subsequent polymerase chain reaction.
Optimisation of the polymerase chain reaction mdrl and P2-M primers chosen were those used by Noonan et al. (1990). The mdrl primers distinguish between the mdrl and mdr2 gene sequences, and both the mdrl and P2-M primers span an intron to control against contamination by amplification of genomic DNA sequences. Primers to 18S rRNA were chosen using a computer program designed by Lowe et al. (1990) -ATGCTCTTAGCTGAGTGTCC and AACTACGACGGTATCTGATC (residues 763-782 and 1055-1074 respectively; Gonzalez & Schmickel, 1986). Primers were made on an oligonucleotide synthesiser (Model 392, Applied Biosystems). The primers yielded products of 167 base pairs (mdrl), 120 base pairs (02-M) and 311 base pairs (18S rRNA). For each set of primers, a series of fixed-condition PCR reactions were performed employing a fixed concentration of input cDNA. The absolute magnesium chloride concentration was varied in 0.5 mM steps in the 0-5 mM range. The optimal magnesium chloride concentrations were determined to be: mdrl, 2.5 mM; P2-M, 3 mM; and 18S rRNA, 1 mM.
Determination of mdrl mRNA levels Typical PCR reaction yields for various cDNA inputs are shown in Figure 1. An initial range-finding experiment was performed to determine the range of serial cDNA dilutions over which PCR amplification is linear for each target species. Serial cDNA dilutions for each species are simultaneously and independently amplified over 25 PCR cycles (94°C for 1 min, 56°C for 1 min, 72°C for 1 min) using otherwise fixed reaction conditions. A 25 gil reaction consisted of the following (final concentrations are stated): 10 gIl of appropriate cDNA dilution, Taq polymerase ( Following amplification, 10 gil aliquots were analysed by electrophoretic separation on 12% polyacrylamide gels at 100 V for 1.5 h. Gels were dried under heat and vacuum and the radioactively labelled PCR products detected and analysed using a PhosphorImager (Molecular Dynamics). A typical autoradiograph image of the separated prod,ucts is shown in Figure 1. For each species, the amount of PCR product (measured as incorporated radioactivity) was plotted against input cDNA dilution (see Figure 2). Regression analysis was performed on the points constituting the linear range of amplification for each species. The ratio of input total cDNA concentrations (X,/X2 in Figure 3) for a given yield in the linear range of amplification is a measure of the ratio of cDNA concentrations present for each species. This is equal to the slope ratio (M1/M2) for the regression lines on the linear part of the yield curve (see Figure 3). Thus, mdrl mRNA levels were expressed as an mdrl/18S ratio. Subsequent replicates were performed using only four serial dilutions (lying within the predefined linear amplification ranges) per species. Analysis was performed in triplicate on all samples. A linear regression correlation coefficient (r2) of greater than 0.95 was used as the criterion for accepting data that fell within the linear range of amplification.

Determination of tumour proliferative status -MIBI analysis
The monoclonal antibody MIBI reacts with the human nuclear cell proliferation-associated antigen recognised by the monoclonal antibody, Ki67, that is expressed in all active parts of the cell cycle (Cattoretti et al., 1992). The extent of MIB1 staining was thus used to assess the proliferative status of a tissue.
Paraffin sections (4 pim thick) were taken and dewaxed in xylene prior to rehydration through alcohol and water. Sections were treated with 0.1 M citrate buffer (pH 6.0) and microwaved for 20 min. Endogenous peroxidase activity was blocked by prior treatment with 3% hydrogen peroxide. The sections were then rinsed in Tris-buffered saline (TBS pH 7.6) cDNA dilution 10°10-1 10-2 10-3 10-4 for 2 min, followed by normal goat serum (Life Science). Sections were incubated with MIB1 antibody (The Binding Site) (diluted 1:50) for 60 min, washed again and peroxidase activity developed using biotinylated goat anti-mouse/rabbit secondary antigen (Dako) for 30 min. After further washing in TBS, sections were incubated in streptavidin AB complex/ horseradish peroxidase (Dako) for 30 min. Sections were then incubated in diaminobenzidine solution for 10 min, washed and counterstained in Carruzi's hamatoxylin, dehydrated and mounted. A negative control was performed for each tissue section by omission of primary antiserum. Tonsillar tissue was used as a positive control. The proportion of tumour cell nuclei staining with MIBI was assessed in random fields from well-preserved areas of tumour. A minimum of 2,000 cells were counted in each case. Nuclei in morphologically malignant cells were considered positive when dark-brown nuclear staining was observed.

Assay validation
For any given tumour RNA sample, the standard error of the mean based on the repeat experiments was typically ± 23% of the mean value. Experiments performed on the H69/LX4 cell line indicate that three separate analyses of a sample utilising independent cell cultures, RNA extractions and cDNA syntheses produces results (mdrl/18S ratios) lying within a range of 1.807 ± 0.911 x l0-' (mean ± s.d.).
To examine any influence that PCR amplification efficiency may have on our results, efficiency studies were performed on four different tissue types with varying mdrl/18S levels. Figure 4a and b shows the results for a typical bladder tumour RNA sample as an example. Reaction efficiency was calculated using the formula Nn = N_ I (1 + a), where Nn is the amount of product after the nth cycle, n is the cycle number and a is the efficiency. 18S rRNA was amplified with the greatest efficiency, followed by mdrl then P2-M ( Figure 4b). This order of efficiency remained constant for tissues analysed. For each product species, the absolute amplification efficiency showed some variation between samples, however relative amplification efficiencies remained consistent between samples, irrespective of tissue type or level of mdrl expression (Table I). The maximum efficiency range corresponds to the linear range of amplification, and its position is dependent on the input cDNA concentration. The assay that we have described utilises cDNA concentrations in the range that gives linear amplification over 25 PCR cycles. Graph showing a log-log plot of product produced versus initial amount of input cDNA for the three series of points shown in Figure 1. The linear ranges of amplification used in subsequent quantifications are highlighted for each species by black arrows. Note that these ranges lie between the ranges of reaction threshold and saturation. A, mdrl; 0, P2-M; 0, 18S. Figure 3 Illustration showing the calculation of relative mRNA levels for two gene product species. Lines labelled 'Gene products 1 and 2' represent the linear ranges of amplification for two given species, extrapolated through the origin. In the linear amplification range, the ratio of input cDNA for a given product (Xl/X2) is a measure of the relative amounts of mRNA for the two genes under consideration. For common extrapolation through zero, this is equal to the slope ratio (M/AM2) for the two lines.  dilution) were amplified over 33 PCR cycles and pi mined at alternate cycles between cycles 15 and 32 shown plotted against cycle number (linear scale typical PCR reaction kinetics of (1) a detection thrl early cycles, (2) a linear range of amplification and ( saturation plateau in the later cycles. b, Changes efficiency over subsequent PCR cycles for the analy a. For each product species, efficiency (I = 100%); the text is plotted against cycle number. Graphs pr efficiency rising to a peak that corresponds to the lir amplification, then tailing off as the reaction rea tion.
gives linear s were found lative to 18S T2, n=2; T3, n=9; T4, n=6; by grade, GI, n=2; G2, n 12; G3, n = 18), with a mean mdrl/18S ratio of 7.34 x 10-6. A 63-fold variation in mdrl/18S levels was observed between individual tumours (highest 3.4 x 10-5; lowest = 5.4 x l0-7). d relative to Relationship of mdrl mRNA levels to stage and grade lative to 18S mdrl/18S ratios were significantly higher in poorly ,vary widely differentiated high-grade (G3) than in welland moderately ierefore con-differentiated low-grade (GI and 2) tumours ( Figure 5). The r subsequent pooled mdrl/18S mean ± s.e. for grades GI and G2 comatios relative bined was 3.41 ± 0.53 x 106 compared with 10.40 ± 2.2 x 106 for the G3 tumours (t-test, P = 0.0057). No low-grade tumours showed high mdrl levels, and while not all highgrade tumours showed elevated mdrl/18S ratios there was a markedly increased incidence of high-expressing tumours in 9our samples this group (Figure 5). With the exception of single his-9; T1. n = 6: tologically atypical tumour which presented as TI G3 (carcinoma in situ), all superficial tumours (Ta and TI) were also low grade, while all invasive tumours (T2, T3 and T4) were high grade. Thus, the mdrl expression pattern observed for a tumour grade essentially extends to tumour stage; based on pooled means, mdrl/18S ratios were marginally significantly higher (P = 0.085) in muscle-invasive tumours than in superficial, non-invasive tumours. Removal of the atypical carcinoma in situ sample from the analysis increases the level of significance (P = 0.012). The two normal tissue samples examined showed higher levels of mdrl mRNA than the superficial tumour group and the levels were similar to those seen in the high-grade invasive tumours.

33
Relationship between mdrl expression and progression, b recurrence and survival For these analyses, tumours were split into high-mdrl mRNA level (mdrl/18S ratio > 1 x 10', n = 8) and lowmdrl mRNA level (mdrl/18S ratio <9.99 x 10-6, n = 24) groups. This distinction was made on the basis of the distribution of expression observed, with all high-mdrl mRNA tumours lying outside the main cluster of points. No evidence was found to implicate mdrl mRNA levels as a predictor of tumour recurrence or progression. No correlation (r = 0.08, n = 13) existed between mdrl/18S ratios and 9-31 31-33 the rate of recurrence. Forty-three per cent (three of seven) high-mdrl-expressing tumours underwent progression, compared with 33% (4 of 12) of low-mdrl-expressing tumours. analysis (Peto et al., 1977) showed no significant difference oduced show (P = 0.36) between survival for the groups of tumours with near phase of high and low mdrl mRNA levels ( Figure 6). However, con-Lches satura-tingency table analysis using Fisher's exact test showed a lower proportion of survivors for patients with high mdrl   24  48  72  96  120  144  168  192  12  36  60  84  108  132  156  180  204 Time (months) Figure 6 Log-rank survival curves for groups of patients with tumours expressing high and low mdrl mRNA levels. Each death is represented by a vertical drop on the graph. Surviving patients are shown as a vertical tick on the graph representing the end of their current follow-up period. The log-rank test gave P = 0.36 for no difference in survival rates between the low-(-) and high-mdrl ( ) mRNA groups (see also text). mRNA tumours, which was marginally significant P= 0.07).
Relationship between mdrl mRNA levels and tumour proliferative status The percentage of proliferating cells within the tumours studied (those displaying the Ki67 antigen) did not correlate with their mdrl/18S ratios (r = 0.06, n = 24).
Other mdrl mRNA determinations As part of general method validation and to put our data into a broader context, mdrl/18S ratios were also determined for some other tissues and cell lines with known low and high levels of mdrl expression. Adrenal tissue had an mdrl mRNA level (mdrl/18S = 2.00 x 10-4) approximately 27 times higher than the mean level of expression in bladder tumours. Hyperexpression (1.58 x 10-3) was detected in the mdr SCLC cell line H69/LX4, while no mdrl expression could be quantified in its drug-sensitive parent line H69 at 25 PCR cycles. The mdr bladder tumour cell line KK47/ADM had 157-fold overexpressed mdrl mRNA levels (mdrl/1 8S ratio 3.88 x 10-4), compared with levels of 2.46 x 10-6 in its parental line KK47.

Discussion
The PCR-based transcription assay Measurement of expression relative to an endogenous internal reference avoids the need for addition and quantification/ titration of an external reference standard, or the manipulation of products post amplification to distinguish them (e.g. the presence of restriction enzyme cleavage site in a synthetic reference standard), both strategies being among those previously described (Becker-Andre & Hahlbrock, 1989;Wang et al., 1989). 18S rRNA is widely used to control for equal loading and transfer in Northern blot methods for RNA analysis. Good evidence exists for the use of rRNA (of which 18S rRNA is one component) as a reference. While variations in total rRNA levels per cell do occur, these are usually accompanied by an equal fluctuation in total mRNA levels (Johnson et al., 1976). Hirsch (1967) observed consistent rRNA levels (80% of total) in normal rat liver, fasted rat liver (with a 50% reduction in total RNA content) and rapidly dividing rat hepatoma. P2-Microglobulin has been commonly used as an internal reference for the determination of mdrl transcript levels in PCR-based studies (Kuwazuru et al., 1990;Noonan et al., 1990). However, our studies have demonstrated a 140-fold variation in P2-M mRNA levels between individual bladder tumours, rendering it completely unacceptable for such a purpose. Whether these observations extend to other tumour systems remains to be investigated. Amplification of target and reference from a common cDNA source controls for variations in efficiency of both RNA extraction and cDNA synthesis. Our efficiency studies indicate that, while absolute amplification efficiencies vary between different species, their relative efficiencies remain constant regardless of tissue type or mdr1/18S ratio and thus have no influence on the results observed. Based on the typical standard error of 23% of the mean of three replicates, this assay is capable of discriminating 1.6fold differences in mRNA levels between samples at the 95% confidence level. The reproducibility studies based on three independent batch cultures, RNA extractions and cDNA syntheses similarly indicate that mdrl mRNA levels from three independent analyses of a given sample fall within a 2.5-fold range, however this may overestimate the situation since the batch cultures were not harvested at uniform cell densities and may also be prone to differing culture conditions, both of which may affect cellular mRNA levels. These limits are small compared with the 63-fold variation that we have observed in mdrl mRNA levels between individual tumour samples. For the assay as described, 10lg of total RNA yields sufficient cDNA to perform at least five replicates of the assay, although there are various ways in which the sensitivity could be further increased if necessary. In comparison, Northern blot analysis may require 10-20 pig of RNA for a single analysis. In the clinical situation, in which sample material is often limited, this may not allow for a repeat analysis. This method offers notable improvements in sensitivity over conventional methods, detecting mdrl transcript in all samples analysed, whereas previous studies using Northern blot analysis have only detected mdrl mRNA positivity in one of six (<17%) bladder tumours (Goldstein et al., 1989). In this context, categorisation of samples as mdrl mRNA positive/negative is misleading since the distinction is made by the arbitrary detection threshold of a partic- As with Northern blot methods, the heterogeneous nature of tumours is one potential problem in this type of analysis. The assay provides an overall mean mRNA level for the whole sample analysed. It is therefore essential that the tumour sample used is representative of the entire sample.
To test the ability of the assay to detect differences in mdrl mRNA levels, we examined adrenal gland as an example of a high-expressing tissue, and also multidrug-resistant lung and bladder tumour cell lines known to overexpress P-glycoprotein. Adrenal gland tissue has been widely reported to have high levels of mdrl mRNA (Fojo et al., 1987;Noonan et al., 1990). Noonan et al. (1990) reported a single bladder tumour to have mdrl mRNA levels 68 times lower than those in adrenal tissue. This is in agreement with our study, which shows the mean bladder tumour mdrl/18S ratio to be 27 times lower than that of an adrenal sample. Likewise, the mdrl/18S ratio for the KK47 cell line, derived from a superficial bladder tumour, lies in the same range as all the superficial bladder tumour samples analysed.
We have thus developed a highly sensitive and reproducibly accurate PCR-based assay for the quantification of low-level gene expression. Should it be necessary, sensitivity of the assay could be even further improved with increased numbers of PCR amplification cycles. Further replicates would also increase confidence in the mean for a given sample. This technique may be applied to the quantification of any low-level gene expression when the cDNA sequence is known, with the accurate quantification of results allowing a much more detailed analysis of the data. The assay could potentially be adapted to work at the DNA level for the determination of gene copy numbers and levels of gene amplification.
Variation in mdrl mRNA levels in TCC of the bladder mdrl/18S ratios were found to vary over a 63-fold range in bladder tumours. The mechanisms underlying this remain to be elucidated. The lack of any correlation between mdrl/18S ratios and tumour Ki67 levels indicates that the mechanism underlying variation in tumour mdrl mRNA levels is not simply a result of tumour proliferation or the rate of cell turnover. Similarly, since these tumours have not been subjected to selection by chemotherapy, it seems unlikely that gene amplification plays any role in the variation of mdrl mRNA levels observed. Transcriptional control mechanisms have been suggested to play a significant role in the regulation of mdrl mRNA levels (Goldstein et al., 1989;Zastawny et al., 1993). Recently it has been demonstrated that wildtype p53 protein represses human mdrl promoter activity while mutant forms (cysteine 135 to serine) of the p53 protein enhance mdrl transcription (Zastawny et al., 1993), thus defining a possible mechanism for increased mdrl mRNA levels in high-grade untreated bladder tumours, in which frequent p53 alterations have been reported (Lunec & Mellon, 1994). The stage and grade associations that we have observed suggest that overexpression could also be a result of the loss of genetic regulation and control associated with high-grade (dedifferentiated) tumours. The level of mdrl mRNA detected in normal bladder raised questions about the contribution of normal tissue contamination to the increased levels of mdrl mRNA found in the high-grade tumours compared with the superficial group. However, because of the careful resection technique, the maximum observed contamination of the invasive tumours by normal tissue of only 10-15% could not have accounted for the elevated levels of mdrl mRNA seen in the high-grade group.

Relationships between mdrl mRNA levels and prognosis/survival
Our results demonstrate that mdrl mRNA levels are clearly elevated in high-grade bladder tumours. A recent immunohistochemistry study also showed increased levels of the mdrl gene product, P-glycoprotein, in high-stage neuroblastomas (Chan et al., 1991). Thus, evidence exists for mdrl mRNA and P-glycoprotein levels being a marker of tumour aggression (invasiveness and dedifferentiation). However, the majority of studies are conducted at the protein level, with this study to our knowledge being the first to report such an association at the mRNA level. Other studies investigating P-glycoprotein levels in bladder cancer (Benson et al., 1991;Naito et al., 1992) have proved inconclusive, showing no clear relationships with stage and grade.
The population distribution of mdrl/18S ratios observed in bladder tumours has allowed us to distinguish groups of highand low-expressing tumours for use in survival analysis.
Log-rank tests showed no significant difference in survival for either group of tumours (Figure 6), however this analysis is complicated by the problem of non-uniform patient treatment and follow-up times. The analysis by Fisher's exact test based on proportions of survivors alone indicates a significant relationship between high mdrl mRNA levels and poor survival. Other studies have similarly reported high mdrl expression to be an indicator of adverse prognosis and poor survival in untreated tumours; at the protein level in softtissue sarcomas (Chan et al., 1990), breast carcinoma (Verrelle et al., 1991), neuroblastoma (Chan et al., 1991) and nonlymphoblastic leukaemia (Campos et al., 1992), and at the message level in acute myeloid leukaemia (Pirker et al., 1991). Our results suggest that it may be worth investigating further the prognostic potential of mdrl mRNA levels in bladder cancer. It may be interesting to compare mdrl mRNA with other known prognostic factors using samples from patient groups undergoing either uniform treatment strategies or receiving no treatment.
We have demonstrated that mdrl mRNA levels are raised in many of the high-grade, invasive bladder tumours that are commonly treated using chemotherapy, and suggest that the 34-fold variation in mdrl mRNA levels observed between individual high-grade tumours could be a significant determinant of their chemotherapeutic response. This is supported by in vitro studies which show that even small variations in mdrl mRNA levels, within the range we have observed, result in significant differences in drug response (Noonan et al., 1990). To evaluate the role of pretreatment mdrl mRNA levels in the determination of chemotherapeutic outcome, full prospective studies have been initiated involving uniform treatment regimens with preand post-chemotherapy tumour biopsies. This will also test whether mdrl mRNA levels are increased following treatment, as has been reported in some tumour systems (Fojo et al., 1987;Noonan et al., 1990).
Many thanks to Dr S. Freemantle for her part in the development of PCR methods. The cell lines used as controls in this study were kindly supplied by Dr Peter Twentyman (H69 and H69/LX4) and Dr Seiji Naito (KK47 and KK47/ADM). This work was generously funded by the North of England Cancer Research Campaign.