Anal. Chem. 2009, 81, 7294–7301
Accurate Measurement of DNA Methylation That Is Traceable to the International System of Units Daniel G. Burke,* Kate Griffiths, Zena Kassir, and Kerry Emslie National Measurement Institute, Australia, 1 Suakin Street, Pymble, NSW 2073, Australia The increased presence of 5-methycytosine at gene promoter regions may be diagnostic of cancer. However, there are many stages in the measurement of gene promoter 5-methylcytosine content where inaccuracies may occur, and this may prevent the use of these measurements for diagnostic or prognostic purposes. A high accuracy LCMS system was developed for measuring the degree of methylation in two 100 base pair amplicons generated by the polymerase chain reaction (PCR) and in which 5-methylcytidine had been synthetically incorporated. Nucleotide monophosphate reference materials were used to calibrate the peak area ratio of cytidine and 5-methylcytidine to their mole ratio in enzymatic hydrolysates of the amplicons, thus enabling metrological traceability of the methylation ratio to the mole. The methylation values obtained agreed closely with the reference values assigned to the materials. A measurement uncertainty budget was completed and showed that the moisture content of the nucleotide monophosphate reference materials was the largest source of uncertainty in the methylation ratio measurement. Measurement of an oligonucleotide supplied with the materials provided evidence that such materials may be used for calibration of DNA methylation ratios without the need for measurement of moisture content. This raises the possibility that submicrogram amounts of appropriately characterized oligonucleotide reference materials could be used to calibrate methylation ratios obtained by contemporary methodologies (such as PCR after bisulfite conversion of genomic DNA) yielding values that are traceable to the International System of Units (SI). Such calibrated gene methylation measurements would then be internationally comparable as requiredforeffectivediagnosticandprognosticmeasurements. DNA methylation at CpG islands of specific genes may be diagnostic of cancer,1,2 and useful for prognostication.3 Various methods are used for measuring the amount or degree of methylation of specific DNA sequences,4-8 but the most com* Corresponding author. Fax: +61 2 9449 1653. E-mail: daniel.burke@ measurement.gov.au. (1) Frigola, J.; Song, J.; Stirzaker, C.; Hinshelwood, R. A.; Peinado, M. A.; Clark, S. J. Nat. Genet. 2006, 38, 540–549. (2) Laird, P. W. Nat. Rev. Cancer 2003, 3, 253–266. (3) Perry, A. S.; Foley, R.; Woodson, K.; Lawler, M. Endocr. Relat. Cancer 2006, 13, 357–377. (4) Hou, P.; Ji, M.; Chen, Z.; Lu, Z. Curr. Anal. Chem. 2006, 2, 309–322. (5) Tryndyak, V.; Kovalchuk, O.; Pogribny, I. P. Anal. Biochem. 2006, 356, 202–207.
7294
Table 1. Purity and Moisture Content of Reference Materials from Supplier Certificates of Analysis compound 5-methyl-2′-deoxycytidine, disodium salt 2′-deoxycytidine-5′monophosphate 2′-deoxyguanosine-5′monophosphate 2′-deoxyadenosine-5′monophosphate thymidine-5′monophosphate
supplier
product puritya moisture number (%) (%)
USB Corporation 19184
>98
Sigma-Aldrich
D7750
>99
3.5
Sigma-Aldrich
D9625
99
5.2
Sigma Aldrich
D6375
98.9
5.9
Sigma Aldrich
T9758
100
> > > >
Table 3. Uncertainty Analysis Results collision dwell (s) cone (V) energy (eV)
112 126 136 152 81
0.20 0.20 0.20 0.20 0.20
19 23 19 23 25
13 13 15 15 13
methylation ratio combined standard uncertainty total degrees of freedom coverage factor expanded uncertainty relative expanded uncertainty
for each of the nucleotide monophosphates are given in Table 2. At the retention time for each nucleotide, the molecular ion was focused into the collision chamber where the collision energy was adjusted for optimum sensitivity of fragment ion; in each case, the fragment ion monitored corresponded to loss of deoxyribosephosphate from the deoxyribonucleotide phosphate except for thymine where the major fragment ion corresponded to cleavage of the deoxyribonucleotide from the phosphate. Analysis was performed by bracketing sample blend injections with calibration blend injections to minimize the effects of instrument drift; five bracketed injections were made for each sample blend. Peak areas were calculated using the proprietary QuanLynx software and were exported to a spreadsheet for calculation of methylation ratios. Calculation of Methylation Ratio. Equation 1 was used to calculate the methylation ratios of the solutions of unknown DNA from peak areas obtained by LC-MS-MS; methylation ratios were calculated for each analysis of test solution using the average peak area ratios from bracketing calibration solutions. The average of the five test solution analyses was used as the measurement. The complete derivation of eq 1 is given in the Supporting Information. The equation describes the use of a reference nucleotide solution with a known methylation ratio to calibrate chromatographic peak area ratio of C to mC. The derivation requires that there is no discrimination between C and mC introduced during transfer from the test solution to the analytical instrument or that the discrimination is the same in test and calibration solutions.
methylation ratio )
100 nC RS P Z × × 1+ RZ PS nmC
(
( ) )
parameter
(1)
CAL
Where nC ) amount (mol) of cytidine in calibration solution, nmC ) amount (mol) of 5-methylcytidine in calibration solution, RZ ) molar response factor in calibration solution, RS ) molar response factor in test solution, PZ ) (PAmC)/(PAC) in the calibration solution, PS ) (PAmC)/(PAC) in the test solution, and (PAmC)/(PAC) ) ratio of peak area of mC to peak area of C from LC-MS-MS analysis. Note that, in this equation, the molar amounts of C and mC appear as a ratio, implying that a mole ratio could be used in place of the two amount of substance values. For the measuring system described here, the mole ratios calculated from gravimetric preparation of calibration solutions given in Table 1 were used in all calculations. Because mole ratios were used, the effects of dilution or incomplete recovery had no effect on the calculation of the analytical result, provided the ratio was not affected. The term RS/RZ in eq 1 arises from any difference in molar response ratio of C to mC that may occur between the test solution
unknown unknown 1 2 25.8 0.99 12.9 2.18 2.1 8.3%
50.3 1.6 10.5 2.23 3.5 7.0%
units % mol/mol % mol/mol none none % mol/mol dimensionless
and the calibration solution. A difference in response can occur with electrospray ionization as the presence of coeluting salts and other organic molecules is known to suppress or enhance ionization. Since this is a critical requirement for accuracy, the equivalence of RS and RZ must be examined. In this work, the test and calibration solutions were functionally equivalent after hydrolysis even though the unknowns were in 0.2 mM Tris-HCl and the calibration solutions were in water since the small amount of Tris-HCl from the unknown study materials represented only 0.2% of Tris-HCl in the final solution analyzed by LC-MS-MS. Differences in response between RS and RZ were further minimized by matching the calibration solution to the expected mole ratios and DNA concentrations in the study material unknown solutions; the target values for total nucleotide concentrations of calibration solutions were those specified as the total DNA concentrations of the unknowns calculated from A260 values. RESULTS AND DISCUSSION The measurement equation, eq 1, converts the observed peak area ratio to a methylation ratio by calibration with a known mole ratio of cytidine to methylcytidine, (nC/nmC)CAL. When the constituents of the calibration solution are identical to the test solution, the factor RS/RZ is unity and the methylation ratio can be accurately calculated using only peak areas and the calibration mole ratio. The calibration mole ratio was given for the oligonucleotide calibrator supplied with the unknown; however, we were not able to validate the identity and purity of the calibrator within the time frame of the CCQM-P94 study, and since an estimate of the measurement uncertainty of the methylation ratio was not given, we were not able to use this oligonucleotide to calibrate our measurements. Instead, calibration solutions were prepared from reference nucleotide monophosphates, and the calibration mole ratio was calculated from the gravimetric amount of substance measurement (mole), i.e., the mass of reference nucleotide monophosphate weighed, for each of the nucleotide monophosphates in the calibration solution. This is not a fundamental requirement for this measurement, and if the mole ratio value of the calibrator could be obtained by other means, e.g., from the verified sequence of an oligonucleotide, then that material may be used to calibrate the measurement and the results would be traceable to the SI, provided the measurement uncertainty of the calibration mole ratio was specified. The effects of coeluting salts or organic molecules on electrospray ionization of cytidine and 5-methylcytidine can be minimized by separating the target molecules from possible interferences through use of a chromatographic column that is selective for Analytical Chemistry, Vol. 81, No. 17, September 1, 2009
7297
Figure 2. Uncertainty components for CCQM-P94 unknown 1 (a) and unknown 2 (b).
these molecules. In this work, we found that even though the capacity factor for cytidine was only about 2, peak areas were consistent throughout an analytical batch indicating that any ionization suppression or enhancement was at least consistent for the analytical conditions used. The effect on molar response ratio of any difference between test and calibration solutions may be evaluated with a blank solution, prepared as for the PCR product solutions. However, in this study, a blank solution was not supplied, so we used the measurement of the calibrator oligonucleotide as a means of evaluating the variability of RS ) RZ since it may be assumed that the differences between the oligonucleotide solution and the calibration solution would be of a nature similar to the differences between the unknown DNA and the calibration solution. Isobaric interference, i.e., a coeluting molecule that produces an ion at the same m/z as used to measure the target molecule, would also cause a bias in this measurement, and a method for evaluating the presence of interfering ions is also needed. This 7298
Analytical Chemistry, Vol. 81, No. 17, September 1, 2009
may be done using ratios of different fragment ions, but in this case, there was only a single major fragment from the collision induced dissociation. Since estimation of possible interference effects was not possible, and since there was no evidence of interference, no adjustment was made to the calculated values or the uncertainty as this factor was considered to be covered by the difference between the measured and theoretical value of the oligonucleotide calibrator. Equation 1 holds when the peak area ratio of cytidine and methylcytidine is determined only by their mole ratio in the solution analyzed and not the amount of substance in the solution, i.e., the peak area ratio must be independent of concentration. Using the conditions detailed above, PZ was 1.858 with coefficient of variation (cv) ) 1.2 over an order of magnitude (cytidine 7-0.7 µg/g) for the mole ratio 1.441 (C/mC), thus proving concentration independence for this ratio. Since the aim of this work was to measure methylation ratios in PCR amplicons supplied in relatively high quantities, no attempt
Figure 3. Comparison to reference value for unknown 1 (a) and unknown 2 (b). The error bars for the NMIA value represent the expanded uncertainty at the 95% confidence interval; for the reference value, an arbitrary uncertainty of 5% was assigned as suggested by the study coordinators.
was made to evaluate this system for measurement of methylation in genomic DNA. Enzymatic Hydrolysis Conditions. Extensive work on optimum hydrolysis conditions has been reported by Yang and Park.20,25 These conditions were used to determine whether the methylation ratio or amount of product varied with hydrolysis time. We found no significant change in amount of nucleotide monophosphate produced or in the ratio of cytidine to methylcytidine over a 2 h period. Variation in Calibration Solutions. The variation due to independent preparation of different calibration solutions from the same reference materials was estimated by preparation of three calibration solutions, and they were used to measure the methylation ratio of a single sample of the matched unknown PCR product. The mean methylation ratio for unknown 1 was 26.6% mol/mol (cv ) 2.7) and for unknown 2 was 51.4% mol/mol (cv ) 2.3). Although ANOVA indicated that the differences in results for the different calibration solutions were significant (95% confidence interval), there was no objective means of readily eliminating this source of measurement bias, so the variability due to this effect was incorporated into the uncertainty estimate. A single calibration solution, the one giving the mean value, was used to measure the methylation value of triplicate subsamples from each of the three vials of unknown DNA. (25) Yang, I.; Park, I. Y.; Jang, S.-M.; Shi, L. H.; Ku, H.-K.; Park, S.-R. Nucleic Acids Res. 2006, 34, e61.
Methylation Ratio Measurements of Study Materials. Three 5 µL subsamples from each of the three vials were hydrolyzed, and the methylation ratio measured as detailed above. The chromatographic peaks for mC and C were well resolved, and the signal-to-noise ratio was greater than 300:1 for these measurements (Figure 1). Each measurement consisted of five LC-MS analyses of the ultrafiltered hydrolysate bracketed by the matching calibration blend; a methylation ratio was calculated for each analysis (using the average of the bracketing calibration blends), and the five analyses were averaged to give the methylation ratio measurement. The methylation ratio value for each unknown was calculated as the mean of the three subsamples from the three vials (nine separate measurements): the methylation ratio of unknown 1 was 25.8% mol/mol (cv ) 1.26), and the methylation ratio of unknown 2 was 50.3% mol/mol (cv ) 0.9). Methylation Ratio Measurements of Oligonucleotide Calibrator. The 36 base oligonucleotide calibrator was analyzed six times using four different calibration solutions; two of the calibration solutions matched unknown 1 and two matched unknown 2. The average methylation ratio from these six measurements was 58.7% mol/mol (cv ) 1.3). The closeness of this measurement to the reference value (60.0% mol/mol) indicated that even if this material had been used as the calibrator in place of the nucleotide monophosphate solutions, the values obtained for the unknowns would have been only marginally different. Howeve, when gravimetrically prepared nucleotide monophosphate solutions were used, the measuring system was linked to the SI through calibration of the balances used to prepare the solutions. It may be possible to trace the oligonucleotide reference value to the SI by rigorous purity measurements that are themselves traceable to the SI, but methodology to achieve this type of traceability has yet to be developed. Uncertainty Estimation. Uncertainty was estimated in accordance with the “Guide to the expression of uncertainty in measurement”.19 Briefly, the standard uncertainty of each term in the measurement equation, eq 1, was calculated and then combined to give the total uncertainty of the measurement; the combined standard uncertainty was multiplied by a coverage factor to give a 95% confidence interval for the result (Table 3). The coverage factor was obtained from the t-distribution with combined degrees of freedom calculated from the Welch-Satterthwaite equation.19 The uncertainty budget contained three terms corresponding to the terms in the measurement equation: 1. uncertainty of the equivalence of molar response ratio in test and calibration solution 2. uncertainty of the peak area ratios in test and calibration solutions 3. uncertainty of the calibration mole ratio The first of these factors, the ratio of molar responses in test and calibration solutions, may have been affected by any differences between the solutes in these solutions. Since the oligonucleotide calibrator was purified in a manner similar to the unknown materials, this may be used to evaluate the effect of solutes on molar response ratios. If the different solutes caused a difference in response between test and calibration solutions, the difference between the measured and reference value of the calibrator would include the solute effect. The standard uncertainty Analytical Chemistry, Vol. 81, No. 17, September 1, 2009
7299
for assigning the molar response ratio as unity could, thus, be calculated as half the difference between the average of the measured values for the calibrator oligonucleotide and the reference value, and the relative uncertainty was calculated as the standard uncertainty divided by the measured value. Other factors such as oligonucleotide purity, calibration solution matching, and method precision would also contribute to the difference between the measured and reference value, so this is an overestimate of the uncertainty for this factor. The second factor could be readily estimated from precision measurements since peak area ratios rather than absolute peak areas were used in all calculations. The uncertainty due to errors in peak area ratios was estimated as the method precision uncertainty (variance of repeatability) from the ANOVA analysis of results from replicates of the unknown samples. Standard uncertainty of method precision was, thus, calculated as the square root of the relative variance within subsamples from the same vials divided by the square root of the total number of independent measurements. The mole ratio of each calibration solution was obtained by dividing the amount of substance C by the amount of substance mC calculated from the mass fractions of the mixed calibration solution. calibration mole ratio )
nC nmC
of reference material first weighed (g), mS1 ) mass of solvent (g, methanol) used to dissolve reference material, m2 ) mass of solution (g) taken for dilution, mS2 ) mass of solvent (g, water) used to dilute first solution, m3 ) mass of diluted solution (g) taken into mixed calibration solution, mS3 ) total mass (g) of mixed calibration solution, and M ) molar mass of reference material.
u(n) ) n ×
(u(P)P )
2
(
( ) ( ) ( ) ) ( ) ( ) (
u(mS2) mS2
( )
nC nC ) × nmC nmC
( ) ( u(nC) nC
2
+
u(nmC) nmC
) ( 2
+
u(MSC) MSC
)
2
(3)
where u(nC/nmC) ) standard uncertainty in the calibration mole ratio, u(nC) ) standard uncertainty of value for amount of substance of 2′-deoxycytidine-5′-monophosphate, u(nmC) ) standard uncertainty of value for amount of substance of 2′deoxy-5-methylcytidine-5′-monophosphate, disodium salt, and u(MSC) ) standard deviation of methylation ratio obtained using replicate calibration solutions. The standard uncertainties in the amounts of substances were calculated by combining the uncertainties of the purity of the reference material and the gravimetric masses used to prepare the calibration solutions. The amount of substance values were calculated as the product of the mass of reference substance and its purity and the mass dilutions in preparation of the mixed calibration solution divided by the molar mass of the reference substance. The overall equations to calculate amount of substance were n)P×
( ) ( ) ( )
m1 m2 m3 109 × × × mS1 mS2 mS3 M
(4)
where n ) amount of substance of reference material (nanomoles), P ) purity fraction of reference material (g/g), m1 ) mass 7300
Analytical Chemistry, Vol. 81, No. 17, September 1, 2009
+
u(mS1) mS1
+
u(m3) m3
2
2
u(mS3) mS3
+
u(m2) m2
+
2
+
2
+
u(M) M
2
)
In practice, the uncertainty in molar mass was insignificant compared to the other terms, so it was not evaluated. The purity fraction of the reference material was calculated using eq 6 so the uncertainty of the purity fraction was obtained by summing the relative uncertainties as given in eq 7.
P)
u(P) ) P ×
u
2
2
(5)
(2)
Uncertainty in the calibration mole ratio, thus, consisted of the sum of the uncertainty in the two values for amount of substance with and an additional factor added to account for the variation observed when multiple calibration solutions were prepared; these were combined as relative uncertainties.
u(m1) m1
+
(100 - W) × PO 100 W ) (u(W)
2
+
(6)
( ) u(PO) PO
2
(7)
where P ) purity fraction of reference material (g/g), W ) Karl Fischer water measurement of reference material (g/100 g), and PO ) organic purity (g/100 g). The Karl Fischer water measurement of the reference material was corrected for bias due to recovery; each batch of Karl Fischer titrations was accompanied by a recovery measurement that consisted of replicate titrations of a dilution of a Hydranal 1 mg/g reference material. The uncertainty in the water measurement, thus, included the uncertainty estimate for the recovery correction. The average coefficient of variation of all recovery measurements of 1.4% was used as the relative standard uncertainty of the recovery correction; the uncertainty of the water measurement (uW) was the sum of standard deviation of the replicate titrations (KF) and the recovery correction (Rec).
uW ) W ×
u(Rec) +( ( u(KF) KF ) Rec ) 2
2
(8)
For both unknowns 1 and 2, the majority of the uncertainty was due to the calibration mole ratio and a major portion of this uncertainty was due to uncertainty of water content in the reference nucleotide monophosphates measured by Karl Fischer titration (Figure 2). Thus, if improvements in measurement uncertainty are needed, then attention must be focused on reducing the uncertainty of the calibration mole ratio. This could be readily achieved by reducing uncertainty in Karl Fisher water measurements through the use of larger masses of reference materials and by subsequent improvements in preparation of calibration solutions. Alternatively, appropriate oligonucleotides could be prepared and rigorously analyzed to obtain a mole ratio
value that is traceable to the SI; this would eliminate the need for water measurement, but purity measurements would still be needed. COMPARISON WITH REFERENCE VALUE AND CONCLUSIONS The measurement of the reference values for study materials unknown 1 and unknown 2 and a comparison of all study participants’ results is reported elsewhere.15 Our results for both DNA samples were extremely close to the reference values, and our uncertainty was on the order of 8% of the value reported (Figure 3). The uncertainty analysis pinpointed simple steps that could halve the uncertainty. The close agreement of the results from the measuring system developed at NMIA and those from other participating laboratories to the reference values validated that accurate measurement of DNA methylation ratios was achieved using nucleotide monophosphate reference materials to calibrate the measuring system. In addition, measurement of the putative calibrator oligonucleotide methylation ratio was also in close agreement to the theoretical value indicating that the oligonucleotide may be used to calibrate the measuring system, though it was not used in this work. That is, single stranded DNA was shown to be a commutable standard for double stranded DNA in this measuring system. The close agreement between the measured and assigned methylation ratio of the study materials also provided evidence that these synthetically produced double stranded DNA materials could be used to calibrate the measuring system. Use of an oligonucleotide or PCR product as a calibrator would allow preparation of reference materials using standard molecular biology techniques, thus eliminating the requirement to measure
the mass of reference material when preparing calibration solutions. However, since mass and purity of reference materials provide the link to the SI, an alternative system for establishing traceability to the SI for synthetically produced oligonucleotides or PCR products would be needed. If lower measurement uncertainty than has been demonstrated in this work should be needed for certification of putative reference material methylation ratios, then more accurate measurements may be readily achieved with this measuring system by reducing the uncertainty of moisture measurement of nucleotide monophosphate reference materials. In conclusion, a system for directly measuring methylation ratios in synthetic DNA materials that gives values that are traceable to the SI has been demonstrated. This approach could be used to measure methylation ratios in synthetic DNA that more closely approximates the methylation patterns found in genomic DNA, and those materials may then be used to calibrate methylation measurements using contemporary technologies such as sequencing of PCR amplicons after bisulfite conversion of genomic DNA. In this way, genomic methylation measurements would be internationally comparable, and this may help in the validation of research findings. SUPPORTING INFORMATION AVAILABLE Derivation of the measurement equation (eq 1) is available in the Supporting Information. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review May 21, 2009. Accepted July 16, 2009. AC901116F
Analytical Chemistry, Vol. 81, No. 17, September 1, 2009
7301
