An algorithm for the determination and quantification of components of nucleic acid mixtures based on single sequencing reactions

Abstract

Background: Determination and quantification of nucleic acid components in a mixture is usually accomplished by microarray approaches, where the mixtures are hybridized against specific probes. As an alternative, we propose here that a single sequencing reaction from a mixture of nucleic acids holds enough information to potentially distinguish the different components, provided it is known which components can occur in the mixture.

Results: We describe an algorithm that is based on a set of linear equations which can be solved when the sequencing profiles of the individual components are known and when the number of sequenced nucleotides is larger than the number of components in the mixture. We have implemented the procedure for one type of sequencing approach, pyrosequencing, which produces a stepwise output of peaks that is particularly suitable for the procedure. As an example we use signature sequences from ribosomal RNA to distinguish and quantify several different species in a mixture. Using simulations, we show that the procedure may also be applicable for dideoxy sequencing on capillary sequencers, requiring only some instrument specific adaptations of protocols and software.

Conclusion: The parallel sequencing approach described here may become a simple and cheap alternative to microarray experiments which aim at routine re-determination and quantification of known nucleic acid components from environmental samples or tissue samples.

Figure 1

Sequence alignments for the seven taxa used in this study covering the region that is probed by the pyrosequencing procedure with 60 dispensation steps (dispensation order: A-T-G-C). The underlined part represents the primer that was used for the sequencing reaction. The length of the sequence recorded by the pyrosequencing procedure depends on the exact order of the nucleotides and the order of the dispensation steps. Hence, it is slightly different for the different sequences.

Figure 2

Results from the mixture experiments. Observed and expected values are plotted for each mixture. The observed values are averages from four replicates, each evaluated with four replicates of the library of profiles. The actual values of the replicates, as well as the standard deviations are listed in supplementary Table 1.

Figure 3

(A) Distribution of the simulated abundance profile of the components in the mix used for assessing the influence of noise in Figure 3. The concentrations were randomly assigned to 99 samples. The 0^th sample had concentration 0 as a negative control. (B) Section of a simulated peak profile of the mix with the abscissa depicted as a time line measured in scan numbers and the ordinate in arbitrary intensity units.

Figure 4

Simulation of the effect of noise on the recovery of the correct concentrations of the components in a mix of 99 samples with different concentrations. The distribution of the concentrations is shown in Figure 3. From top to bottom: 1, 5 and 10 % noise level. The graphs represent a direct comparison of given (X axis) and found (Y axis) values. A solid line indicates detection limit calculated as 3 times standard deviation of the negative control. Solutions below this line are not reliable.