Full Genotyping of a Highly Polymorphic Human Gene Trait by Time-Resolved Fluorescence Resonance Energy Transfer

The ability of detecting the subtle variations occurring, among different individuals, within specific DNA sequences encompassed in highly polymorphic genes discloses new applications in genomics and diagnostics. DQB1 is a gene of the HLA-II DQ locus of the Human Leukocyte Antigens (HLA) system. The polymorphisms of the trait of the DQB1 gene including codons 52–57 modulate the susceptibility to a number of severe pathologies. Moreover, the donor-receiver tissue compatibility in bone marrow transplantations is routinely assessed through crossed genotyping of DQB and DQA. For the above reasons, the development of rapid, reliable and cost-effective typing technologies of DQB1 in general, and more specifically of the codons 52–57, is a relevant although challenging task. Quantitative assessment of the fluorescence resonance energy transfer (FRET) efficiency between chromophores labelling the opposite ends of gene-specific oligonucleotide probes has proven to be a powerful tool to type DNA polymorphisms with single-nucleotide resolution. The FRET efficiency can be most conveniently quantified by applying a time-resolved fluorescence analysis methodology, i.e. time-correlated single-photon counting, which allows working on very diluted template specimens and in the presence of fluorescent contaminants. Here we present a full in-vitro characterization of the fluorescence responses of two probes when hybridized to oligonucleotide mixtures mimicking all the possible genotypes of the codons 52–57 trait of DQB1 (8 homozygous and 28 heterozygous). We show that each genotype can be effectively tagged by the combination of the fluorescence decay constants extrapolated from the data obtained with such probes.


Introduction
The completion of the Human Genome Project has evidenced a large number of subtle variations (polymorphisms) among the genomes of different individuals of the same population [1]. Gene polymorphism is responsible for the expression of many variants of a protein among different subjects of the same species. To be classified into polymorphism, variations of gene sequences must appear in at least 1% of the population. In the latter case, the genetic variants are called alleles. The great majority of DNA polymorphisms consist in single-base variations, i.e. single-nucleotide polymorphisms (SNPs) [1]. Most SNPs occur in introns, thus they are silent, i.e. they have no effect on the individual phenotypes. However, SNPs, though rarely being directly pathogenic, are frequently correlated to the occurrence of multigenic diseases [2]. Moreover, certain SNPs result in mutations of genes that are pivotal in tumorigenesis. An example is the tumor suppressor gene p53 [3]. More recently, SNPs have been also exploited as markers in pharmacogenomics [4]. For all the above reasons, the possibility to perform wide range population screening of the relevant polymorphic genes would provide a powerful tool for disease prediction.
A particularly interesting highly polymorphic locus within the human genome is the major histocompatibility complex, located in the short arm of chromosome 6. It comprises many genes coding for the human leukocyte antigen (HLA) proteins that, originally discovered for their role in the rejection of transplanted organs, play a crucial role in the triggering of the immune response against pathogens. Indeed, they guide the development and the activation of T lymphocytes. There are two types of HLA molecules, namely class I and class II, that activate functionally distinct T cell populations, CD8+ T cells and CD4+ T cells, respectively [5]. The expression of certain HLA molecules has been associated with the susceptibility to several autoimmune diseases, including celiac disease [6][7][8][9], multiple sclerosis [10,11], Hashimoto thyroiditis [12], Graves disease [12], narcolepsy [13], selective immunoglobuline A (IGA) as well as common variable immunodeficiencies (CVI) [14], sarcoidosis [15], primary biliary cirrhosis [16], autoimmune [17] and viral [18] hepatitis, insulin-dependent diabetes mellitus (IDDM) [12,19,20]. For all the above pathologies, genetic markers are found in the DQB1 gene, namely in the short though highly polymorphic trait encompassing codons 52-57. The latter trait is manifested in eight allelic variants: DQB1-0201, DQB1-0301, DQB1-0501, DQB1-0302, DQB1-0402, DQB1-0502, DQB1-0602, and DQB1-0503, which in turn correspond to 36 different genotypes (8 homozygous and 28 heterozygous). At least for some pathologies there seems to hold a link between the risk/protection factors and the presence in position 57 of a neutral residue. The latter is encoded in sequences DQB1-0201 and DQB1-0302 (alanine), DQB1-0501 (valine), DQB1-0502 (serine) instead of an aspartic acid residue (provided by all the remaining sequences). However, the susceptibility patterns are peculiar to the different pathologies and not easily explainable on the mere basis of the coded aminoacids. A compendium of disease-correlated DQB1 genotypes and of their Asp/non-Asp nature is compiled in Table 1.
From the above picture, it is clear that being able to determine the genotype of an individual with respect to the highly polymorphic DBQ1 gene trait coding for residues 52-57 would be desirable for various clinical purposes. The ambition of introducing wide-scale screening protocols calls for the development of rapid high-throughput DQB1 typing technologies. Nowadays, molecular typing of polymorphic genes may be performed with single nucleotide sensitivity by applying several different techniques [21]. However, as a rather common feature, virtually all these techniques require amplification of the gene trait to be typed by means of Polymerase Chain Reaction (PCR) as the initial step. Although it is out of doubt that PCR has revolutionized biology, it is not the ideal technique to be applied to the screening of vast populations, due to lack of both readiness and costeffectiveness. During the last decade, molecular typing techniques based on exploitation of the differential fluorescence emission of fluorophore-labeled allele-specific oligonucleotide probes upon recognition of their genomic targets have raised increasing interest because of their particularly neat response (i.e. high signal-to-noise ratio and low probability of false-positives detection), conferring to fluorescence-based assays notable readiness and ease of interpretation [22][23][24][25][26][27]. The common rationale of the majority of such techniques is that light emission by the fluorophore functionalizing the oligonucleotide probe (hereafter called the fluorescence donor, D) can be quenched by suitable dyes (hereafter referred to as fluorescence acceptors, A) in a strongly donor-acceptor-distance dependent way by virtue of a non-radiative decay mechanism of the D from the fluorescent excited state, called fluorescence resonance energy transfer (FRET). The latter occurs whenever the absorption spectrum of the A is significantly superimposed to the fluorescence emission spectrum of the D.
Among others [23,24], FRET-based molecular typing methods making use of D-A dual-labeled probes have proven to be particularly sensitive [22,[25][26][27]. Such probes can be designed to be very slightly fluorescent while not annealed to their genomic target. In general, single-stranded DNA has a much shorter persistence length than double-stranded DNA. Thus, due to hydrophobicity of bases the non-hybridized probe coils in a compact structure in which the 59 and 39 ends (and thus the D and the A) are near. Upon hybridization to the target, the probe assumes a rigid longitudinal structure, and the D, brought further apart from the A, emits fluorescence. Palindromic sequences may be included at the opposite ends of a probe (in this case called Molecular Beacon) in order to stabilize the single-strand in a hairpin structure, in which fluorescence is optimally quenched. In most instances a number of allele specific probes equal to that of possible allelic variants are used. Annealing conditions allowing hybridization to the template DNA only of the exactly matching probe are then determined, and the response of a typing experiment consists in emission of a detectable fluorescence signal exclusively by the specimen added with such probe [22]. However, during the last five years, we validated an alternative approach based on quantitative analysis of FRET-tuned fluorescence signals [25][26][27][28][29]. In the latter approach, a reduced number of probes (ideally a single probe) are used, all of which being capable of hybridizing to any of the possible target sequences (i.e. to each possible polymorphic variant of the gene of interest) with sufficient selectivity with respect to other genomic regions. Allele-specific annealing thus occurs through formation of duplexes whose structural details depend on the probe-target degree of complementarity, thus, in the end, from the target sequence. Routinely, FRET efficiencies are evaluated through relative intensity measurements. Attempts of pursuing molecular typing of polymorphic genes by exploiting this method have been made [23,24] and have brought to a proof of principle of the feasibility of detecting the polymorphisms of the ABL portion of the BCR-ABL oncogene [24]. However, this approach is plagued by major limitations. Firstly, quantitative assessments of the fluorescence intensity are biased by variations, among specimens, in the probe as well as in template DNA concentrations. Moreover, application to non-purified genomic material is complicated by superposition to the relevant fluorescence signal emitted by the probe of an unpredictable amount of fluorescence produced by endogenous fluorophores. Finally, detectors endowed with light-intensityproportional responses must be used, whose low quantum efficiency makes PCR amplification hardly avoidable. The above-mentioned drawbacks can be overcome by quantitating the variations in FRET efficiency through measurements of the fluorescence decay time t D of D. Indeed, t D can be determined by means of time-correlated single-photon counting (TCSPC). TCSPC works by detecting and timing one photon per excitation pulse at most. The subsequent reconstruction of the statistical distribution of the detection times relative to the excitation pulses yields the fluorescence decay pattern, whose fitting provides the value of t D . For this reason, TCSPC is intrinsically independent from the D concentration and takes advantage of the usage of Geiger-mode light detectors, such as single-photon avalanche diodes, which are sensitive to even extremely deem fluorescence pulses, virtually reduced to few/single photons. We recently showed that by this method it is possible to pursue the recognition of the DQB1-0201 genotype, conferring susceptibility to, e.g., celiac disease and IDDM, by working on template DNA contained in untreated (i.e. unamplified and non-purified) cell extracts [27]. To this aim we designed an oligonucleotide probe dual labelled at its 59 and 39 ends, respectively, with a fluorescent dye and with a non-emitting FRET acceptor, and with sequence complementary to the region of DQB1-0201 allele encompassing codons 51-57 and the first nucleotide of codon 58. Namely, the fluorophore was tetramethyl-rhodamine (TAMRA) while the acceptor was black-hole quencher 2 (BHQ2). The latter probe, hereafter Probe1, was made to hybridized to oligonucleotides mimicking any of the eight possible allelic variants of that region of DQB1. The t D of TAMRA attached to Probe1 annealed to each of the target oligonucleotides was measured with 30 ps resolution. Because each target oligonucleotide was characterized by distinctive mismatches with respect to Probe1, we expected the resulting probe/target duplexes to have slightly different conformations, hopefully reflected in significantly different t D values. Indeed, the allele of interest, DQB1-0201, yielded a t D value distinct from the most similar of the others for as much as 42 standard deviations. Moreover, we managed to reveal the presence of the DQB1-0201 allelic variant also in heterozygous conditions, and to discriminate the latter from the homozygous case. Furthermore, Probe1 demonstrated to be sufficiently selective to specifically recognize the DQB1 trait of interest within the whole genome. Finally, we were able to replicate the t D results by using unpurified template DNA contained in bare cell lysates. However, we failed to obtain a full discrimination of alleles. Namely, even in the preliminary calibration measurements undergone on the oligonucleotide targets, Probe1 yielded hardly distinguishable t D values when annealed to DQB1-0503, DQB1-0602, and DQB1-0402-like sequences. With the goal of breaking such a degeneracy on t D values and pursuing a complete genotyping of the DQB1 trait of interest, more recently we performed additional TCSPC experiments aimed at characterizing the fluorescence decay of TAMRA attached to DNA duplexes obtained by annealing the same target oligonucleotides of above with a second oligonucleotide probe (Probe2), complementary to DQB1-0503 [28]. We obtained notably different t D values for Probe2 hybrids with DQB1-05031, DQB1-0602, and DQB1-0402-like targets.
In this work we present a systematic study of the fluorescence decay patters produced by Probe1 and Probe2 upon reaction with oligonucleotide samples mimicking all the possible genotypes of the DQB1 trait encompassing codons 52-57 (i.e. 8 samples containing a single target oligonucleotide variant and the 28 samples resulting by mixing 1:1 concentrations of any possible pair of oligonucleotides). We demonstrate that, while the ''homozygous'' samples can be discriminated by cross-combining the t D value obtained with Probe1 with that obtained with Probe2, for the ''heterozygous'' ones it is still possible to extract an average decay time value from the decay data obtained with both probes, ,t D .. The (,t D,Probe1 .; ,t D,Probe2 .) pair are an exhaustive tag-parameter allowing to discriminate among the different heterozygous genotypes.

Oligonucleotides and Sample Preparation
In Figure 1 the eight target oligonucleotides reproducing the possible sequences of the 22-bases polymorphic trait encompassing codons 51 to 57 and the first base of codon 58 are depicted. In the same figure, the TAMRA-BHQ2 dual-labelled probes Probe1 and Probe2, with sequences complementary to the DQB1-0201 and DQB1-0503 allele-mimicking oligos, respectively, are also shown. Mismatch-sites of the target oligonucleotides with respect to Probe1 are black-shaded, while those with respect to Probe2 are light-gray shaded. Mismatching sites common to both probes are indicated by dark-grey shading. The target oligonucleotides were purchased by TIB Molbiol (Genova, Italy) and were desalted and further purified by double reverse-phase procedure. The duallabelled probes were purchased by Eurofins MWG Operon (Ebersberg, Germany) and purified by high-performance liquid chromatography. A nominal amount of 20 nanomoles of the lyophilized oligonucleotides was resuspended in appropriate volumes (indicated by the suppliers) of Tris-HCl EDTA buffer at pH 7.6 and 10 mM ionic strength, to obtain 50 mM concentrated solutions. Equal amounts (10 mL) of probe and target stock solutions were mixed to obtain the ''homozygous'' solutions. To attain the solutions mimicking the heterozygous genotypes, e.g. DQB1-0XXX/DQB1-0YYY, 5 mL of DQB1-0XXX target oligo and 5 mL of DQB1-0YYY were mixed to 10 mL of probe. The annealing protocol, which is fully described elsewhere [27], consisted in heating the samples to 98uC in order to denaturate any residual secondary structures of the single-stranded oligonucleotides, cooling at the rate of 1uC/min to the annealing temperature and allowing the probe to react with the complementary oligonucleotide for 10 minutes. For ''heterozygous'' samples, the cooling was stopped for 10 minutes at the annealing temperatures of both targets. Finally, the solutions were cooled to 25uC at 3uC/min rate. The thermostat of a ThermoQuest-Finnigan gas chromatograph (San Jose, CA, USA) was used as the temperature controller. The obtained solutions were diluted in phosphate buffer at pH 7.6 and 150 mM ionic strength to a final probe concentration of 250 nM, to produce the sample for TCSPC analysis.

Time-Correlated Single-Photon Counting Apparatus
The fluorescence excitation/detection apparatus is extensively described elsewhere [26]. Briefly, a 113 MHz repetition rated train of excitation pulses (pulse duration 6.4 ps) at 532 nm is provided by a Nd: VAN cw-mode-locked laser (GE-100, Time Bandwidth Products, Zurich, CH) and attenuated by a factor of 10 3 by means of neutral density filters, in order to achieve single-photon regime (photon counts rate ,100 kHz for all samples). Fluorescence is collected at 90u to the excitation beam through a long-wavelengthpass filter with cut at 550 nm, and focused onto the sensitive area of a single-photon avalanche diode (PDM50, Micro-photondevices, Bolzano, IT) by a 20X microscope objective. The avalanche pulses give the START signal to an integrated TCSPC board (which is a single card of a SPC 152 module, Becker & Hickl, Berlin, DE). The STOP signal is provided by the subsequent laser pulse, as detected by a fast photodiode internal to the laser. The START-STOP lag-times occurring in a time window of 9 ns are digitized with 2.44 ps/channel resolution. The full-width at half-maximum duration of the detected excitation pulse is ,30 ps.

Data Analysis
The decay patterns of all the probe/target duplex samples and of the single-stranded probe were measured for 5 minutes. The data processing protocol devised in order to get read of the contributions to the fluorescence decays of unquenched TAMRA molecules and non-hybridized (i.e. single-stranded) probe molecules, which is detailed in previous works [25][26][27] and briefly summarized in Fig. 2, was standardized and automatized by compiling a Matlab routine. We further subtracted a constant background value accounting for non-time-correlated environmental light, which was evaluated by acquiring a photon lag-times pattern with the sample substituted by a water-filled cell for the same measuring time, and averaging the counts values over 1000 channels far from the pulse response peak. Preliminary to the new experiments on heterozygous genotypes mimicking samples, we repeated (in one parallel only) measurements on homozygous genotypes mimicking samples. The so obtained data, after being ''cleaned'' by using the Matlab routine, were fitted to single exponential decays such as in previous works [27,28].
The ''clean'' heterozygous data ideally represent the decay distributions built up uniquely by the fluorescence photons emitted by TAMRA molecules labelling intact (i.e. BHQ2-functionalized) probes correctly hybridized to a target oligonucleotide. Two different approaches were applied to analyze such decays. Firstly, two-exponential fits were performed, with time constants fixed at any possible pair of t D values measured for homozygous alleles. Ex-post discrimination of the correct genotype by comparison of the results of the above fits was attempted. Secondly, the ''clean'' heterozygous decay distributions were fitted to single exponential decays. The extracted t D values, representing a sort of average decay constants, were used to tag each genotype. In both cases, due to the negligible spread of the instrumental response function (IRF, ,30 ps) with respect to the measured t D values (.2 ns), no IRF deconvolution algorithm was applied. However, the fitting interval was chosen to be of 2800 channels, starting from 200 channels (i.e., 488 ps, thus 16 IRF full-width at half maximum) after the detected excitation pulse peak. This choice, besides assuring to start the fitting far enough from excitation in order to be able to bypass deformations induced in the decay histograms by convolution with the IRF, allowed to exclude the board nonlinearity region (i.e., the transient in which the current flowing through the board time-to-amplitude converter switches on and off).
It should be mentioned that putative systematic errors might arise from an incorrect choice of the fitting interval. However, fitting parameters values stable within few percent were obtained for exemplary datasets upon variation of the initial channel of the fitting interval in the range from 170 to 600 channels after the detected excitation pulse peak, with the last channel fixed at 3000 channels after excitation. Moreover, because our measurements are differential in nature and the t D values are simply used as tags of the corresponding genotypes throughout the manuscript, without attempting any absolute evaluation of the donor-toacceptor distance, systematic errors of any source would not invalidate our conclusions.
All the experiments on heterozygous samples were performed in triplicate, and the values reported below are the average on the three parallel samples. The pertaining standard deviations were used as experimental errors. The built-in Matlab fitting tool which we exploited within our routine automatically yields confidence intervals for the fitting parameters values, which were in all cases smaller than such standard deviations.

Results and Discussion
Firstly, we shortly report on the results yielded by analysis of the repeated experiments on homozygous samples. The t D values obtained from the new datasets were equal, within the experimental errors, to those published in the past. This stems in support of both the repeatability and reliability of the typing protocol and the correctness of the new automatized data processing routine. The t D values resulting by averaging the old with the new fitting parameters, with errors given by standard deviations, are summarized in Table 2. The new data are reported in parenthesis, while the old ones are published in [28].
We now come to describe in details the results obtained on the heterozygous genotype mimicking samples. Theoretically, for both probes and for any given ''heterozygous'' sample DQB1-0XXX/ DQB1-0YYY, after subtraction of the unquenched TAMRA and non-hybridized probe transients and of the constant background, there remain only two exponential components building up the ''cleaned'' fluorescence decay. The first component is produced by the part of the probe annealed to the target oligonucleotide bearing the sequence of the DQB1-0XXX allele. This should decay with the time constant t D,XXX tagging the homozygous DQB1-0XXX genotype, reported in Table 2. The other compo- nent is due to the emission of the remaining probe molecules, which are hybridized to oligonucleotides reproducing DQB1-0YYY and thus decay with the time constant t D,YYY tagging DQB1-0YYY in the same Table. In practice, fitting the ''heterozygous'' datasets to a double exponential without introducing constrains leaded to non-convergence. More specifically, the two time constants tended to approach a common value , t D ., being min(t D,XXX ; t D,YYY ) , ,t D . ,max(t D,XXX ; t D,YYY ), which can be interpreted as a sort of average decay time, while the equilibrium values of the relative amplitudes were not reached. We attribute this fact to the rather similar (although statistically differentiable) time constants tagging the different allelic variants. Nonetheless, for all the ''heterozygous'' samples, fitting the ''cleaned'' decay distributions to a single exponential model yields fits significantly less accurate than those obtained for the homozygous-genotypes-mimicking samples (see [29] for a detailed discussion on this topic). Furthermore such single-exponential fits yield ,t D . values very similar to those obtained by means of the unconstrained double-exponential fit and not ascribable to any of those reported in Table 2. This result allows us to conclude that our decay distributions are sensitive to the homozygous/heterozygous nature of the template DNA to be typed.
We thus implemented a data analysis procedure based on constrained double-exponential fit of the data pertaining to the heterozygous samples. Namely, for a subset of 10 out of these datasets we performed 28 fits, in which the time constants were fixed successively to any possible pair t D,XXX ; t D,YYY of t D values measured for homozygous alleles, and the x 2 -value was minimized leaving only the values of the corresponding initial amplitudes f XXX and f YYY as free fitting parameters. A first non-trivial result of this procedure, which is worth being mentioned, is that, upon fixing the t D,XXX and t D,YYY values to those proper of the correct allelic pair, fits significantly more accurate than single-exponential fits with free ,t D . are obtained (see Fig. 2c), particularly the bad agreement between the single-exponential fitting curve and the tail of the experimental decay pattern). Moreover, as discussed at length in the File S1, in which we also provide all the fitting results obtained for the exemplary genotype DQB1 0201/0502, most of the heterozygous genotypes can be excluded on the basis of compelling physical arguments on the retrieved initial amplitude values. However, for all the ten probed genotypes, the fitting with the t D,XXX and t D,YYY values proper of the correct genotype yields physically meaningful amplitudes values for both probes. In Table 3 we report the f XXX,Probe1 ; f YYY,Probe1 ; and x 2 values obtained for the genotypes surviving to the constrained fit analysis with both Probe1 and Probe2 for the exemplary DQB1 0201/ 0502 dataset. The corresponding values for Probe2 are reported in Table 4. Notably, for both probes the t D,XXX ; t D,YYY pair yielding the best x 2 value is t D,201 ; t D,502 , i.e. the fit is maximally accurate when the correct genotype is assumed. However, in this exemplary case (and in seven others out of the ten scrutinized), a few other genotypes (three in the case under examination of the DQB1-0201/DQB1-0502 sample) survive the self-consistency analysis on the relative amplitudes values and yield fits of comparable (although somewhat worse) accuracy. In other terms, a univocal Figure 2. Main steps of the data cleaning procedure shown for an exemplary dataset, namely one of those obtained with the DQB1-0201/0301 genotype mimicking oligonucleotide mixture annealed to Probe2. Upper panel: the decay patterns of the single-stranded probe and of the duplex are synchronized and the probe decay is fitted to a three-exponential-components model. Central panel: the longest lived component of the fit, which is attributed to non-quenched TAMRA emission, is subtracted to both the probe and the duplex distributions, which are then normalized to their integrals and fitted to a 2-exponential and a 3-exponential model, respectively. Lower panel: the weighted subtraction of the non-hybridized probe fraction to the duplex decay yields the net decay of TAMRA molecules attached to intact (i.e. bearing also BHQ2) and correctly hybridized probes. The latter can either be interpolated with a single-exponential or fitted to a constrained 2-exponential model in which the decay time constants values are fixed to those measured for any pair of alleles, the relative amplitudes are allowed to vary and their values analyzed in order to derive the correct genotype (see text for details). In the figure the 2-exponential fitting curve obtained by using the decay time constant values pertaining to the correct pair of alleles is reported. doi:10.1371/journal.pone.0107310.g002 attribution of genotype is not feasible by means of this analysis, which was thus abandoned.
Interestingly, the average decay time values ,t D . (see fifth column in Tables 3 and 4) calculated from any of the surviving set of fitting parameters as Þare very similar, and equal within the experimental errors to the ,t D . values yielded by both single exponential fit (256669 ps and 2279611 ps for the DQB1-0201/DQB1-0502 sample and Probe1 and Probe2, respectively) and unconstrained double exponential fits. This fact suggests that fitting convergence is mainly dictated by retrieval of the correct average decay time value. Hence, we decided to use the ,t D . value restituted by single exponential fit as the representative tag-parameter also for the heterozygous sample, in spite of the somewhat lower quality of fits, which in turn leads to higher uncertainties on the average ,t D . values calculated over the three parallel samples. The results of such single-exponential fits are reported in Table 5.
We first comment on the results obtained with Probe1. Although the ,t D . values obviously do not allow unambiguous discrimination of the various heterozygous genotypes (the latter was not possible with this probe even limiting the analysis to the homozygous genotypes), relevant information concerning the Asp/non-Asp nature of residue 57 can be extracted from these data. In Fig. 3 we show the dendrogram resulting from cluster analysis of the Probe1 data obtained assuming Euclidean metric.
(Dt i,j :DSt D T i {St D T j D) as the distance and applying the complete-linkage cluster joining criterion. In the latter, the distance between two clusters is defined as the maximum distance between one element belonging to the first and one element belonging to the second cluster. The heterozygous genotypes bearing both alleles coding for Asp residues cluster up in a common family (red branches of the dendrogram). Even more relevantly, the genotype associated to maximum proneness to Graves disease and to incremented proneness to celiac disease, i.e. DQB1-0201/DQB1-0302, clusters on its own (black branch). In addition, all and only the heterozygous genotypes conveying per-se susceptibility to celiac disease (i.e. DQB1-0201/DQB1-XXX and DQB1-0501/DQB1-0302) make up the second (green) family of the dendrogram. It is also worth noting that the genotypes containing the maximally disease-correlated allele DQB1-0201 (conferring susceptibility to celiac and Graves diseases, IDDM, Hashimoto thyroiditis, IGA, and CVID, and protection with respect to sarcoidosis) all cluster in the above two families. Finally, the blue cluster contains all the DQB1-0302/DQB1-0XXX genotypes different from DQB1-0201/DQB1-0302, and only one additional ''extraneous'' genotype, namely DQB1-0501/ DQB1-0301. We recall that the presence of the DQB1-0302 allele is correlated to increased risk factor for IDDM, IGA and CVID and to protection with respect to Hashimoto thyroiditis. We also attempted cluster analysis of the Probe2 ,t D . values, but the results (reported in the dendrogram of Fig. 4) are less suggestive from a diagnostic standpoint, as the segregated families do not contain genotypes endowing with susceptibility to a common disease. Moreover, also with this probe unambiguous recognition of any possible genotype is not feasible (e.g. the DQB1-0501/ DQB1-0302 genotype is not distinguishable from DQB1-0602/ DQB1-0302, DQB1-0201/DQB1-0501 from DQB1-0201/ DQB1-0602, DQB1-0201/DQB1-0402 from DQB1-0301/ DQB1-0503, and DQB1-0501/DQB1-0402 from DQB1-0301/ DQB1-0502, as evidenced by the negligible distance values visualized in the dendrogram).  The next step consisted in simultaneously exploiting the ,t D . data pertaining to both probes. With this aim, we devised the graph shown in Fig. 5, in which each genotype is represented by a point whose abscissa is given by the ,t D,Probe2 . value and whose ordinate is ,t D,Probe1 .. Moreover, for any heterozygous genotype DQB1-0XXX/DQB1-0YYY, the fraction F XXX,Probe1 of Probe1 hybridized to the DQB1-0XXX allele can be estimated graphically from Fig. 4. The relevant segments are indicated in Fig. 4 for the same DQB1-0201/DQB1-0502 sample also used to exemplify the constrained double-exponential fit procedure. Namely, F XXX,Probe1 (F 201,Probe1 in the example) is found by dividing the difference (thick vertical segment in Fig. 4) of the ordinates Y XXX,YYY and Y YYY,YYY of the points corresponding to the heterozygous genotype (Y 201,502 in the case under discussion) and to the homozygous DQB1-0YYY/DQB1-0YYY genotype (Y 502,502 in the example) by the difference (thin vertical segment)  Tables 4 and  5 for the DQB1-0201/DQB1-0502 sample). In Fig. 5, the rectangles surrounding the points indicate the confidence intervals (6s) in the determination of ,t D,Probe2 .  (rectangle base) and ,t D,Probe1 . (rectangle height), respectively. Thus, superposition of two or more rectangles would denounce ambiguous tagging of the corresponding genotypes. As it is apparent from Fig. 5, this situation holds for no genotype. This demonstrates our ability of recognizing any possible genotypic variant of the DQB1 trait encompassing codons 52-57 through combined analysis of the TCSPC data obtained with Probe1 and Probe2 within a confidence of one standard deviation. It must be said that such a degree of confidence is rather poor for a clinical test, corresponding to a probability to measure a ,t D . value outside the rectangles .0.33. However, although not decisive, the above discussed method might be exploited as a preliminary screening tool, by virtue of its applicability on negligible amounts of untreated cell lysate, which constitutes a unique prerequisite for achieving superior readiness and cost-effectiveness with respect to traditional, PCR-based genotyping techniques. Moreover, if we limit our analysis to selected genotypes of cardinal clinical relevance, discrimination is feasible to a much higher degree of confidence. Namely, the genotypes DQB1-0201/DQB1-0201, DQB1-0201/DQB1-0302, and DQB1-0402/DQB1-0402, which are correlated with the maximum risk factors for celiac disease and IDDM, for Graves disease, and for biliary cirrhosis, respectively, are unambiguously typed even if the required confidence interval is extended to 63s, corresponding to a probability to measure a ,t D . value outside the rectangles ,0.003. Furthermore, with the same degree of confidence the ambiguity in recognition of DQB1-0602/DQB1-0602 genotype, encoding enhanced proneness to multiple sclerosis, sarcoidosis, and narcolepsy, involves three other possible genotypes only, namely DQB1-0502/DQB1-0503, DQB1-0602/DQB1-0501, and DQB1-0503/DQB1-0402. Finally, the clustering properties inherited by the one-dimensional , t D,Probe1 . topology are conserved also by the distance distribu-tions given by the two-dimensional Euclidean metrics applied on the (,t D,Probe1 .; ,t D,Probe2 .) vectors. As a consequence, the genotypes originally comprehended within the green cluster, and predisposing to Graves and celiac disease, remain recognizable within the confidence interval of 62s, with the only ambiguity between the pathogenically relevant genotype DQB1-0201/ DQB1-0502 and the non-predisposing genotype DQB1-0302/ DQB1-0301. Obviously, removal of such degeneracy by submission of the patients suspected to bear the former genotype to further genetic screening with more refined techniques would be much easier than ab-initio traditional genotyping. The same can be said for the blue cluster, for which extension of the confidence interval to 62s only introduces ambiguity in the recognition of the pathogenically relevant genotypes DQB1-0302/DQB1-0402 and DQB1-0302/DQB1-0502, whose rectangles partially superimpose to those of the non-predisposing genotypes DQB1-0501/ DQB1-0501 and DQB1-0501/DQB1-0503, respectively.

Conclusions
We designed two oligonucleotide probes specific for the highly polymorphic trait of the DQB1 gene encompassing codons 52-57, dual-labelled at their opposite ends with a fluorescence resonance energy transfer donor-acceptor pair. A full characterization of the fluorescence response of the donor when the probes are made to hybridize to oligonucleotide mixtures mimicking all the possible genotypes of that trait (8 homozygous and 28 heterozygous) was undertaken. We demonstrated that each genotype can be effectively tagged by the fluorescence decay constants extrapolated from the time-correlated single-photon counting data obtained with both probes. The probability of misassignments is .5% (corresponding to a Dt D /2s value ,2) for 18 out of 36 genotypes. Here, each genotype is represented by a point whose abscissa is given by the ,t D,Probe2 . value and whose ordinate is ,t D,Probe1 .. The half-base and half-height of the rectangles surrounding each point represent s(,t D,Probe2 .) and s(,t D,Probe1 .), respectively. Discrimination is warranted by lack of superposition among different rectangles. doi:10.1371/journal.pone.0107310.g005 We are aware that such a poor degree of reliability in the attribution of genotypes makes the method unfit to serve as a selfsufficient diagnostic tool in clinical practice. Nonetheless, the readiness and cost-effectiveness of the method, combined with its proneness to be performed on untreated cell lysates, make it a valuable alternative to current genotyping techniques for preliminary wide-range population screening for genetic susceptibility to autoimmune diseases as severe as multiple sclerosis, celiac disease, narcolepsy, diabetes and several others.