Multiplexed, Particle-Based Detection of DNA UsingFlow Cytometry With 3DNA Dendrimers for Signal

AmplificationMary Lowe,1* Alex Spiro,1 Yu-Zhong Zhang,2 and Robert Getts3

1Physics Department, Loyola College in Maryland, Baltimore, Maryland2Molecular Probes, Eugene, Oregon3Genisphere, Hatfield, Pennsylvania

Received 12 May 2003; Revision Received 12 December 2003; Accepted 20 January 2004

Background: Complex mixtures of DNA may be found inenvironmental and medical samples. There is a need fortechniques that can measure low concentrations of targetDNAs. For a multiplexed, flow cytometric assay, we showthat the signal-to-noise ratio for fluorescence detectionmay be increased with the use of 3DNA dendrimers. Asingle fluorescent DNA molecule per bead could be de-tected with conventional flow cytometry instrumentation.Methods: The analyte consisted of single-stranded (ss)DNA amplicons that were hybridized to capture probeson the surface of fluorescent polystyrene microspheres(beads) and initially labeled with streptavidin-R-phyco-erythrin (single-step labeling). These beads have a lowreporter fluorescence background and high efficiency ofDNA hybridization. The DNA/SA-RPE complex was thenlabeled with 3DNA dendrimers and SA-RPE. The beadcomplexes were detected with a Luminex 100 flow cy-tometer. Bead standards were developed to convert theintensity to the number of SA-RPE labels per bead and thenumber of dendrimers per bead.

Results: The dendrimer assay resulted in 10-fold fluores-cence amplification compared with single-step SA-RPE label-ing. Based on concentration curves of pure target ss-ampli-cons, the signal-to-noise ratio of the dendrimer assay wasgreater by a factor of 8.5 over single-step SA-RPE labeling.The dendrimer assay was tested on 16S ribosomal DNAamplified from filter retentates of contaminated groundwa-ter. Multiplexed detection of a single dendrimer-labeled DNAmolecule per bead was demonstrated.Conclusions: Multiplexed detection of DNA hybridiza-tion on a single molecule level per bead was achievedwith conventional flow cytometry instrumentation. Thisassay is useful for detecting target DNAs at low concen-trations. © 2004 Wiley-Liss, Inc.

Key terms: flow cytometry; beads; microspheres; den-drimers; DNA; multiplex; single molecule

A particle-based method for rapid multiplex DNA identifi-cation has been reported by several groups [see, e.g. (1–9)].The technique involves flow cytometry detection of analyteDNAs hybridized to capture probes on the surface of poly-styrene microspheres (beads). The beads are coded withfluorescent classification dyes impregnated within a polysty-rene matrix. Each classification code corresponds to aunique probe targeted at a specific DNA. The reporter flu-orophore is associated with successful hybridization on thebead surface. Because conventional flow cytometers, such asthe Becton Dickinson FACSCaliber (1,4) and the Luminex100 [e.g., (2,3)] may be used, this multiplex technology hasthe potential to be implemented in a wide range of labora-tories.

When low amounts of analyte DNA have to be detected, asin the case of rare members of a microbial community inenvironmental samples, the detection threshold becomes acritical issue. Two ways to improve the detection limit in-

clude: (i) increasing the efficiency of probe-target hybridiza-tion (10), and/or (ii) increasing the signal-to-noise ratio (SNR)for fluorescence detection. In the present study, we focusedon improving the sensitivity of fluorescence measurementsby increasing the SNR for labeling of captured DNAs on thesurface of fluorescent beads.

A natural limit of sensitivity is single molecule detec-tion, in which a very small reporter fluorescence signal is

Contract grant sponsor: National Science Foundation; Contract grantnumber: BES-0116765; Contract grant number: DOE NABIR DE-FG02-01ER63264.

*Correspondence to: Mary Lowe, Physics Department, Loyola Collegein Maryland, 4501 N. Charles Street, Baltimore, MD 21210.

E-mail: mlowe@loyola.eduPublished online 29 June 2004 in Wiley InterScience (www.interscience.

wiley.com).DOI: 10.1002/cyto.a.20021

© 2004 Wiley-Liss, Inc. Cytometry Part A 60A:135–144 (2004)

distinguished from the background. The ability to interro-gate one DNA molecule with a single label was demon-strated in several contexts. For example, recent advancesin flow cytometry have led to the capability to detect asingle fluorophore molecule in solution based on time-correlated single photon counting under ultrashort laserpulse excitation (11–15). This technique provided excel-lent discrimination between a single molecule fluores-cence burst and noise due to light scattering and fluores-cent contaminants in the surrounding fluid. However, theapparatus is difficult for widespread use. Another exampleis fluorescence microscopy imaging with total internalreflection excitation. This yielded low background scat-tered light and good SNR sufficient to detect a single dyemolecule on the glass surface (16). Finally, on the surfaceof an unstained bead, a DNA molecule with a single labelwas detected using confocal fluorescence microscopy andtime tracing of single-step photobleaching (17). The beadswere photobleached before DNA attachment to removeautofluorescence from the polymer matrix that could con-tribute to the background. However, none of these ap-proaches is suitable for a multiplexed, bead-based assay, inwhich significant contribution to the background is asso-ciated with classification dyes.

Another approach for single molecule detection is tolabel one DNA molecule with many fluorophores. Inflow cytometry, the typical minimum detection level ofstandard instruments is equivalent to 250 – 800 dye mol-ecules (1,18 –20). When instrumental noise is combinedwith other sources of noise such as classification dyes,nonspecific binding of fluorophores to the bead sur-face, and fluorescent impurities in the fluid, distinguish-ing one DNA molecule from the noise requires severalhundred dye molecules per DNA. Typical target DNAsthat can be captured by bead-probes are �1,000 nucle-otides (nt) in length. To develop a general assay for anytarget length in this range, we explored post-hybridiza-tion fluorescence amplification by coupling the cap-tured DNA with 3DNA dendrimers (21–24), followed byfluorescence labeling.

In the present study, fluorescence amplification withdendrimers, and DNA detection on a single moleculelevel per bead were conducted with a new set of 14bead types stained with either red (R) or infrared (IR)dyes. The beads were designed to minimize the back-ground in the spectral range of the reporter fluoro-phore at wavelengths �600 nm and to have high hy-bridization efficiency of DNA on the bead surfaces. Themultiplexed assays were tested with 16S ribosomalDNA amplified from microorganisms filtered fromgroundwater. Various fluorescent bead standards weredeveloped to convert the intensities of the beads to thenumber of fluorescent labels per bead and the numberof dendrimers per bead. The lower detection limit cor-responding to a single dendrimer-labeled DNA moleculeper one bead was shown. The assay can be used withany commercially available flow cytometer.

MATERIALS AND METHODSBeads

In this study, R and IR fluorescent, carboxylated micro-spheres (5.5-�m-diameter) were from Molecular Probes(Eugene, OR). The beads were stained with long wave-length R or IR BODIPY dyes. Qualified batches of thebeads were shown by analysis with a Becton DickinsonFACScan flow cytometer (BD Biosciences) to have narrowdistributions in diameter and fluorescence intensity, andwere highly monodispersed. Fluorescence spectra of theR and IR beads were measured with an F-4500 fluores-cence spectrophotometer (Hitachi). Both the R and IRbeads have emission spectra between 630 nm and 750 nmwith maxima at 681 nm and 696 nm, respectively. Bothtypes of beads have little to no fluorescence emission inthe range �600 nm, minimizing fluorescence spillover tothe reporter channel during flow cytometric analysis.

Fourteen bead types, six R and eight IR intensity levels,were developed and labeled Rn and IRn, where n indi-cates the intensity level, and R1 and IR1 correspond to thehighest levels. Plain (i.e., undyed) beads 5.60 �m in diam-eter (Bangs Laboratories) and 14 kinds of R and IR beadswere combined in a multiplexed assay consisting of 15bead populations.

The key features of the dye choice and the stainingprocedure were: (i) to produce distinguishable beadsbased on two color classification in the spectral region�600 nm; (ii) to minimize fluorescence spillover at thereporter wavelength 575 nm; and (iii) to avoid modifica-tion of the carboxylated surface of the beads that couldreduce hybridization efficiency and that could increasemultimerization of the beads and nonspecific binding ofthe labels to the beads. The hybridization efficiency, non-specific binding, and multimerization of these newly de-veloped beads were tested in comparison with unstainedbeads, which have similar behavior to Luminex beads (2).

PCR, Capture Probes, Hybridization, and Single-Step Labeling

Two main kinds of template DNAs for the polymerasechain reaction (PCR) amplifications were used: (i) plasmidDNA inserted with the 16S rRNA gene from Desulfobacterpostgatei or Geobacter sulfurreducens, and (ii) commu-nity DNA extracted from microorganisms filtered with anAnodisc from 2 L groundwater at the UMTRA B-01-16-1well in Rifle, Colorado (sampling date June 17, 2002). ThePCR primers were 338Fbs2, containing a 5� biotin, and907R (2). The PCR product was purified with the QIA-quick PCR purification kit (Qiagen) and quantified from a1% agarose gel and a titration of Low DNA Mass Ladder(Life Technologies catalog no. 10068-013). Single-strandedamplicons were prepared according to (1).

The capture probe 6019 for Desulfobacter was 5�[unilink-er]TCTTGCTGTTTCAAGTGCAC. Capture probe 3005 forGeobacter was 5�[unilinker]TACCCGCAACACCTAGTACT(25). All capture probes were attached to the bead surfacesaccording to (1).

136 LOWE ET AL.

The analyte consisted of biotinylated, ss-amplicons. Di-rect hybrid capture of target DNA and single-step labelingwith streptavidin-R-phycoerythrin (SA-RPE, MolecularProbes) are described in (2). The number of beads variedbetween 3,000 and 12,000 of each bead type per hybrid-ization reaction. After washing and resuspension of thebeads in 1� TMAC buffer, the signals were measured witha Luminex 100 flow cytometer.

3DNA Dendrimers and Labeling Procedure

Five types of biotinylated dendrimers were used withdifferent numbers of biotinylated arms and Alexa Fluor546 labels (Molecular Probes) per dendrimer (Table 1).DNA dendrimers were prepared as previously described(21,22), except that the core dendrimer preparations usedin this study, 2-layer [2n(x)] and 4-layer (4n(x)], werelabeled with either biotinylated oligonucleotides [2n(x)b,4n(x)b, and 4n(x)bb] or a combination of biotinylatedoligonucleotides and Alexa Fluor 546-labeled oligonucle-otides [2n(x)bA and 4n(x)bA]. Excess unlabeled oligonu-cleotides were removed by size exclusion chromatogra-phy, and the concentration and specific fluorescencewere determined.

After single-step SA-RPE labeling and washing of thehybridized beads, the dendrimers were bound to theDNA/SA-RPE complex by combining the beads with2.14 fmoles of dendrimers in 46°C 1� TMAC in a totalvolume of 12.5 �l, vortexing, and incubating at 46°C for40 min. The beads were then washed in 500 �l of 46°C1� TMAC buffer. When biotinylated dendrimers 2n(x)b,4n(x)b, and 4n(x)bb were used, an additional layer ofSA-RPE was applied by combining the beads with 12 �l of20-�g/ml SA-RPE, prepared in 1� TMAC buffer at roomtemperature, and incubating at 46°C for 10 min. Thebeads were then washed two times in 46°C 1� TMACbuffer and resuspended in 70 �l 1� TMAC buffer. Figure1 is a schematic representation of the final structure of thebead with the SA-RPE-labeled dendrimer.

To determine the background signal due to nonspecificbinding of labels (SA-RPE and dendrimers) to the beadsurface, background beads were prepared by not expos-ing the bead probes to target amplicons. But in eachexperiment, they were labeled in the same way as thebeads with target DNA analyte.

Preparation of Surface-Labeled Fluorescent BeadStandards

Three kinds of bead standards were prepared by cap-turing biotinylated oligos followed by labeling. In thisstudy, 20 pmols of oligo 338Fbs2 was diluted in 17 �lwater. A capture probe complementary to 338Fbs2 wasattached to plain beads. The oligo and 100,000 beadprobes were combined in a total volume of 51 �l 1�TMAC buffer, vortexed, incubated at 46°C for 1 hour,washed, and resuspended:

1. SA-RPE bead standard. Four pmols of SA-RPE werecombined with 50,000 hybridized beads in 12 �l 1�TMAC buffer, incubated for 10 min at 46°C, and washedtwo times in 46°C 1� TMAC buffer.

2. Two-layer, biotinylated, Alexa Fluor 546-labeled,2n(x)bA bead standard. In this procedure, 4 pmols ofstreptavidin (Molecular Probes) was combined with25,000 hybridized beads in 12 �l 1� TMAC buffer, incu-bated at 46°C for 10 min, and washed two times in 46°C1� TMAC buffer. 50 �l containing 60.5 fmoles 2n(x)bA,prepared in 1� TMAC buffer at room temperature, wasadded to the beads, vortexed, and incubated at 46°C for40 min. The beads were then washed in 46°C 1� TMACbuffer.

Table 1Five Types of Dendrimers Used in This Study

Dendrimer

Characteristics

Mr

Diameter(nm)Structure

No. ofbiotins

No. of AlexaFluor 546

2n(x)b Two layer, spherical 16 None 106 802n(x)bA Two layer, spherical 16 32 1.35 � 106 8044n(x)b Four layer, spherical 125 None 1.2 � 107 1804n(x)bA Four layer, spherical 125 250 1.2 � 107 1804n(x)bb Four layer, spherical 250 None 1.2 � 107 180

FIG. 1. Fluorescence amplification assay using 2n(x)b, 4n(x)b, and4n(x)bb dendrimers. For 2n(x)bA and 4n(x)bA, the final SA-RPE labelingstep was not conducted

137MULTIPLEXED DETECTION OF DNA USING DENDRIMERS

3. Four layer, biotinylated, Alexa Fluor 546-labeled,4n(x)bA bead standard. A total of 25,000 hybridized beadswere first labeled with streptavidin, and then labeled with9.45 fmoles of 4n(x)bA dendrimers, according to the pro-cedure described in step 2 above.

Calibration of Bead Standards

Fluorescence spectra of SA-RPE, SA-RPE-labeled oligo,and fluorescently labeled dendrimers were measured atroom temperature with an F-2500 Fluorescence Spectro-photometer (Hitachi).

The fluorescent bead standards were calibrated in termsof the number of fluorescent labels per bead using anupright episcopic fluorescence microscope (Nikon Labo-phot-2) equipped with a CCD camera (First MagnitudeSpecIm-3, 2 � 2 binning mode) and an argon laser (LaserPhysics 150M-006-00B). A laser beam (514 nm) was di-rected into the microscope and compressed by the micro-scope objective lens onto the sample plane. The fluores-cence from the sample was selected by two glass-cuttingfilters and an IR cutoff interference filter (LS-600, Corion)that allowed � � 560–590 nm to pass through.

The images of the surface-labeled bead standards inTMAC buffer were compared with solutions of the corre-sponding fluorophore conjugates in TMAC buffer. Thebead standards (5.6 �m in diameter) were placed betweentwo quartz coverslips (0.2-mm thickness). Solutions offluorophore conjugates with known concentration wereprepared, mixed with plain beads, and placed betweentwo quartz coverslips. The background sample consistedof TMAC buffer. In all samples, the beads functioned asspacers so that the thickness of the fluid layer between thecoverslips was fixed at 5.6 �m. For the bead standards, thefield of view typically contained several tens of individu-ally resolved surface-labeled beads with a diameter of25 pixels. The intensity of the labeled beads was typically10 times higher than the background fluorescence fromthe buffer and from plain beads. After subtracting thebackground fluorescence, the total intensity of each fluo-rescent bead was obtained and averaged among 80–100beads.

For the fluorophore conjugate solutions, the bead-freeareas were processed to obtain the average intensity froma cylindrical volume with a diameter of 25 pixels (i.e.,138-�m3 volume). From this, the labeled beads were cal-ibrated in terms of the number of fluorophore conjugatesper bead.

For the SA-RPE bead standard, the mean intensity valueper bead was found to be equivalent to 4.58 � 104 SA-RPEmolecules. The concentration of the known solution ofSA-RPE was determined from absorption of the dye at565 nm with an extinction coefficient of 1,960,000 cm�1

M�1. The average fluorophore-to-protein ratio (F/P) for SA-RPE is 1.1 based on high-performance liquid chromatogra-phy (HPLC) and absorption measurements (280 nm for pro-tein, 565 nm for PE). Fluorescence spectroscopicmeasurements showed neither quenching nor spectralchanges when SA-RPE molecules in solution were bound tothe biotinylated oligo (338Fbs2) or 4n(x)bb dendrimers.

Therefore, we believe that flow cytometry calibration withthe SA-RPE bead standard reflects the correct number ofSA-RPE molecules per bead for all samples labeled withSA-RPE.

For the Alexa Fluor 546-labeled bead standards [2n(x)bAand 4n(x)bA], comparison with known dendrimer solutionsindicated 1.97 � 104 and 2.25 � 103 dendrimers per bead,respectively. The concentrations of dendrimer solutionswere determined initially from radioactive tagging of thedendrimers. This was followed by measuring the absorbanceat 260 nm to calculate the extinction coefficient of variousdendrimer assemblies [ � 31 0.5 �g/(ml � OD � cm)].Thereafter, absorbance was used to determine the concen-tration of dendrimer solutions (22).

The surface-labeled bead standards were only stable fora few days. Therefore the bead standards were used tocalibrate a commercial fluorescent bead kit (LinearFlowOrange Flow Cytometry Intensity Calibration Kit, Molec-ular Probes L-14815) with a stability of 1 year. The inten-sities of the commercial beads were converted into thenumber of labels per bead. This calibration kit was rou-tinely used in all flow cytometry measurements of theDNA samples.

Flow Cytometry Measurements

The beads were analyzed with a Luminex 100 flowcytometer. Cytometry data were processed with FCS Ex-press software version 1.065 (De Novo Software) and within-house software. For all calculations, the mean back-ground signals from the no-DNA control beads were sub-tracted from the mean intensities of the bead samples. Theresulting intensity values of the beads were converted intothe number of fluorescent labels per bead with the beadstandards. For example, on the flow cytometer, the SA-RPE bead standard was used to calibrate the signal fromthe hybridization reactions in terms of SA-RPE units (i.e.,the number of SA-RPE molecules on the bead surface).

Statistical Analysis of Flow Cytometry Data

The number n of beads, the mean signal values �s, thestandard deviation s of the beads, and the uncertainty ofthe mean � (� sn�1/2) were determined for a given beadpopulation. Typically n was �500–800. The signal due tothe classification dyes and instrument background weredetermined with separate measurements and described by�bd, sbd, and �bd values. In most experiments, statisticaltests were accomplished by running N replicate hybrid-ization reactions and averaging the �s values. The averagesignal of the replicates (Ms) and the standard deviation(SDs) for reactions with analyte DNA were determined.The replicate reactions with no analyte DNA (i.e., back-ground reactions), but labeled with the same procedure,were used to find the average background signal (Mb) andstandard deviation (SDb). The signal due to DNA wasdefined by M � Ms � Mb. The lower detectable level ofDNA signal was found by computing the t-test betweenDNA and no-DNA groups of replicate samples. The stan-dard error (SE) in the difference of two replicate means,the degrees of freedom (df), the experimental t-values (t �

138 LOWE ET AL.

M/SE), and the critical t0 value for significance level �0.05 were determined by the Satterthwaite approximation(26). The statistically significant difference between Ms

and Mb was determined with the condition that t � t0.A similar procedure was used to determine the lower

detection limit based on a single hybridization reactionand a no-DNA reaction. The DNA signal was determinedby � � �s � �b. (As above, subscripts s and b refer to theDNA and no-DNA samples, respectively). The uncertain-ties in the mean �s and �b were determined, and thet-values were calculated by t � �/�d , where �d � (�s

2 ��b

2)1/2. Since �s, �b, and �d do not reflect variations insample preparation, they are the minimum values ex-pected for SDs, SDb, and SE, respectively.

RESULTSMapping R and IR Beads With The Luminex 100

Flow Cytometer

The fluorescence maxima of R and IR beads are shiftedby 15 nm, sufficient to distinguish the beads on the clas-sification 1 (CL1, 658 10 nm) and classification 2 (CL2,� 720 nm) channels of the Luminex 100 flow cytometer.The bead map, shown in Figure 2a, includes 14 coloredbead types and plain (unstained) beads. It was producedby selecting the monomers in the side-scatter (doubletdiscriminator) histogram and gating the bead populationson the CL1–CL2 dot plot using FCS Express software.

Background Signal of the Beads

In the reporter channel of the Luminex 100, the high-intensity R and IR beads have a significant backgroundsignal due to the classification dyes. Nominally, the re-porter channel of the Luminex 100 is designed to detectemission at 575 12 nm. However, there is leaking ofinfrared radiation from the classification dyes through the

spectral filters into the reporter PMT. The leakage waseliminated by placing an IR cutoff interference filter (LS-600) in front of the PMT. In Figure 2b, the backgroundmeans �bd and the standard deviation sbd for 15 beadpopulations are shown with and without the LS-600 filter.For all stained beads, the background was diminished. Themost significant reduction occurred with the higher levelof classification dyes. For example, the background for R1beads was reduced from �bd � 1550 SA-RPE (sbd � 285SA-RPE) to �bd � 87 SA-RPE (sbd � 66 SA-RPE). For theplain beads, the values �bd � 28 SA-RPE and sbd� 35SA-RPE are mostly due to the instrument background.

Hybridization Efficiency and Percentage ofMonomers of R and IR Beads

The test was conducted by hybridizing G. sulfurreducensand D. postgatei analyte DNAs to capture probes 3005 and6019, respectively, on R, IR, and plain beads. Fifteen beadtypes were multiplexed and exposed to the analyte. Thesignals due to captured DNA were normalized by the signalfor plain beads. The relative hybridization efficiency for all Rand IR beads ranged from 118% for R1 to 172% for IR5 (100%for plain beads). The percentage of monomers after hybrid-ization ranged from 88% for R1 to 99.5% for IR6 beads.

Determining the Optimal Dendrimer andFluorescence Amplification

(1) Gain. For the dendrimer assay shown in Figure 1,the gain relative to single-step labeling is given by:

G � 1 � NPEP (1)

where P � Ndr/NDNA; NPE is the number of SA-RPE mole-cules bound to one dendrimer after the final SA-RPE label-ing; NDNA is the number of SA-RPE-labeled DNA molecules

FIG. 2. a: CL1–CL2 intensity dotplot for fifteen bead populations de-tected with a Luminex 100 flow cy-tometer. The beads used in this studydo not fall on the Luminex bead map.Rectangular gates are drawn aroundeach bead population during postpro-cessing of the raw data acquisitionfiles. b: Spillover of the bead intensi-ties into the reporter channel withand without an LS-600 filter in front ofthe reporter PMT.

139MULTIPLEXED DETECTION OF DNA USING DENDRIMERS

per bead; and Ndr is the number of dendrimers per bead.The gain G was determined by flow cytometry measure-ments of the bead samples before and after fluorescenceamplification.

(2) Size of dendrimer and number of biotins per den-drimer. A study was conducted to determine how size ofthe dendrimer and the number of biotins per dendrimeraffect Ndr. Two types of Alexa Fluor 546-labeled dendrim-ers, 2n(x)bA and 4n(x)bA, were used (Table 1). Desul-fobacter ss-amplicons (570 nt in length) were initiallyhybridized to IR5/6019 beads and labeled with SA-RPE.Prior to dendrimer labeling, the samples were split, andNDNA� 140 was determined using the SA-RPE bead stan-dard (Table 2). The beads were then labeled with 2n(x)bAand 4n(x)bA. In each reaction, there were 12,000 beadsand 2.14 fmoles of dendrimers. The fluorescence signalassociated with Alexa Fluor 546 was converted into thenumber of dendrimers using the 2n(x)bA and 4n(x)bAbead standards. The number Ndr of dendrimers per beadwas calculated to be 190 for 2n(x)bA and 49 for 4n(x)bA.The percentage P was determined to be 136% for 2n(x)bAand 35% for 4n(x)bA. In other words, for 2n(x)bA, morethan one biotinylated dendrimer was bound on average toan SA-RPE/DNA complex. (In solution, one streptavidinmolecule can bind four biotins.)

It was assumed that dendrimers of the same size andnumber of biotins would have the same binding efficiencyto DNA. That is, under the same reaction conditions statedabove with NDNA � 140, Ndr � 190, P � 136% for 2n(x)band Ndr � 49, P � 35% for 4n(x)b (Table 2). By measuringG, NPE was found to be �1 for 2n(x)b and �8 for 4n(x)b.

(3) 4n(x)bb dendrimer. The 4n(x)bb dendrimer hastwo times as many biotinylated arms as 4n(x)bA but hasthe same size and mass. The P � 35% value for 4n(x)bAwas therefore assumed to be the lower limit of P for4n(x)bb. Measurements of G using the same reaction

conditions as above indicated G � 9.3. Assuming P � 35%for 4n(x)bb, NPE was found to be � 23. An alternatemethod for determining NPE was to note that for 4n(x)bA,there were 8 SA-RPE per 125 biotins. Assuming the samepercentage of SA-RPE labeling, then 4n(x)bb should have16 SA-RPE per 250 biotins, i.e., NPE � 16. Therefore, therange of NPE for 4n(x)bb was estimated to be 16–23(Table 2).

To maximize the gain G, the quantity NPEP must bemaximized. A comparison between 2n(x)b and 4n(x)bdendrimers showed that the NPE value increased by thesame factor as the number of biotins. However, P de-creased when larger dendrimers were used. Nevertheless,the gain for 4n(x)b was greater than for 2n(x)b. There-fore, we believed that greater amplification could beachieved with larger dendrimers with more biotinylatedarms. For all subsequent experiments, we used the4n(x)bb dendrimer.

Dynamic Range of Amplification

The amplification of the 4n(x)bb dendrimer assay wasdetermined over a wide range of analyte DNA concentra-tions. Five thousand IR5/6019 beads were exposed to0–950 fmoles of Desulfobacter analyte ss-amplicons,shown in Figure 3. The signal �s was measured for single-step labeling (curve 1) and the dendrimer assay (curve 2).The no-DNA background was 101.8 58.4 SA-RPE units(curve 1) and 217 98 SA-RPE units (curve 2). The gainof the DNA signal � was G � 10 for C � 10 fmoles(NDNA � 250 SA-RPE units). Near saturation (e.g., C �950 fmoles, NDNA � 7050 SA-RPE units), the gain de-creased to G � 4. In both cases, the dependence of � onC can be approximated by a saturation type curve:

� � kaImaxC/�1 � kaC� (2)

Table 2Values of G, NPE, Ndr, and NDNA for Five Types of Dendrimers

DendrimerSize

(nm)No. ofbiotins

No. of SA-RPE-labeled DNA per

bead (NDNA),SA-RPE units

Totalsignal,k

SA-RPEunits

Gain(G)

No. ofdendrimersper bead

(Ndr) P,f %

No. ofSA-RPE

perdendrimer

(NPE)

2n(x)bA 80 16 140i 205a 190c 136 N/A4n(x)bA 180 125 140i 300a 49c 35 N/A2n(x)b 80 16 140i 340 2.4b 190d 136 1g

4n(x)b 180 125 140i 525 3.8b 49d 35 8g

4n(x)bb 180 250 260j 2420 9.3b 91f 35d 16–23h

aThe value reflects the total signal contributed by NDNA molecules and Alexa Fluor 546-labeled dendrimer in equivalent SA-RPE units.bG � (total signal)/NDNA.cThe values were obtained using the Alexa Fluor bead standards.dThe value was assumed to be the same as for Alexa Fluor 546-labeled dendrimers of the same size.eP � Ndr/NDNAfCalculated from Ndr � PNDNAgThe value was calculated from eq. 1.hSee text.iThe hybridization reaction contained 6.7 fmoles ss-amplicons.jThe hybridization reaction contained 8.0 fmoles ss-amplicons.kBackground was subtracted.

140 LOWE ET AL.

where Imax is the maximum signal corresponding to com-plete saturation. For single-step labeling, ka (fmoles�1) isrelated to the association constant (2). With dendrimeramplification, ka is a parameter describing probe-targethybridization and multistep labeling. Parameter values forcurve 1 are ka � 4.4 x 10�3 fmoles�1 and Imax � 7400SA-RPE units. For curve 2, ka � 1.3 x 10�2 fmoles�1 andImax � 3 x 104 SA-RPE units. The higher value of ka forcurve 2 is indicative of earlier saturation. Most likely this isbecause the maximum number of 180-nm dendrimers thatcan be placed on the bead surface is approximately 3,000,a value less than the number of DNA molecules per beadmeasured at high concentration (NDNA �7050).

Detection on a Single Molecule per Bead Level

The lower detection limit of the dendrimer assay wasdetermined with replicate hybridization reactions ofDesulfobacter analyte ss-amplicons and five thousand IR5/6019 beads. These beads have a background mean Mbd �

91.8 SA-RPE units and SDbd � 2.3 SA-RPE units. Theamounts of target DNA were 0 (“no DNA”), 0.031, and0.125 fmoles. Four replicates were prepared for eachconcentration. For each set of replicates, Ms, Mb, M, andSE were determined, and the t-test was applied for �0.05. The results are shown in Table 3. For single-steplabeling (i.e., no amplification), the experimental t-valueswere less than the critical t0; therefore, there was nostatistically significant difference between the DNA andno-DNA samples. With dendrimer amplification, DNA onthe bead surface could be detected for 0.031 and 0.125fmoles. The values for M were converted into the numberNdr of dendrimer-labeled DNA using the result of 24 SA-RPE molecules per one DNA, i.e., one SA-RPE moleculedirectly attached to the target and 23 SA-RPE on thedendrimer (see above section). At 0.031 fmoles, the signaldue to DNA was M � 27 SA-RPE units, equivalent to Ndr �1.1. Since a great excess of dendrimers was added to ahybridization reaction, this result strongly indicates thatone dendrimer-labeled DNA molecule per bead was de-tected on average. Since P � 35% for 4n(x)bb (Table 2),we believe that on average, each bead has three SA-RPE/DNA complexes on the surface, of which one is labeledwith a dendrimer. For 0.125 fmoles of target DNA, 2.5dendrimer-labeled DNA molecules were detected on aver-age per bead.

Multiplex DNA Detection in an EnvironmentalSample

A bead-based assay using dendrimer amplification wasdeveloped to target microbial groups (e.g., genus, family)known to reduce uranium. In many applications, thesemicrobial groups may be present at very low levels, andwould require a sensitive assay to detect their presence. Atest was conducted on a filtered biomass sample fromWell B-01 at the UMTRA site. Fifteen capture probe se-quences were either obtained from the literature or de-signed. The microbial groups are shown in Table 4.

The results for single-step labeling and dendrimer am-plification were compared. Hybridization reactions wereprepared in triplicate using 0 fmoles analyte DNA and 67fmoles environmental sample ss-amplicons. A total of45,000 beads (i.e., 3,000 beads of each type) were used ineach reaction. Table 4 shows the DNA signals M, standarderrors SE, and the results of the t-test for each captureprobe. With no amplification, only four bead-probes IR8/

FIG. 3. Concentration curves using 16S rDNA ss-amplicons from Des-ulfobacter postgatei. The vertical axis is the reporter signal �s (i.e., nobackground subtraction). Curve 1, single-step SA-RPE labeling. Curve 2,4n(x)bb dendrimer amplification assay. Experimental data (circles andstars). The solid line is a fit.

Table 3Detection of a Single Dendrimer-Labeled DNA Molecule on a Bead Surface

Amount ofanalyte DNA(fmoles)

Single-step SA-RPE labeling Amplification with dendrimers

Ma SEa t t0 t � t0 Ma SEa t t0 t � t0

No. of dendrimer-labeled DNA per

beadb

0.031 0.20 1.46 0.14 2.45 No 27.0 4.49 6.02 2.78 Yes 1.10.125 5.80 2.35 2.47 2.78 No 60.0 10.5 5.71 3.18 Yes 2.5

aIn SA-RPE units.bAverage number detected per bead.

141MULTIPLEXED DETECTION OF DNA USING DENDRIMERS

6012, IR3/6013, R3/3010, and plain/3005 exhibited de-tectable signals. The DNA signal was �10 SA-RPE mole-cules per bead.

After amplification, 14 probes yielded detectable sig-nals. The bead-probe R4/6023 (non-detectable signal be-fore amplification) had on average less than one den-drimer-labeled DNA molecule per bead. The lowestdetectable signals (for R6/6017, R5/6022, and plain/3005)were equivalent to 1–2 dendrimer-labeled DNA moleculesper bead. For three bead-probes IR8/6012, IR3/6013, andR3/3010 (detectable signal before amplification), the gainwas G � 8–12. Unexpectedly low amplification G � 1.2was obtained for plain/3005. Based on subsequent analy-sis by cloning and sequencing methods of the target DNA,there was likely weak binding of capture probe 3005 tothe target DNA. It is possible that most of the target DNAmay have been removed from the beads by the extra washsteps in the dendrimer assay resulting in a reduced gain.

Signal-to-Noise Ratio and Limit of Detection

Replicate hybridization and labeling reactions enablethe determination of variability in sample preparation.However, depending on the application, there may bepractical restrictions in the number of replicates that canbe prepared. It is therefore useful to determine fromsingle reactions (i.e., no replicates) the SNR and detectionlimit.

A minimum of two different reactions are required todetermine the DNA signal and the noise: (a) a no-DNAreaction for determining the background (mean �b anduncertainty �b), and (b) a hybridization reaction withanalyte DNA for determining �s and �s . The SNR ort-value is given by t � �/ �d (see Materials and Methods).

The SNR for the experimental data shown in Figure 3 wascalculated. For large amounts of target DNA (C �30 fmoles, NDNA � 1,000), the SNR was t � 100 andapproximately the same for both data sets with and with-out amplification. However, dendrimer amplification im-proved the SNR by a factor of 8.5 at low target amounts(C � 2 fmoles) when the signal due to DNA (NDNA � 60;see Fig. 3) was comparable or less than the IR5 beadbackground (�bd � 92 SA-RPE, sbd � 54 SA-RPE). Figure 4compares the SNR with and without amplification in thelow analyte concentration range. The lower limit of de-tection Clim was estimated from the linear regression ofthe SNR curves and their intersection with the criticalvalue t0 � 1.96. From this, Clim � 0.25 fmoles of analyteDNA for single-step labeling and Clim � 0.03 fmoles withdendrimer amplification.

For no replicates, it can also be noted that the lowerlimit of DNA detection can be calculated from �lim � t0�d.Since a typical value of �d for amplified samples at lowconcentrations (C � 0.125 fmoles) was �d � 6 SA-RPE, thedetection limit corresponded to �lim � 12 SA-RPE orNdr � 0.5 on average. This result is in agreement withvalues obtained for four replicate samples in which onedendrimer-labeled DNA was detected at C � 0.031 fmoles,and t was about two times greater than t0 (Table 3).Therefore, the interpretation is that t0 is equivalent toabout one dendrimer-labeled DNA per two beads on av-erage.

DISCUSSIONUsing conventional flow cytometry instrumentation,

the natural limit of fluorescence detection for this type ofbead-based assay—a single DNA/fluorophore conjugate

Table 4Multiplexed Detection of an Environmental PCR Product Using 15 Bead Probes*

Bead/probeTarget microbial

group

Single-step labeling Amplification with dendrimers

Ma SEa Presence (t/t0) Ma SEa Presence (t/t0)Demdrimer-labeled

DNA per beadb

plain/3005 Geobacteraceae 21.9 2.6 Yes (1.9) 26.9 7.8 Yes (1.1) 1.1IR8/6012 Desulfococcus 10.7 2.1 Yes (1.8) 96.5 15.6 Yes (1.5) 4.0R6/6017 Desulfovibrio 4.3 4.2 No (0.3) 40.6 5.8 Yes (2.2) 1.7IR6/3004 Geothrix 2.0 3.6 No (0.2) 105.5 11.7 Yes (2.3) 4.4IR7/6011 Shewanella 1.6 3.1 No (0.2) 64.4 8.3 Yes (2.6) 2.7IR5/6019 Desulfobacter 3.2 2.1 No (0.5) 78.0 7.6 Yes (2.6) 3.2R5/6022 Desulfotomaculum 1 3.9 4.2 No (0.2) 33.0 9.8 Yes (1.2) 1.4R4/6023 Desulfotomaculum 2 7.6 6.7 No (0.3) 21.1 41.9 No (0.1) 0.9IR4/6000 Clostridium 1 0.0 3.5 No (0.0) 89.6 15.0 Yes (1.6) 3.7IR3/6013 Desulfobulbus 18.1 5.2 Yes (1.1) 143.9 16.0 Yes (2.8) 6.0IR2/6021 K39Al (Clos) 0.0 5.2 No (0.0) 180.0 36.5 Yes (1.6) 7.5R3/3010 533FA “universal” 99.9 15.9 Yes (1.5) 1209.1 52.3 Yes (5.5) 50.4IR1/6004 Escherichia coli 13.9 10.0 No (0.5) 118.5 15.5 Yes (2.4) 4.9R2/3006 SJA-19c 0.2 8.3 No (0.0) 113.5 18.0 Yes (2.2) 4.7R1/3008 Unculturedd 0.9 9.8 No (0.0) 85.0 16.1 Yes (1.5) 3.5

*The results from single-step labeling and dendrimer amplification are compared. A t-test ( � 0.05) was used to determine thepresence or absence of captured target DNA.

aIn SA-RPE units.bAverage number detected per bead.cLow G � C gram-positivedMaybe Proteobacteria, found in TCE-contaminated site.

142 LOWE ET AL.

per bead—was achieved. This was accomplished by (i)reducing the background emission of the beads, and (ii)using 3DNA dendrimers to amplify the fluorescence sig-nal. For low concentrations of analyte DNA, consisting of16S rDNA ss-amplicons (570 nt), the signal-to-noise ratioof the dendrimer assay was greater by a factor of 8.5 oversingle-step labeling. With the dendrimer assay, the mini-mum detectable amount of analyte DNA was loweredfrom hundreds to tens of attomoles.

In general, if a system obeys a Poisson distribution, theSNR is proportional to the square root of the signal (26).Then, for amplified signals, the SNR is proportional toG1/2, where G is the gain. However, for flow cytometrydata with very low DNA signals �, the noise (�d) does notsimply depend on �1/2, but is mostly due to classificationdyes and instrument background which is not routinelymeasured (18). In this case, the SNR is essentially propor-tional to G. This enables the dendrimer assay to improvethe detection limit. For low concentrations of target DNA(Figs. 3 and 4), the gain was G � 9–10, and the increasein SNR was by a factor of 8.5.

The well-defined concentration curves (Fig. 3) suggestthat quantification of target DNA in a complex mixture ofDNAs is possible with a dendrimer assay using similarmethods described in (2). In practice, however, the den-drimer assay is more complex to execute than single-steplabeling, and would require more time to quantify targetDNAs. For analyzing microbial communities in environ-mental samples, we believe the best use of this assay is todetermine the presence or absence of target DNAs at verylow concentrations. This could occur when target micro-

organisms have a low relative abundance or when thebiomass collected from the environment is small, such asthe case shown in Table 4.

While detection of a single molecule per bead wasachieved, the true analytical sensitivity of the method isdetermined by the number of target molecules in theoriginal unknown sample. For environmental studies, thesample consists of microorganisms collected from, forexample, groundwater or sediment, from which commu-nity DNA is extracted. It is useful to estimate whether thedendrimer assay has sufficient sensitivity to detect DNAfrom a target organism without a PCR step. Based on 2 Lgroundwater from the Rifle site, genomic DNA from �109

cells could be extracted with a volume of 50 �l, as deter-mined by fluorometry. For detection by the dendrimerbead assay, the DNA must be shorter, single-stranded, andbiotinylated. This can be accomplished by single-strandsynthesis using a biotinylated, phosphorothioated primer(e.g., 338Fbs2) and extension with Taq. Since environ-mental DNA extracts often contain contaminants that in-hibit Taq, we estimate that about 5 �l of the DNA extractcan be used, resulting in �108 fragments in a typicalreaction volume of 100 �l from which �107 single-stranded fragments can be obtained after the exonucleaseprocedure. If a generous assumption is made that a targetorganism is 10% of the total population, 106 target frag-ments (�2 attomoles) are in the analyte. This is too low tobe detected using the current protocols for the dendrimerassay.

Several improvements could be made. An importantarea is the hybridization efficiency. In the current assay,there are 3,000 beads in the hybridization reaction. At thedetection limit of 30 attomoles (2 x 107 molecules), only�9,000 ss-amplicons could be captured for a hybridiza-tion efficiency of about 0.05%. To detect 106 target frag-ments, the hybridization efficiency needs to be increasedto 1%. If the number of beads could also be reduced by afactor of 10, direct detection is achievable. Other areas ofimprovement include better protocols for purification ofthe DNA extract; a few PCR cycles; or inclusion of biotin-ylated-dNTPs in the extension reaction.

For these studies, a number of surface-labeled beadstandards were developed to relate the flow cytometryfluorescence intensities to the number of labels (dendrim-ers or SA-RPE molecules) on the surface. These standardsenabled the determination of an optimal dendrimer struc-ture for fluorescence amplification, and the quantificationof the number of dendrimer-labeled DNA molecules onthe bead surface. From this, single molecule detection inmultiplexed DNA measurements could be evaluated alongwith the dynamic range of amplification.

The intensity of the 2n(x)bA bead standard was foundto be equivalent to 1.97 x 104 dendrimers per bead. Basedon close packing of 80 nm dendrimers on a bead surfacewith diameter 5.60 �, the number of dendrimers per beadis at most 2.02 x 104. This suggests that maximum bindingof the 80-nm dendrimers occurred and that calibration ofthe standard was correct. For the 180-nm dendrimers, thebinding efficiency was less than that of the 80 nm struc-

FIG. 4. Dependence of SNR on the amount of ss-target DNA without(curve 1) and with (curve 2) amplification. Experimental data (circles andstars). Linear regression (solid line). Critical t0 (dashed line).

143MULTIPLEXED DETECTION OF DNA USING DENDRIMERS

ture; the 4n(x)bA standard had 2.25 x 103 dendrimers perbead, lower than the maximum number of 4.12 x 103

based on close packing estimates. Because 4n(x)bA hasmore biotins than the 80-nm structure, this result suggestsmass-limited diffusion of the larger dendrimer.

Finally, 14 kinds of R and IR fluorescent beads weredeveloped in this study. The beads were designed to havea low background on the reporter channel of the Luminex100 flow cytometer with an additional filter and to pro-vide good DNA hybridization efficiency. In all tests, theDNA signal and formation of multimers for all R and IRbeads was comparable or superior to unstained beads.These beads may be used on any conventional flow cy-tometer. It should be noted that these beads do not fall onthe Luminex bead map, and must be gated following dataacquisition to determine the mean reporter signal for eachpopulation. While this can be done with commercial andpublic-domain postprocessing software packages, we de-veloped software to analyze the data more efficiently.

ACKNOWLEDGMENTSThe authors thank J. Kadushin for preparation of the

dendrimers, and P. Kutt for flow cytometry analysis soft-ware.

LITERATURE CITED1. Spiro A, Lowe M, Brown D. A bead-based method for multiplexed

identification and quantitation of DNA sequences using flow cytom-etry. Appl Environ Microbiol 2000;66:4258–4265.

2. Spiro A, Lowe M. Quantitation of DNA sequences in environmentalPCR products by a multiplexed, bead-based method. Appl EnvironMicrobiol 2002;68:1010–1013.

3. Dunbar SA, Jacobson JW. Application of the Luminex LabMAP inrapid screening for mutations in the cystic fibrosis transmembraneconductance regulator gene. Clin Chem 2000;46:1498–1500.

4. Iannone MA, Taylor JD, Chen J, Li M-S, Rivers P, Slentz-Kesler KA,Weiner MP. Multiplexed single nucleotide polymorphism genotypingby oligonucleotide ligation and flow cytometry. Cytometry 2000;39:131–140.

5. Taylor JD, Briley D, Nguyen Q, Long K, Iannone MA, Li M-S, Ye F,Afshari A, Lai E, Wagner M, Chen J, Weiner MP. Flow cytometricplatform for high-throughput single nucleotide polymorphism analy-sis. BioTechniques 2001;30:661–669.

6. Ye F, Li MS, Taylor JD, Nguyen Q, Colton HM, Casey WM, Wagner M,Weiner MP, Chen J. Fluorescent microsphere-based readout technol-ogy for multiplexed human single nucleotide polymorphism analysisand bacterial identification. Hum Mutat 2001;17:305–316.

7. Armstrong B, Stewart M, Mazumder A. Suspension arrays for high

throughput, multiplexed single nucleotide polymorphism genotyp-ing. Cytometry 2000;40:102–108.

8. Cai H, White PS, Torney D, Deshpande A, Wang Z, Marrone B, NolanJP. Flow cytometry-based minisequencing: a new platform for high-throughput single-nucleotide polymorphism scoring. Genomics2000;66:135–143.

9. Nolan JP, Sklar LA. Suspension array technology: evolution of theflat-array paradigm. Trends Biotechnol 2002;20:9–12.

10. Chandler DP, Stults JR, Anderson KK, Celuba S, Schuck BL, BrockmanFJ. Affinity capture and recovery of DNA at femtomolar concentra-tions with peptide nucleic acid probes. Anal Biochem 2000;283:241–249.

11. Keller RA, Ambrose WP, Goodwin PM, Jett JH, Martin JC, Wu M.Single-molecule fluorescence analysis in solution. Appl Spectrosc1996;50:12A-32A.

12. Enderlein J, Robbins DL, Ambrose WP, Goodwin PM, Keller RA. Thestatistics of single molecule detection: an overview. Bioimaging 1997;5:88–98.

13. Goodwin PM, Cai H, Jett JH, Ishaug-Riley SL, Machara NP, Semin DJ,Van Orden A, Keller RA. Application of single molecule detection toDNA sequencing. Nucleosides Nucleotides 1997;16:543–550.

14. Machara NP, Goodwin PM, Enderlein J, Semin DJ, Keller RA. Efficientdetection of single molecules eluting off an optically trapped micro-sphere. Bioimaging 1998;6:33–42

15. Enderlein J, Robbins DL, Ambrose WP, Keller RA. Molecular shotnoise, burst size distribution, and single-molecule detection in fluidflow: effect of multiple occupancy. J Phys Chem A 1998;102:6089–6094.

16. Ambrose WP, Goodwin PM, Nolan JP. Single-molecule detection withtotal internal reflection excitation: comparing signal-to-backgroundand total signals in different geometries. Cytometry 1999;36:224–231.

17. Osborn MA, Furey WS, Klenerman D, Balasubramanian S. Single-molecule analysis of DNA immobilized on microspheres. Anal Chem2000;72:3678–3681.

18. Steen HB. Noise, sensitivity, and resolution of flow cytometers. Cy-tometry 1992;13:822–830.

19. Shapiro HM. Practical flow cytometry. 3rd ed. New York: Wiley-Liss;1995.

20. Steinkamp JA, Yoshida TM, Martin JC. Flow cytometer for resolvingsignals from heterogeneous fluorescence emissions and quantifyinglifetime in fluorochrome-labeled cells/particles by phase-sensitive de-tection. Rev Sci Instrum 1993;64:3440–3450.

21. Nilsen TW, Grazel J, Prensky W. Dendritic nucleic acid structures. JTheor Biol 1997;187:273–284.

22. Stears RL, Getts RC, Gullans SR. A novel, sensitive detection systemfor high-density microarray using dendrimer technology. PhysiolGenom 2000;3:93–99.

23. Orentas RJ, Roskopf SJ, Casper JT, Getts RC, Nilsen TW. Detection ofEpstein-Barr virus EBER sequence in post-transplant patients withDNA dendrimers. J Virol Methods 1999;77:153–163.

24. Wang J, Jiang M, Nilsen TW, Getts R. Dendritic Nucleic Acid Probesfor DNA Biosensors. J Am Chem Soc 1998;120:8281–8282.

25. Emrich SJ, Lowe M, Delcher A. A Web-based software tool for select-ing optimal DNA oligos. Nucleic Acids Res 2003;31:3746–3750.

26. Rosner B. Fundamentals of biostatistics. 4th ed. Belmont, CA: Wads-worth; 1995.

144 LOWE ET AL.