JournalofMicroscopy,Vol.246,Pt32012,pp.229–236 doi:10.1111/j.1365-2818.2012.03604.x Received2August2011;accepted30January2012 Interferometer-based structured-illumination microscopy utilizing complementary phase relationship through constructive and destructive image detection by two cameras L. SHAO∗,‡, L. WINOTO∗, D.A. AGARD∗,†, M.G.L. GUSTAFSSON ‡,§ & J.W. SEDAT∗ ∗KeckAdvancedMicroscopyLaboratory,DepartmentofBiochemistryandBiophysics,Universityof California,SanFrancisco,California,U.S.A. †HowardHughesMedicalInstitute,ChevyChase,Maryland,U.S.A. ‡JaneliaFarmResearchCampus,HowardHughesMedicalInstitute,Ashburn,Virginia,U.S.A. Keywords. Imagereconstruction,interferometer,I5S,opticaltransfer function,structured-illumination,widefieldfluorescencemicroscopy. Summary dimensions.Itcombinesstructured-illuminationmicroscopy (SIM; Heintzmann & Cremer, 1999; Gustafsson et al., Inaninterferometer-basedfluorescencemicroscope,abeam 2000; Gustafsson, 2000; Gustafsson et al., 2008) with splitter is often used to combine two emission wavefronts an interferometer using two opposing objective lenses interferometrically.Therearetwoperpendicularpathsalong (Hell & Stelzer, 1992; Gustafsson et al., 1999). In SIM, which the interference fringes can propagate and normally the sample is illuminated by a periodic pattern, which only one is used for imaging. However, the other path mixes with the otherwise undetectable high-resolution also contains useful information. Here we introduced a information and, in spatial frequency space, moves it into secondcameratoourinterferometer-basedthree-dimensional thepassbandofthemicroscopeintheformofmoire´ fringes. structured-illumination microscope (I5S) to capture the The down translated high-resolution information is then fringesalongthenormallyunusedpath,whichareoutofphase computationallyrecoveredtoproduceanimagewithextended byπrelativetothefringesalongtheotherpath.Basedonthis resolution (Heintzmann & Cremer, 1999; Gustafsson et al., complementaryphaserelationshipandthewell-definedphase 2000; Gustafsson, 2000; Gustafsson et al., 2008). SIM can interrelationships among the I5S data components, we can double the conventional resolution in all three dimensions, deduceandthencomputationallyeliminatethepathlength resultinginaresolutionofabout100nmlaterallyand300nm errorswithintheinterferometerloopusingthesimultaneously axially(Gustafssonetal.,2008).I5SisanextensionofSIMin recorded fringes along the two imaging paths. This self- whichasecondobjectivelensisusedandplacedontheopposite correction capability can greatly relax the requirement for side of the sample, and the two apertures are combined eliminatingthepathlengthdifferencesbeforeandmaintaining interferometrically (Hell & Stelzer, 1992; Gustafsson et al., thatstatusduringeachimagingsession,whicharepractically 1999;Shaoetal.,2008).Asaresult,theaxialresolutionis challengingtasks.Experimentaldataisshowntosupportthe greatlyimprovedbothbecausethesetoflight-gatheringangles theory. of the microscope is enlarged and because the highest axial spatialfrequenciesintheilluminationpatternareincreased Introduction due to the interference of the illumination beams coming Wereportedearlieraresolution-enhancingschemeforthree- fromthetwoopposingobjectives.Anisotropic100-nm-scale dimensional (3D) widefield fluorescence microscope, I5S resolutionisachievedbythistechniqueinalldimensions(Shao (Shao et al., 2008), that greatly improves the resolution etal.,2008). of optical microscopy to nearly 100 nm in all three InI5S,acollimatedlaserilluminationbeamisdiffractedby atransmissionphasegratingandthecentralthreediffraction Correspondenceto:JohnW.Sedat,DepartmentofBiochemistryandBiophysics, orders are split by a beam splitter, resulting in six mutually UniversityofCalifornia,SanFrancisco,CA94158,U.S.A.Tel:+1-415-476-4156; coherent illumination beams being incident on the sample, fax:+1-415-514-4242;e-mail:[email protected] one triplet of beams passing through each objective lens §Deceased (Fig. 1). The illumination pattern that results from the (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety 230 L. SHAO ET AL. Conventional illumination M4 M3 Obj m Sample CCD 2 Obj s M2 M1 CCD 1 Dichroic m=-2 -1 0 1 2 Fig.2. PrinciplesofI5Sexplainedinfrequencyspace.(A)Thegeneralized Grating pupilfunction(GPF)ofanI5S/I5Mmicroscope,i.e.theportionsofthe wavefrontdetectablethroughthetwoobjectivelenses.Inaconventional widefieldmicroscope,onlyoneofthetwoshellsisdetectable.Thesupport Multi-mode fiber (i.e.regionofnonzerovalue)oftheintensitydetectionOTF,whichisthe Fig.1. TheschematicdrawingofanI5Smicroscope.Theillumination autocorrelationoftheGPF,isshownin(B).Incomparison,thedetection OTFofaconventionalmicroscopeisjustthemiddleregion.(C)Thesixdots lightpassesfirstthroughatransmissiongrating,whichdiffractsitinto representthesixcollimatedilluminationbeamsbeingincidentuponthe threebeams(greenlines),andthenthroughabeamsplitter,whichsplits sample.Theirinterferencecreatesa19-componentilluminationintensity eachbeamanddirectsthreebeamstoeachofthetwoopposingobjective pattern(D)organizedintofiveverticallinescorrespondingtom=−2to lenses. The same beam splitter combines the two beams of emission light(red)fromthesampleontothecamera.Normally,onlyonecamera 2.(E)TheeffectiveI5SOTFsupportisshowninky–kzcross-section.As areference,theconventionalOTFsupportisdrawnindashedcontour (CCD1)isusedtorecordtheemissionlightfromoneofthetwobeamsplitter in(E). exitports.HereweintroducedasecondimagingpathtorecordonCCD2 theemissionlightfromtheotherexitport.Themovableobjectivelenscan 3D acquisition, the Fourier transform D˜ of an observed 3D bepositionedinX,YandZwithrespecttothestationaryobjectivelens. imageD canbeexpressedasfollows(Gustafssonetal.,2008; MirrorsM3andM4canbetranslatedtogethertoadjustthepathlength Shaoetal.,2008): difference.Thegratingcanberotatedandlaterallytranslatedtocontrol theorientationandlateralphaseoftheilluminationpattern.MirrorsM3 (cid:2)2 ismadepartiallytransmissivetoletconventionalilluminationcomein D˜(k)= Om(k)S˜(kxy−mp)eimϕ, (2) fromtheresothatI5SdetectionOTFcanbemeasured. m=−2 where S˜(k)isthesampleinformation, O (Figs3A–C)isthe m interference of these six beams consists of 19 frequency mth-ordereffectiveopticaltransferfunction(OTF)obtainedby components (Fig. 2D) and can be expressed as (Shao et al., convolvingthedetectionOTFwiththeaxialFouriertransform 2008) of I . The axially extended support, or the nonzero valued m (cid:2)2 region, of O in I5S is attributed to both the high axial m I(rxy,z)= Im(z)ei(m2πp·rxy+mϕ), (1) illuminationfrequencies(Fig.2D)andadetectionOTFwith m=−2 twoaxially extendedsupportregions(orthesidebands;Fig. where r represents the lateral coordinates (x, y), p is the 2B)inadditiontotheusualOTFsupportregion(orthecentral xy lateral component of the wave vector of each beam not band) of the single-objective microscopy (Gustafsson et al., coincident with the optical axis k (or the side beams), I (z) 1999).Equation(2)showsthatI5Simprovestheresolution z m is the mth-order axial illumination pattern in real space laterally via frequency shift of the sample information and correspondingtothemthaxialrowofilluminationfrequencies axiallyviatheextendedsupportintheeffectiveOTFs. shownasdotsinFigure2Dandϕisaphaseangle(Gustafsson et al., 2008; Shao et al., 2008). Given the form of the Imagereconstructionandpathlengthdifference illuminationpatterninEq.(1)andtheconditionthattheaxial positionoftheilluminationpatternisfixedinrelationtothe TherawdataofSIM,includingI5S,isthesummationoffive focalplaneofthemicroscopeasthesampleisrefocusedduring components (Eq. (2)). These components can be separated (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 DUAL-CAMERA I5S MICROSCOPY 231 relationshipbetweenthefringesatthetwobeamsplitterexits (tobediscussednext),theactualpathlengthdifferencewithin the interferometer loop can be estimated from a pair of 3D imagessimultaneouslyacquiredthroughthetwoexits.With thisquantityknown,onecantheneitherchooseamatching set of OTFs from a library of OTFs or, in a more attractive solution, synthesize a set of OTFs that can account for the estimatedpathlengthdifference,andusethosetoreconstruct thedataset.Asaresult,therequirementforachievingaspecific pathlengthdifferenceforeachexperimentcanberelaxed. Pathlengthdifferenceestimation First, we discuss how the path length difference can be Fig.3. ExperimentallymeasuredandrotationallyaveragedI5SOTFsOm estimated from an I5S data set. Given a known path length ofEq.(2)form=0,±1and±2(A—C).(D)Demonstratesthatm=0 difference δ, the detection OTF of I5S can be divided into andm=±1componentspartlyoverlapwhentheyareplacedwherethey three regions based on different phases: b, 0 and −b, where belonginfrequencyspace. b = 2πδ/λem and λem is the emission wavelength (Fig. 4A). These values are because of the fact that the detection OTF given a series of images taken with five different lateral isanautocorrelationofthegeneralizedpupilfunction(GPF; pattern’s phases (ϕ in Eqs (1) and (2); Gustafsson et al., Fig. 2A) and the relative phase difference between the two 2008;Shaoetal.,2008).Becausethelateralpatterninour shellsofGPF(i.e.thetwowavefrontscollectedseparatelyby implementation is one-dimensional (1D), the same pattern the two objective lenses) is b. Likewise for the illumination needs to be applied multiple times with different angles (i.e. components(Fig.4B),threegroupscanbedistinguishedbased p in Eqs (1) and (2) with the same magnitude but different onthecomponents’phases:a,0and−a,wherea=2πδ/λ ex orientations)soastoisotropicallyextendthelateralresolution. andλ istheexcitationwavelength.BecausetheeffectiveI5S ex Afinalimagewith100nmresolutioninallthreedimensions OTFO istheconvolutionofthedetectionOTFwiththemth m (Fig. 2E) is reconstructed by positioning all the separated illumination order I˜ (Eq. (2) and Fig. 3) and phases of the m information components back to their original locations in operandsaddinconvolution,thephasesindifferentregions frequencyspaceandthencorrectingforamplitudedifferences ofI5SeffectiveOTFsarevariousadditivecombinationsof0, viaageneralizedWienerfilter(Gustafssonetal.,2008;Shao ±aand±b.Thecolour-codedmapsofphaseintheseOTFsare etal.,2008). showninFigures4C–E. For I5S reconstruction to work, it is required that the Becausethemth-ordercomponentoftheI5Srawdataisa raw image is acquired under the same conditions as the subregionofthesample’sFouriertransform—corresponding measuredeffectiveOTFs.InI5Soranyinterferometricset-up, tothesupportofO shiftedbymp(Eq.(2))—multipliedbyO , m m oneconditionpronetovariationindifferentimagingsessions thiscomponent’sphasemapistheadditionofthesample’sand isthepathlengthdifferenceoftheinterferometerloop.Before theOTF’sphases.Furthermore,componentsofdifferentorders everyexperiment,significanttimemustbespentondigitally partiallyoverlapwitheachotherinfrequencyspace(Fig.3D). adjustingthispathlengthdifferencetoreachzero,whichgives For example, the central line of the m = 1 component (the the maximum axial fringe contrast. Even if this preimaging dashed line in Fig. 4D) should be placed p distance (Eqs (1) adjustment is performed very accurately, instrumental drift and (2)) from the central line of the m = 0 component; i.e. willhavecauseddifferentpathlengthdifferences,i.e.different tothedashedlineinFigure4C.Ifweignoretheportionsof imagingconditions,intheactualdatasetandinthedataset theeffectiveOTFsderivedfromthesidebandsofthedetection takenfortheOTFdetermination,resultinginaxiallyringing OTFforthemomentandmovethem=1componenttoits artefactsinthereconstructionbecauseoftheunmatchedOTF originalpositioninfrequencyspace,asillustratedinFigure5, used.TomakeI5Seasiertouse,weneedaschemethatcan the green coloured part of the m = 1 component partially relaxthisverystringentrequirementonpathlengthdifference overlapswiththeredcolouredpartofthem=0component consistencywithoutinducingartefacts. (shadedareasinFig.5).Withinthisoverlap,therawimage’s AnotherdisadvantageofthecurrentI5Sset-upisthatonly m=0and1componentsatacertainfrequencykarerelated theimaginginterferencefringesatoneofthebeamsplitter’s by two exit ports are recorded, meaning half of the emission Oˆ (k−p)D˜ (k)= Oˆ (k)D˜ (k−p)eia, (3) lightisdiscarded.Capturingbothhalvesoftheemissionlight 1 0 0 1 simultaneously would significantly improve the signal-to- where Oˆ and Oˆ are OTFs that are acquired with zero 0 1 noiseratio.Furthermore,becauseofthecomplementaryphase path length difference and D˜ is the mth-order component m (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 232 L. SHAO ET AL. (a+b a b b-a 0 a b b a a b Fig.4. ThemapsofphasevaluesinI5SdetectionOTF(A),illuminationstructures(B)andtheeffectiveOTFsform=0(C),±1(D)and±2(E).Thereason whymostilluminationcomponentsarestretchedaxiallycomparedtoFigure2Dwasexplainedindetailinearlierpublications(Gustafssonetal.,2008; Shaoetal.,2008).Theregionscoloureddifferentlycorrespondtodifferentphasesequaltovariouslinearcombinationsofaandb,whereaandbarethe phasedifferencescausedbythepathlengthdifferenceintheinterferometerloopfortheexcitationandemissionwavelengths,respectively.Inreality,the dashedlinesin(C)and(D)coincide,i.e.theyrepresentthesamelateralfrequency. Figure 5 before the path length difference can be estimated usingEq.(3). Estimationbasedoncomplementaryphaserelation It can be proved (see Appendix) that in I5S, when the two emission wavefronts collected from the two objective lenses are coherently combined at the beam splitter, the phase of Fig.5. Whenthem=1component(Fig.4D)isplacedwhereitbelongsin the interference fringes at the primary exit port (i.e. the frequencyspace,itpartlyoverlapswiththem=0componentasindicated illuminationentryport,Fig.1)ofthebeamsplitterisexactlyπ bythe shadedareas. Shownhere areonlythe regionsbecauseof the differentthanthatattheotherexitport(thesecondaryexit). convolutionofcentralbandofthedetectionOTF(Fig.4A,green)and thecentralthreeandtwofrequenciesofthem=0and1illumination This results in the following relation between the detection structure,respectively(Figs2Dand4B). OTFsacquiredattheprimaryandthesecondaryexits:when thephasesare±binthesidebandsoftheformer,thephases are±(b+π)inthesidebandsofthelatter.ThenifI5Seffective O (k)S˜(k−mp)separatedfromEq.(2)(thezeropathlength OTFsareacquiredthroughthetwoexits,thesameπ phase m differencerequirementon Oˆ canbesatisfiedevenforOTFs differenceexistsbetweenthetwosetsofOTFsinthoseregions m acquiredwithnonzeropathlengthdifference,tobediscussed because of the convolution of the illumination components inOTFseparationandsynthesis).Linearregressioncanthenbe withthesidebandsofthedetectionOTF.Furthermore,asan appliedtosolveEq.(3)forausingallthefrequencieswithin image in frequency space is a multiplication of the object’s the overlap. This seems to be a straightforward method to Fourier transform by the OTF, this π phase difference in estimate the path length difference, except that it requires thosesideband-derivedregionsisalsotherebetweentwoI5S a clean overlap between only the two central band-derived datasetsacquiredsimultaneouslythroughthetwoexitsfor regions of the m = 0 and 1 components pointed to by the any sample. Consequently, the addition (or subtraction) of arrows in Figure 5. Unfortunately, superimposed on those thesetwodatasetsresultsinonlythecentral(orside)band- two central band-derived regions are other signals derived derived regions being nonzero in frequency space, provided from the sidebands of the detection OTF (Figs 3 and 4C–E), thatcorrectintensitynormalizationbetweenthetwoimages which make Eq. (3) invalid. Hence, we first need a solution hasbeenapplied.Iftheadditionimageisusedforextracting for obtaining the uncontaminated overlaps as depicted in the overlap shown in Figure 5, the overlap would not be (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 DUAL-CAMERA I5S MICROSCOPY 233 with each other to compensate the residual misalignment between the imaging paths. The misalignment, mainly becauseoflateralshift,rotationandmagnification,onlyneeds tobecalibratedonceeveryfewmonthsusingscatteredbeads samples.Theresultofthecalibration,atwo-dimensionalaffine transform matrix, was saved for aligning the image pairs in subsequent experiments. The intensity ratio between the imagepairafterbackgroundsubtractionwasalsocalibrated and saved for future use. After alignment and intensity correction, each raw image was separated into the five information components (Eq. (2)), each corresponding to a differentmofEq.(2)(Shaoetal.,2008),andthem=0and1 componentsoftheimagesacquiredbyonecamerawereadded totheircounterpartsacquiredbytheother,resultinginthe new m = 0 and 1 components with little sideband-derived contamination. The overlap between the red subregion of the m = 0 component and the green subregion of the m = 1 component (Fig. 5) was then extracted and used for the estimationofthepathlengthdifferencebasedonEq.(3). OTFseparationandsynthesis Given the estimated path length difference, we still need a schemeforimagereconstructiontotakeadvantageofit.Here weintroducesuchaschemebasedonthefactthatI5Seffective OTFsaretheconvolutionofthedetectionOTFOwiththe1D axialilluminationstructuresI˜ (Eq.(2)).Supposeameasured m I5S OTF can be decomposed into O and I˜ . Once the path m length difference is estimated from a raw image, then it is trivialtomodifythephasesofOandI˜ bymultiplyinge±ia to m theI˜ componentsofnonzerophases(Fig.4B)ande±ibtothe m Fig.6. TheeffectiveI5SOTFswereacquiredthroughtheprimaryand sidebandsof O(Fig.4A).Thephase-modulatedOand I˜m can secondarybeamsplitterexitandrotationallyaveraged,andtheiradditions then be convolved together to synthesize a set of OTFs that andsubtractions,form=0,±1and±2,areshownin(A)–(C)and(D)–(E), canaccountforthepathlengthdifferencespecifictotheraw respectively.TheregionsderivedfromthesidebandsofthedetectionOTF image. aremostlymissingintheadditions,althoughthosefromthecentralband ToseparateI5SOTFsintotwoconvolvingparts,oneofthe aremissinginthesubtractions,astheoreticallypredicted. partshastobeknown.Thereisawaytomeasurethedetection OTFandI5SeffectiveOTFsalmostsimultaneouslyaslongas contaminatedbythesuperimposedsideband-derivedsignals the mirror M3 in Figure 1 also transmits partially (20% is and therefore, the path length difference can be cleanly whatweused)sothataconventionalilluminationpathcanbe estimatedasdescribedinEq.(3). introducedthroughthatmirror.Withthisnewset-up(Fig.1), Toimplementtheaboveidea,asecondimagingpathwas the3DI5SeffectivePSFanddetectionPSFcanbeacquiredinan addedtoI5Stocapturetheinterferencefringesthroughthe interleavedmannerbyalternatingbetweentheconventional secondaryexitofthebeamsplitter(CCD2inFig.1).Apairof andthestructured-illuminationpathateachdefocusstep.We I5Spointspreadfunctions(PSF)wasacquiredsimultaneously canthenobtainrotationallyaveragedI5SeffectiveOTFsand bythetwoCCDsandtherotationallyaveragedeffectiveOTFs detectionOTFfromthesePSFs. were generated (Gustafsson et al., 2008; Shao et al., 2008). The OTF separation problem is formulated as solving Thesummations and subtractionsof thesetwosetsof OTFs for I˜ (k ) in O (k)= O(k)⊗I˜ (k ) given O (k) and O(k). m z m m z m areshowninFigure6,which,asexpected,containmostlythe Because I˜ (k ) is a 1D function, this problem is equivalent m z subregionsderivedonlyfromthecentralbandandsidebands to solving it in a series of 1D deconvolution problems ofthedetectionOTF,respectively. O (k ,k )= O(k ,k )⊗I˜ (k ) for each lateral frequency m z xy z xy m z To estimate the path length difference in each data set, a k . Linear convolution can be transformed into a circular xy pairofrawdatasetswasacquiredsimultaneouslybythetwo convolution, which can then be formulated as a vector (in cameras. The image pair was then computationally aligned this case the unknown I˜ (k )) being multiplied by a cyclic m z (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 234 L. SHAO ET AL. I5S illumination structure (Fourier space) described above can be applied to the I5S effective OTFs measured by the two cameras. The estimated path length order 0 differencecanthenberemovedfromOandI˜mand,therefore, 1 order 1 wecansynthesizetheI5SeffectiveOTFswithzeropathlength e 0.8 order 2 difference (Oˆm in Eq. (3)) by convolving the zero phase- plitud 0.6 mpaotdhulelantgetdhOdiaffnerdenI˜mc,ewinheiachchaIr5eSndeaetdaesdetto(Ecqo.r(r3e)c).tlFyoersetaimchatoef m A 0.4 thethreelateralpatternorientations,data-specificI5Seffective OTFs are synthesized based on the estimated path length 0.2 differenceasdescribedaboveandthenusedinreconstructing 0 thenormalI5Sdatasetacquiredthroughtheprimarybeam −10 −5 0 5 10 splitterexit. k (µm−1) z Fig. 7. I5S illumination structure axial profiles in frequency space, obtained by deconvolving measured I5S detection OTF from the I5S Results effectiveOTFs.Black,redandbluecurvescorrespondtoorderm=0, Figure 7 shows the results of separating I5S illumination ±1and±2components,respectively. structuresfromthemeasuredI5SeffectiveOTFsasdiscussed above.ThethreecurvesinFigure7aretheamplitudesofI5S matrix, if the vectors have a sufficiently short support or illuminationstructureversusk infrequencyspaceform=0, z are padded with enough zeros (Bertero & Boccacci, 1998, ±1and±2(Eqs(1)and(2)). Chapter2.7).Astackoflinearequations,oneforeachk ,can Two pairs of dual-camera I5S images of a single-bead xy thusbeformulatedintoanoverdeterminedsystemforsolving samplewereacquired.Nonzeropathlengthdifferenceswere I˜ (k ).Theaxialilluminationstructuresthussolvedandthe intentionallyintroducedwhenacquiringthesedatasets.The m z measureddetectionOTFarestillmodulatedbyanunknown reconstructionsusing astandard setofeffective OTFstaken phasebecauseoftheunknownpathlengthdifferenceduring with little path length difference present show the expected dataacquisition.Toestimatethispathlengthdifference,the ringing artefacts in the axial intensity profile through the same summation–subtraction and overlap extraction steps bead centre (Figs 8A and C). The different polarities of the A B 1 1 0.8 0.8 y 0.6 y 0.6 sit sit n n e 0.4 e 0.4 nt nt I I 0.2 0.2 0 0 −50 0 50 −50 0 50 Z Z C D 1 1 0.8 0.8 Fig. 8. Results showing the effects of path length y 0.6 y 0.6 difference estimation and correction. (A) and (C) sit sit n n The axial profiles of I5S reconstruction of a single nte 0.4 nte 0.4 bead,imagedwithintentionalpathlengthdifferences I I 0.2 0.2 estimatedtobe0.15and−0.9radians,respectively. (B)and(D)ResultsusingsynthesizedOTFsbasedon 0 0 theestimatedpathlengthdifferencescorresponding to(A)and(C),respectively.Therearesignificantlyless −50 0 50 −50 0 50 ringingartefactsin(B)and(D)thanin(A)and(C). Z Z (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 DUAL-CAMERA I5S MICROSCOPY 235 significantly enhance the real-world applicability of I5S to Secondary exit Is biology. The basic principle demonstrated in this paper lays thegroundworkforaschemethatcanachievesuchagoal.It I p doesnot,however,affectotherfactorslimitingtheapplication Primary exit of I5S to biology, including the requirement of refractive indexmatchingamongthemountingmedium,theimmersion medium and the sample itself, and the requirement that thetwoobjectivelensesarecloselymatchedintheiroptical properties. In our demonstration, the fluorescent bead was I2 (A2) I1 (A1) mounted in the same medium as the high refractive index Point (1.51)immersionmedium. source As mentioned earlier, the image pair recorded on two camerascanalsobeusedsimplyforimprovingthesignalto Fig.9. Proofofπphasedifferencebetweentheinterferencefringesatthe noise.Thiscanbeaccomplishedbyincludinginthegeneralized primaryandthesecondarybeamsplitterexits.SeeAppendixfordetails. Wiener filter (Gustafsson et al., 2008) the additional terms correspondingtotheimagesandOTFsacquiredbythesecond asymmetryinthoseplotsindicateeitherpositiveornegative cameraoncetheimagepairisaligned. path length difference in the loop. For each data set, we performed phase difference estimation (the results are 0.15 Acknowledgements and –0.9 radians for data set shown in Figs 8A and C, We dedicate this paper to the late Dr. Mats Gustafsson. respectively)anddataset-specificOTFsynthesisasdescribed This work was supported in part by the Keck Laboratory above,andthenreconstructedtheimagesacquiredthrough for Advanced Microscopy, the National Institutes of Health theprimarybeamsplitterexitusingthesynthesizedOTFs.The (GM25201toJ.W.S.andGM31627toD.A.A.),theHoward resultsarealsodisplayedasaxialintensityprofilesinFigures HughesMedicalInstitute(D.A.A.)andbytheNationalScience 8BandD,whichexhibitmuchlessringingthanFigures8A Foundation through the Center for Biophotonics, an NSF andC. Science and Technology Center that is managed by the UniversityofCalifornia,Davis,underCooperativeAgreement Discussion No.PHY0120999. Weintroducedamethodtoestimatethepathlengthdifference present in each I5S data set with the assistance of a References secondcamerarecordingfluorescenceemission’sinterference fringes through the secondary exit of the beam splitter. Bertero, M. & Boccacci, P. (1998) Introduction to Inverse Problems in The estimation method exploits the π phase difference Imaging.InstituteofPhysicsPublishing,Philadelphia,PA. relationshipbetweentheimaginginterferencefringesatthe Born, M. & Wolf, E. (1980) Principles of Optics: Electromagnetic Theory ofPropagation,InterferenceandDiffractionofLight,6thedn.Pergamon primary and secondary beam splitter exits and the well- Press,Oxford. defined Fourier space phase maps for all the information Gustafsson,M.G.(2000)Surpassingthelateralresolutionlimitbyafactor components constituting an I5S data set. Another way to oftwousingstructured-illuminationmicroscopy.J.Microsc.198,82– think about the π phase difference is to recall that when 87. theprimarybeamsplitterexithasconstructiveinterferences Gustafsson,M.G.,Agard,D.A.&Sedat,J.W.(1999)I5M:3Dwidefieldlight information,thenewsecondarybeamsplitterexitwillhave microscopywithbetterthan100nmaxialresolution.J.Microsc.195, destructive interferences information. With the estimated 10–16. phase difference, a data set specific set of OTFs can then Gustafsson,M.G.,Shao,L.,Carlton,P.M.etal.(2008)Three-dimensional be synthesized, because I5S effective OTFs are separable resolutiondoublinginwidefieldfluorescencemicroscopybystructured- intoadetectionOTFconvolvingwiththeaxialillumination illumination.Biophys.J.94,4957–4970. structures, and used for reconstructing the data set. We Gustafsson,M.G.L.,Agard,D.A.&Sedat,J.W.(2000)Doublingthelateral resolution of wide-field fluorescence microscopy using structured- have used a single bead imaging experiment to prove the illumination.Proc.SPIE.3919,141–150. principle. Reconstruction using a synthesized, as opposed Hecht,E.(1998)Optics,3rdedn.Addison-Wesley,Reading,MA. to a standard, set of OTFs successfully removes most of the Heintzmann, R. & Cremer, C. (1999) Laterally modulated excitation axialringingartefactsthatarebecauseofanunknownpath microscopy:improvementofresolutionbyusingadiffractiongrating. length difference. One major difficulty of I5S or any other Proc.SPIE.3568,185–196. interferometricmicroscopeisthestrictrequirementonpath Hell,S.&Stelzer,E.H.K.(1992)Propertiesofa4piconfocalfluorescence lengthdifferenceadjustment.Relaxingthisrequirementwill microscope.J.Opt.Soc.Am.A.9,2159–2166. (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236 236 L. SHAO ET AL. Shao,L.,Isaac,B.,Uzawa,S.,Agard,D.A.,Sedat,J.W.&Gustafsson,M.G. Thereflectedhalfofwavefront1interferesattheprimaryexit (2008)I5S:widefieldlightmicroscopywith100-nm-scaleresolutionin withthetransmittedhalfofwavefront2,whosephaseisϕ2+ threedimensions.Biophys.J.94,4971–4983. θ,andthustheinterferenceintensityis(Born&Wolf,1980, Chapter7.2)eitherI =I /2+I /2+(I I )1/2·cos(ϕ1+π – p 1 2 1 2 ϕ2 – θ), denoted Case 1a, or I = I /2 + I /2 + (I I )1/2· Appendix p 1 2 1 2 cos(ϕ1+θ –ϕ2),denotedasCase1b.Likewise,thereflected Hereweprovetheπphasedifferencebetweentheinterference half of wavefront 2 either has phase ϕ2 + π or ϕ2 + fringes at the primary and secondary exit port of the beam 2θ, depending on where the reflection occurs, and when it splitter. Two parts of a wavefront with intensities I1 and I2 interferes with the transmitted half of wavefront 1 at the combineatthebeamsplitteraftertraversingthetwoarmsof secondary exit, whose phase is ϕ1 + θ, the interference the interferometer. Each of them is split into halves by the intensityiseitherI =I /2+I /2+(I I )1/2·cos(ϕ1+θ–ϕ2 s 1 2 1 2 beam splitter, with one-half transmitting through and the –π),denotedCase2a,orI =I /2+I /2+(I I )1/2·cos(ϕ1– s 1 2 1 2 otherhalfreflectedbythebeamsplitter(Fig.9).Thereflected ϕ2–θ),denotedCase2b.Becauseofenergyconservation,i.e. halfofwavefront1eitherhasphaseϕ1+π,whereϕ1isthe thatI +I mustequalI +I ,itismandatorythatCase1aand 1 2 p s phasebeforethesplit,ifthereflectionoccursattheglass–glue 2borCase1band2aco-occursothatthecos(·)termscancel interface, or ϕ1 + 2θ if the reflection occurs at the glue– out. In either case, it means the phases of the interference glassinterfaceaccordingtotheStokesrelations(Hecht,1998, fringes at the primary and secondary exit port are different Chapter4.10),whereθisthephasedelaybecauseoftheglue. byπ. (cid:3)C 2012TheAuthors JournalofMicroscopy(cid:3)C 2012RoyalMicroscopicalSociety,246,229–236