TH

EJ

OU

RN

AL

OF

CE

LL

BIO

LO

GY

JCB: FEATURE

!

The Rockefeller University Press $8.00The Journal of Cell Biology, Vol. 172, No. 1, January 2, 2006 9–18http://www.jcb.org/cgi/doi/10.1083/jcb.200507103

JCB 9

Seeing is believing

?

A beginners’ guide to practical pitfalls in image acquisition

Imaging can be thought of as themost direct of experiments. You seesomething; you report what you see.If only things were truly this simple.Modern imaging technology hasbrought about a revolution in thekinds of questions we can approach,but this comes at the price of increas-ingly complex equipment. Moreover,in an attempt to market competingsystems, the microscopes have oftenbeen inappropriately described aseasy to use and suitable for near-beginners. Insufficient understandingof the experimental manipulationsand equipment set-up leads to theintroduction of errors during imageacquisition. In this feature, I reviewsome of the most common practicalpitfalls faced by researchers duringimage acquisition, and how they canaffect the interpretation of the experi-mental data.

This article is targeted neither to themicroscopy gurus who push forwardthe frontiers of imaging technologynor to my imaging specialist col-leagues who may wince at the overlysimplistic comments and lack of detail.Instead, this is for beginners whogulp with alarm when they hear the

word “confocal pinhole” or sigh asthey watch their cells fade and die infront of their very eyes time and timeagain at the microscope. Take heart,beginners, if microscopes were ac-tually so simple then many people(including myself) would suddenly beout of a job!

All data are subject to interpretation

Deliberate scientific fraud exists, but inmodern microscopy a far greater numberof errors are introduced in complete in-nocence. As an example of a commonproblem, take colocalization. Upstairs inthe lab, a researcher collects a predomi-nantly yellow merged image on a ba-sic microscope, naturally interpreted ascolocalization of green and red signals.But on the confocal microscope, thereis no yellow in the merged images.

short article cannot be an exhaustive“how to” guide, I have also assembled abibliography of a few highly recom-mended textbooks and microscopy websites, which readers should consult formore extensive treatments of the criticalissues introduced here.

Sample preparation

“Garbage in

!

garbage out” is the uni-versal motto of all microscopists. A wor-rying tendency today is to assume thatdeconvolution software or confocal mi-croscopes can somehow override thestructural damage or suboptimal immuno-labeling induced by poor sample prepa-ration. The importance of appropriatefixation, permeabilization, and labelingmethods for preserving cellular morphol-ogy or protein localization is well knownto electron microscopists (Hayat, 2000),but often underestimated in optical mi-croscopy (Fig. 1).

Many labs use one standardizedprotocol for labeling with all antibodies,irrespective of whether the targets aremembrane- or cytoskeleton-associated,nuclear or cytosolic. However, inappro-priate fixation can cause antigen redistri-bution and/or a reduction in antigenicity.It is therefore important to test each anti-body on samples fixed in a variety ofways, ranging from solvents such asmethanol to chemical cross-linkingagents such as paraformaldehyde andglutaraldehyde (for protocols see Bacal-lao et al., 1995; Allan, 1999), althoughglutaraldehyde fixation often reduces an-tigenicity and increases background au-tofluorescence. Consult textbooks fornotorious pitfalls: phalloidin labeling isincompatible with methanol fixation,while microtubules are inadequatelyfixed by formaldehyde. Moreover, cer-tain cell types, such as yeast cells, re-quire specialized fixation protocols(Hagan and Ayscough, 1999).

Permeabilization is also critical inachieving a good compromise betweenantigen accessibility and ultrastructuralintegrity. Specific detergents will producedifferent effects (for example, Saponintreatment produces smaller holes in

“When you employ the microscope, shake off all prejudice, nor harbour any favorite opinions; for, if you do, ‘tis not unlikely fancy will betray you into error, and make you see what you wish to see.”

Henry Baker,

chapter 15, "Cautions in

viewing objects" of The

Microscope Made Easy, 1742.

How can this be? Many factorscontribute. Here, I take the readerthrough the imaging process, from sam-ple preparation to selection of the im-aging and image-processing methods.Throughout, we will be on the look-outfor problems that can produce mislead-ing results, using colocalization as themost common example. Because one

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

http://www.jcb.org/cgi/content/full/jcb.200507103/DC1Supplemental Material can be found at:

JCB • VOLUME 172 • NUMBER 1 • 200610

membranes than Triton exposure), and itis also important to test the effects ofpre-, simultaneous, or post-fixation per-meabilization. Be aware that tissue pro-cessing, and particularly “air drying”steps, may introduce tissue distortionsthat will affect dimensions and measure-ments. Many sample preparation prob-

lems are of course avoided by imagingliving cells, though live cell work intro-duces a whole range of new potential ar-tifacts (see

Important considerations forlive cell imaging).

What type of mountant should you use?

Of the many types of homemade andcommercial mounting media, no oneproduct is ideal for all applications.Mounting media that harden (oftencontaining polyvinyl alcohol) are use-ful for long-term sample storage andare preferred for imaging using a wide-field (compound) microscope becausethe sample flattens as the mountanthardens. For that very reason, however,those that remain liquid (typicallyglycerol-based) are preferable whenthree-dimensional (3D) information isdesired. These require a sealant aroundthe coverslip for stability and to pre-vent desiccation.

Anti-fade agents are used to sup-press photobleaching, but an anti-fadethat is incompatible with specific fluo-rochromes can quench their signal sig-nificantly and/or increase backgroundfluorescence. Consult the mountant’smanufacturer for compatibility informa-tion because the anti-fade’s identitymay not be revealed in the datasheet.For GFP and its derivatives it is advis-able to avoid anti-fades altogether, un-less the sample is also labeled with afluorochrome prone to photobleaching.Reports differ as to whether nail var-nish, when used as a coverslip sealant,reduces GFP fluorescence, but usersshould be aware of the potential prob-lem. A nondetrimental alternative sealantis VALAP, a 1:1:1 mixture of Vaseline,lanolin, and paraffin.

Optical properties of the microscope that you need to know about

Few students and post-doctoral re-searchers will have the opportunity tochoose the microscope they will use, orto influence the selection of specificcomponents for purchase. However,there are certain factors that users cancontrol, and they should consider thesechoices when configuring the micro-scope for their own experiments.

All objectives are not equal...

The objective lens is the most criticalcomponent of a microscope and yet fewresearchers grasp the differences be-tween specific objective classes.

For example, most scientists can tellyou the magnification of an objectivelens, but few will know its numerical ap-erture (light-gathering ability). Yet it’s thenumerical aperture (NA) that determinesthe resolving power of the lens (Fig. 2),while magnification is only then useful toincrease the apparent size of the resolvedfeatures until they can be perceived by thehuman eye. Thus, a 40

"

1.3 NA objec-tive lens will be able to resolve far finerdetails than a 40

"

0.75 NA lens, despitetheir similar magnification. The intensityof the signal also increases steeply withincreasing NA (Fig. 3). Therefore, the ob-jective’s NA, as well as its magnification,should always be provided in the Materi-als and methods section of publications.

Why would anybody then choosean objective of lower NA? The answer isthat other features of the objective mayprove more critical for a particular sam-ple or application. For example, NA isproportional to the refractive index ofthe immersion medium, thus oil immer-Figure 1. Poor sample preparation. A z-stack of

optical sections, 18.2 #m in total thickness, wascaptured from a mouse brain tissue slice using aconfocal microscope (LSM 510; Carl Zeiss Mi-croImaging, Inc.) with a C-Apochromat 40"/1.2NA objective (1-#m optical slice thickness, 50z-sections collected at 0.33-#m intervals). Thetissue was labeled with three colors: blue (DAPI),marking the nuclei; green (GFP), marking thedendrites; and red (Cy3), marking the microtu-bule-associated protein MAP2 by indirect anti-body labeling. A and B show merged xy imagestaken from the top and middle of the stack, re-spectively; C shows a merged xz reconstructionof the stack (i.e., a “slice” through the stack, per-pendicular to the image plane). The GFP andDAPI labeling extend throughout the whole tissueslice, but the antibody labeling (red) is restrictedto a few sections at the top and bottom, althoughall of the dendrites should contain MAP2 label.Thus, incomplete penetration of antibodies canlead to false data interpretation.

The objective lens is the most critical component of a microscope and yet few researchers grasp the differences between

specific objective classes.

sion objectives can have a higher NAthan water immersion objectives, anddry objectives have the lowest NA. Butfor certain applications water immersionobjectives have distinct advantages overoil (see section “The problem of spheri-cal aberration”) and high NA also comesat the expense of reduced working dis-tance (how far the objective lens canfocus into your sample), which maybe problematic for thicker specimens.Other important factors to consider in-clude design for use with or withoutcoverslips, corrections for flatness of

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

PRACTICAL PITFALLS IN IMAGE ACQUISITION • NORTH 11

field and for chromatic aberrations, andtransmission of specific wavelengths(particularly UV or IR light) (for de-tailed explanations see Keller, 1995;Murphy, 2001).

It is important to consider how res-olution will affect colocalization analy-sis. We consider two fluorochromes tobe “colocalized” when their emittedlight is collected in the same voxels (3Dpixels). If the distance separating two la-beled objects is below the resolutionlimit of the imaging system, they will ap-pear to be colocalized. Thus, users may“see” colocalization using a low resolu-tion imaging system where a higher reso-lution system might achieve a visibleseparation of labels that are in closeproximity but are not actually colocal-ized (Fig. 4). The NA of the objectivelens, good refractive index match, andappropriate sampling intervals (smallpixel sizes) will all affect resolution, andconsequently, colocalization analysis.Note also that colocalization never indi-cates that two proteins are actually inter-acting, but only that they are locatedwithin close proximity.

Know your fluorochromes and filter sets

Colocalization can only be claimed inthe certain absence of “cross-talk” (or

“bleed-through”) between selected fluoro-chromes. Choosing fluorochromes withwell-separated excitation and emissionspectra is therefore critical for multiplelabeling. Consider the use of any twofluorochromes together. If their excita-tion peaks overlap, the wavelength ofexciting light selected for the first mayalso excite the second, and vice versa. Iftheir emission spectra also overlap, thefluorescence emitted by each may pass

through both the emission filter selectedfor the first channel and that selectedfor the second. Thus one fluorochromemay also be detected in the other’s de-tection channel, a phenomenon knownas cross-talk or bleed-through. Be par-ticularly suspicious of cross-talk if yourtwo fluorochromes appear to be 100%colocalized.

Certain fluorochromes, such asCyanine 3, are excellent for single label-ing but can be problematic for multiplelabeling because of spectral overlap withgreen emitters like fluorescein or AlexaFluor 488. Conversely, Alexa Fluor 594is well separated from standard greenemitters, but is shifted too close tothe far-red region to be useful formost green/red/far-red triple imaging(Rhodamine Red-X is better suited tothis). It pays to stock a range of second-ary antibody conjugates or dyes in orderto tailor the combination toward spe-cific protocols. Moreover, the brighterand more stable fluorochromes that arecontinually being developed may provevastly superior to the reagents your labhas used for the past 20 years!

It is equally important to considerwhich filter sets are available on yourmicroscope before selecting your fluo-rochromes. Long-pass filter sets, col-lecting all emissions past a certainwavelength, are generally less useful formultiple labeling than band-pass filters,

Figure 2. NA determines resolution. Differential interference contrast (DIC) images of two differentdiatoms (A/B and C/D), captured using a microscope (IX-70; Olympus) fitted with a U Plan-Apo100" objective lens with an adjustable NA (0.5–1.35). For A and C the NA was set to 1.35,whereas for B and D it was set to 0.5. In all cases the focus was adjusted to give the best possible image.The rows of separately resolved “spots” in A appear as blurred lines in B, and the fine details in thesecond diatom (C) cannot even be seen using the lower NA setting (D). Thus it is NA, not magnification,that determines the resolving power of the objective lens.

Figure 3. NA determines intensity. A triple-labeled MDCK epithelial cell monolayer was imaged byconfocal microscopy (LSM 510; Carl Zeiss MicroImaging, Inc.). Image A was acquired using a10"/0.25 NA objective (A-Plan), and B using a 10"/0.45 NA objective (C-Apochromat). Singleoptical sections were acquired using identical acquisition settings, and with pinholes of 1 Airy Unitfor the red channel and adjusted to maintain constant optical slice thickness in the other channels.Image A was acquired first using the minimum laser power and gain necessary to visualize signal ineach channel. The image acquired with the higher NA objective (B) was so much brighter that allchannels are highly saturated. This is because intensity is proportional to the fourth power of the NA.

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

JCB • VOLUME 172 • NUMBER 1 • 200612

which collect emissions in a specificrange (Fig. 5), and the narrower therange of the band-pass filter, the better itcan separate fluorochromes with closeemission spectra.

Single-labeled controls should al-ways be used to assess bleed-through.On confocal microscopes an additionaltest involves collecting images with eachlaser line deactivated in turn (you shouldnow see no emission in that laser line’scorresponding detection channel, unlessthere is cross-talk). Some cross-talkproblems can be overcome on confocalmicroscopes by the use of sequentialscanning (also known as multitracks orwavelength-switching line scans). In thismode, rather than exciting the samplewith multiple laser lines at once and col-lecting the emissions simultaneously,first one laser line is activated and itscorresponding emission collected, fol-lowed by the second laser line and itscorresponding emission. However, thiswill not solve the problem if there is sig-nificant overlap between both excitationand emission spectra.

Equally problematic is the overlapof specific fluorescence with backgroundautofluorescence, particularly in planttissues, in animal tissues rich in highlyautofluorescent proteins such as lipofuscinand collagen, and in cultures containinglarge numbers of dead or dying cells. Un-labeled samples are necessary to establishthe levels and locations of autofluores-cence, and narrow band-pass filters

maximize the collection of specific signalcompared with autofluorescence. Modernspectral imaging systems can be invalu-able for separating specific fluorescentsignal from autofluorescence, as well asfor separating fluorochromes with exten-sive spectral overlap (Fig. 5 C).

Three-dimensional (3D) microscopy

Standard compound or “wide-field”microscopes are often preferable for im-aging thin cell or tissue specimens asmore signal is presented to the detector.However, wide-field images can provideclear lateral (x- and y-axis) informationbut only limited axial (z-axis) informa-tion. For specimens thicker than a fewmicrometers, or when precise axial in-formation is required, an instrument thatremoves out-of-focus blur and permitsyou to distinguish between the signalsin thin “optical” slices may thereforeprove superior. Current technologies toachieve optical sectioning include con-focal microscopy, which uses one ormore pinhole apertures to prevent out-of-focus light from reaching the detec-tor; multi-photon microscopy, in whichexcitation only occurs in the plane offocus; and deconvolution algorithms,which are used to “restore” images fromany type of microscope to a closer ap-proximation of the original object. Eachtechnology has distinct advantages forspecific applications, which are bestunderstood from detailed comparisons

(Shaw, 1995; Murray, 2005) or by con-sulting local experts for advice. Thisarticle will concentrate largely on con-focal microscopy, which is the mostcommon approach. The following sec-tions will consider how to establish thecorrect optical conditions to acquiremeaningful 3D microscopy data.

The importance of pinhole size in confocal microscopy

The size of the confocal pinhole aperturedetermines the thickness of the opticalsection; that is, the thickness of sampleslice from which emitted light is collectedby the detectors. In most laser scanningconfocal systems the pinholes have an ad-justable diameter. Small pinhole diametersgive thinner optical sections and thereforebetter z-axis resolution, which is importantfor colocalization analysis. However,the signal intensity is decreased, sowhen z-axis information is not required,or photobleaching is a problem, a largerpinhole diameter may be preferred.

Stating either the pinhole diameteror the optical section thickness in publi-cations facilitates a more informed dis-cussion of 3D localization (includingcolocalization). Confocal images are gen-erally collected using a pinhole aperturesetting around 1 Airy Unit, a diameterthat achieves a good balance between re-jection of out-of-focus light and signalcollection. For multicolor imaging it iscritical to achieve the same optical sec-tion thickness in all channels, which isaccomplished by adjusting the pinholesize for the different wavelengths. Beaware that regular maintenance to ensurealignment of the pinholes is critical, as apoorly aligned pinhole can result in lat-eral shift and a “double” image wherethe same pattern is visualized in consec-utive z-sections.

The problem of chromatic aberration

The property of different wavelengths oflight being focused to different positionswithin your sample is known as chro-matic aberration. This can lead to anapparent lack of colocalization in theimage stack of fluorochromes that arecolocalized in the actual sample. Thusall microscopists need to be aware ofthis phenomenon.

Figure 4. High resolution images are necessary for colocalization studies. MDCK cells were labeledfor tubulin (green; Alexa Fluor 488) and cytokeratins (red; Rhodamine Red-X), and imaged by confocalmicroscopy (LSM 510; Carl Zeiss MicroImaging, Inc.) using an A-Plan 10"/0.25 NA objective (A) oran $-Plan-Fluar 100"/1.45 NA objective (B). Merged images show single optical sections, collectedwith the pinhole set to 1 Airy Unit for the red channel and adjusted to give the same optical slice thick-ness in the green channel. Using the low resolution and magnification objective (A) the merged imageappears largely “yellow,” suggesting colocalization of the two components. However, the higher reso-lution image (B) reveals that they are actually localized in extensive but nonoverlapping cytoplasmicnetworks. Thus, low resolution images can give false indications of colocalization.

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

PRACTICAL PITFALLS IN IMAGE ACQUISITION • NORTH 13

Lateral (xy-axis) chromatic aberra-tions are generally corrected within themicroscope, but note that full compen-sation is only achieved with propermatching of optical components. Somemanufacturers use only the tube lens toimpart corrections, whereas others usethe objectives; thus, combining objec-tives and microscopes from differentmanufacturers can introduce aberra-tions. Lateral chromatic shifts can alsobe caused by mechanical shifts betweendifferent filter cubes or dichroic mirrors.

Corrections for axial (z-axis)chromatic aberrations are more diffi-cult. Objective lenses are corrected forchromatic aberrations across a certainwavelength range, the extent of whichdepends on the type and age of the ob-jective (improved lenses are developedevery year). Most users are unawarethat the majority of objective lensescurrently in use are fully correctedacross only the (approximately) green-to-red range of emission wavelengths.Thus, two fluorochromes outside theseranges (such as DAPI and Cy5) couldbe focused to z-positions several hun-dreds of nanometers apart, even if theirtargets are colocalized in the actual

specimen. When compounded by fur-ther aberrations such as spherical aber-ration, they could appear well over amicrometer apart in the z-axis, and thusin different z-slices in your image stack(Fig. 6).

How do we check for chromaticaberration? One option is to image thetiny (e.g., 0.1-

#

m diameter) multicolor“Tetraspeck” beads available from Mo-lecular Probes and see whether the dif-ferent colors of each bead show up inthe same z-position or not. Anothermethod is to use two secondary anti-bodies, both directed against the sameprimary antibody but conjugated to dif-ferent fluorochromes (those used inyour double-labeling experiment), andsee whether the signals are superim-posed in the z-axis or whether one al-ways appears below the other. Rulingout severe chromatic aberrations in yourmicroscope set-up by these methodspermits you to be more confident ofyour data interpretation. When aberra-tions are found, try using fluorochromescloser together in emission wavelength.Alternatively, certain software pro-grams will permit you to “shift” the im-age in one channel relative to the other

(applying the exact shift calculatedfrom multicolor bead images).

The problem of spherical aberration

Spherical aberration describes the phe-nomenon whereby light rays passingthrough the lens at different distancesfrom its center are focused to differentpositions in the z-axis. It is the majorcause of the loss in signal intensity andresolution with increasing focus depththrough thick specimens.

Spherical aberrations occur aslight rays pass through regions of differ-ent refractive index (for example, fromthe coverslip to the tissue, or betweenregions of different refractive indexwithin the sample itself). The effects in-clude a reduction in intensity and sig-nal-to-noise in the plane of interest anddistortions in the 3D image, with finefeatures appearing smeared out alongthe z-axis (Fig. 7). The aberrations be-come worse as you focus deeper intothe sample. Corrections for sphericalaberrations within the objective lens it-self are only effective when certain pre-conditions are met. Thus, aberrationsare increased by factors such as the useof the wrong coverslip thickness or typeof immersion oil, too thick a layer ofmounting medium, the presence of airbubbles in the immersion or mountingmedium, or simply a temperaturechange. Note that most objective lensesfor high resolution fluorescence workare calibrated for use with 0.17-mmthick glass, to which no. 1 1/2 coverslipscorrespond most closely. The specimenshould be mounted on or as close to thecoverslip as possible (avoid multiwellslides with a nonremovable gasket thatplaces the coverslip many micrometersfrom the cells). The coverslip must alsobe mounted flat, as an angled coverslipwill result in distorted optical properties(remove excess mounting medium bybriefly placing small pieces of torn filterpaper against the edge of the coverslipafter mounting).

In selecting an objective lens forimaging thicker samples, you need toconsider the balance between the effectsof spherical aberrations and NA on yoursignal intensity. A high NA oil immer-sion lens may be optimal for use with

Figure 5. Emission filters must be selected to match your fluorochrome combination. Question: howmany different fluorochromes are present in this sample? Labeled MDCK cells were imaged using aconfocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) fitted with a Plan-Apochromat 63"/1.4 NA objective lens. All three images were acquired using 543-nm excitation and an HFP 488/543 main beamsplitter. Image A was acquired using a 560-nm long-pass filter. Actin filaments, nuclei,and cell–cell junctions all appear red. Image B was acquired using a sequential scan (multitrack), us-ing a 560–615 band-pass emission filter for the red channel and a 650-nm long-pass filter for theblue channel. It is now evident that the nuclear stain is far red, while the cytoplasmic labeling is red.Image C was acquired using a spectral imaging device (here the META detector; Carl Zeiss Micro-Imaging, Inc.), collecting the emission as a series of !11-nm spectral bands across a total range of552–723 nm. Unmixing software separated the total emission into three separate components, herepseudocolored green, red, and blue to aid visualization. The answer? Three: Tetramethyl Rhodamine(TRITC; labeling actin), Rhodamine Red-X (labeling desmosomes), and To-Pro3 (labeling nuclei).Conclusion: band-pass filters are generally preferable to long-pass filters for multiple labeling, andare often sufficient to distinguish between fluorochromes with well-separated spectra. Fluorochromeswith highly overlapping spectra (such as TRITC and Rhodamine Red-X, or GFP and YFP) requirespectral imaging systems for clean separation.

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

JCB • VOLUME 172 • NUMBER 1 • 200614

thin specimens, because the glass cover-slip, whose refractive index is matchedto that of the immersion oil, becomesthe predominant sample component.However, a water immersion lens, de-spite its lower NA, may achieve betterimages from a thick, largely aqueousspecimen due to the better match of re-fractive index between immersion me-dium and sample. Methods of minimizingspherical aberrations range from thedevelopment of objectives with adjust-able correction collars to the use of im-mersion oils with differing refractiveindices (Fig. 7; and for a detailed andhighly readable explanation of opticalaberrations and their practical correc-tion see Davis, 1999).

Establishing your acquisition settings

Appropriate acquisition settings are criti-cal for obtaining meaningful and quanti-fiable data as well as “pretty” images.You must distinguish between acquiringall information in the raw data, and laterpresenting the data in a way that conveysthe result more clearly.

All settings should be estab-lished using the real sample (or a posi-tive control), and then the negativecontrol is imaged using identical set-tings. The “autoexpose” or “find” func-tions should never be used for negativecontrols, as the camera or detectorswill attempt to compensate for the lowsignal.

Figure 6. The effect of chromatic aberrations on images. A–D show merged images of 0.1-#m multi-color Tetraspeck beads (Molecular Probes), acquired through the blue and green emission channels.Image A shows an xy-plane maximum projection of a deconvolved stack of images, acquired at50-nm z-intervals using a DeltaVision image restoration system (IX70 microscope; Olympus) and aU Plan-Apo 100"/1.35 NA objective. The lateral shift between blue and green signals is minimal.However, when rotated to view from the z-axis (B), the green and blue images appear separated by!600 nm, due to chromatic aberrations. After applying a 600-nm corrective z-axis shift to the blueimage the two colors now appear superimposed (C and D). E–H show images, collected and pre-sented under identical conditions, of some nuclear structures (marked by Alexa Fluor 488; green) inthe Schizosaccharomyces pombe nucleus (labeled with DAPI; blue). Most of the green spots are super-imposed over the nucleus in the xy-plane (E), but few in the z-axis (F). After applying the chromaticcorrection calculated for the beads (G and H), the majority of spots are seen to be nuclear. Thus,chromatic aberrations can lead to incorrect conclusions by uninformed users. Note also that this is agood-case scenario, using a top resolution system, one of the most highly corrected objective lensesavailable five years ago, and minimal spherical aberrations. Also, a far greater shift would be seenusing fluorochromes spectrally further apart, such as DAPI and Cy5. Happily, the objective manufac-turers have recently introduced a new range of objectives with excellent chromatic corrections acrossthe whole spectrum.

Keep the acquisition settings constant between specimens to be compared quantitatively and particularly between

sample and control.

First, keep the acquisition settingsconstant between specimens to be com-pared quantitatively and particularlybetween sample and control.

Second, it is important to distin-guish between an image that is useful forvisualization alone, and an image fromwhich meaningful quantitative data canbe extracted. For quantitative micros-copy the exposure time and/or gain(brightness) and offset (by which pixelsbelow a certain threshold are defined asbeing black) should be adjusted to usethe entire dynamic range of the detectors.Too high a gain results in saturated pix-els, which cannot be quantified becausethe dataset is clipped at the maximumend of the dynamic range. Conversely,an inappropriately large offset, oftenused to hide “background” cellular fluo-rescence, clips the data at the minimumend of the dynamic range and again pro-hibits quantitative measurements. Moresignificantly, how do you distinguishnonspecific background signal from alow, ubiquitous level of your proteinwith real biological significance?

Most researchers lean toward abright, high contrast image, and thus willinvariably saturate their images. To avoidthis, use acquisition software featuressuch as an “autohistogram” display, or a

“range indicator” or “glow scale” look-uptable, to establish the settings more objec-tively. Once the correctly acquired data issaved (and always stored in the raw for-mat for future reference!), brightness andcontrast or scaling adjustments can thenbe applied for a more visually pleasingpresentation.

How do you avoid saturation withsamples containing some particularlybright but other very weak regions? Thecollection of more grayscale levels (12-bitinstead of 8-bit data) will help. You can

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

PRACTICAL PITFALLS IN IMAGE ACQUISITION • NORTH 15

then present two pictures showing differ-ent scalings applied to the same image,adjusted for the bright features in one andthe finer, less intense details in the second.In more severe cases, image each areausing two different acquisition settings,then present the two images side by side.

Great care must be taken to ensureadequate sampling (pixel/voxel dimen-sions) of your data, in all axes. Accord-ing to the Nyquist sampling theorem,your spatial sampling intervals must bemore than two times smaller than thesmallest resolvable feature in your speci-men. If this sampling requirement is notmet, you will have gaps in your data andalso spurious features can be introducedinto your image by a process called“aliasing” (for explanation see Webb and

Dorey, 1995). Thus, most confocal mi-croscope software packages suggest theuse of a z-interval around half the opticalslice thickness (which is usually calcu-lated for you). This is sufficient for de-tecting all resolvable features, althoughthe use of even smaller z-intervals is ad-vantageous for deconvolution or forcreating smoother volume renderings.

In the xy-plane you should aim torelate the pixel size to the resolution ofthe system by adjusting the optical zoomsetting (if available) or by using binning(a process by which the signals fromneighboring pixels are combined into onevalue). The use of smaller pixel spacingthan this, known as “oversampling,”results in longer acquisition times thatcan cause greater photodamage to your

Figure 7. The effect of spherical aberrations on images. Mitotic spindles of HeLa cells were stained formicrotubules (green), centrosomes (red), and chromosomes (blue). Z-stacks of images were acquiredusing a U Plan-Apo 100"/1.35 NA objective on a DeltaVision microscope. Deconvolution and volumerendering were performed using SoftWoRx software. A–C show projections viewed from the xy axis(the view seen down the microscope) and D–F viewed from the z axis (rotated around the y-axis by90%). Here I used oils of different refractive index (r.i.) to demonstrate the spherical aberrations causedby r.i. mismatch between lens immersion medium and sample. Using the correctly matched oil for thissample (B and E; r.i. 1.518), the spindle fibers and centrosomes are cleanly resolved even in the z-axisview (E). Using mismatched oils (r.i. 1.504 or 1.530), some effects on the xy images can be seen(A and C; note the more blurred spindle fibers) but the z-axis views are more markedly compromised(D and F), with elongated, blurred structures and streaks fanning out toward the edges.

samples. The image frame resolution(e.g., 1024

"

1024) should be set highenough to submit images for publicationat a suitable size while maintaining 300dpi resolution (Rossner and Yamada,2004). Note that the use of an optical(real) zoom during image acquisition, tomagnify specific features, will avoid the“pixilated” appearance of low magnifi-cation images to which digital zoom hassubsequently been applied. Beyonda certain optical zoom, however, theuser will enter “empty magnification,”where that objective’s maximum reso-lution has been reached and so no addi-tional information is being obtained.

Choosing the “right” cell to image and publish

One of the greatest microscopy chal-lenges is the choice of which cell(s) topresent as a “typical” image. You mayhave preconceived ideas concerningyour protein’s localization, and subcon-sciously scan the sample to find the cellmost closely fitting your expectations. Insome cases this is a valid approach—forexample to search for microinjectedcells, for the sole expressers of a gene orfor cells at a particular stage of the cellcycle. But there is a strong risk of focus-ing in on one cell and ignoring 10,000strikingly different ones around it.

The use of an optical (real) zoom during image acquisition, to magnify specific features, will avoid the “pixilated” appearance of low magnification images to which digital zoom has

subsequently been applied.

The more passionate we are aboutour experiment, the more we mustdoubt our ability to be truly objective.So ask an unbiased colleague to blindlabel the samples or to help collect orevaluate the data. The use of a motorizedstage to image multiple, random posi-tions can also help avoid bias. Samples

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

JCB • VOLUME 172 • NUMBER 1 • 200616

containing cells with varying expres-sion patterns or morphology should bepresented as a low magnification over-view beside high magnification viewsof representative, contrasting regions.Most importantly, a statistical analysisof cell numbers exhibiting particularcharacteristics will strengthen your datainterpretation.

Transient transfections are particu-larly problematic for localization stud-ies. A common mistake is to seek outtransfected cells displaying the strongestexpression levels, but here the high con-centration of expressed protein may in-terfere with the balance of other proteinsor cellular processes. Weak expressersare generally a better choice, in particu-lar those showing limited signal localiza-tion. When available, antibodies to theendogenous protein can be used to assayfor a normal distribution pattern. Aber-rant localization may also be indicatedby the abnormal distribution of a partnerprotein that should colocalize with thetagged component.

Presenting and interpreting your images

You must decide how to present yourdata in the most appropriate form. With3D or 4D data this typically involves achoice between a single z-slice or a pro-jection of multiple slices. A single slicemust be presented when colocalizationand/or z-resolution are in question, but aprojection may better illustrate the conti-nuity of a 3D network.

A merged image is often inade-quate for demonstrating colocaliza-tion. A green-emitting fluorochromeand a red-emitting fluorochrome couldbe completely colocalized, but if one isbrighter than the other the merged im-age may not appear yellow. Colocaliza-tion is better demonstrated using the“line profile” function included inmany software packages, where an in-tensity plot for each channel is createdalong a line drawn across the image.Algorithms are available for calculat-ing the degree of colocalization, buttake care when establishing parame-ters such as threshold levels (for practi-cal tips and caveats for colocalizationstudies see Smallcombe, 2001; for de-tailed methods of quantifying colocal-

ization see Manders et al., 1993; Costeset al., 2004).

All images should be presentedwith scale bars. Many software pack-ages include automatic scale bar calcu-lation and pasting onto exported images.You can also image a stage microme-ter to calculate the total magnificationof a given system, which will be theproduct of the magnification of the ob-jective and of other components suchas the tube lens and relay optics to thecamera.

Quantification of images—why is it useful and when is it appropriate?

Quantifying image data is necessary forthe transition from anecdotal observationto an actual measurement. Quantitationis also an important means of avoidingsubjective bias and presenting the overallpattern of the data. It is rarely straight-forward in practice, requiring stringentacquisition conditions.

Images to be quantified should beacquired and exported in 12-bit or highergrayscale format, rather than the stan-dard 8-bit (or 24-bit color) format suit-able for most image presentation. Imageprocessing can then be used, beforequantification, to correct aberrations thathave been introduced into the imagestack during acquisition. You need to beaware of which image processing ma-nipulations are consistent with quantifi-cation, and which are not. Constrainediterative 3D deconvolution algorithms,for instance, maintain the total signal in-tensity within an image stack, whereasnearest neighbor algorithms are subtrac-tive and therefore do not.

Relative quantitation, such as com-paring the signal intensity between one re-gion of interest and another, or betweenthe sample and a control (assuming con-stant acquisition settings), is simpler thanabsolute quantitation, but even this as-sumes a number of prerequisites such aseven illumination across the entire field.Calibration slides (made from coloredplastic and available from companies suchas Applied Precision, Chroma TechnologyCorp., and Molecular Probes) can beimaged to determine irregularities in il-lumination and apply corrections. Whencalculating changes in signal intensity

over time you must compensate for gen-eral photobleaching as well as for temporalfluctuations in laser power or lamp illumi-nation. Laser power can be particularlyvolatile immediately after switching onthe system, so a warm-up period of 30–60minutes is recommended. A monitor di-ode or photosensor (if available on thesystem) and/or standardized samples areuseful for normalizing experiments forfluctuations in excitation intensity.

Absolute quantification presentsa greater challenge, requiring the re-searcher to have a good understanding ofboth the spectral and physical propertiesof the specific fluorochrome/fluorescentmolecule and the appropriate choice ofmicroscope optics and settings (Pawley,2000). Important properties of the fluo-rochrome that must be taken into consid-eration include the extinction coefficient,the quantum yield, the photobleachingrate and properties, the chromophorefolding kinetics, and the pH sensitivity(this can substantially affect measure-ments of proteins moving into and out ofsubcellular compartments).

The most critical components of thefluorescence microscope to consider forquantitative imaging are the objectivelens (including its NA and its spectraltransmission properties), the emission fil-ter, and the detector (Piston et al., 1999).An emission filter that is well matched tothe spectrum of your fluorescent probewill result in a better signal-to-noise ratio.A narrow band-pass filter is usually pref-erable to maximize collection of specificsignal while minimizing the contributionof autofluorescence. Linear detectors(including the majority of cooled CCDcameras and photomultiplier tubes) willfacilitate quantitation better than non-linear ones (such as intensified CCDcameras). Standardized samples of knownfluorochrome concentration can be usedto establish appropriate gain and offsetsettings for the detectors. Saturation ofthe fluorophore, which occurs particu-larly when using laser excitation, alsointroduces nonlinearity into the measure-ments, making calibration of the systemvery difficult. Thus, it is recommended touse the lowest laser power that gives asufficient signal-to-noise ratio.

Wide-field microscopy is often in-appropriate for quantitation because you

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

PRACTICAL PITFALLS IN IMAGE ACQUISITION • NORTH 17

collect emitted light from the whole sam-ple depth without knowing the thicknessof each cell or structure. The applicationof 3D deconvolution algorithms to animage stack can overcome this problemfor thin samples, but not for thick orhighly fluorescent samples. Confocal mi-croscopy is generally more quantifica-tion-friendly for samples over 15–20

#

min depth because of the defined opticalsection thickness. However, deeper focalplanes will show reduced signal intensitydue to absorption and scatter, necessitatingfurther, more complex corrections.

Four-letter methods: FRAP and FRET

So far this article has concentrated onbasic image acquisition. This next sec-tion will highlight a few danger areasassociated with some more complextechniques used to monitor the kineticsof protein trafficking or protein–proteininteractions in living cells.

The most common technique formonitoring protein kinetics is fluores-cence recovery after photobleaching(FRAP). Many confocal and deconvolu-tion microscope systems have incorpo-rated remarkably user-friendly FRAProutines into their acquisition software.Unfortunately, interpreting the data isnot always as simple. It is essential to“normalize” for general photobleachingby monitoring control cells that were nottargeted. Furthermore, be aware that ex-citation light bright enough to bleachfluorescent molecules in a short timeperiod can severely disrupt cellular ultra-structure. For quantitative FRAP, youmust decide in advance which of the nu-merous available models will be used foranalyzing the recovery curves, as thischoice may affect the experimental design(Rabut and Ellenberg, 2005).

Fluorescence (or Forster) resonanceenergy transfer (FRET) describes thenonradiative transfer of photon energyfrom a donor fluorophore to an acceptorfluorophore when they are less than 10 nmapart. FRET thus reveals the relativeproximity of fluorophores far beyond thenormal resolution limit of a light micro-scope. However, since FRET also relieson additional prerequisites, such as a cer-tain relative orientation of donor and ac-ceptor (Herman et al., 2001), an absence

of FRET cannot always be interpreted asthe fluorophores being more than 10 nmapart. Positive signals can also be mis-leading, as FRET may occur where anytwo noninteracting proteins are highlyconcentrated in localized areas. Twocommon methods for measuring FRETinclude sensitized emission and acceptorphotobleaching, and the most appropriatefor your experiment will depend on fac-tors such as the need for time-lapse im-aging, the ability to bleach regions ofinterest, and laser line availability. Insensitized emission studies, rigorous pos-itive and negative controls are required tocorrect for factors such as cross-talk anddirect excitation of the acceptor by thedonor’s excitation.

Important considerations for live cell imaging

When working with live cells, the rulesfor optimal imaging of fixed cells drop inpriority. Phototoxicity and photobleach-ing become your biggest enemies and ef-forts focus on keeping the cells alive andbehaving in a “normal” physiologicalmanner (see the invaluable “Live CellImaging—A Laboratory Manual”, R.D.Goldman and D.L. Spector [eds.], for de-tailed coverage). This requires appropriateenvironmental conditions (temperature,media, CO

2

, and possibly perfusion) andalso optical considerations (such as thereduced phototoxic effects of longer exci-tation wavelengths). A major issue dur-ing time courses is the focal drift causedby thermal fluctuations in the room.Environmental chambers that enclose theentire microscope are generally morethermo-stable than smaller imagingchambers, but the latter can be more con-venient for perfusion or for experimentsrequiring rapid temperature shifts.

Another serious challenge lies in ac-quiring images fast enough to capturerapid biological events and to accuratelyportray dynamic structures. This is partic-ularly tricky when using multiple probes,as the labeled target may move in the timerequired to switch between filter posi-tions. Solutions to this include the use of asimultaneous scan mode on a confocalmicroscope (in the absence of cross-talk),or positioning an emission splitter in frontof a CCD camera to collect both signalssimultaneously on the camera chip.

With fixed cells you typicallymaximize your signal-to-noise ratio vialonger exposure times in wide-field sys-tems, or in confocal microscopes by us-ing higher laser power or by increasingthe time the laser dwells on each pixel(using averaging or slower scan speeds).Optical zoom and Nyquist samplingare applied for optimal image quality.Approaches to minimize photobleachingin live cell imaging include a reductionin exposure time or laser power and pixeldwell time (you can compensate by us-ing higher gain), increased pinhole diam-eters, the use of lower magnifications,and sub-optimal spatial sampling in xy-and z-axes. Binning your signal and theuse of faster camera readout rates willenable more rapid imaging. The conse-quence of these compromises may in-clude poorer resolution and reducedsignal-to-noise ratios.

As the imaging proceeds, how doyou avoid misleading data by checkingfor normal physiological behavior of yourcells? Here are a few clues: (1) Have theymaintained their typical morphology, orare they shrunken, blebbing, rounded up,or coming off the coverslip? (2) Are cellsand organelles still moving around at acustomary rate? Acquiring simultaneousor sequential transmitted light and fluo-rescent images is an excellent way of as-sessing this; (3) If the sample is returnedto the incubator after the experiment,will the cells carry on dividing or the em-bryos survive? Continuing cell divisionis perhaps the most critical indication ofhealthy cells.

In conclusion

Given the complexities discussed above,how can we all share the responsibilityfor ensuring that published imaging datais an accurate representation of the truth?Researchers need to learn enough aboutspecimen preparation and the availableimaging equipment to establish appropri-ate settings and collect optimal images.The wealth of modern information re-sources (see online supplemental mate-rial, available at http://www.jcb.org/cgi/content/full/jcb.200507103/DC1) enablesall users to grasp at least basic micros-copy concepts. Central imaging facilitiescan provide more advanced informationrequired for specific applications and can

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from

JCB • VOLUME 172 • NUMBER 1 • 200618

help to ensure the use of appropriate im-aging systems. The researcher’s supervisorshould rigorously critique both the rawand processed data, and needs to appreci-ate that high quality microscopy requiresa significant time investment. Manufac-turers must strive to design hardware thatprovides constant imaging conditions,and software that includes user-friendlytools for image analysis, and to ensurethat researchers purchase the most appro-priate equipment for their needs. Finally,scientific journals should set stringentguidelines, and manuscript reviewersmust be critical of imaging data presen-tation, to guarantee that publicationscontain sufficient experimental detail topermit the readers to properly interpretthe images and to repeat the experiments.Only by such a collective effort can westrive to present the true picture.

colleagues and vendors, and in particular to VicSmall, who has the most exacting eye for detail ofanyone I know, yet who always encouraged meto view even my ugliest experimental samples as“not

entirely

uninteresting.”

References

Allan, V.J. 1999. Basic immunofluorescence.

In

Protein Localization by Fluorescence Mi-croscopy—A Practical Approach. V.J. Al-lan, editor. Oxford University Press, Oxford,UK. 1–26.

Bacallao, R., K. Kiai, and L. Jesaitis. 1995. Guid-ing principles of specimen preservation forconfocal fluorescence microscopy.

In

Hand-book of Biological Confocal Microscopy.2nd edition. J.B. Pawley, editor. PlenumPress, New York. 311–325.

Costes, S.V., D. Daelemans, E.H. Cho, Z. Dobbin,G. Pavlakis, and S. Lockett. 2004. Auto-matic and quantitative measurement of pro-tein-protein colocalization in live cells.

Biophys. J.

86:3993–4003.Davis, I. 1999. Visualizing fluorescence in

Dro-sophila

—optimal detection in thick speci-mens.

In

Protein Localization by Fluores-cence Microscopy—A Practical Approach.V.J. Allan, editor. Oxford University Press,Oxford, UK. 133–162.

Hagan, I.M., and K.R. Ayscough. 1999. Fluores-cence microscopy in yeast.

In

Protein Lo-calization by Fluorescence Microscopy—A Practical Approach. V.J. Allan, editor.Oxford University Press, Oxford, UK.179–206.

Hayat, M.A. 2000. Principles and Techniques ofElectron Microscopy: Biological Applica-tions. 4th edition. Cambridge UniversityPress, Cambridge, UK. 564 pp.

Herman, B., G. Gordon, N. Mahajan, and V. Cen-tonze. 2001. Measurement of fluorescenceresonance energy transfer in the opticalmicroscope.

In

Methods in Cellular Imaging.A. Periasamy, editor. Oxford UniversityPress, Oxford, UK. 257–272.

Keller, H.E. 1995. Objective lenses for confocal mi-croscopy.

In

Handbook of Biological Confo-cal Microscopy. 2nd edition. J.B. Pawley,editor. Plenum Press, New York. 111–126.

Manders, E.M.M., F.J. Verbeek, and J.A. Aten. 1993.Measurement of co-localization of objects indual-color confocal images.

J. Microsc.

169:375–382.Murphy, D.B. 2001. Lenses and geometrical optics.

In

Fundamentals of Light Microscopy andElectronic Imaging. Wiley-Liss, Inc., NewYork. 43–60.

Murray, J.M. 2005. Confocal microscopy, deconvo-lution, and structured illumination methods.

In

Live Cell Imaging—A Laboratory Man-ual. R.D. Goldman and D.L. Spector, edi-tors. Cold Spring Harbor Laboratory Press,Cold Spring Harbor, NY. 239–279.

Pawley, J. 2000. The 39 steps: a cautionary tale ofquantitative 3-D fluorescence microscopy.

Biotechniques.

28:884–887.Piston, D., G.H. Patterson, and S.M. Knobel.

1999. Quantitative imaging of the greenfluorescent protein (GFP).

Methods CellBiol

. 58:31–48.Rabut, G., and J. Ellenberg. 2005. Photobleaching

techniques to study mobility and moleculardynamics of proteins in live cells: FRAP,iFRAP, and FLIP.

In

Live Cell Imaging—ALaboratory Manual. R.D. Goldman andD.L. Spector, editors. Cold Spring HarborLaboratory Press, Cold Spring Harbor, NY.101–126.

Rossner, M., and K.M. Yamada. 2004. What’s in apicture? The temptation of image manipu-lation.

J. Cell Biol.

166:11–15.Shaw, P.J. 1995. Comparison of wide-field/decon-

volution and confocal microscopy for 3Dimaging.

In

Handbook of Biological Confo-cal Microscopy. 2nd edition. J.B. Pawley,editor. Plenum Press, New York. 373–387.

Smallcombe, A. 2001. Multicolor imaging: the im-portant question of co-localization.

Biotech-niques.

30:1240–1242, 1244–1246.Webb, R.H., and C.K. Dorey. 1995. The pixelated

image.

In

Handbook of Biological ConfocalMicroscopy. 2nd edition. J.B. Pawley, editor.Plenum Press, New York. 55–67.

For a more extensive list of useful mi-croscopy resources, including highlyrecommended textbooks, web sites, andpractical courses, please see the on-line supplemental material, available athttp://www.jcb.org/cgi/content/full/jcb.200507103/DC1.

Alison J. Northnortha@rockefeller.edu

Enormous thanks to Kirk Czymmek, Michael David-son, Scott Henderson, and Jyoti Jaiswal for theirextremely helpful and encouraging comments onthis article, and also to Lily Copenagle andZhenyu Yue (formerly of the Rockefeller University)for allowing me to use the images I collected usingtheir samples, both good and bad! I am im-mensely grateful to all those over the years whohave patiently mentored me in microscopy,whether during my research positions, at numerousmicroscopy courses, or simply in discussions with

“Remember that truth alone is the matter that you are in search after; and if you have been mistaken, let not vanity seduce you to persist in your mistake.”

Henry

Baker, The Microscope Made

Easy, 1742.

on May 16, 2007

ww

w.jcb.org

Dow

nloaded from