Femtosecond pump-probe microscopy generates virtual cross … · Femtosecond pump-probe microscopy generates virtual cross-sections in historic artwork Tana Elizabeth Villafanaa,

Femtosecond pump-probe microscopy generatesvirtual cross-sections in historic artworkTana Elizabeth Villafanaa, William P. Brownb, John K. Delaneyc, Michael Palmerc, Warren S. Warrena,1,and Martin C. Fischera

aDepartment of Chemistry, Duke University, Durham, NC 27708; bConservation Department, North Carolina Museum of Art, Raleigh, NC 27607; and cScientificResearch Department, National Gallery of Art, Washington, DC 20565

Edited by Michael D. Fayer, Stanford University, Stanford, CA, and approved December 23, 2013 (received for review September 12, 2013)

The layering structure of a painting contains a wealth of informationabout the artist’s choice of materials and working methods, butcurrently, no 3D noninvasive method exists to replace the takingof small paint samples in the study of the stratigraphy. Here, weadapt femtosecond pump-probe imaging, previously shown in tis-sue, to the case of the color palette in paintings, where chromo-phores have much greater variety. We show that combining thecontrasts of multispectral and multidelay pump-probe spectroscopypermits nondestructive 3D imaging of paintings with molecular andstructural contrast, even for pigments with linear absorption spectrathat are broad and relatively featureless. We show virtual cross-sectioning capabilities in mockup paintings, with pigment separa-tion and nondestructive imaging on an intact 14th century painting(The Crucifixion by Puccio Capanna). Our approach makes it possibleto extract microscopic information for a broad range of applicationsto cultural heritage.

nonlinear imaging | pigment spectroscopy

Identifying an artist’s choice of materials (e.g., support, pigments,binders, and varnishes in a painting) and working methods can

lead to greater understanding of past cultures and enhance theability of conservators to preserve that culture. In a painting, thisinformation is contained in its layered structure, and it is generallystudied by the physical removal of a small paint sample, whichcan be characterized by a plethora of analytical techniques (1).The sample needs to be representative of the painting but assmall as possible (typically <0.5 mm), and only local informationis obtained. Nondestructive analysis by traditional macroscopicmethods, such as X-radiography, near-infrared reflectance im-aging, and UV-visible fluorescence photography, can providesome information about a painting’s support, compositionalpaint changes, underdrawings, paint and varnish applications,and restorations (1). Materials can be identified in situ on themicroscopic scale using Raman (2–4) or the macroscopic scalewith reflectance imaging spectroscopy (5, 6) and X-ray fluores-cence intensity mapping (7). Unfortunately, none of these tech-niques contain quantitative depth-resolved material information.Methods that could offer 3D information are under active re-search, such as confocal X-ray fluorescence, absorption near-edgestructure imaging (8), optical coherence tomography (9), andterahertz imaging (10), but they are not yet widely used in con-servation science laboratories because of their limitations: X-ray–based techniques have absorption limited depths, whereas opticalcoherence tomography and terahertz imaging produce imagecontrast that is largely based on refractive index mismatches andtherefore, only provide structural contrast.In general, conventional (linear) optical imaging into the

paint layer of a painting is limited in its depth penetration byabsorption and scattering from the pigment particles. In biologyand biomedical applications, nonlinear imaging can provideoptical sectioning in highly scattering and absorbing samples(11, 12). Traditional nonlinear imaging has found a few appli-cations to cultural heritage; recent research includes the 3Dimaging of wood and varnishes in a violin with second harmonic

generation and two-photon excited fluorescence (13) and themapping of oil and varnish interfaces with third harmonic gen-eration (14). However, most inorganic pigments neither fluorescenor generate appreciable harmonic light, leaving these techniqueslimited in their scope for cultural heritage.Near-infrared femtosecond pump-probe optical microscopy

expands the range of detectable molecular signatures (15) toinclude signals from excited state absorption, ground state de-pletion, and stimulated emission (16). This microscopy techniquewas mainly developed for biomedical imaging and has been usedto provide high-resolution images for the biological pigmentshemoglobin (17, 18), eumelanin, and pheomelanin (19, 20) thatare present in skin (21) or ocular cancer (22).Extension of pump-probe microscopy from biological pig-

ments to samples of artist’s pigments has yielded promisingpreliminary results (23). However, achieving pump-probe con-trast in fine art objects is more challenging than skin imaging,because artist colorants range from organic dyes to inorganicminerals, with colors spanning the entire visible spectrum. Incontrast, the pigments in a sample of skin tissue are mainlylimited to hemoglobin, eumelanin, and pheomelanin, which allprovide image contrast with a single pump-probe wavelengthcombination (in this case, 720 and 810 nm, respectively). Here,we show that an increased spectral range of both the pump andthe probe beams, from near-IR to visible, and a variable timedelay of the pump-probe pulses help to address the complexityintroduced by the large range of possible pigments in the paintlayers and allow for in situ 3D imaging of paintings with mo-lecular specificity. We first show virtual cross-sectioning capa-bilities in historically relevant mockup paintings and use specific

Significance

We show that a nonlinear microscopy technique (femtosecondpump-probe microscopy) allows for nondestructive 3D imagingof paintings with molecular and structural contrast. Until now,studying the layering structure of a painting has generally re-quired the physical removal of a cross-section sample. Pump-probe imaging has previously been shown on biological tissue,but applications to cultural heritage are more challenging: thevariety of pigments in the artist’s palate is enormous comparedwith the biological pigments present in skin. Nonetheless, weshow virtual cross-sectioning capabilities in mockup paintingsand nondestructive imaging on an intact 14th century painting.This work represents a comprehensive collaborative effort be-tween laser and biomedical imaging experts and scientists andconservators in national museums.

Author contributions: T.E.V., W.S.W., and M.C.F. designed research; T.E.V., J.K.D., andM.P. performed research; W.P.B., J.K.D., M.P., and W.S.W. contributed new reagents/analytic tools; T.E.V., J.K.D., M.P., and M.C.F. analyzed data; and T.E.V., W.S.W., andM.C.F. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: [email protected].

1708–1713 | PNAS | February 4, 2014 | vol. 111 | no. 5 www.pnas.org/cgi/doi/10.1073/pnas.1317230111

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0

pump-probe signatures to provide pigment separation. We thenperform in situ 3D imaging on a 14th century painting (TheCrucifixion by Italian artist Puccio Capanna) to highlight ourability to noninvasively image and create virtual cross-sections ofcomplex pigment layers. Although we focus on historic paintings,our approach can be applied to a wide range of cultural heritageobjects and provides information extremely relevant to currentareas of interest in conservation science.

ResultsApproach. Pump-probe microscopy uses a sequence of ultrafastpulses (typically 0.2 ps in duration) to first electronically excitemolecules and then probe their response at a later time (up toabout 100 ps). A pump pulse moves a fraction of the ground statepopulation into electronic excited states, creating a correspond-ing hole in the ground state spectral distribution. In response tothe excitation, the population distributions in both ground andexcited states rearrange (excited state population tends toeventually relax back to the ground state). The changes in pop-ulation can be monitored by applying a second delayed (probe)pulse. Different molecular processes have different effects on theprobe pulse as a function of pump intensity and pump-probedelay. For example, in sequential two-photon absorption, theprobe is absorbed only by molecules in the excited state; hence,the presence of the pump increases the probe absorption (theabsorption then diminishes for longer delays). In contrast, forground state depletion, the probe is absorbed by moleculesremaining in the ground state, which has been partially depletedby the pump; hence, the presence of the pump decreases theprobe absorption (probe absorption increases back to the equi-librium value for long delays). Pump-probe spectroscopy hasbeen a mainstay of chemical physics for decades using high-powered lasers; however, at the powers that we are willing to useon important artwork, the differences in absorption might typi-cally be 1 part in 106 parts or a tiny signal on a large background.A schematic of our experimental setup (Fig. 1) shows our solu-tion to this challenge (17, 18). The pump is an intensity-modu-lated, mode-locked pulse train, which is synchronized andcombined with an unmodulated probe pulse train and coupledinto a laser-scanning microscope. Nonlinear interactions in thefocal volume within the sample will cause the modulation totransfer from the pump to the probe. The modulation frequencyis several megahertz, chosen to overcome the noise spectrum oflaser fluctuations. Pump-probe microscopy, like other nonlinearimaging methods, is much less affected by light scattering thanconventional microscopy; the signal is proportional to theproduct of the intensities of the two lasers, causing scattered lightto produce much less signal and giving the method its power in3D imaging.

Pump-Probe Specificity in Quinacridone Red and Ultramarine Blue. Ina typical painting, the 3D structure could consist of single tomultiple colorants in layers, mixtures, or a combination of layersand mixtures. Given the limited available colorants in Italianrenaissance paintings compared with contemporary works, pur-ples were often made using combinations of red pigments, suchas kermes or red madder (both substituted anthraquinone),mixed or layered with blue mineral pigments, such as natural ul-tramarine or azurite. The combination of kermes and natural ul-tramarine (lapis lazuli, which during the time, was more expensivethan gold) gives a rich purple. Such a combination would besuitable for the robes of major characters of a painting, such as theMadonna’s robe, whereas a combination with the cheaper azuritecan give a danker muted purple useful for less prominent figures.To test our virtual cross-section capabilities to separate mix-

tures vs. layering of pigments, we began by creating a set ofmockup paintings that features historically relevant pigmentpairings. In one case, a blue pigment (synthetic ultramarine) has

been covered with a thin glaze of red pigment (quinacridone red,a modern transparent, light-stable replacement for the naturalsubstituted anthraquinone) to create a purple appearance. Inanother case, a blue pigment (lapis lazuli) has been mixed withthe same red pigment to create a similar purple appearance. Totake a virtual cross-section, we first determined the pump-probewavelength combinations and interpulse delays that would fullyseparate ultramarine blue from quinacridone red by imaginga physical cross-section from the layered mockup at differentwavelength combinations. In the future, we can build a pump-probe library with cross-section samples from a variety of his-torical artworks that have already been characterized with cur-rently accepted analytical techniques. The spectroscopy resultsare seen in Fig. 2. The ground and excited state dynamics foreach pigment are specific to that pigment, providing structuredand complex pump-probe signatures. At a pump-probe wave-length combination of 615/810 nm, the signal in quinacridone redis positive (i.e., the amount of detected probe light decreaseswhen the pump is turned on) and decays in time. In ultramarineblue, the signal is negative, also decaying in time. The combina-tion of positive and negative pigment-specific transient absorp-tion signals provides an ideal case for creating a virtual cross-section. Interestingly, shifting to pump-probe wavelengths of 655/810 nm, the transient absorption amplitudes for these pigmentsare reversed (although much weaker in magnitude for quinacri-done red), although the linear absorption barely differs at ourchoice of pump wavelengths. The temporal decay characteristics

Fig. 1. Schematic of experimental setup. Pump-probe imaging uses an in-tensity-modulated pump pulse train and a nonmodulated probe pulse trainseparated by a variable time delay. Nonlinear interactions at the sample causethe pump modulation to transfer to the probe, which is detected by a lock-inamplifier. The pump is filtered out, and a series of images is collected, eachwith a different interpulse delay. AOM, acousto-optic modulator.

Villafana et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1709

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0

of the pigments also vary with pump-probe wavelengths, pro-viding yet another method of pigment separation.Bright-field microscopic images of the physical cross-sections

taken from the two mockup paintings (Fig. 3) show that, in thelayered case, the red glaze layer is ∼5-μm thick and the syntheticultramarine is ∼25-μm thick, whereas the mixed sample has onelayer with a varying thickness of 25–90 μm. Pump-probe images(at 615/810 nm) of the physical cross-sections were taken atdifferent time delays to allow assignment of pigments. The pig-ments are assigned false colors according to their pump-proberesponse; the red glaze (colored red) has a positive response, andthe ultramarine blue (colored cyan) has a negative response.There is no signal from the acrylic binder. The pump-probeimages give similar results in terms of both the distribution ofpigments and the layer thickness to the structure of the cross-section obtained from the bright-field microscope.

Virtual Cross-Sections of the Mock Paintings. Having found appro-priate imaging parameters for the two pigments, we createdvirtual cross-sections from the mockups by taking a series of enface images (xy images perpendicular to the beam axis) at dif-ferent depths (z direction) using the wavelength combination of615/810 nm with an interpulse delay of 0.1 ps. We generatedvirtual cross-sections from this volume set by selecting a dataslice in the xz or yz direction. In Fig. 4, we false-colored theimages according to the previously discussed methodology; thered glaze has been colored red, and the blue pigment has beencolored cyan. Although one mock painting features a layered

structure and the other mock painting features a mixture, bothcases present a purple color; the reflectance spectra acquiredusing fiber optic reflectance spectroscopy (FORS) from eachmock-up painting are similar (Fig. 4) and give no indication ofeither painting’s stratigraphy. However, in both cases, the pump-probe virtual cross-sections not only distinguish between thelayered and mixed stratigraphy but also, highlight variations inthe paintings caused by the artist’s brushwork. In the virtualcross-section image in Fig. 4, Right Inset displays data acquiredwith a high-N.A. objective, clearly resolving the thin red glazelayer. At this particular pump wavelength (615 nm), our imagingdepth in lapis lazuli was limited to roughly 10 μm because ofabsorption by ultramarine, but tuning the pump wavelength to710 nm increased the penetration through this pigment sixfold,which is discussed below, and let us image through the entireultramarine layer in the mockup (at the expense of a negligiblesignal from red glaze).Our technique has several advantages over the removal of

a physical cross-section other than its nondestructive nature. Bymapping out an entire volume, we can create virtual slices fromthe entire field of view in any direction and visualize differencesin brushwork or abrupt changes in layering that may not be ev-ident in a physical cross-section, in which accessible informa-tion is dependent on the sampling orientation. In addition, wecan sample many areas anywhere in the painting, which is notpossible when acquiring physical cross-sections (generally con-servators do not remove samples form pristine areas of thepaintings).

Nondestructive Investigation of Intact Artwork. The Crucifixion waspainted by Puccio Capanna in roughly 1330 on a wooden panelusing various pigments in egg tempera with gold leaf. We fo-cused on two areas in the painting: the rich blue of the VirginMary’s robe and the light blue robe of a floating angel that isoutlined with purple shading.Prior cross-sectional analysis of the Virgin Mary’s robe indi-

cates that the robe has been painted with a thick (up to 60 μm)layer of lapis lazuli. Thus, examination of the robe presenteda unique opportunity to test our depth penetration capabilities ina real work of art in a relatively uncomplicated setting; previouswork indicates no pump-probe signal in egg tempera binder orother binder materials. The results are presented in Fig. 5.Pump-probe imaging in the center of the robe gave virtual cross-sections consistent with the known thickness of the lapis lazuli,

Quinacridone Red

Ultramarine Blue

A B C

Fig. 2. Multispectral pump-probe investigation of quinacridone red and ultramarine blue. (A) Linear absorption spectra of quinacridone red (red curve) andultramarine blue (blue curve), with dashed lines indicating the different wavelengths that we used in the pump-probe imaging of the pigments. The pump-probe delay traces for (B) ultramarine blue and (C) quinacridone red at a variety of pump-probe wavelengths (indicated in the legend in nanometers) showvastly different time responses. All wavelength combinations were taken at a total power of 7 mW.

Layered Sample Mixed Sample

Pu

mp

-P

rob

eB

rig

ht

Fig. 3. Pump-probe separation of quinacridone red and ultramarine blueusing physical cross-section samples. Bright-field and pump-probe images ofthe physical cross-sections from the layered and mixed mockup painting. Thepump-probe images were taken at an interpulse delay of 0.1 ps and awavelength combination of 615/810 nm with a total power of 5 mW.Quinacridone red is false-colored red, whereas ultramarine blue is cyan.The pump-probe images are 365 × 90 μm in size.

1710 | www.pnas.org/cgi/doi/10.1073/pnas.1317230111 Villafana et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0

highlighting the ability of this method to noninvasively imagedeeply into pigment layers.To test the ability to obtain a virtual cross-section in a historic

painting, we imaged in an area with known layering—specifically,the outline of the floating angel. Fig. 6 shows a bright-field imageof a physical cross-section taken from this region of the angel’srobe. Bright-field microscopy and scanning electron microscopywith energy-dispersive X-ray spectroscopy (SEM-EDS) analysisof this cross-section indicates a very delicate and thin layering ofpigments containing, from top to bottom, a faded red glaze,a mixture of lead white and lapis lazuli, iron oxide, an organiccoating, gold leaf, an iron-rich mordant (a mixture of pigmentsand oil used to adhere the gold leaf) (24), and a gypsum ground.To image this area, we tuned the pump-probe pulses to a wave-length of 710/810 nm, for which we obtain signals from lapislazuli, iron oxide, mordant, and gold (gold has a strong two-photon absorption response at this wavelength combination). Weclassified the temporal dynamics in the pump-probe signals usingphasor analysis, a method that is commonly used to visualizedecay times in fluorescence lifetime measurements (25) and wasrecently adapted to pump-probe work (26). We identified threedistinct decay behaviors, consistent with previously observed datain lapis lazuli, iron oxide, and gold. Fig. 6 also displays thephysical cross-section as a false-colored pump-probe image, cor-relating well with the bright-field image. At the chosen wave-length combination, we do not see a signal in the faded red glaze,lead white, organic coating, or gypsum, except for a few mineralimpurities that may be present in those layers. We obtain signalfrom gold; however, the gilding is very thin and could not bespatially resolved. Also, at this wavelength combination, ironoxide and mordant showed signals with identical decay behaviors.The optical image was taken after pump-probe imaging, in-dicating no visible damage. We then imaged an area of the intactpainting adjacent to the sample site. At this location, we acquiredvolume data with a fixed pump-probe delay of 0.2 ps, which yieldspositive pump-probe signals from iron oxide, gold, and mordantand negative signals from lapis lazuli. Because the pump-probedynamics of iron oxide/mordant and gold are very similar, theycan only be cleanly separated by acquiring data at many denselysampled time delays, which with our current setup, was not fea-sible during the loan period of the painting. Hence, in these

pump-probe images, we color-coded the positive signals orange,encompassing any of the three materials, and negative signalscyan (lapis lazuli). At the probed location, we found a composi-tion that is slightly different from the physical cross-section. Theen face images show positive signal on the surface (most likelyfrom iron oxide), negative signal in the center from lapis lazuli,and positive signal again underneath the lapis lazuli, which ismost likely gold with possible contributions from mordant (thegilding in this region is heavily cracked, exposing the mordantunderneath). This view is supported by virtual cross-sectionsextracted from this dataset. The virtual xz slice and even more so,the maximum intensity projection of the volume along the y di-rection suggest either a mixture or very thin layers of iron oxidewith lapis lazuli and gold leaf with mordant underneath it.

DiscussionThese results represent a large step in the nondestructive 3Danalysis of pigments and their composition in historical artworks. In an intact 600-y-old painting, we have shown that ourtechnique can noninvasively image through a relatively thicklayer of paint and map multilayer structures. With our currentmicroscope design, we can easily image volume data at fixed timedelays or 2D images at varying pump-probe delays. Typical ac-quisition times of the three-parameter datasets in Fig. 6 were 30–60 min. Future improvements in detection sensitivity shoulddecrease imaging time markedly, leading the path to acquisitionof 4D datasets (3D space and delay) dense enough for 3D pig-ment-specific mapping. The achievable imaging depth dependson the structure and layering of the artwork. Certain flexibility isafforded by the choice of pump and probe wavelengths. For eachpigment, there will be tradeoffs between signal strength (pumpingclose to an absorption line best excites electronic states but resultsin the largest linear absorption) and contrast (not all wavelengthcombinations yield usable signal). Efforts to establish a pump-probe database for the most common pigment types are currentlyunderway. Some materials might not yield distinct pump-probesignals (a fact that we use to our advantage by rendering bindersinvisible), in which case combination of our microscopy techniquewith other optical contrasts in the same microscope might bebeneficial. For example, to study organic glazes, binders, and

Pump-Probe Images

x

z

x

z

En-face (xy) Layered En-face (xy) Mixed

Fig. 4. Reflectance spectra and virtual cross-sections of mock paintings.(Left) The linear reflectance spectrum from each painting indicates thepresence of quinacridone red (peak at 600 nm) and ultramarine blue (in-creased reflection at 700 nm) but does not indicate which painting is layeredand which is mixed. Inset shows photographs of the (Upper) layered and(Lower) mixed mockups. (Right) A volume set of pump-probe images of theintact mock paintings was taken at a wavelength combination of 615/810nm, fixed interpulse delay of 0.1 ps, and total power of 3 mW with a 20× 0.7N.A. objective. One image of each set is shown (false-colored red for qui-nacridone red and cyan for ultramarine blue). Virtual cross-section (xz)images immediately reveal the composition difference between the layeredand mixed samples. Inset on the virtual cross-section of the layered samplewas obtained with a higher-resolution 60× 0.9 N.A. objective. Each en face(xy) image is 365 × 365 μm, and the virtual cross-sections are 365 × 90 μm.

The Crucifixion Virtual Cross-Section

x

z

En-face (xy)

Fig. 5. Virtual cross-section of the Virgin Mary’s blue robe in Puccio Cappana’sThe Crucifixion. (Left) The painting was imaged in an area of Mary’s robecontaining only a single layer of lapis lazuli with a wavelength combination of720/810 nm and an interpulse delay of 0.2 ps with a total power of 2.7 mW.(Upper Right) The en face image shown (365 × 365 μm) was from roughly30 μm beneath the surface of the robe. Here, the images have been false-colored cyan for lapis lazuli and magenta for mineral impurities that occur withnatural lapis lazuli. (Lower Right) The virtual xz cross-section (365 × 60 μm)highlights the thickness of the lapis lazuli used to paint Mary’s robe.

Villafana et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1711

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0

varnishes, it is possible to incorporate nonlinear fluorescence orharmonic generation contrast, which was shown useful in somerecent 3D imaging work (13, 14).Our work with the iron-rich pigments in The Crucifixion fur-

ther suggests interesting applications for earth pigments in a va-riety of objects from pottery (27) and ancient relics (28, 29) toGreek statuary, which was not white (as believed for centuries)but brightly colored (30–32). Other potential applications in-clude the in situ mapping of degradation products. For example,The Joy of Life by Henri Matisse (1905) contains large areas ofcadmium yellow that have degraded to browns and whites (33),an issue that has also affected masterworks by Van Gogh,Picasso, and others (34). Mapping degradation products couldaid in understanding degradation processes (35, 36). Finally, 2Dwavelet analysis of van Gogh’s brushstrokes has been applied to101 high-resolution grayscale scans (37), and our work suggests

that extensions to 3D brushstroke imaging are possible, pre-senting opportunities and challenges for data mining (38).Pump-probe microscopy, in conjunction with current techniques

in conservation science, could dramatically impact the study andunderstanding of our cultural heritage. Ultimately, the generalapplication of our technique in situ will require a portable nonlinearmicroscope, but all of the miniaturization technology needed forsuch a device is being developed for biomedical applications (39);also, appropriate laser sources have recently dropped drastically incost (40). We have shown that it is possible to leverage large so-cietal investment in biomedical and molecular imaging to enableapplications with a broader impact.

MethodsApproach. A Ti:Sapphire mode-locked laser operating at a repetition rate of80 MHz, with a wavelength in the near-IR and pulse duration of roughly150 fs, pumps an optical parametric oscillator, with an output in the visible tothe near-IR of a similar pulse duration. The pump pulse train is intensity-modulated at 2 MHz using an acousto-optic modulator. The probe pulse isunmodulated, and the interpulse delay is controlled by an adjustable opticalpath length in the probe arm. The two beams are overlapped on a dichroicmirror and sent collinearly into a laser-scanning microscope. The pulses arefocused onto the sample with a 20× 0.7 N.A. or 60× 0.9 N.A. air objective.Any nonlinear interaction with the sample will transfer the modulationfrom the pump to the probe, and changes in the probe intensity aredetected by a photodiode and a lock-in amplifier with a reference at themodulation frequency.

Mock Paintings of Quinacridone Red and Ultramarine Blue. Layered painting.Quinacridone red (1310; Golden) and synthetic lapis lazuli (45000; Kremer)were prepared in an acrylic medium. The synthetic lapis lazuli was paintedonto a glass slide that had been preparedwith a gesso ground. After the layerdried, a thin coat of quinacridone red was painted on top and allowed to dry.A small sample was extracted with a scalpel from the painting and mountedin Wards Bio-Plastic for the cross-section.Mixed painting. Quinacridone red (1310; Golden) and Afghan lapis lazuli(15300; Kremer) were prepared in an acrylic medium, mixed together ina roughly 1:1 ratio, and painted onto a glass slide that had been preparedwith a gesso ground. A small sample was extracted with a scalpel from thepainting and mounted in Wards Bio-Plastic for the cross-section.FORS analysis. A fiber optic spectroradiometer, FS3 (ASD Inc.), was used toobtain FORS spectra from the mock paintings. The spectrometer operatesfrom 350 to 2,500 nm, with a spectral sampling of 1.4 nm from 350 to 1,000nm. The spectral resolution at 700 nm is 3 nm. The light source of a leaf probehead (ASD Inc.) was used at a distance of 20 cm to illuminate the samples(∼400 lx), and the fiber was placed ∼1 cm from the object, giving an ∼3-mmspot size at the painting. We averaged two spectra, with a total acquisitiontime of <5 s per point.

Nondestructive Investigation of Intact Artwork. The Crucifixion was the cen-tral compartment of one panel of a diptych altarpiece. The pigments—typical of an early Renaissance palette—include pure lapis lazuli, azurite,vermillion, red lake, red lead, terra verte, white lead, black, and earth colors.The medium is estimated to be egg yolk, and the panel support was iden-tified as poplar. The gold was applied to the embroidered decoration bymordant gilding and the gold field by water gilding onto a red mordant. Asmall sample was extracted with a scalpel from the angel’s robe andmounted in Wards Bio-Plastic for the cross-section.

ACKNOWLEDGMENTS. We acknowledge Prathyush Samineni for his help inbeginning this research with the first paint samples and Jesse Wilson forhelpful discussions and providing the schematic in Fig. 1. We thank the NorthCarolina Museum of Art and the National Gallery of Art for their collabora-tion and contribution of various art pieces. This material is based on worksupported by National Science Foundation Grant CHE–1309017.

1. Steward B (2007) Analytical Techniques in Materials Conservation (Wiley,New York).

2. Clark RJH (2007) Raman microscopy as a structural and analytical tool in the fields ofart and archaeology. J Mol Struct 834–836:74–80.

3. Vandenabeele P, Edwards HGM, Moens L (2007) A decade of Raman spectroscopy inart and archaeology. Chem Rev 107(3):675–686.

4. Brambilla A, et al. (2011) A remote scanning Raman spectrometer for in situ mea-surements of works of art. Rev Sci Instrum 82(6):063109.

5. Delaney JK, et al. (2010) Visible and infrared imaging spectroscopy of Picasso’s Harlequinmusician:Mappingand identificationof artistmaterials in situ.Appl Spectrosc64(6):584–594.

6. Dooley KA, et al. (2013) Mapping of egg yolk and animal skin glue paint binders inEarly Renaissance paintings using near infrared reflectance imaging spectroscopy.Analyst (Lond) 138(17):4838–4848.

7. Dik J, et al. (2008) Visualization of a lost painting by Vincent van Gogh using syn-chrotron radiation based X-ray fluorescence elemental mapping. Anal Chem 80(16):6436–6442.

Virtual Cross-Section Generated from En-face Volume Cube

Cross-Section Maximum Intensity Projection

0 m (Surface) - 5 m - 12 m

En-face Pump-Probe Images at 0.2 ps Delay

Physical Cross-Section and Pump-Probe Delay Behavior

Ab

sorp

tio

n (

arb

. u.)

Pump-Probe Delay (ps)

Iron Oxide/MordantGold

Lapis Lazuli

Fig. 6. Investigation of the angel’s purple robe in Puccio Cappana’s TheCrucifixion. (Top) A bright-field image of the physical cross-section taken fromthe angel’s robe is shown. A pump-probe delay dataset was acquired (40interpulse delays, pump-probe = 710/810 nm, total power = 1.5 mW, size is545 × 55 μm). From this set, we created a false-colored image according to thepump-probe delay behavior (cyan for lapis lazuli, red for the two iron-richpigments above and below the gold layer, and yellow for gold). Cumulativepump-probe traces of all identified pigments in the image are shown in Right.Note that the gold layer is thinner than the resolution of our microscope, andthe gold-labeled trace likely contains some contribution from the adjacentmordant. (Middle) A pump-probe volume dataset was taken with a fixed0.2-ps delay in the angel’s robe (pump-probe = 710/810 nm, total power =1.5 mW). The images from this set have been false-colored according to thesignal at this delay: cyan for negative signal (corresponding to lapis lazuli) andorange for positive (iron oxide/mordant and gold; with only a single delay,these three materials cannot be separated). Each image is 185 × 185 μm.(Bottom) An xz slice taken from the volume data shows a positive componentmixed within the lapis lazuli layer (most likely iron oxide) with another posi-tive component underneath (most likely gold and possibly, underlying mor-dant that we image through microscopic cracks in the gold layer). Thiscomposition is seen more clearly in a maximum intensity projection of theentire volume cube. The virtual cross-section dimension is 185 × 50 μm.

1712 | www.pnas.org/cgi/doi/10.1073/pnas.1317230111 Villafana et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0

8. Kanngießer B, Malzer W, Mantouvalou I, Sokaras D, Karydas AG (2012) A deep view incultural heritage—confocal micro X-ray spectroscopy for depth resolved elementalanalysis. Appl Phys A Mater Sci Process 106(2):325–338.

9. Targowski P, Iwanicka M (2012) Optical coherence tomography: Its role in the non-invasive structural examination and conservation of cultural heritage objects—a re-view. Appl Phys A Mater Sci Process 106(2):265–277.

10. Adam AJL, Planken PCM, Meloni S, Dik J (2009) TeraHertz imaging of hidden paintlayers on canvas. Opt Express 17(5):3407–3416.

11. Helmchen F, Denk W (2005) Deep tissue two-photon microscopy. Nat Methods 2(12):932–940.

12. Denk W, Strickler JH, Webb WW (1990) Two-photon laser scanning fluorescencemicroscopy. Science 248(4951):73–76.

13. Latour G, Echard J-P, Didier M, Schanne-Klein M-C (2012) In situ 3D characterizationof historical coatings and wood using multimodal nonlinear optical microscopy. OptExpress 20(22):24623–24635.

14. Nevin A, et al. (2010) Multi-photon excitation fluorescence and third-harmonic gen-eration microscopy measurements combined with confocal Raman microscopy for theanalysis of layered samples of varnished oil films. Appl Phys A Mater Sci Process100(3):599–606.

15. Warren WS, Fischer MC, Ye T (2007) Cellular imaging—novel nonlinear contrast im-proves deep-tissue microscopy. Laser Focus World 43(6):99–103.

16. Ye T, Fu D, WarrenWS (2009) Nonlinear absorption microscopy. Photochem Photobiol85(3):631–645.

17. Fu D, et al. (2007) High-resolution in vivo imaging of blood vessels without labeling.Opt Lett 32(18):2641–2643.

18. Fu D, Matthews TE, Ye T, Piletic IR, Warren WS (2008) Label-free in vivo optical im-aging of microvasculature and oxygenation level. J Biomed Opt 13(4):040503.

19. Fu D, Ye T, Matthews TE, Yurtsever G, Warren WS (2007) Two-color, two-photon, andexcited-state absorption microscopy. J Biomed Opt 12(5):054004.

20. Fu D, et al. (2008) Probing skin pigmentation changes with transient absorption im-aging of eumelanin and pheomelanin. J Biomed Opt 13(5):054036.

21. Matthews TE, Piletic IR, Selim MA, Simpson MJ, Warren WS (2011) Pump-probe im-aging differentiates melanoma from melanocytic nevi. Sci Transl Med 3(71):71ra15.

22. Wilson JW, et al. (2013) Imaging microscopic pigment chemistry in conjunctival mel-anocytic lesions using pump-probe laser microscopy. Invest Ophthalmol Vis Sci 54(10):6867–6876.

23. Samineni P, deCruz A, Villafaña TE, Warren WS, Fischer MC (2012) Pump-probe im-aging of historical pigments used in paintings. Opt Lett 37(8):1310–1312.

24. Gettens RJ, Stout GL (1966) Painting Materials: A Short Encyclopaedia (Dover, NewYork).

25. Redford GI, Clegg RM (2005) Polar plot representation for frequency-domain analysisof fluorescence lifetimes. J Fluoresc 15(5):805–815.

26. Robles FE, Wilson JW, Fischer MC, Warren WS (2012) Phasor analysis for nonlinearpump-probe microscopy. Opt Express 20(15):17082–17092.

27. Sciau P, et al. (2011) Reverse engineering the ancient ceramic technology based onX-ray fluorescence spectromicroscopy. J Anal At Spectrom 26(5):969–976.

28. Edwards HGM, Villar SEJ, David AR, de Faria DLA (2004) Nondestructive analysis ofancient Egyptian funerary relics by Raman spectroscopic techniques. Anal Chim Acta503(2):223–233.

29. Colombini MP, et al. (2004) Integrated analytical techniques for the study of ancientGreek polychromy. Talanta 63(4):839–848.

30. Gurewitsch M True colors. Smithsonian 2008 July. Available at www.smithsonianmag.com/arts-culture/true-colors.html. Accessed January 4, 2014.

31. Brinkmann V, Wunsche R (2007) Gods in Color: Painted Sculpture of Classical Antiquity(Glypotothek, Munich).

32. Panzanelli R, Schmidt ED, Lapatin K (2008) The Color of Life: Polychromy in Sculpturefrom Antiquity to the Present (Getty Research Institute, Los Angeles).

33. Mass JL, et al. (2013) The photodegradation of cadmium yellow paints in HenriMatisse’s Le Bonheur de vivre (1905-1906). Appl Phys A Mater Sci Process 111(1):59–68.

34. Thoury M, et al. (2011) Near-infrared luminescence of cadmium pigments: In situidentification and mapping in paintings. Appl Spectrosc 65(8):939–951.

35. Van der Snickt G, et al. (2009) Characterization of a degraded cadmium yellow (CdS)pigment in an oil painting by means of synchrotron radiation based X-ray techniques.Anal Chem 81(7):2600–2610.

36. Van der Snickt G, et al. (2012) Combined use of synchrotron radiation based micro-X-ray fluorescence, micro-X-ray diffraction, micro-X-ray absorption near-edge, andmicro-fourier transform infrared spectroscopies for revealing an alternative degradationpathway of the pigment cadmium yellow in a painting by Van Gogh. Anal Chem84(23):10221–10228.

37. Johnson CR, et al. (2008) Image processing for artist identification. IEEE Signal Pro-cessing Mag 25(4):37–48.

38. Chen G, Little AV, Maggioni M, Rosasco L (2011) Applied and Numerical HarmonicAnalysis (Birkhauser, Basel).

39. Ghosh KK, et al. (2011) Miniaturized integration of a fluorescence microscope. NatMethods 8(10):871–878.

40. Durfee CG, et al. (2012) Direct diode-pumped Kerr-lens mode-locked Ti:sapphire laser.Opt Express 20(13):13677–13683.

Villafana et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1713

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Dec

embe

r 17

, 202

0