Archives of Biochemistry and Biophysics 504 (2010) 40–49

Contents lists available at ScienceDirect

Archives of Biochemistry and Biophysics

journal homepage: www.elsevier .com/ locate /yabbi

Validation model for Raman based skin carotenoid detection

Igor V. Ermakov, Werner Gellermann *

Department of Physics and Astronomy, University of Utah, Salt Lake City, UT 84112, United States

a r t i c l e i n f o a b s t r a c t

Article history:Available online 1 August 2010

Keywords:Human skinAntioxidantsCarotenoidsRaman spectroscopyHPLCValidation

0003-9861/$ - see front matter � 2010 Elsevier Inc. Adoi:10.1016/j.abb.2010.07.023

Corresponding author. Fax: +1 801 581 4801.E-mail address: werner@physics.utah.edu (W. Gel

Raman spectroscopy holds promise as a rapid objective non-invasive optical method for the detection ofcarotenoid compounds in human tissue in vivo. Carotenoids are of interest due to their functions as anti-oxidants and/or optical absorbers of phototoxic light at deep blue and near UV wavelengths. In the mac-ular region of the human retina, carotenoids may prevent or delay the onset of age-related tissuedegeneration. In human skin, they may help prevent premature skin aging, and are possibly involvedin the prevention of certain skin cancers. Furthermore, since carotenoids exist in high concentrationsin a wide variety of fruits and vegetables, and are routinely taken up by the human body through the diet,skin carotenoid levels may serve as an objective biomarker for fruit and vegetable intake. Before theRaman method can be accepted as a widespread optical alternative for carotenoid measurements, directvalidation studies are needed to compare it with the gold standard of high performance liquid chroma-tography. This is because the tissue Raman response is in general accompanied by a host of other opticalprocesses which have to be taken into account. In skin, the most prominent is strongly diffusive, non-Raman scattering, leading to relatively shallow light penetration of the blue/green excitation lightrequired for resonant Raman detection of carotenoids. Also, sizable light attenuation exists due to thecombined absorption from collagen, porphyrin, hemoglobin, and melanin chromophores, and additionalfluorescence is generated by collagen and porphyrins. In this study, we investigate for the first time thedirect correlation of in vivo skin tissue carotenoid Raman measurements with subsequent chromatogra-phy derived carotenoid concentrations. As tissue site we use heel skin, in which the stratum corneumlayer thickness exceeds the light penetration depth, which is free of optically confounding chromophores,which can be easily optically accessed for in vivo RRS measurement, and which can be easily removed forsubsequent biochemical measurements. Excellent correlation (coefficient R = 0.95) is obtained for this tis-sue site which could serve as a model site for scaled up future validation studies of large populations. Theobtained results provide proof that resonance Raman spectroscopy is a valid non-invasive objectivemethodology for the quantitative assessment of carotenoid antioxidants in human skin in vivo.

� 2010 Elsevier Inc. All rights reserved.

Introduction

Carotenoid molecules play an important protective role in theskin’s antioxidant defense system [1]. The eight most concentratedcarotenoid antioxidants in human skin are lycopene, a-carotene,b-carotene, lutein, zeaxanthin, cryptoxanthin, phytoene and phy-tofluene, with lycopene and the carotenes accounting for about60–70% of total carotenoid content [2]. They are thought to act asscavengers for free radicals [3], singlet oxygen [4], and other harmfulreactive oxygen species [5,6] formed by metabolic processes or byexcessive exposure of skin to the UV components of sunlight.

If unbalanced due to a lack of antioxidants, the destructive effectsof reactive oxygen species and free radicals can lead to skin malig-nancies and disease. In animal models, carotenoids have been shownto inhibit carcinoma formation in the skin [7]. In humans, it has been

ll rights reserved.

lermann).

shown that skin carotenoid levels are strongly and significantly cor-related with carotenoid levels in plasma [8]. As is found in plasma,skin carotenoid levels are lower in smokers than in nonsmokers. Car-otene levels in skin are known to increase with supplementation [9],and supplemental b-carotene is used to treat patients with erythro-poietic protoporphyria, a photosensitive disorder [10]. Supplemen-tal carotenoids have also been shown to delay erythema in normalhealthy subjects exposed to UV light [11,12]. There is limited evi-dence that they may be protective against skin and other malignan-cies [13], but more research is required to confirm these findings.

Since carotenoids are lipophilic molecules, they are well placedin the skin to act as chain-breaking antioxidants protecting epider-mal polyunsaturated fatty acids from oxygen peroxidation [14].Other dermal antioxidants such as superoxide dismutase, glutathi-one peroxidase, alpha-tocopherol, ascorbic acid, and melaninswork in collaboration with carotenoids to provide skin with adefensive mechanism against free radical attack and oxidativestress [15]. Because these molecules work as a network, definitive

Fig. 1. Comparison of light excitation/detection paths in RRS based detection ofcarotenoids in the human retina (top panel) and skin (bottom panel). In the ocularcase, the excitation light has to traverse only a transparent nerve fiber layer, NFL,prior to excitation of the carotenoid containing layer, MP. RRS responses from theMP layer, indicated as black arrows, can be obtained in simple 180�, single-path,backscattering geometry free of confounding chromophores. Additional chromoph-ores (circles) exist, but they are concentrated in a posterior layer, the retinalpigment epithelium, RPE, and their fluorescence contributions (white arrows) to thetotal optical response can be simply subtracted. In the dermal case, several

I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49 41

measurement of a subset of these antioxidants provides an indica-tion of the relative strength of the whole system.

The effectiveness of this protective network can be diminishedeither by excessive generation of free radicals or by insufficientantioxidant molecules being supplied to the skin. The result is astate of oxidative stress where important skin constituents are ex-posed to free radical damage and associated structural and chem-ical degenerative effects. If an individual is measured and found tohave a lower than normal skin carotenoid levels, that person’s anti-oxidant defense system would likely be relatively ill-equipped tobalance oxidative processes compared to an individual havinghigher levels of antioxidants. Skin antioxidant measurements pro-vide an opportunity for intervention strategies such as increasingthe dietary intake of fruits and vegetables, smoking cessation,and/or prescribing dietary antioxidant supplements.

The gold standard technique for measuring carotenoids is thebiochemical method of high performance liquid chromatography(HPLC).1 Requiring chemical decomposition of the sample of inter-est, HPLC works well for the measurement of carotenoids in serum,where it has been used to assess carotenoid antioxidant status fol-lowing the collection of blood samples. Serum measurements, how-ever, are more indicative of short-term dietary intakes of carotenoidsrather than steady state accumulations in skin tissue sites or drops inconcentration due to the influence of external oxidative stress fac-tors such as smoking and UV light exposure. Skin carotenoid HPLCmeasurement have been carried out, too, but it requires highly inva-sive tissue biopsies. Optical detection approaches for skin carote-noids could potentially overcome these limitations by rapidly andobjectively measuring carotenoid content directly in the skin tissuesites of interest in a completely non-invasive fashion. The methodscould be used to assess microscopically small tissue volumes ifneeded, could be used to track carotenoid status over time, and couldallow for inter-subject comparisons of skin carotenoid levels in vivo.It is clear, however, that any optical detection method would have tobe correlated with HPLC results of excised skin tissue, a rather non-trivial task, in order to gain acceptance as a viable detectionalternative.

A first optical approach for skin carotenoid detection usedreflectance spectroscopy [11,16]. Dermal carotenoid absorptionand related carotenoid concentrations could be derived from themeasured reflectance spectra, correlations with plasma carotenoidlevels could be demonstrated, and skin uptake of b-carotene couldbe tracked in supplementation experiments. A difficulty of thereflectance approach lies in the analytical derivation of concentra-tion data since the latter cannot be simply derived with Beer’s lawdue to the unknown path length of the reflected light in the tissue.To overcome this problem, a non-linear mapping model was intro-duced that provides a one-to-one mapping relation between reflec-tance and absorption spectra and that takes into account tissueinhomogeneity [16]. In a later publication, the spectral reflectancemethod was compared with objective skin color measurements viadetermination of tri-stimulus chromaticity values, with a reason-ably good correlation, and it could be demonstrated that carote-noids reduce photosensitivity in Caucasian populations [17].

Another optical approach for the detection of tissue carotenoidsis based on resonance Raman scattering (RRS) spectroscopy. Initially,we used this method for the detection of carotenoids in the humanretina (see Ref. [18] and references therein). In healthy subjects,carotenoids are typically very highly concentrated in the macular re-gion of the retinal area, and are thought to protect this critical tissueregion via optical filtering and antioxidant action. The macularcarotenoids are located just below the transparent outer nerve fiber

potentially confounding chromophore species (open squares and circles) existsimultaneously besides the carotenoids in stratum corneum (SC), epidermis (EP)and dermis (D). Varying in general between tissue sites and subjects in strength andcomposition, their influence on the RRS responses has to be minimized. In thisstudy this is achieved by limiting the light excitation to the stratum corneum layer.

1 Abbreviations used: HPLC, high performance liquid chromatography; RRS, reso-nance Raman scattering.

layer of the retinal layer structure. Their high concentration and theabsence of potentially confounding absorbers in the excitation lightpath prior to the carotenoid containing layer provide a highly favor-able RRS excitation/detection scenario, as illustrated in Fig. 1 (upper

Fig. 2. (a) RRS spectrum of b-carotene in a solution (acetone); (b) optical responseobtained from living human skin; and (c) RRS carotenoid spectrum obtained aftersubtraction of background from spectrum (b). All spectra were obtained withexcitation at 488 nm. Note that RRS spectra are virtually undistinguishable at roomtemperature at the used spectral resolution (�1.3 nm).

42 I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49

panel). Since the blue excitation light can proceed without attenua-tion to the carotenoid containing tissue layer of interest, the RRS re-sponse, obtained in 180� backscattering geometry, can be taken as adirect measure of the macular carotenoid level. The portion of theexcitation light that is not absorbed by the carotenoids, traversesinto the deeper retinal layers, where it generates fluorescence fromother chromophores (like lipofuscin in the retinal pigment epithe-lium), but this fluorescence contribution to the overall detected lightresponse can be treated as a simple superposition and therefore besubtracted. In fact, our validation studies with excised ocular tissuestructures and eye cups demonstrated a very high correlation be-tween Raman and HPLC methods [19].

Subsequently we suggested RRS as a feasible method also forthe non-invasive quantitative detection of carotenoids in humanskin [20,21]. However, this tissue poses a much more challengingoptical excitation/detection scenario, as sketched in Fig. 1 (lowerpanel). First difficulties arise since on average skin carotenoid con-centrations are about two orders of magnitude lower relative tothe human macula, and since skin tissue is highly heterogeneous.Furthermore, a variety of other chromophores (like collagen, por-phyrin, hemoglobin, and melanin) are simultaneously excited withthe carotenoids of interest, thus producing competing absorptionand associated fluorescence (‘‘autofluorescence”) events. Addition-ally, the outer skin layer, the stratum corneum, produces highlydiffusive scattering events for all light components, including exci-tation light as well as all fluorescence and Raman light componentstraveling back towards the detector from within the excited tissuevolume. The overall spectral response in skin is therefore a com-plex superposition of a weak carotenoid RRS response with strongabsorptions from other chromophores, strong fluorescence (auto-fluorescence) contributions, and strong scattering. Since all non-Raman contributions to the total optical response are generatedin the same tissue layers/volume as the carotenoid RRS response,they cannot be simply subtracted. Furthermore, their respectiveconcentrations and compositions can be expected to vary betweensubjects and tissue sites. In principle, one would therefore have tomeasure their combined absorption strength separately with otherspectroscopy methods in order to derive a suitable correction fac-tor for the RRS carotenoid measurement of the tissue site ofinterest.

One strategy to avoid the confounding influences is to limit thedermal RRS measurements to the palm of the hand or the heel ofthe foot, tissue sites that have a relatively thick stratum corneumlayer and therefore prevent the excitation light from penetratinginto deeper layers. Furthermore, these tissue sites are relativelyfree of melanin, independent of ethnicity. Using this approach,we could demonstrate that RRS is able to track carotenoid concen-trations over time and to monitor concentration changes occurringas a result of dietary modifications and/or carotenoid supplemen-tation [22]. Also, using sequential excitation with two lasers atgreen and blue excitation wavelengths, which differ significantlyin respective lycopene excitation efficiencies, we could demon-strate the possibility to selectively deduce skin lycopene levels[23]. Furthermore, we developed portable dermal carotenoidinstruments for first use in a clinical setting [2], and as platformsfor mass produced devices suitable to track carotenoid uptake inthe nutritional supplement industry [22,24]. The main correlationresults between RRS derived skin carotenoid levels and carotenoidsupplements and vegetable consumption have been confirmed inindependent studies [25,26].

Recently it was demonstrated that dermal carotenoid RRS mea-surements can be carried out with LEDs, instead of lasers, as exci-tation sources, resulting in the development of a more robustinstrument configuration with high thermal tolerance [27]. Onthe clinical side, we have begun to extend dermal carotenoid mea-surements to the field of neonatology, where the method avoids

the drawing of blood samples in infants. This will make it possibleto investigate correlations between infant tissue carotenoid levelsand oxidative stress related degenerative disorders, as well as todevelop effective antioxidant containing infant formula [28].

However, before RRS can be accepted as a reliable biomarker forhuman research, the method needs to be scrutinized more thor-oughly in view of the caveats discussed above. In particular, forany chosen set of instrumentation parameters and tissue type in-volved in RRS based carotenoid detection, data is critically neededon validity as compared to chemical analysis of excised tissue.

Resonance Raman method, linearity of response, influence ofexcitation wavelength on skin carotenoid spectra, and heel skinmeasurements

RRS based carotenoid detection takes advantage of the strongelectronic absorption bands of carotenoids in the blue/green spec-tral region (peak at �450 nm, �80 nm width). This absorption iscaused by strong electric dipole-allowed vibronic transitions ofthe carotenoid molecule’s conjugated p-electron from the 1Ag

ground state to the 1Bu excited state. Optical excitation into thisabsorption leads to resonantly enhanced Raman scattering withmore than 1000-fold increase of the Raman scattering cross sectionrelative to non-resonant excitation. The Raman spectrum of carot-enoid molecules is characterized by two prominent Stokes lines at1159 and 1524 cm�1, shown in Fig. 2(a) for a solution of beta car-otene in acetone. The lines originate, respectively, from the C–Csingle-bond and the C@C double-bond stretch vibrations of themolecule’s conjugated carbon backbone [29]. A weaker peak at1008 cm�1 is attributed to rocking motion of the molecule’s methylside groups [29].

Due to the unique ordering of the excited electronic states,carotenoids exhibit an unusually weak intrinsic fluorescence. Thisallows one to detect the RRS response relatively free of interferingintrinsic carotenoid fluorescence signals. In RRS spectra of pure

I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49 43

carotenoid solutions, the intensity ratio between Raman signal andbackground fluorescence is as high as �1:1 at the spectral positionof the C@C Raman peak, and the intensity of each Stokes line variesproportionally with the concentration of the molecules. Therefore,in principle, the Raman peak height can be used as a measure ofcarotenoid content. Quantitatively, the relation between RRS lightintensity, IR, and excitation light intensity, Iexc, is given by

IRðkÞ ¼ IexcrðkÞNðEiÞ: ð1Þ

Here, N(Ei) is the population density (concentration) of the mole-cules, and rðkÞ is the Raman scattering cross section, a constantwhose magnitude depends on the excitation and collection geome-try, and excitation wavelength. In optically thick media, as in theskin, a deviation from the linear Raman response of IR versus con-centration N can be expected if strong absorbers are present. In thiscase one would have an excitation/detection scenario with impededlight propagation and self-absorption of the Stokes Raman lines.

A cross section of human skin is shown schematically in Fig. 3.The outermost layer of the skin, the stratum corneum, is relativelyuniform and bloodless. It is entirely composed of cells with missingnuclei, and features low melanin content irrespective of ethnicity.The stratum corneum is followed by deeper epidermal and dermallayers. In contrast to the stratum corneum, these two deeper layersare rather heterogeneous due to the presence of blood capillaries,hair follicles, sebaceous glands, and sweat glands. Furthermore,dermis and epidermis are rich in confounding blood chromophoresand melanin, with blood components varying rather rapidly. As aconsequence, time varying attenuation effects can be expectedfor both excitation light and Raman light paths in these layers.

The simplest light propagation scenario for RRS based skincarotenoid detection can be realized by limiting the RRS skin carot-enoid detection method to the stratum corneum layer, as sketchedin Fig. 3, with its smaller subset of potentially confounding chro-mophores due to the absence of melanin and blood capillaries.Since the light penetration depth in the visible wavelength regionis typically only �400 lm, this can be realized by choosing thepalm of the hand or the heel of the foot as tissue probing sites; sites

Fig. 3. Illustration of skin morphology and excited tissue volume within thestratum corneum layer.

which have the thickest stratum corneum layer (between 500 and1000 lm) among skin tissue. A further important advantage ofpalm and heel tissue sites lies in the high carotenoid concentra-tions of the respective stratum corneum layers relative to otherskin sites. This is due to the high lipid-to-protein ratio found inthose tissue sites and to the fact that carotenoids are lipophilicby nature.

Skin carotenoid measurements based on simple reflection spec-troscopy appears to support these assumptions. Following dietarysupplementation, statistically significant changes could be ob-served in stratum corneum carotenoid content. Also, it could bedemonstrated that dermal carotenoid levels measured at varioustissue sites are highly correlated with serum carotenoid levels [11].

To validate the RRS carotenoid detection method, we comparedin this study in vivo RRS results for heel skin tissue with HPLC re-sults obtained after removal of a thin sliver of tissue. Heel tissuesites are ideally suited for this type of experiment since the stra-tum corneum skin layer is extremely thick, typically ranging be-tween 1 and 2 mm. Also, the stratum corneum is essentiallybloodless, so it is easy for participants to self-excise a tissue sampleof sufficient weight (10–50 mg) for HPLC analysis without any bod-ily damage.

The Raman instrumentation and associated light delivery/col-lection probe used in this study is shown schematically in Fig. 4.It consists of a portable, fiber-based, computer-interfaced instru-ment with high light throughput. The excitation light, which orig-inates from a 488 nm argon laser, is routed via fiber, band-passlaser line filter, and dichroic beam splitter to the tissue site of inter-est, as illustrated in Fig. 4 for a heel skin site. A typical measure-ment uses a 2 mm spot size and 10 mW power for 10 s exposuretime. The Raman-scattered light is collected in a 180� backscatter-ing geometry, then routed through the same beam splitter into aseparate detection path which contains a holographic Rayleighlight rejection filter, a fiber bundle, and a small high-throughputspectrograph coupled to a cooled two dimensional silicon arraycamera.

Typical heel skin spectra are obtained in near real time usingspecially developed software and displayed on a computer moni-tor, as shown in plots (b) and (c) in Fig. 2. The RRS carotenoid re-sponses ride as relatively sharp spectral features on an intense,spectrally broad autofluorescence background (Fig. 2, spectrumb), the latter clearly demonstrating the presence of chromophoresother than carotenoids in the light excitation path. Since the inten-sity of this background is nearly 100 times higher than the caroten-oid RRS signal, it is necessary to use a detector with high dynamicrange to retrieve the carotenoid RRS response with sufficient accu-racy. For data retrieval, the autofluorescence background is mod-eled with a fourth order polynomial and subtracted from the rawspectrum [20]. As a result one obtains the isolated RRS spectrumof the tissue site’s carotenoids featuring the three RRS lines charac-teristic for carotenoids (Fig. 2, spectrum c). Their spectral positionsare undistinguishable from pure b-carotene in liquid organic sol-vents (Fig. 2, spectrum a).

The RRS peaks result from contributions of all skin carotenoidspecies absorbing at the excitation wavelength. Since their abso-lute positions and bandwidth are indistinguishable at room (hu-man body) temperature, it allows us to use the absolute peakheight of the C@C signal at 1524 cm�1 as a measure of the overallcarotenoid content at the tissue site. Results can be obtained withsafe light excitation intensities [30], even though the RRS effect isrelatively weak.

In order to investigate the optical measurement conditions forhighest repeatability of in vivo results we first developed a skinphantom. Consisting of a mixture of glycerol, fine aluminum oxidepowder, Coumarin 540A dye, and b-carotene, it simulates thespectral response of skin in terms of light scattering, carotenoid

Fig. 4. Left: Schematic diagram of RRS instrument and fiber-based light delivery/detection probe module used in this study (left panel, not to scale). Right: Illustration of RRSmeasurement of heel skin tissue site.

Fig. 5. RRS intensity of C@C carotenoid peak, measured under excitation at 488 nmversus exposure time (a) and excitation power (b), respectively. Both plots demon-strate linearity of RRS response under the chosen experimental conditions.

44 I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49

RRS response, and spectral shape as well as strength of the overlap-ping autofluorescence. Measurements of the phantom carotenoidsrevealed an extremely high repeatability with a standard deviationbelow 1% for 10 consecutive measurements. Next we compared theresult with in vivo repeatabilities. Standard deviations betweenhuman subjects obtained in vivo for 10 consecutive RRS scans persubject were seen to fall into a wide range between 0.5% and15%, with an average around 4%. Since instrumentation aspectscan be excluded, the cause for the strong influence of the samplingsite on the reproducibility has to lie in a spatially non-uniform skincarotenoid distribution in those subjects. However, most subjectsappear to have a relatively uniform carotenoid distribution, witha resulting standard deviation of RRS intensities better than 4%.In each case, the repeatability figures can be held to a minimumby tightly controlling the positioning of the probe on the tissue siteof interest.

We checked the linearity of the RRS response by measuringadjacent skin tissue sites under variation of exposure times and la-ser power levels. The results of these experiments are summarizedin Fig. 5. In one experiment, we measured a selected heel skin sitefive times with fixed exposure time, then repeated the measure-ments five times while increasing the exposure time by 5 s in eachstep. In a second experiment, we measured a selected heel skin sitefive times again with fixed exposure time, then repeated the mea-surements six times while increasing the excitation laser power by2 mW in each step. The plots show the average of each measure-ment set along with its error bar (standard deviation), and astraight line through the points as guide to the eye. As can be seenfrom the plots, both relationships demonstrate nearly perfect lin-ear behavior, and are therefore in agreement with Eq. (1). Thisproves that under the used instrument and measurement condi-tions and limitation of the optical light paths to the outer stratumcorneum, the absorption and fluorescence effects of the residualchromophores are sufficiently small.

A further important parameter for validity considerations of theRaman method is the spectral position of the laser excitation wave-length within the carotenoid absorption band. It is desirable to useinstead of the argon laser a solid state laser for excitation. Currentsolid state laser lines useful for carotenoid excitation exist at 488and 473.2 nm. Because solid state lasers are far superior in termsof wall plug efficiency, are much more compact and generate less

heat as well as noise, we explored their RRS performance relativeto the argon laser. The bandwidths of the solid state laser linesare larger than that of the argon laser line. As a consequence, theresulting RRS carotenoid line widths are broader as well, and

I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49 45

may not be as clearly separable from the autofluorescence back-ground. Furthermore, the autofluorescence intensity increasesstrongly when changing excitation from 488 to 473.2 nm. Thequestion arises therefore, as to whether it is necessary to validatethe RRS method separately for each excitation condition.

To find out, we compared in a first step a compact solid state la-ser emitting at 488 nm with an argon laser emitting at the samewavelength. We found that the solid state laser RRS results arefully interchangeable with its ‘‘gas counterpart”, i.e. the carotenoidexcitation efficiency and spectral quality were essentially identical[I.V. Ermakov, W. Gellermann, unpublished results, 2004]. Thisshows that laser line width differences can be ignored betweenthe two sources. Next, we investigated changes in the carotenoidRRS response regarding spectral quality and repeatability whenusing a solid state laser at 473 nm as excitation source.

In Fig. 6 we compare the raw RRS carotenoid spectra obtained forthe palm of the same human subject with 473 nm and 488 nm exci-tation wavelength. All other excitation parameters, i.e. power den-sity at the skin, exposure time, etc., were kept unchanged. Thespectra were corrected for the spectral throughput of probe andspectrograph. As seen from Fig. 6, the C@C Raman peak componentsof the skin spectra have almost the same amplitude, as can be ex-pected from Raman theory, i.e. the resonance Raman excitation effi-ciency follows the absorption profile. The absorption coefficients of

Fig. 6. Fluorescence spectrum (top) and Raman spectrum (bottom) of skin tissuesite measured with excitation at 473 nm (curves a) and 488 nm (curves b). Theexposure times and excitation intensities in both cases were identical and equal to30 s and 10 mW, respectively. Strong fluorescence components occurring under473 nm and 488 nm excitation mask carotenoid RRS lines. Using a detector withhigh dynamic range and subtraction of background fluorescence, RRS signals can beretrieved. Raman spectrum excited with 473 nm appears to be much noisiercompared to that excited with 488 nm due to higher quantum noise associated withhigher fluorescence background.

all skin carotenoids change only by several percent when movingfrom 488 to 473 nm since all individual carotenoid absorption bandsare very wide and the change in excitation wavelengths occurs nearthe spectrally smooth absorption maxima. However, contrary to theRaman peak intensity, the intensity of the skin fluorescence back-ground increases by a factor of three for 473-nm excitation, andtherefore reduces the already small ratio between the useful C@C Ra-man signal to the autofluorescence background even further, toabout 0.3%. In comparison, at 488 nm excitation, the average ratioreaches 1%. The lowered Raman to fluorescence ratio deterioratesthe signal-to-noise ratio for the C@C RRS signal due to increased shotnoise, which now is about

ffiffiffi3p

times higher compared to excitationwith 488 nm. To compensate for this reduction, we increased theexposure (integration) time accordingly to determine whether it ispossible, in this way to retain the same signal-to-noise ratio achiev-able with 488 nm excitation. Indeed, the results showed that compa-rable performance figures can be obtained in either case, and that theincreased exposure time at 473 nm is a small sacrifice, outweighingall drawbacks associated with air-cooled argon laser or the currentlymore expensive solid state alternative at 488 nm.

For RRS/HPLC comparisons we recruited eight volunteer sub-jects (36–68 year age range) including both genders (4 males, 4 fe-males). All subjects were Caucasians with light to medium skinpigmentation, thus excluding any attenuation effects on RRS de-rived skin carotenoid levels which would occur if significant mela-nin concentration levels were to exist in the probed tissue volume.Also, all subjects were free of any skin disorders. The back of the leftheel of each subject was brought in contact with the optical win-dow of the Raman probe, marked, and measured three times. TheRRS results are listed in Table 1. Within 15 min of the RRS measure-ments, a sliver of tissue was self-excised at the center of the markedarea, using a sterile single-use regular blade. The skin samples wereimmediately weighed, deposited in an individual air tight vial, andstored at dry ice temperature for later extraction and HPLC analysis.For RRS/HPLC correlation we used Microsoft Excel’s LINEST functionand Sigmaplot (Systat Software, Inc.) graphics.

Biochemical carotenoid detection method

The gold standard HPLC technique for skin carotenoid analysisincludes effective transfer of the analyte from the tissue to a suit-able solvent (‘‘extraction”), spatial separation of different analytesfrom the extract mixture, and an objective, usually optics-baseddetection of the spatially separated chemical components. Whendealing with a complex mixture of several analytes, one sometimeshas to perform the second step several times using different HPLCtechniques to isolate a specific carotenoid, such as different col-umn types, direct/reverse phase HPLC, etc. Each step of HPLC issubject to a rigorous calibration where the known concentrationsof all chemicals in question are measured to find the effective ‘‘sys-tem response” for each. Thus, HPLC is a complex multi-step,destructive, and time consuming technique that needs relativelylarge tissue volume (�10 mg) to satisfy detection accuracy.

The collected skin samples were weighed after excision, imme-diately frozen and shipped at dry ice temperature for later HPLCanalysis [Craft Technologies, Inc., Wilson, NC]. The analysis proto-col involved the following procedures [31]. The skin samples wereweighed again and transferred into tubes with 430 ll PBS and70 lL collagenase (70 mg/mL PBS). They were incubated for 1 hat 37 �C. Glass beads, 2 mL of 50% methanol/50% THF and 100 lLof tocol (internal standard) were added. The samples were mixedto macerate the tissue and extract the carotenoids. Samples weresaponified overnight at 4 �C by addition of 100 lL 10% pyrogalloland 1 mL of 40% potassium hydroxide. Samples were extractedtwo times with 5 mL of hexane and dried over sodium sulfate.The combined hexane extract was dried in a SpeedVac™. The

Table 1RRS carotenoid measurements of human heel skin sites: statistics and comparison with HPLC derived results.

Subject ID Raman measurements, counts HPLC, ng/mg

Individual measurements Average STD STD, %

S1 38,600 34,668 33,600 35,623 2633 7.4 1.427S2 31,000 31,152 31,331 31,161 166 0.5 1.137S3 13,500 14,500 14,500 14,167 577 4.1 0.748S4 12,598 12,288 12,068 12,318 266 2.2 0.748S5 17,377 18,788 18,070 18,079 706 3.9 0.676S6 25,907 26,228 23,290 25,142 1612 6.4 1.263S7 13,064 15,311 11,464 13,280 1933 14.6 0.561S8 24,264 25,241 24,838 24,781 491 2.0 1.064

46 I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49

sample for astaxanthin analysis was extracted with acetone ratherthan CH3OH/THF and was hydrolyzed using cholesterol esterase. Itwas separated by HPLC using a Diol column.

The HPLC system consisted of a computer data system, an auto-sampler maintaining samples at 20 �C, a column heater at 31 �C,and a diode array detector (ThermoSeparation Products, Fremont,CA). The separation was performed isocratically on a SpherisorbODS2 column (3 lm, 4.0 � 250 mm with titanium frits, ES Indus-tries, West Berlin, NJ) protected by a Javelin™ guard column con-taining a similar stationary phase (Thermo Electron Corp,Bellefonte, PA). The mobile phase consisted of acetonitrile/diox-ane/isopropanol/triethylamine (80/15/5/0.1) at a flow rate of1.0 mL/min. The alcohol component contained 100 mM ammo-nium acetate. The diode array detector with light pipe flow cellwas programmed to monitor tocol and phytoene at 290 nm, phy-tofluene at 325 nm, and carotenoids at 450 nm.

Linear calibration curves were prepared for three concentra-tions of analytes which spanned the physiological levels of micro-nutrients in serum. The calibrants included astaxanthin, lutein,zeaxanthin, a-cryptoxanthin, b-cryptoxanthin, lycopene, a-caro-tene, b-carotene, phytoene, and phytofluene.

A typical HPLC chromatogram is presented in Fig. 7, with theabsorption of the carotenoid extract at the carotenoid maximumabsorption wavelength, shown as a function of time. The area undereach peak of the chromatogram is associated with the concentrationof a specific carotenoid or group of carotenoids. Carotenoid speciesare separated spatially in HPLC instrumentation due to the specificand different retardation times as they traverse the length of the col-umn. The degree of retardation depends on the chemical structureand mass of the carotenoid species and leads to the specific retentiontime (time when a specific analyte elutes) at the chromatogram.Good signal-to-noise ratios for the carotenoid peaks allow us todetermine their concentration in skin with high accuracy.

Fig. 7. Typical HPLC chromatogram of caroten

Results and discussion

The HPLC results for all eight subjects are reported in Table 2 asamounts in ng/1 mg tissue for each carotenoid species found in thesample. The particular HPLC analyses concentrated on 13 majorcarotenoid species, namely lutein, zeaxanthin, cis-lutein/zeaxan-thin, a-cryptoxanthin, b-cryptoxanthin, trans-lycopene, cis-lyco-pene, a-carotene, trans-b-carotene, cis-b-carotene, canthaxanthin,phytoene, and phytofluene.

Phytoene and phytofluene were excluded from further consider-ation since they do not contribute to the RRS signal under blue lightexcitation. These shorter-chain carotenoids absorb in the UV rangesuch that for effective Raman excitation one has to use a narrow line-width UV source [30]. It is worthwhile to note that the UV activecarotenoids are found in significant amounts in the skin [2]. Thisstudy confirms the finding, with the tissue sample of one subject(S4) containing as much as 48% phytoene relative to the total carot-enoid content. The column farthest to the right in Table 2 lists the to-tal concentration of skin carotenoids absorbing in the visible. Sincethese contribute to the Raman signal, they are compared with theRRS data of Table 1.

Comparing the two sets of data, one needs to be aware of thefact that either set is prone to a number of systematic errors. Forinstance, any HPLC analysis is set up to ‘‘see” a definite and lim-ited subset of the species, thus not including minor carotenoidsand a potentially large class of oxidized carotenoids. This is incontrast to the RRS effect which would pick up any C@C signalfrom those unaccountable molecules as well. On the other hand,under single wavelength excitation, RRS measurements give a va-lue, R, comprising the contribution of a number of carotenoidspecies,

R ¼ Iexc

X

i

riNi; ð2Þ

oids extracted from heel skin tissue site.

Tabl

e2

Caro

teno

idco

mpo

siti

onof

hum

ansk

inas

mea

sure

dw

ith

HPL

C.

Subj

ect

IDTi

ssu

ew

eigh

t,m

gLu

tein

Zeax

anth

inci

s-Lu

tein

/ze

axan

thin

a-C

rypt

oxan

thin

b-C

rypt

oxan

thin

tran

s-Ly

cope

ne

cis-

Lyco

pen

ea

-Car

oten

etr

ans-

b-C

arot

ene

cis-

b-C

arot

ene

Phyt

oen

ePh

ytofl

uen

eTo

tala ,n

g/m

g

S138

.20.

198

0.08

20.

008

0.04

50.

233

0.20

00.

110

0.08

50.

314

0.12

40.

262

0.04

01.

427

S261

.30.

072

0.03

90.

007

0.02

50.

111

0.17

40.

129

0.08

20.

351

0.14

70.

130

0.05

51.

137

S345

.90.

076

0.05

10.

007

0.02

10.

052

0.19

00.

119

0.05

30.

116

0.06

30.

141

0.03

50.

748

S423

.20.

064

0.05

00.

003

0.01

80.

041

0.24

30.

140

0.05

00.

088

0.05

10.

683

nd

0.74

8S5

45.6

0.12

70.

052

0.00

60.

032

0.10

10.

137

0.08

50.

025

0.09

00.

021

0.01

3n

d0.

676

S614

.20.

142

0.03

90.

017

0.03

30.

169

0.35

00.

262

0.04

20.

157

0.05

20.

059

nd

1.26

3S7

47.6

0.10

90.

031

0.00

60.

020.

052

0.11

50.

084

0.03

30.

084

0.02

70.

015

nd

0.56

1S8

80.

106

0.04

60.

009

0.03

70.

153

0.27

60.

232

0.02

40.

132

0.04

90.

042

nd

1.06

4

nd

–n

otde

tect

able

.C

anth

axan

thin

not

dete

ctab

lein

all

sam

ples

.a

Tota

lco

nce

ntr

atio

nof

caro

ten

oid

mol

ecu

les

insk

insa

mpl

esex

clu

din

gU

V-a

bsor

bin

gca

rote

noi

dsph

ytoe

ne

and

phyt

oflu

ene.

I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49 47

where the contribution of each species is not strictly proportional toits concentration Ni but, in general would have to be corrected forslightly differing Raman excitation cross sections ri. In the currentmodel, we assume that all carotenoid species in the skin contributeto the Raman signal with the same strength,

R ¼ IexcrX

i

Ni; ð3Þ

where r is effective or average Raman excitation cross section/effi-ciency. To justify this approach, we note that the RRS section ri(k)usually follows the absorption profile ei (k), and that the specificabsorption coefficients ðemax

i Þ and ðemaxi Þ differ from one another by

no more than 10% [32], so emaxi � emax

j and ri � rj = r.On the optical side, we observed that human skin is not always

uniform in carotenoid content, even though this in not detectableby the naked eye. This is demonstrated by Fig. 8 and Table 3, whichcontain the results of 10 Raman measurements for three of the vol-unteer subjects described above. Five RRS measurements were ta-ken at the same skin site, with the optical probe not movingbetween measurements; subsequently, five measurements wereperformed on adjacent skin sites. The optical probe was intention-ally removed after each measurement, and replaced prior to com-pleting the following measurement. For each series ofmeasurements, the standard deviation was calculated. Similar dataon the skin phantom is also reported in Table 3 for reference pur-poses. The skin phantom can be considered as ideal sample interms of homogeneity, as it delivers the best repeatability with arelative standard deviation better than 0.5%. For living skin, therepeatability is still excellent (about 1%) as long as the opticalprobe remains on the selected site of the skin. However, whenremoving the probe and placing it back onto the site, we effectivelychange the sampling site due to the positioning error. This error, inour estimate did not exceed the diameter of the sampling spot(�2 mm), but causes a significant and reproducible span in stan-dard deviation of the RRS measurements, varying between 1%and 4%. The consistent and relatively high standard deviation ob-served in subject ‘‘B” serves as indirect evidence for a non-uniformcarotenoid distribution in human skin in some subjects. If thisinhomogeneity is on the same scale as the sampling mismatch be-tween RRS measurements, lower repeatability figures result inthese cases.

In spite of the presence of these unavoidable errors andassumptions, we obtained excellent correlations between Ramanand HPLC data, as seen from Fig. 9, with R as high as 0.95 orR2 = 0.91. Note that the regression line even passes through the ori-gin, proof that the influence of the additional chromophores to theRRS method can be ignored under the used instrument configura-tion and measurement parameters. Over a wide range of physio-logical concentrations of skin carotenoids, an excellent linearitytherefore exists between skin carotenoid levels measured withthe RRS method versus HPLC. This result validates RRS spectros-copy as an objective and accurate method for skin carotenoidmeasurements.

We can use the HPLC–Raman correlation results to directly cal-ibrate our RRS instrument in carotenoid concentration unitsAccording to the regression analysis, the cumulative skin caroten-oid content, C, measured in lg/g of skin tissue is linked to theheight of the C@C carotenoid Raman peak, I, in the instrument via

C ½lg=g� ¼ 4:3 � 10�5 � I ½counts�: ð4Þ

Integrating Eq. (4) with our data acquisition software, it is possibleto obtain skin carotenoid content in convenient lg/g units in realtime.

Fig. 8. Examples of standard deviations for five consecutive carotenoid RRSmeasurements obtained for a tissue phantom and for three subjects, obtained withoptical probe fixed on the tissue site (closed circles) and optical probe removedbetween measurements (open circles). Test–retest RRS results suggest differentdegrees of carotenoid homogeneity in living human skin. With fixed probe,standard deviations for all subjects are small and very comparable. With reappliedprobe, values of standard deviations are higher and differ significantly.

Table 3Test–retest skin carotenoid RRS measurements obtained for tissue phantom and threesubjects in five consecutive measurements per sample site. The data suggest differentdegrees of carotenoid homogeneity in human skin.

Meas.conditions

Probe fixed on tissue site Probe removed betweenmeas.

Measurements,counts

STD,%

Measurements,counts

STD,%

Phantom 25,701 0.31 25,117 0.4126,005 25,93425,946 25,35526,373 26,39825,296 26,094

Subject A 52,619 0.98 47,633 1.048,075 52,74952,397 53,21549,218 48,32547,326 50,219

Subject B 27,938 0.92 25,441 3.8428,777 34,71531,304 29,26229,840 37,42630,614 24,014

Subject C 29,474 1.05 31,184 2.1427,461 31,82831,105 25,00328,823 26,05027,539 27,755

Fig. 9. Heel skin carotenoid levels of eight subjects measured with RRS methodin vivo, versus HPLC results of subsequently excised tissue samples. Dotted lineindicates corresponding linear regression crossing the origin, with resulting highcorrelation coefficient R = 0.95.

48 I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49

Conclusion

In conclusion, we directly compared RRS based in vivo skincarotenoid detection with the biochemical gold standard methodof HPLC. Even though the methodologies are totally different, aremarkably high correlation (R = 0.95) is obtained for skin tissuesites in which the light penetration is confined to the stratum cor-neum layer. With the specified, carefully chosen light excitationand detection conditions, the influence of intrinsically uncontrolla-ble, potentially confounding factors on the RRS carotenoid re-sponse can be held to a negligible subset of chromophores. Theremaining autofluorescence background in the total dermal opticalresponse therefore can be treated as a simple offset in the reduc-

tion algorithm for the RRS carotenoid levels under these measure-ment conditions. This holds for two tested excitation wavelengthsof 488 nm and 473.2 nm, and by inference, for any wavelengthlying on the long-wavelength shoulder of the carotenoid absorp-tion range.

Along with previously reported, more indirect evidence for areasonably good correlation between RRS skin carotenoid levelsand levels in serum, these correlation experiments validate theRRS method and establish it as a rapid, non-invasive optical meth-odology for measurement of carotenoid antioxidant status in vivo.It is possible to use the method for rapid quantitative assessmentof dermal carotenoid levels in clinical and field settings in largepopulations under the specified experimental conditions. Thisshould make it possible to apply the method as biomarker for fruitand vegetable intake in large-scale dietary intervention studies aswell as in clinical studies investigating correlations between der-mal carotenoid levels and exposure to oxidative stress.

The observed very high correlation for tissue sites with thickstratum corneum indicates that it may be possible to extend der-mal RRS carotenoid detection also to skin tissue sites with highercontributions of additional chromophores, such as abdominal skin.This would come with a penalty of lower accuracy, and would haveto be validated in a similar RRS/HPLC correlation study for thatmore complex tissue. However, this might well be acceptable inclinical studies looking only for relatively large changes in dermalcarotenoid content between subjects, or in the same subjects overtime in response to oxidative stress exposure. Lastly, the obtainedvalidation results will make it possible to cross-correlate otheroptical carotenoid detection methods, such as reflectance spectros-copy, with the Raman method to evaluate their accuracy in themeasurement of skin carotenoids.

References

[1] N. Krinsky, S.T. Mayne, H. Sies (Eds.), Carotenoids in Health and Disease,Marcel Dekker, New York, 2004.

[2] T.R. Hata, T.A. Scholz, I.V. Ermakov, R.W. McClane, F. Khachik, W. Gellermann,L.K. Pershing, J. Invest. Dermatol. 115 (2000) 441–448.

[3] F. Böhm, J.H. Tinkler, T.G. Truscott, Clin. Cancer Res. 1 (1995) 98–99.[4] C.S. Foote, R.W. Denny, J. Am. Chem. Soc. 90 (1968) 6233–6235.[5] A. Farmillo, F. Wilkinson, Photochem. Photobiol. 18 (1973) 447–450.[6] P.F. Conn, W. Schalch, T.G. Truscott, J. Photochem. Photobiol. B 11 (1991) 41–

47.[7] L.C. Chen, L. Sly, C.S. Jones, R. De Tarone, L.M. Luca, Carcinogenesis 14 (1993)

713–717.[8] Y.M. Peng, Y.S. Peng, Y. Lin, T. Moon, D.J. Roe, C. Ritenbaugh, Nutr. Cancer 23

(1995) 233–246.

I.V. Ermakov, W. Gellermann / Archives of Biochemistry and Biophysics 504 (2010) 40–49 49

[9] M.R. Prince, J.K. Frisoli, Am. J. Clin. Nutr. 57 (1993) 175–181.[10] M.M. Mathews-Roth, Ann. NY Acad. Sci. 691 (1993) 127–138.[11] W. Stahl, U. Heinrich, H. Jungmann, H. Sies, H. Tronnier, Am. J. Clin. Nutr. 71

(2000) 95–798.[12] J. Lee, S. Jiang, N. Levine, R. Watson, Proc. Soc. Exp. Biol. Med. 223 (2000) 170–

174.[13] E. Giovannucci, J. Natl. Cancer Inst. 91 (1999) 317–331.[14] H.I. Krinsky, S.M. Deneke, J. Natl. Cancer Inst. 69 (1982) 205–210.[15] D.P.T. Steenvorden, G.M.J. Beijerbergen van Henegouwen, J. Photochem.

Photobiol. B 41 (1997) 1–10.[16] W. Stahl, U. Heinrich, H. Jungmann, H. Tronnier, H. Sies, Methods Enzymol. 319

(2000) 494–502.[17] S. Alaluf, U. Henrich, W. Stahl, H. Tronnier, S. Wiseman, J. Nutr. 132 (2002)

399–403.[18] I.V. Ermakov, M. Sharifzadeh, P.S. Bernstein, W. Gellermann, in: J.T. Landrum

(Ed.), Carotenoids – Physical, Chemical and Biological Functions andProperties, CRC Press, Atlanta, GA, 2009.

[19] P.S. Bernstein, M.D. Yoshida, N.B. Katz, R.W. McClane, W. Gellermann, Invest.Ophthal. Vis. Sci. 39 (1998) 2003–2011.

[20] I.V. Ermakov, M.R. Ermakova, R.W. McClane, W. Gellermann, Opt. Lett. 26(2001) 1179–1181.

[21] W. Gellermann, R.W. McClane, N.B. Katz, P.S. Bernstein, Method and Apparatusfor Noninvasive Measurement of Carotenoids and Related ChemicalSubstances in Biological Tissue, US Patent #6,205,354, 2001.

[22] C.R. Smidt, W. Gellermann, J.A. Zidichouski, Fed. Am. Soc. Exp. Biol. J. 18A(2004) 480.

[23] I.V. Ermakov, M.R. Ermakova, W. Gellermann, J. Lademann, J. Biomed. Opt. 9 (2)(2004) 332–338.

[24] W. Gellermann, J.A. Zidichouski, C.R. Smidt, P.S. Bernstein, in: L. Packer, U.Obermueller-Jevic, K. Kraemer, H. Sies (Eds.), Carotenoids and Retinoids, AOCSPress, Champaign, Illinois, 2005 (Chapter 6).

[25] C.R. Smidt, D.S. Burke, Current Topics in Nutraceutical Research 2 (2004) 79–91.

[26] S. Rerksuppaphol, L. Rerksuppaphol, J. Med. Assoc. Thai 89 (2006) 1206–1212.[27] S.D. Bergeson, J. B Peatross, N.Y. Eyring, J.F. Fralick, D.N. Stevenson, S.B.

Ferguson, J. Biomed. Opt. 13 (2008) 044026-1–044026-6.[28] G.M. Chan, C. Rau, W. Gellermann, M. Ermakova, Retinopathy of prematurtiy

and carotenoids in human milk fed infants, Abstract, 2006 Meeting of theAmerican Academy of Pediatrics, Washington, DC, 2006.

[29] Y. Koyama, in: G. Britton, S. Liaaen-Jensen, H. Pfander (Eds.), Carotenoids,Spectroscopy, vol. 1B, Birkhäuser, Basel, 1995, pp. 135–146,.

[30] I.V. Ermakov, M. Sharifzadeh, M.R. Ermakova, W. Gellermann, J. Biomed. Opt.10 (6) (2005) 064028.

[31] The methods are essentially the same as described in N.E. Craft,Chromatographic Techniques for Carotenoid Separation, Current Protocols inFood Analytical Chemistry F2.3.1–F2.3.15, John Wiley & Sons, Inc., 2001.

[32] G. Britton, S. Liaaen-Jensen, H. Pfander (Eds.), Carotenoids, vol. 1B, BirkhaeuserVerlag, Basel, Switzerland, 1995, p. 57.