Detecting Biochemical Changes in the Rodent Cervix During Pregnancy

Using Raman Spectroscopy

ELIZABETH VARGIS,1 NAOKO BROWN,2 KENT WILLIAMS,3 AYMAN AL-HENDY,4 BIBHASH C. PARIA,2

JEFF REESE,2 and ANITA MAHADEVAN-JANSEN1

1Department of Biomedical Engineering, Vanderbilt University, VU Station B, Box 351631, Nashville, TN 37235, USA;2Division of Neonatology, Vanderbilt University Medical Center, Nashville, TN, USA; 3Department of Gastroenterology,Nationwide Children’s Hospital, Columbus, OH, USA; and 4Department of Obstetrics and Gynecology, Meharry Medical

College, Nashville, TN, USA

(Received 14 November 2011; accepted 1 March 2012; published online 13 March 2012)

Associate Editor James Tunnell oversaw the review of this article.

Abstract—The goal of this research is to determine whetherRaman spectroscopy (RS), an optical method that probes thevibrational modes of tissue components, can be used in vivoto study changes in the mouse cervix during pregnancy. Ifsuccessful, such a tool could be used to detect cervicalchanges due to pregnancy, both normal and abnormal, inanimal models and humans. For this study, Raman spectrawere acquired before, during and after a 19-day mousegestational period. In some cases, after Raman data wasobtained, cervices were excised for structural testing andhistological staining for collagen and smooth muscle. Vari-ous peaks of the Raman spectra, such as the areas corre-sponding to fatty acid content and collagen organization,changed as the cervix became softer in preparation for laborand delivery. These findings correspond to the increase incompliance of the tissue and the collagen disorganizationvisualized with the histological staining. The results of thisstudy suggest that non-invasive RS can be used to studycervical changes during pregnancy, labor and delivery andcan possibly predict preterm delivery before overt clinicalmanifestations, potentially lead to more effective preventiveand therapeutic interventions.

Keywords—Raman spectroscopy, Structural testing, Biomed-

ical photonics, Gestation, Parturition.

INTRODUCTION

The physiologic changes during pregnancy that re-sult in labor and delivery are part of complicatedprocesses that are not fully understood. Current stud-ies correlate maternal steroid hormonal changes and

paracrine molecules with gestation and parturition(pregnancy and labor).6,19,29 These small moleculesregulate many of the changes in the tissues of both themother and baby by altering the biochemical compo-sition of the cervix to prepare for labor and deliveryand by promoting fetal development. However,although pregnancy has been studied for centuries, theinteractions of these molecules and the exact mecha-nisms governing the regulation and progression ofpregnancy remain unknown. Researching human par-turition is hindered since direct testing on pregnantwomen is limited and the hormonal pathways regu-lating animal pregnancy and labor vary substantiallyfrom those of humans. Much research into thematernal reproductive tract is currently focused onidentifying important markers that trigger specificchanges during pregnancy. Yet, it may be equallyimportant to study the downstream effects such mol-ecules have on maternal tissues, which changethroughout pregnancy in preparation for birth.

During pregnancy, there are many well-documentedphysical changes in maternal tissues. The cervix, forexample, is initially rigid and stays tightly closed toprotect the fetus within the uterus and withstand itsincreasing weight. Closer to delivery, the cervixundergoes ripening, leading to its effacement (thinning)and dilation. As the uterus contracts with increasingfrequency and intensity, the fetus is expelled throughthe softened cervix. These contractions also speed upthe dilation process, with each contraction dilating thecervix as much as 30%.24 Many of these propertiesrequire changes to the components of the extracellu-lar matrix (ECM) of the tissues, which consists offibrillar collagen, proteoglycans, hyaluronan, elastin

Address correspondence to Anita Mahadevan-Jansen, Depart-

ment of Biomedical Engineering, Vanderbilt University, VU Station

B, Box 351631, Nashville, TN 37235, USA. Electronic mail: anita.

mahadevan-jansen@vanderbilt.edu

Annals of Biomedical Engineering, Vol. 40, No. 8, August 2012 (� 2012) pp. 1814–1824

DOI: 10.1007/s10439-012-0541-4

0090-6964/12/0800-1814/0 � 2012 Biomedical Engineering Society

1814

and water.15,24 These components undergo a complexbiochemical reorganizing and remodeling processthroughout pregnancy, including a realignment of thecollagen structure.34,44

A number of small molecules have been implicatedin regulating such processes.6 Factors, such as estrogenand progesterone,6,19 platelet-activating factor,38

prostaglandins16,18 and interleukin-8,37 may all play arole in controlling the process that prepares the cervixfor labor and delivery. Other research groups havefocused on detecting changes in some of these indi-vidual factors.9,14,32 However, few correlations havebeen found between specific molecules and biochemicalchanges in cervical tissues. Gestation and parturitionare complex processes and it is likely that multipleinteracting pathways control the progression of preg-nancy, resulting in uterine contractions and cervicalripening.6 As these pathways are discovered, healthcare providers may be able to more accurately deter-mine pre-symptomatically when a woman is about togive birth and if necessary, intervene to delay pretermlabor and premature birth.

Here we propose using an optical method to quan-tify changes downstream of small biomolecules in thecervix. Rather than removing tissue or measuring howa single molecule changes in blood or bodily secretions,optical methods can provide non-invasive, real-time,automated measurements from bulk tissues as a whole.These measurements can contain a wide range ofinformation, including collagen content and organi-zation, changes due to hormonal fluctuations, and ameasure of cervical length. For example, Garfieldet al.12 developed electromyography to measurechanging action potentials of the uterus throughoutpregnancy. Kuon et al.23 used a collascope to measurethe autofluorescence of the cervix and correlated theresults with the cervical surface area, finding thatduring pregnancy, surface area increases as autofluo-rescence increases. Jokhi et al.17 utilized electricalimpedance spectroscopy to measure cervical resistivityto determine the onset of labor, with predictive valuesof 0.65–0.8. Second-harmonic generation has also beenused to measure changes in collagen structure duringgestation.2 Other groups have applied optical methods,such as Fourier-transform infrared (FTIR),39 reso-nance,40 and light-scattering26 spectroscopy to measuresalivary changes and the presence of nitric oxide orfetal nucleated red blood cells, respectively. To ourknowledge, this is the first report using Raman spec-troscopy (RS), a sensitive optical method, to studybiochemical changes in tissue during pregnancy.

Raman spectroscopy is based on the Raman effectby which energy can be exchanged between incidentphotons and scattering molecules. When an incidentphoton collides with certain molecules, energy may be

transferred either from the molecule to the photon orvice versa. The energy differences of the scatteredphotons are indicative of the molecules set intovibration. A Raman spectrum then consists of a seriesof peaks, which represent the different vibrationalmodes of the scattering molecules. These peaks arespectrally narrow and molecular-specific, such that theobserved peaks may be associated with specific bondsin specific molecules. Many biological molecules havedistinguishable spectra, so that one can determine atissue’s biochemical composition from its Ramanspectrum. For example, one relevant biochemicalchange during pregnancy is the softening and ripeningof the cervix due to changes in collagen. This change,among other changes in elastin, glycogen and watercontent, can be detected with RS.20,21,41,42 Otherchanges that RS is likely to be sensitive to are changesin collagen cross-linking, water content and hormonalvariations, many of the important factors that fluctu-ate during pregnancy, labor and delivery, which havebeen observed in other studies.1

We have previously demonstrated the potential ofRS to detect subtle changes in tissue biochemistry fromthe cervix.21,41 This technology has been applied todetect cervical precancer lesions in vivo and can dis-tinguish between normal, inflammation, low-gradedysplasia and high-grade dysplasia areas with classifi-cation accuracy rates of over 97%. Also, it has beendemonstrated that RS is sensitive to subtle changes inhormonal levels, the permanent effects of pregnancyand delivery, as well as malignancy-associatedchanges.42 The only report on using RS for obstetricsresearch is for studying preeclampsia, an abnormalgestation-related increase in maternal blood pressure.4

However, previous work indicates that the technologyand knowledge exists to develop RS as a tool forstudying the cervix throughout pregnancy.12,21,23,41 Inthis pilot study, RS was used to monitor andunderstand cervical changes in normal mice duringpregnancy.

The goal of this paper is to characterize changes inthe cervix of normal mice throughout pregnancy usingRS. To achieve this goal, two sets of studies wereconducted. First, Raman spectra were acquired fromthe cervix of non-gravid (not pregnant) mice to de-velop a baseline of the normal cervix during murineestrus cycles. Once the effect of normal cycling wascharacterized, the primary study was conducted andspectra were acquired from normal mice at multipletime points before, during and after pregnancy. Toverify the changes observed in the Raman spectra,after Raman data was acquired, the cervix was har-vested to understand the structural properties of thetissue and to visualize cellular changes with histo-logical staining. Logistic regression algorithms and

Biochemical Changes in the Cervix 1815

statistical analyses were then used to determine if sig-nificant differences existed in the Raman measure-ments, force–displacement testing and histologicalstaining as a function of pregnancy time point.

MATERIALS AND METHODS

Animals and Tissue Collection

Adult virgin female and male mice (strain:CD1(ICR)) were purchased from Charles River Lab-oratories, Raleigh, NC. They were maintained in a12 h:12 h light/dark cycle in the Vanderbilt UniversityAnimal Facility with unlimited access to water andfood. All animal maintenance, handling and proce-dures were performed in accordance with NationalInstitutes of Health guidelines for the care and use oflaboratory animals and were approved by the Van-derbilt University Institutional Animal Care and UseCommittee (IACUC). Two sets of experiments wereconducted: a study to observe changes in the cervix dueto hormonal changes alone and a study to examinechanges due to pregnancy. Non-pregnant (or non-gravid (NG)) mice with normal cyclicity were evalu-ated first to examine the effect of hormonal changesduring the estrous cycle on the cervix. Raman mea-surements were obtained from 3 sets of NG mice, with3–4 mice in each set (n = 11). Measurements weretaken from each set for 5 days straight over the course of3 weeks to acquire data. The stages of the estrous cyclethat were measured include proestrus, estrus, metestrusand diestrus.8 Prior to the Raman measurements, vagi-nal washings were obtained by rinsing the vaginal canalwith saline and the estrous stage was determined byidentifying cell types and their relative abundancepresent in smears under a stereo-microscope.

Timed matings were carried out by housing onenormal fertile male with three normal females to studythe cervical changes related to pregnancy. On the fol-lowing day, females were evaluated at 0900 for thepresence of a copulatory plug with gestation day 1defined by the presence of the plug. Data from anymouse that was not visually pregnant by day 12 wasnot used in any of the following results. Animals in thiscolony typically give birth on the evening of gestationday 19. A set of mice was followed before, during andafter pregnancy (n = 47). Female mice were anaes-thetized with isoflurane and Raman measurementswere acquired prior to mating (NG), on days 1, 4, 12,15, 18, and 19 of pregnancy, and on post-partum day 1(PP1). At the same time points, Raman measurementswere acquired and then the mouse was euthanized(overdose of isoflurane inhalation) to collect the cervixfor force–displacement testing or histological staining.The cervix was dissected and trimmed to 5 mm in

length to maintain consistency for the force–displace-ment testing. Cervices were dissected with the uterusand vaginal tissue still attached in order to control fororientation during sectioning and imaging.

Raman Instrumentation and Data Processing

Raman spectra were collected in vivo using a por-table RS system consisting of a 785 nm diode laser(I0785MM0350MS, Innovative Photonic Solutions,Monmouth Junction, NJ), a beam-steered fiber opticprobe without its casing (785 nm, Emvision, Loxa-hatchee, FL), an imaging spectrograph (Holospecf/1.8i, Kaiser Optical Systems, Ann Arbor, MI), and aback-illuminated, deep-depletion, thermo-electricallycooled charge-coupled device (CCD) camera (Pixis256BR, Princeton Instruments, Princeton, NJ), allcontrolled with a laptop. The fiber optic probe (10 cmlong and approximately 2.1 mm in diameter) delivered80 mW of incident light to the tissue at an integrationtime of 2–3 s with all room lights and the computermonitor turned off. The system provided a spectralresolution of 8 wavenumbers (cm21).

Spectral calibration of the system was performedeveryday using a neon-argon lamp and naphthaleneand acetaminophen standards to correct for day-to-day variations. A National Institute of Standards andTechnology (NIST)-calibrated tungsten lamp was usedto account for the system’s wavelength-dependentresponse. The spectra were processed for fluorescencesubtraction and noise smoothing using the modifiedpolynomial fit and Savitzky-Golay methods, describedpreviously.25 Following data processing, each spec-trum was normalized to its mean spectral intensityacross all Raman bands to account for intensity vari-ability. The code developed to process the data mini-mized the error introduced by the fluorescencesubtraction and the background removal. The samecode with the same parameters was used for all thesamples such that the same amount of minimal errorwas introduced to all the spectra. Finally, the spectrawere normalized in order to perform relative compar-isons across the data.

Tissue Structural Properties

Structural properties of excised cervical tissues wereevaluated using modifications of the methods devel-oped by Harkness and Harkness13 to correlate struc-tural tissue properties to biochemical Ramanmeasurements. Briefly, the cervix was mounted ontotwo hooks inserted through the cervical canal. Thehooks were made of stainless steel (22 Ga, 316L AISIgrade), approximately 0.8 mm thick and 7 mm long.One hook was attached to a stationary hook and the

VARGIS et al.1816

other hook was attached to a force transducer and amicrometer. Tissues were incubated in a water-jack-eted bath containing Krebs solution at 37 �C bubbledwith 95% O2/5% CO2. The force transducer was cal-ibrated using standard weights to set the minimum(~21.03 V) to 0 g and the maximum (~9.04 V) to 50 g.

Baseline cervical dilation was determined byincreasing the distance between the two hooks until asmall increase (~0.1 g, remained constant) was de-tected with the force transducer. The cervix was incu-bated in Krebs solution and held at this displacementfor 10 min before measurements were recorded. Thenthe inner diameter of the cervix was increased in 1-mmincrements at 4-min intervals to produce cervical dis-tention. The amount of force at each 1-mm displace-ment and every 4 min was recorded (displacement rate:0.0042 mm/s). The diameter was increased until thetissue tore. Force was plotted as a function of cervicaldiameter. A moving average function was used tocalculate the slope of the force–displacement curve;this slope was then used as cervical stiffness. A steeperslope indicates an increased resistance to stretchor a reduced compliance. Results are displayed asmean ± standard error (SE) from five independentsamples in each group.

Statistical Analysis

Cycling status was determined by the cells observedin vaginal smears. Pregnancy dating was determined bythe presence of a cervical plug.8 These are two goldstandards often used, however, there can be errorsassociated with such techniques based on the expertiseof the user. Measurements were excluded from anymouse originally enrolled in the study that ended upnot being pregnant.

For this study, a logistic regression method calledSparse Multinomial Logistic Regression (SMLR) wasused to tease out subtle differences among spectraacquired from different samples.22 In brief, SMLR is aBayesian machine-learning framework that computesthe posterior probability of a spectrum belongingto each tissue class based on a labeled training set.For these analyses, a composite spectrum averagingRaman measurements from each mouse at each timepoint was used. A range of input parameters to SMLRwas tested. The settings that provided the most accu-rate classification while also maximizing sparsity werea Laplacian prior, a direct kernel, a lambda value of0.01, and no additional bias term. To avoid bias,SMLR employs a leave-one-mouse-out at each timepoint. The algorithm uses all but one of the labeledspectra to create a classification algorithm that is usedto classify the left out spectrum. In the next iteration, adifferent spectrum is left out, creating a new training

algorithm that is then used to classify the left outspectra. This process is completed until all of thespectra are classified. A Student’s t test was performedto compare individual peaks from d4 to all the othertime points. A p-value <0.1 was considered significant.

For the force–displacement testing studies, a Krus-kal–Wallis one-way analysis of variance (ANOVA)was performed, followed by Dunn’s Method for posthoc analysis and pairwise comparison to control forerrors among the same sample group. The Kruskal–Wallis is the nonparametric alternative to ANOVAand has been used in previous tissue structural studiesof the rodent cervix.34

Tissue Processing and Masson’s Trichrome Staining

Cervical tissues were removed and stained to cor-relate the Raman spectra to important molecular andcellular changes resulting from pregnancy. First, ex-cised cervico-uterine tissue sections were immediatelysnap-frozen in liquid nitrogen after removal and storedat 280 �C for later use. Next, the tissues wereembedded in optimal cutting temperature (OCT)embedding medium (Tissue-Tek, Quiagen, Valencia,CA). Embedded tissues were cryosectioned into 12 lmthick slices and thaw-mounted onto poly-L-lysinecoated slides. Tissues were then fixed in Bouin’s fixa-tive, followed by Masson’s trichrome staining perprotocol (Sigma-Aldrich, St. Louis, MO). Trichromestaining labels collagen fibrils (blue), nuclei (black),and smooth muscle and cytoplasm (red). Tissuesections were imaged and recorded under 209

magnification.

RESULTS

Cycling Study

The first step of this study was to determine if RS issensitive to changes in the NG cervix due to hormonalcycling alone. Raman spectra (120 spectra from n = 11mice) were acquired from the cervix of NG mice atfour different cycling time points: proestrus, estrus,metestrus, and diestrus (Fig. 1). Figure 1 shows somevariations among the different time points. Themajority of these differences are in the region between1200 and 1400 cm21. This area has been shown inprevious work to tentatively correspond to collagen,amide III, and lipid content.3,11,27,28 Logistic regression(SMLR) was used to determine if any differences ex-isted due to cycling alone. Ninety-two of the 120spectra or 77% classified correctly, meaning thatSMLR was able to correctly classify the spectra asbelonging to its corresponding time point within theestrous cycle only 77% of the time (Table 1).

Biochemical Changes in the Cervix 1817

Pregnancy Study

After characterizing the NG cervix using RS,Raman spectra were acquired from multiple timepoints during pregnancy (day 1, 4, 12, 15, 18, 19, andPP1, Fig. 2, 317 spectra from n = 47 mice). For somemice, acquiring spectra before mating or on day 1(after the presence of the copulatory plug) led to mis-carriage (discussed below). Accordingly, measurementswere acquired starting on day 4 of the gestationalperiod. Within these spectra, there are many regionsthat appear different across the entire range of 990–1800 cm21. Differences are most visible in the spectrabetween NG and day 19 mice. Figure 3 shows changesin peak intensity (a–c) and peak width (d) as a functionof pregnancy in four important peaks. It has beendemonstrated previously that some of these regionscorrespond to certain biochemical structures, such asproline and tyrosine (Fig. 3a, 1200 cm21),3,27,28 CH2

and lipids (Fig. 3b, 1308 cm21),11 and the CH3CH2

bending modes found in protein side chains of multipletissue types (Fig. 3c, 1450 cm21).11,27 Figure 3d showsthe full width at half max (FWHM, ± S.E.) of the1650 cm21 peak, which has tentatively been assignedto amide I, lipids, and collagen.28 For this region, theFWHM, as opposed to peak intensity, is a more

accurate indicator of the organization and polarizationof the amide I bonds in the tissue. A Student’s t testwas performed to determine if there was any statisticalsignificance between measurements from day 4 and theother time points, noted with a star. PerformingSMLR on spectra across 6 time points: NG, early(days 1, 4 and 12), day 15, day 18, day 19 and PP1,resulted in a classification accuracy of over 94% (Ta-ble 2). Combining the early time points (d1, d4, d12)led to an increase in classification accuracy. Non-gravid data was used from the cycling study and day 1data was used from any mouse that maintained itspregnancy after a day 1 measurement.

Tissue Structural Properties

Force–displacement testing was performed toexamine the physical properties of the cervix as itchanges during pregnancy. The two plots in Fig. 4demonstrate how the structural properties of the cervixchange over the course of pregnancy and 24 h afterdelivery. In Fig. 4a, force is plotted as a function of

FIGURE 1. Normalized average Raman spectra of NG mice atvarious points during the menstrual cycle.

TABLE 1. SMLR output for cycling study.

Classification accuracy = 77%

SMLR output

Proestrus (%) Estrus (%) Metestrus (%) Diestrus (%) Di to pro (%)

Cycling timepoint

Proestrus 79 6 11 – 4

Estrus 14 78 7 1 –

Metestrus – 8 83 6 3

Diestrus 2 2 7 80 9

Di to pro 13 4 5 11 67

Di to pro are measurements taken between diestrus and proestrus.

FIGURE 2. Normalized average spectra from the cervix ofpregnant mice at 5 time points during their pregnancy and at 1time point 24 h after delivery (PP1). Highlighted regions ingray were focused in Fig. 3.

VARGIS et al.1818

displacement or cervical diameter, showing how sam-ples react and resist to a specific amount of elongation.For this study, cervical stiffness (g/mm), or the amountof tension (g) divided by displacement (mm) using amoving average function is plotted for samples at each

time point (Fig. 4b), similar to previous studies.34 Thisplot corresponds to the slope of the linear portion ofthe Fig. 4a and the inverse of the compliance of thecervix. Any result with a p-value <0.1 was defined asbeing statistically significant. Results from day 19 are

FIGURE 3. Bar graphs of specific peak intensities from Raman spectra that change over the course of pregnancy. These Ramanshift peaks are potentially correlated to fatty acids and lipids (a), protein side chains (b), and amino acids (c). Bar graph of full-width at half-max shows the change of the amide I band, which may correspond to collagen (d). Any peak that was significantlydifferent (p-value <0.1) compared to d4 was marked with *.

TABLE 2. SMLR output results of pregnancy and postpartum study.

Classification accuracy = 94%

SMLR output

NG (%) Early (%) d15 (%) d18 (%) d19 (%) PP1 (%)

Pregnancy time point

NG 99 1 – – – –

Early 2 97 1 – – –

d15 – 9 84 3 – 4

d18 – – 2 92 6 –

d19 – – – 3 96 1

PP1 2 3 – – – 95

Early data includes days 1, 4, and 12.

Biochemical Changes in the Cervix 1819

statistically significant compared to NG, day 1, day 4and PP1. Cervical stiffness results from both NG andPP1 mice are also significantly different (p-value <0.1)than days 15, 18 and 19.

Histological Staining

Trichrome staining of samples from multiple timepoints (NG, d4-19, and PP1) during mouse pregnancyis shown in Fig. 5. Before, at the start of and afterpregnancy, a dense collagen network, stained blue, isthe most prominent tissue component (d4, white ar-row). Smooth muscle cells, stained red, also appear toform tight bundles within the collagen and along theedges of the uterus and cervix. As pregnancy pro-gresses, the dense collagen network appears to becomedisorganized, especially on day 19 (d19, black arrow).While the cervix is not completely recovered by PP1,the blue-stained regions in PP1 are thicker comparedto d19, signifying a reorganization of collagen within24 h of delivery.

DISCUSSION

The goal of this research is to demonstrate the fea-sibility of using RS to detect changes in the cervixduring normal pregnancy. Before studying pregnancy,cervical changes associated with estrous cycling of NGmice were studied using RS to create a baseline forcomparison. Next, Raman spectra were acquired fromthe cervix of pregnant mice prior to structural testingand histological staining to study cervical changesduring pregnancy using three sets of data. We foundthat biochemical changes that were observed in theRaman spectra of late-stage pregnancy mice correlatewith the compliance of the cervical tissue and the his-tological results. These results will provide researcherswith a better understanding of the biochemical func-tion of the cervix during pregnancy and the mecha-nisms regulating pregnancy, labor and delivery.

The cycling data was obtained to determine ifovarian steroid hormonal changes during the estrouscycle would have an effect on NG data. Our priorhuman studies showed that the cervical changes duringthe menstrual cycle could be identified within Ramanspectra, making it necessary to study the hormonalfluctuations of the NG cervix of mice using RS prior tocomparison to a gravid cervix.20,21 Furthermore, ourlab has never used RS on the cervix of mice. If thesteroid hormonal effect of the cycling time point hadan effect on the spectra, then multiple NG categorieswould need to be considered when comparing NGspectra to spectra obtained from pregnant mice.However, unlike the human data, which classified withan accuracy of over 98% based only on cycling timepoint,20 the mouse cycling data classified with anaccuracy of only 77% (Table 1), suggesting that thereare fewer distinctions in Raman spectra from themouse cervix due to cycling time point alone. Thedifference in classification accuracies between mice andhumans may be due to the smaller variations that existin the shorter 4-day mouse cycle compared to anapproximately 28-day human cycle. In the humanstudy, only two phases were considered, the prolifer-ative and secretory phases, as opposed to the fourdistinct phases studied in the mouse. From these re-sults, it was concluded that a single category combin-ing all the spectra acquired from NG mice could beused in the analysis that followed.

The measurements acquired during pregnancydemonstrate that many changes in the cervix can beobserved in the Raman spectra (Fig. 2). As pregnancyprogresses, there is a sharp loss of spectral integrityseen by the lack of distinct peaks and valleys that ismost prominent on day 19, resulting from theincreasing amount of elastin, collagen reorganizationand dilation. Measurements taken on day 19 were

FIGURE 4. Measurements from biomechanical testing ofcervical tissue. (a) Tension as a function of time, displayingthat the cervix can withstand increased displacement aspregnancy progresses. (b) Cervical stiffness, measured fromthe slope of the graph in (a) showing that the cervix becomesless stiff during pregnancy and quickly regains its strength24 h after delivery. Day 19 is significantly different than NG, d1and PP1. NP and PP1 are significantly different than d15, d18and d19. *p < 0.1; **p < 0.05.

VARGIS et al.1820

acquired 6–12 h prior to delivery and represent themaximum change of the cervix during pregnancy. Ta-ken within 24 h after delivery, the postpartum spectra(PP1) appear similar to spectra acquired from NGmice and mice early in pregnancy.

Changes in the Raman spectra associated with thechanges over the course of pregnancy are observed inthe four plots in Fig. 3. Figures 3a and 3b show spe-cific peaks decreasing in intensity as a function ofgestational age. These peaks may potentially correlateto carbon bonds and lipids, as discussed previously.11

It follows that as pregnancy continues, a decrease inthe concentration of many cervical components occurs.While many of the peak intensities decrease as preg-nancy continues, there are some that increase, such asthe 1200 cm21 peak (Fig. 3c), which has been shown tocorrelate to proline, tyrosine, and amino acids, whichincrease in concentration closer to labor and deliv-ery.3,28,31 The FWHM of the 1650 cm21 amide I peakprovides information about the collagen content,

organization and polarization (Fig. 3d). The wideningof the band may indicate an increasingly disperseddistribution of peptide carbonyl stretching during thecourse of pregnancy, signifying a change in the orien-tation of collagen fibers.43 Many studies have demon-strated that the total amount of collagen does notnecessarily decrease during pregnancy; instead, itsreorganization contributes to cervical softening andripening.34,44

A classification accuracy of over 94% was foundwhen SMLR was used to classify the spectra frommultiple time points during pregnancy (Table 2),showing that the biochemical components changingduring pregnancy are substantial, and distinctive dif-ferences at specific gestational ages can be utilized toclassify the spectra. Spectra acquired from mice on day15 of their pregnancy had the lowest classification rate(85%). Although Raman measurements for each daywere taken at the same time, some mice may havestarted the cervical softening process before others,

FIGURE 5. Trichrome staining images acquired at 203 from NG, d4, d12, d15, d19 and PP1 cervical tissue samples. Note thedensely packed collagen structures (blue) from d4 (white arrow) compared to the disorganized and sparse areas in d19 (blackarrow). Samples 24 h after delivery (PP1) more so resemble early time points.

Biochemical Changes in the Cervix 1821

effectively introducing a higher amount of variation inday 15 measurements compared to other time points.34

Although the goal of this study is not to classifyspectra based on gestational age, these results verifythat the changes occurring in the Raman spectra dur-ing pregnancy can be teased out using SMLR.

The force–displacement testing, performed afterRaman data was collected, verified that the cervix isable to withstand a larger displacement (Fig. 4a) whilebecoming softer (Fig. 4b) throughout pregnancy. Thesteep slope found in measurements taken at early timepoints and post-partum (Fig. 4a) indicates anincreased resistance to stretch. Many of the changesseen in the Raman data acquired during pregnancy(Figs. 3a, 3b) correspond with the changes seen withthe force–displacement testing (Fig. 4). The intensitiesof the two Raman peaks displayed in Figs. 3a, 3bcorrespond to fatty acids, lipids and protein sidechains. These constituents play an important role inmaintaining the cervix’s rigidity by contributing to thestrength of the ECM and reinforcing the tissue’sstructure.15,24,33–35 The decrease in their concentrationcorresponds to an increase in compliance as the cervixprepares for birth. The similarities between importantRaman peaks and these tests show that RS is correctlyidentifying many of the important biochemical com-ponents responsible for the changes in the stiffness ofthe cervix.

After in vivo Raman spectra were acquired, tissueswere excised for histology as a reference for compari-son to the Raman data. The results of the Masson’strichrome staining (Fig. 5) provide one explanation forthe changes observed in the Raman spectra and force–displacement results and are similar to previous studiesthat used histology to understand collagen organiza-tion in the cervix during pregnancy.10,45,46 As preg-nancy continues, the concentration of collagen appearsto remain consistent, while its organization is altered.These results correspond to the loss of spectral integ-rity in the Raman measurements on days 18 and 19(Fig. 2). The FWHM plot of the amide I peak from theRaman spectra (Fig. 3d) also matches the results fromthe staining. Although there is no change in theintensity of the 1650 cm21 peak (corresponding to thetotal amount of collagen), there are differences due toprotein orientation, polarization and solubility. Col-lagen reorganization and increased solubility are twofactors that result in increased compliance of the cervixas it prepares for labor and delivery,15,34 an outcomewhich has been verified in all experiments. The paral-lels among the Raman measurements, the structuralproperties, and the histological staining indicate thatRS is a quantifiable method for assessing the structuralproperties and cellular components of the cervix dur-ing pregnancy without excising tissue.

There were some drawbacks in this current studydesign that can be overcome in future studies. First,vaginal cytology was used to quickly determine thecycling point. While the results of the cycling andpregnancy studies suggest that the changes in a NGcervix are less noticeable than the changes resultingfrom pregnancy, it may be beneficial for future cyclingstudies to use serum testing to ensure that the correctcycling point is reported. Furthermore, the mouse isnot the ideal model for studying estrous cycling sincethe phases can be shorter or longer than 24 h, as somestudies have suggested.30 Acquiring Raman data fromother rodents, such as the guinea pig, may providemore accurate results.7 Repeated in vivo Raman mea-surements of the NG cervix sometimes led to pseudo-pregnancy; such measurements were excluded infurther analyses. Also, in the initial experimental de-sign, data were acquired before mating and on the firstday of gestation. However, taking Raman measure-ments at these time points sometimes resulted in theloss of the pregnancy about day 12. It was determinedthat starting on gestational age day 4 led to full termpregnancies without complications. Future studiesmay consider decreasing the size of the fiber opticprobe to prevent pseudopregnancies and miscarriages.

Throughout pregnancy, the epithelium of the cervixbecomes thicker. The probe used to obtain thesemeasurements contains forward-looking optics whichtakes volumetric measurements approximately 1 mmin depth. For this in vivo study, the placement of theprobe onto the mouse cervix is not known, however,each measurement is an average of at least 3 differentmeasurements where the probe was completelyremoved from the cervix and vaginal canal and thenreplaced. Few regional variations in the spectra werefound in the measurements obtained consecutivelyfrom the cervix, suggesting that data can be takenreliably without having to control from site-specificity,matching previous human studies in our lab.36 Sincethis method is meant to be performed without surgeryor removing tissues, knowing the placement of theprobe is not essential in this initial study. Futurestudies will consider the effect of the thickening of theepithelium and the placement of the probe on the data.

The results from this study demonstrate that RS canbe a useful tool to non-invasively and accurately studybiochemical changes in the cervix during pregnancy.The impact of these results provides a new avenue forobstetrics research that does not rely on removingtissue and is not focused on one molecule. Instead, RScan be used to identify cervical changes resulting fromthe interactions of multiple biomolecules, hormonalagents, feedback loops, etc., thereby providingresearchers with new ways to understand the progres-sion of pregnancy. The next steps are to use RS to

VARGIS et al.1822

study what occurs during labor and normal humanpregnancy. The long-term goal of this research is notonly to use RS to understand how pregnancy affectsthe human cervix, but to develop a method for deter-mining patients at risk for preterm labor and delivery.Preterm birth, defined as labor prior to 37 weeks, is aserious medical complication, affecting over 1 in 8pregnancies in the US and 75% of infants that haveperinatal death are born premature.5 Even with currentadvances in health care and research, there are limiteddiagnostics in place for predicting preterm birth.5

Developing an effective and non-invasive method thatidentifies women who are at-risk for preterm birthwould have a tremendous impact on the medicalcommunity, enabling providers to identify patients atrisk for preterm labor, thereby improving the man-agement of these patients.

ACKNOWLEDGMENTS

The authors acknowledge the financial support ofthe National Institutes of Health (Grant No. R01-CA-095405, AMJ and HD 044741, BCP) and a predoctoralfellowship (Grant No. T32-HL7751-15) for EV. Spe-cial thanks go to Stan Poole and Wais Folad for theirhelp with the structural testing experiments, XiahongBi for conversations about Raman peak assignmentsand Amy Rudin for proofreading this paper.

CONFLICT OF INTEREST

None of the authors of the above manuscript hasdeclared any conflict of interest within the last threeyears which may arise from being named as an authoron the manuscript.

REFERENCES

1Akins, M. L., K. Luby-Phelps, R. A. Bank, and M.Mahendroo. Cervical softening during pregnancy-regu-lated changes in collagen cross-linking and composition ofmatricellular proteins in the mouse. Biol. Reprod. 84:1053–1062, 2011.2Akins, M. L., K. Luby-Phelps, and M. Mahendroo. Sec-ond harmonic generation imaging as a potential tool forstaging pregnancy and predicting preterm birth. J. Biomed.Opt. 15:026020, 2010.3Arp, Z., D. Autrey, J. Laane, S. A. Overman, and G. J.Thomas, Jr. Tyrosine Raman signatures of the filamentousvirus Ff are diagnostic of non-hydrogen-bonded phenox-yls: demonstration by Raman and infrared spectroscopy ofp-cresol vapor. Biochemistry 40:2522–2529, 2001.4Basar, G., U. Parlatan, and S. Seniak. Investigation ofpreeclampsia by Raman spectroscopy. SPIE 1267:378,2010.

5Behrman, R. E., Butler, A. S., and Institute of Medicine(U.S.). Committee on Understanding Premature Birth andAssuring Healthy Outcomes. Preterm Birth: Causes, Con-sequences, and Prevention. Washington: National Acade-mies Press, 2007.6Challis, J. R. G., S. G. Matthews, W. Gibb, and S. J. Lye.Endocrine and paracrine regulation of birth at term andpreterm. Endocr. Rev. 21:514–550, 2000.7Croix, D., and P. Franchimont. Changes in the serumlevels of the gonadotrophins progesterone and estradiolduring the estrous cycle of the guinea pig. Neuroendocri-nology 19:1–11, 1975.8Deb, K., J. Reese, and B. C. Paria. Methodologies to studyimplantation in mice. Methods Mol. Med. 121:9–34, 2005.9Erez, O., F. Gotsch, S. Mazaki-Tovi, E. Vaisbuch, J. P.Kusanovic, C. J. Kim, T. Chaiworapongsa, D. Hoppen-steadt, J. Fareed, and N. G. Than. Evidence of maternalplatelet activation, excessive thrombin generation, andhigh amniotic fluid tissue factor immunoreactivity andfunctional activity in patients with fetal death. J. Matern.Fetal Neonatal Med. 22:672–687, 2009.

10Fosang, A. J., C. Handley, V. Santer, D. Lowther, and G.Thorburn. Pregnancy-related changes in the connectivetissue of the ovine cervix. Biol. Reprod. 30:1223, 1984.

11Frank, C. J., R. L. McCreery, and D. C. B. Redd. Ramanspectroscopy of normal and diseased human breast tissues.Anal. Chem. 67:777–783, 1995.

12Garfield, R. E., H. Maul, W. Maner, C. Fittkow, G. Olson,L. Shi, and G. R. Saade. Uterine electromyography andlight-induced fluorescence in the management of term andpreterm labor. J. Soc. Gynecol. Investig. 9:265–275, 2002.

13Harkness, M. L., and R. D. Harkness. Changes in thephysical properties of the uterine cervix of the rat duringpregnancy. J. Physiol. 148:524–547, 1959.

14Hellemans, P., J. Gerris, and P. Verdonk. Fetal fibronectindetection for prediction of preterm birth in low risk wo-men. BJOG 102:207–212, 1995.

15House, M., D. L. Kaplan, and S. Socrate. Relationshipsbetween mechanical properties and extracellular matrixconstituents of the cervical stroma during pregnancy. Se-min. Perinatol. 33:300–307, 2009.

16Ji, H., T. L. Dailey, V. Long, and E. K. Chien. Prosta-glandin E2-regulated cervical ripening: analysis of proteo-glycan expression in the rat cervix. Am. J. Obstet. Gynecol.198(536):e1–e7, 2008.

17Jokhi, R., B. Brown, and D. Anumba. The role of cervicalElectrical Impedance Spectroscopy in the prediction of thecourse and outcome of induced labour. BMC PregnancyChildbirth 9:40, 2009.

18Junqueira, L. C., M. Zugaib, G. S. Montes, O. M. Toledo,R. M. Krisztan, and K. M. Shigihara. Morphologic andhistochemical evidence for the occurrence of collagenolysisand for the role of neutrophilic polymorphonuclear leu-kocytes during cervical dilation. Am. J. Obstet. Gynecol.138:273–281, 1980.

19Kamel, R. M. The onset of human parturition. Arch.Gynecol. Obstet. 281:975–982, 2010.

20Kanter, E. M., S. Majumder, G. J. Kanter, E. M. Woeste,and A. Mahadevan-Jansen. Effect of hormonal variationon Raman spectra for cervical disease detection. Am. J.Obstet. Gynecol. 200(512):e1–e5, 2009.

21Kanter, E. M., E. Vargis, S. Majumder, M. D. Keller, E.Woeste, G. G. Rao, and A. Mahadevan-Jansen. Applica-tion of Raman spectroscopy for cervical dysplasia diag-nosis. J. Biophotonics 2:81–90, 2009.

Biochemical Changes in the Cervix 1823

22Krishnapuram, B., L. Carin, M. A. Figueiredo, and A. J.Hartemink. Sparse multinomial logistic regression: fastalgorithms and generalization bounds. IEEE Trans. PatternAnal. Mach. Intell. 27:957–968, 2005.

23Kuon, R. J., S. Q. Shi, H. Maul, C. Sohn, J. Balducci, L.Shi, and R. E. Garfield. A novel optical method to assesscervical changes during pregnancy and use to evaluate theeffects of progestins on term and preterm labor. Am. J.Obstet. Gynecol. 205:82.e15–82.e20 (2011).

24Leppert, P. C. Anatomy and physiology of cervical ripen-ing. Clin. Obstet. Gynecol. 38:267, 1995.

25Lieber, C. A., and A. Mahadevan-Jansen. Automatedmethod for subtraction of fluorescence from biologicalRaman spectra. Appl. Spectrosc. 57:1363–1367, 2003.

26Lim, K.-H., S. Salahuddin, L. Qiu, H. Fang, E. Vitkin, I.C. Ghiran, M. D. Modell, T. Takoudes, I. Itzkan, E.B. Hanlon, B. P. Sachs, and L. T. Perelman. Light-scat-tering spectroscopy differentiates fetal from adult nucleatedred blood cells: may lead to noninvasive prenatal diagnosis.Opt. Lett. 34:1483, 2009.

27McColl, I. H., E. W. Blanch, L. Hecht, N. R. Kallenbach,and L. D. Barron. Vibrational Raman optical activitycharacterization of poly(L-proline) II helix in alanine oli-gopeptides. J. Am. Chem. Soc. 126:5076–5077, 2004.

28Miura, T., and G. Thomas, Jr. Raman spectroscopy ofproteins and their assemblies. Subcell. Biochem. 24:55,1995.

29Nathanielsz, P. W. Life Before Birth: The Challenges ofFetal Development. New York: W.H. Freeman, 1996.

30Nelson, J., L. Felicio, P. Randall, C. Sims, and C. Finch. Alongitudinal study of estrous cyclicity in aging C57BL/6Jmice: I. Cycle frequency, length and vaginal cytology. Biol.Reprod. 27:327–339, 1982.

31Palejwala, S., D. E. Stein, G. Weiss, B. P. Monia, D.Tortoriello, and L. T. Goldsmith. Relaxin positively regu-lates matrix metalloproteinase expression in human loweruterine segment fibroblasts using a tyrosine kinase signalingpathway. Endocrinology 142:3405–3413, 2001.

32Polettini, J., J. Peracoli, J. Candeias, J. Araujo Junior, andM. Silva. Inflammatory cytokine mRNA detection by realtime PCR in chorioamniotic membranes from pregnantwomen with preterm premature rupture of membranes.Eur. J. Obstet. Gynecol. Reprod. Biol. 144:27–31, 2009.

33Puskas, J. E., and Y. Chen. Biomedical application ofcommercial polymers and novel polyisobutylene-basedthermoplastic elastomers for soft tissue replacement. Bio-macromolecules 5:1141–1154, 2004.

34Read, C. P., R. Word, M. A. Ruscheinsky, B. C. Timmons,and M. S. Mahendroo. Cervical remodeling during preg-nancy and parturition: molecular characterization of thesoftening phase in mice. Reproduction 134:327, 2007.

35Reinwald, S., Y. Li, T. Moriguchi, N. Salem, and B. A.Watkins. Repletion with (n23) fatty acids reverses bonestructural deficits in (n23)-deficient rats. J. Nutr. 134:388–394, 2004.

36Robichaux-Viehoever, A., E. Kanter, H. Shappell, D.Billheimer, H. Jones III, and A. Mahadevan-Jansen.Characterization of Raman Spectra measured in vivo forthe detection of cervical dysplasia. Appl. Spectrosc. 61:986–993, 2007.

37Sennstrom, M. K. B., A. Brauner, Y. Lu, L. M. M.Granstrom, A. L. Malmstrom, and G. E. Ekman. Inter-leukin-8 is a mediator of the final cervical ripening inhumans. Eur. J. Obstet. Gynecol. Reprod. Biol. 74:89–92,1997.

38Sugano, T., H. Narahara, K. Nasu, K. Arima, K. Fujisa-wa, and I. Miyakawa. Effects of platelet-activating factoron cytokine production by human uterine cervical fibro-blasts. Mol. Hum. Reprod. 7:475, 2001.

39Sultana, R. R., S. N. Zafarullah, and N. H. Kirubamani.Saliva signature of normal pregnant women in each tri-mester as analyzed by FTIR spectroscopy. Indian J. Sci.Technol. 4:477–480, 2011.

40Suzuki, T., C. Mori, H. Yoshikawa, Y. Miyazaki, N.Kansaku, K. Tanaka, H. Morita, and T. Takizawa.Changes in nitric oxide production levels and expression ofnitric oxide synthase isoforms in the rat uterus duringpregnancy. Biosci. Biotechnol. Biochem. 73:2163–2166,2009.

41Vargis, E., T. Byrd, Q. Logan, D. Khabele, and A. Ma-hadevan-Jansen. Sensitivity of Raman spectroscopy tonormal patient variability. J. Biomed. Opt. 16:117004-1–117004-9, 2011.

42Vargis, E., E. M. Kanter, S. K. Majumder, M. D. Keller,R. B. Beaven, G. G. Rao, and A. Mahadevan-Jansen. Ef-fect of normal variations on disease classification ofRaman spectra from cervical tissue. Analyst 139:2981–2987, 2011.

43Wang, Y. N., C. Galiotis, and D. Bader. Determination ofmolecular changes in soft tissues under strain using laserRaman microscopy. J. Biomech. 33:483–486, 2000.

44Yu, S. Y., C. A. Tozzi, J. Babiarz, and P. C. Leppert.Collagen changes in rat cervix in pregnancy-polarized lightmicroscopic and electron microscopic studies. Exp. Biol.Med. 209:360, 1995.

45Zhao, L., P. J. Roche, J. M. Gunnersen, V. E. Hammond,G. W. Tregear, E. M. Wintour, and F. Beck. Mice withouta functional relaxin gene are unable to deliver milk to theirpups. Endocrinology 140:445, 1999.

46Zhao, L., C. S. Samuel, G. W. Tregear, F. Beck, and E. M.Wintour. Collagen studies in late pregnant relaxin nullmice. Biol. Reprod. 63:697–703, 2000.

VARGIS et al.1824