Heterogeneous microstructural changes of the cervix influence cervical funneling

Abstract

The cervix acts as a dynamic barrier between the uterus and vagina, retaining the fetus during pregnancy and allowing birth at term. Critical to this function, the physical properties of the cervix change, or remodel, but abnormal remodeling can lead to preterm birth (PTB). Although cervical remodeling has been studied, the complex 3D cervical microstructure has not been well-characterized. In this complex, dynamic, and heterogeneous tissue microenvironment, the microstructural changes are likely also heterogeneous. Using quantitative, 3D, second-harmonic generation microscopy, we demonstrate that rat cervical remodeling during pregnancy is not uniform across the cervix; the collagen fibers orient progressively more perpendicular to the cervical canals in the inner cervical zone, but do not reorient in other regions. Furthermore, regions that are microstructurally distinct early in pregnancy become more similar as pregnancy progresses. We use a finite element simulation to show that heterogeneous regional changes influence cervical funneling, an important marker of increased risk for PTB; the internal cervical os shows ~6.5x larger radial displacement when fibers in the inner cervical zone are parallel to the cervical canals compared to when fibers are perpendicular to the canals. Our results provide new insights into the microstructural and tissue-level cervical changes that have been correlated with PTB and motivate further clinical studies exploring the origins of cervical funneling.

Graphical abstract

1. Introduction

Preterm birth (PTB), defined as a live birth before 37 weeks gestation, is a significant health issue affecting more than 15 million infants annually worldwide [1]. The consequences of PTB can be life-long, costly, and severe. Premature infants are at higher risk for many developmental disabilities and diseases, and more than one million die each year because of complications of PTB [1,2]. Despite these consequences, current screening tools are not sufficiently sensitive to accurately assess the risk of PTB and thus do not allow for effective medical intervention in many cases [3,4]. Therefore, there is an immediate need for development of new tools and technologies that enable understanding and prediction of PTB.

Cervical insufficiency, or the inability of the cervix to retain the fetus, is one of the pathways leading to PTB, though its origins are not understood. Cervix is a barrier between the uterus and vagina, retaining the fetus during pregnancy and allowing birth at term. Critical to these functions, the cervix undergoes a remodeling process in preparation for delivery [5,6]. The immune cell activity, such as the increased number of macrophages that resembles an inflammatory process, influences the softening and ripening of the tissue [7,8]. It is also well established that the collagen fiber microstructure of the cervix becomes progressively more disorganized during pregnancy and, as a result, the cervix softens [9–15]. Abnormal cervical remodeling, defined as a remodeling that is temporally or spatially irregular, can disrupt this process and lead to cervical insufficiency [16,17]. Cervical funneling, defined as dilation of the internal os of the cervix accompanied by protrusion of amniotic membranes, is one of the markers of cervical insufficiency and is highly associated with risk of PTB when observed in the second trimester [18–20]. Clinically, cervical funneling is evaluated using translabial or transvaginal ultrasound examination. The protrusion of the amniotic membranes into the internal cervical os is measured along the lateral border of the funnel, and a protrusion greater than 5 mm is considered cervical funneling [18,21]. Characterization of the microstructural changes of collagen and evaluation of the role of microstructural changes in alteration of the mechanical properties and function of the cervix will help to answer important questions associated with PTB.

Stress and strain distributions can be non-uniform across the cervix, depending on loading conditions, geometry, and material properties such as collagen microstructure. The remodeling process may therefore locally alter the tissue microstructure for specific mechanical functions. While heterogeneity of the non-pregnant cervix microstructure has been confirmed in both rats and humans [22,23], the evolution of tissue microstructure in different cervical regions during pregnancy has not been fully investigated. The inner cervical zone, in particular, has a microstructure quite distinct from that of the outer zone [24]. Yet the biomechanical function and evolution of this region during pregnancy have not been studied. Comprehensive characterization of region-specific microstructural changes of the cervix that occur during remodeling may reveal important changes that influence behavior of the entire cervix and PTB.

The biological function and mechanical behavior of tissue depend on the composition and arrangement of its constituents, particularly the collagen microstructure [25–28]. Several techniques have been used to image the collagen fiber microstructure of cervix. Optical microscopy combined with exogenous stains, e.g., picrosirius red [29], standard fluorescence microscopy [30], polarized light microscopy [14], spectral-domain optical coherence tomography (OCT) [31], and X-ray diffraction [22] are among the most frequently used techniques for studying the cervix microstructure and cervical remodeling. Despite being important and informative, these studies do not quantify the evolution of the 3D microstructural properties of the cervix during pregnancy. Although previous studies on pregnant and non-pregnant cervices have shown a significant correlation between collagen fiber organization, degradation of collagen cross-links, and mechanical properties of the tissue [22,24,32–34], comprehensive 3D quantification of region-specific changes of cervical collagen microstructure during pregnancy is still lacking.

We use second-harmonic generation (SHG) microscopy to image the collagenous cervical tissue and its complex 3D microstructure. Inherent confocality and deep optical sectioning capability allow for high-resolution 3D imaging. While SHG microscopy has been previously used to image the cervix, there were important limitations such as lack of quantification, a focus on non-pregnant tissue, or use of 2D imaging [35–37]. Our previous work on non-pregnant cervix [24], as well as our results from other soft tissues [38], and the work of others [39–43], show that 3D microstructural properties of tissue correlate well with mechanical properties. The 3D characterization of microstructure can therefore provide new insight into the changes of the cervix during pregnancy and illustrate their potential functional benefits.

To understand regional variations in cervical remodeling and their role in the mechanical function of the cervix, we probe the 3D microstructural properties of the rat cervix at multiple gestational stages. We find that microstructural remodeling of the cervix does, in fact, occur non-uniformly across different regions of the mid-cervix; the initially distinct cervical regions converge to a more homogeneous microstructure as pregnancy progresses. Consistent with other reports, we find the collagen fibers become more disperse, but the changes are not uniform across the cervical regions.

Interestingly, the dominant orientation of the collagen fibers in the inner zone has a strong correlation with the gestation day, but a similar correlation does not exist in other cervical regions. In addition to this spatiotemporal characterization, we use finite element (FE) simulation to understand how regional changes of the tissue influence larger-scale displacements of the cervix. We suggest, for the first time, that the abnormal cervical remodeling of the inner cervical zone could be linked to cervical funneling and therefore risk of PTB; the intact inner zone collagen fibers late in pregnancy could act to transfer load from the uterus to the cervix and thus dilate the internal os. Based on our data from the rat model, the remodeling of the inner cervical zone could play a role in pregnancy maintenance that allows the cervix to stay closed despite increased forces from the growing fetus. These observations motivate further clinical studies in humans. The results of this study facilitate a more comprehensive characterization of cervical remodeling and illustrate how microstructure may influence PTB, and more broadly, behavior of collagenous tissues.

2. Materials and methods

2.1. Sample preparation

All procedures were approved by the Institutional Animal Care and Use Committee at the University of Illinois at Urbana-Champaign. 12-week-old non-pregnant (n=2), 15-days pregnant (n=2), 17-days pregnant (n=2), and 21-days pregnant (n=3) Sprague Dawley rats were euthanized following relevant guidelines and regulations. Cervical tissue was harvested within one hour and stored at −80 °C. Excised tissue was thawed at room temperature and the tissue surrounding the cervix removed. Cervical tissue was embedded in optimal cutting temperature compound (OCT) and cryosectioned at −20 °C (Leica Biosystems, Wetzlar, Germany). Samples were cut at mid-cervix, perpendicular to the cervical canals, for transverse images. Selected samples consisting of non-pregnant (n=1) and 17-days pregnant (n=1) were later cut parallel to the canals for coronal images after being imaged in the transverse plane. OCT was rinsed and cleared after sectioning, and samples were imaged immediately. Details regarding sample preparation are provided in a previous publication [24].

2.2. SHG imaging

A confocal microscope (Zeiss LSM 710, Oberkochen, Germany) was used to image the cervices. A femtosecond laser (MaiTai DeepSee, eHP, Spectraphysics, Newport Corporation, CA, USA) producing 70-fs centered at 780 nm was used to illuminate the sample, and a 20x 0.8 NA objective lens and motorized stage were used to image the entire cross-section of the tissue. Images were then automatically stitched in Zen software (Zeiss, Oberkochen, Germany). Imaging the entire cross-section allowed navigation of the anatomical regions for 3D imaging. A 40x 1.2 NA water-immersion objective acquired three volumetric images (212x212x50 μm) from each of the outer zone, inner zone, and septum (n=9 images from each cervix), scanning selected regions with a 330-500 nm step size (50 μm total depth). Each volumetric image was divided into 4 equal sub-images (50x50x50 μm) to better capture the heterogeneity of the samples. Together, 36 data points were collected from each cervix, resulting in 324 total measurements from cervical regions at different stages of pregnancy. Samples were placed on a No. 1.5 cover glass-bottom dish (MatTek, Ashland, MA) with PBS buffer solution to keep samples hydrated during imaging.

2.3. SHG image quantification

Each image in the stack was down-sampled to make pixel size equal in all orientations and reduce computation cost. Three image parameters, Φ, circular variance (CV), and spherical variance (SV), were computed for each of the 50x50x50 μm regions using orientation data from FT-SHG analysis [44–47]. The Φ parameter represents the average angle of the collagen fibers relative to the transverse plane within the volume, where 0° and 90° refer to the fibers parallel to the transverse and coronal planes, respectively. CV quantifies dispersion of the collagen fibers in 2D, in the imaging plane, with a value between 0 (aligned in one direction) and 1 (completely dispersed). The image in the center of the stack was used to calculate the CV. SV indicates the dispersion of the collagen fibers in the 3D space, considering both in-plane and out-of-plane orientations. Modified definitions of CV and SV were used in this study to adjust the analysis for fibers (details here [24]).

2.4. Statistical analysis

Statistical analyses, including principal component analysis (PCA), were performed in R 2020 software (R Foundation for Statistical Computing, Vienna, Austria [48]). A one-way ANOVA followed by a post hoc Tukey analysis was used to indicate which groups differed. Spearman’s rank correlation was used to evaluate monotonic relationships between image parameters and gestational day. Three significance levels (p < 0.01, 0.05, and 0.1) are reported in the text and figures. Measurements from the same cervix are treated as independent data because they capture the inherent biological variability of the samples [49].

2.5. FE Modeling

We developed a model of the cervix and uterus in COMSOL Multiphysics V 5.6 (COMSOL AB, Stockholm, Sweden) to investigate how collagen fiber microstructure influences the cervix, specifically at the internal os as a measure of cervical funneling. The model does not attempt to replicate a precise gestational age. Instead, it simulates the cervix during roughly the third pregnancy trimester in humans. We used a simple 2D geometry first to avoid complicating the analysis with geometric uncertainties. The 2D axisymmetric geometry was partitioned into five domains: inner cervical zone, outer cervical zone, uterus, inner transition zone, and outer transition zone (Fig.S1, Supplementary). The material properties changed smoothly in the transition zone to prevent creating a sharp boundary between the cervix and uterus. Constant pressure was applied on the inner boundary of the uterus to simulate the contact of fetal membrane and uterine wall. Since our model did not include the fetal membrane, we assumed a pressure boundary close to what is observed by ultrasound as illustrated in [50] and the boundaries of the applied pressure did not move throughout simulation. The top boundary of the uterus was constrained along the Z-axis, but was free to move radially along the R-axis (Fig.S1, Supplementary). The outer boundaries of the cervix and uterus are assumed to move without any constraints. Unstructured triangular elements meshed the entire 2D domain (tetrahedral elements are the equivalent element type for 3D models) [51]. A mesh convergence study was performed to verify the results are independent of the mesh size (Fig.S2, Supplementary).

A structural hyperelastic model included the influence of the collagen fiber organization in the simulation. While we quantify the changes of the cervix microstructure at the fiber level (~1-20 μm), this energy-based approach does not rely on representative volume elements, and therefore, does not incorporate length scales of the fibers.

In this model, the strain energy function is the sum of energy in the non-fibrillar extracellular matrix and the fibrillar collagen, as described below

$$\mathrm{W}={\mathrm{W}}_m+{\mathrm{W}}_f$$ (1),

where W_m and W_f are the energy functions of the matrix and fibrillar collagen, respectively. We used a Neo-Hookean material model for the matrix potential

$${\mathrm{W}}_m=\frac{\mu }{2}(I_1-3)-\mu \mathit{ln}J+\frac{\lambda }{2}{(\mathit{ln}J)}^2$$ (2),

where μ and λ are Lamé constants, I₁ is the first invariant of the right Cauchy-Green deformation tensor, J = det(F) is the Jacobian, and F is the deformation gradient.

We used the Exponential Power Law model [52] to describe the strain energy function of the fibrillar collagen

$${\mathrm{W}}_f=\frac{\xi }{\beta }{(I_n-1)}^{\beta }$$ (3),

where ξ and β are material parameters, $I_n=n_o\cdot C\cdot n_o$ is the square of the fiber stretch, n_o is the initial fiber direction, and $C=F^{T}F$ is the Green’s deformation tensor.

To introduce fiber dispersion into this equation, we substitute I_n with the generalized structure tensor [53], H, defined as

$$H=\kappa I+(1-3\kappa )I_n$$ (4),

where κ is a dispersion parameter that describes the degree of anisotropy, and I is the identity tensor, κ can vary between 0 (no fiber dispersion) and 1/3 (completely random fiber organization). The generalized W_f can be described by the following equation

$${\mathrm{W}}_f=\frac{\xi }{\beta }{(\kappa I+(1-3\kappa )I_n-1)}^{\beta }$$ (5).

The material properties μ, λ, ξ, and β were chosen from a well-established model described in reference [54] and are listed in Table 1. The trend of changes in SV and Φ, calculated by quantification of the SHG images, informed the trend of changes in κ and n_o, respectively, in the material model. This illustrative model aims to show the general dependence of the dilation of internal cervical os on the collagen microstructure of cervical regions; the absolute values resulting from the simulation should not be considered independently due to the uncertainties in geometry, material properties, and boundary conditions of the model.

Table 1:

Material properties used in the constitutive equations of uterus, inner cervical zone, and outer cervical zone

Anatomical location	Properties
	μ (kPa)	λ (MPa)	ξ (kPa)	β	κ	no
Uterus	2	333.33	100	3	0.32	0°
Inner cervical zone	0.67	1000	18	3	0-0.2	0-90°
Outer cervical zone	0.67	1000	18	3	0-0.2	0°

We also used a publicly available 3D MRI-based geometry, provided by Fang et al. [55], to model deformation of the cervix. We aimed to confirm that our conclusions from the 2D simulation are valid for a more realistic, nonsymmetric geometry. The same Exponential Power Law material model was used for the 3D geometry, but the transition zone was not included. Also, only two extreme cases of the inner zone fiber orientation, 0 and 90°, were investigated. Other modeling steps were the same as the 2D case.

3. Results

3.1. Pregnancy-associated increase in collagen fiber dispersion is heterogeneous across the cervix

Research has shown that cervical collagen fibers become more dispersed (disorganized) as pregnancy progresses [9–15]. However, our results show that these changes are heterogeneous across the tissue. We divided the rat cervix into three regions at mid-cervix based on location and structure: the inner zone, outer zone, and septum. Briefly, the outer zone surrounds the cervical canals; the septum connects the anterior and posterior parts of the outer zone; and the inner zone is the region immediately adjacent to the canals (Fig.1A). In a previous publication [24] we used different terms for the inner and outer zones, referring to them as near-septum and ring, respectively. Here, we also reference human anatomy and therefore use consistent terminology for cervical regions in both humans and rats. The septum only exists in rats. In this paper, the term “cervical regions” only refers to the distinct anatomical regions of the mid-cervix, not subregions throughout the length of the cervix.

Figure 1. Illustration of the three cervical regions and representative images at different gestational stages.

A: Schematic of the rat cervix and cross-section in the transverse plane highlighting different cervical regions. B: Representative SHG image of the entire cross-section of a 15-days pregnant cervix. C-F: outer zone, and G-J: inner zone images of non-pregnant (NP), 15-days pregnant (PG15), 17-days pregnant (PG17), and 21-days pregnant (PG21) cervices. Representative images were taken at comparable locations.

We performed SHG microscopy to determine region-specific changes in collagen fiber organization across the three cervical regions. Figure 1B shows the representative SHG image of a 15-days pregnant cervix with the regions labeled. Representative images of the outer and inner zones at different stages of pregnancy are shown in Fig. 1C–F and Fig.1G–J, respectively. Qualitatively, we note that collagen fibers in the outer zone become more crimped and disorganized as pregnancy progresses. Collagen fibers in the inner zone also appear more disorganized in later stages and occupy a larger proportion of the transverse imaging plane on day 21, compared to earlier stages.

We used Fourier transform SHG (FT- SHG) analysis to quantify changes in microstructure [35,47]. CV and SV provide a quantitative description of the degree of collagen fiber dispersion in 2D and 3D, respectively, where larger CV and SV indicate more disorganized collagen. The 3D microstructural parameters provide a more accurate description of collagen fiber organization since they can capture the orientation of both in-plane and out-of-plane fibers. Nevertheless, we characterize both CV and SV to facilitate comparison of our results with those of others, considering 2D imaging is more readily available and more widely used. Figure 2I compares the CV across gestational days and across cervical regions within a gestational day. CV increases monotonically as pregnancy progresses, indicating a decreased level of collagen fiber organization; this observation is consistent with previous studies quantifying the level of collagen fiber alignment in rat and human cervix [29,36,56,57]. The inner zone is significantly different from other regions in all pregnancy stages (p<0.05), and CV is consistently highest in the inner zone and lowest in the outer zone, meaning fibers appear more disorganized in the inner zone based on this 2D parameter. While the Spearman correlation between gestational day and CV is statistically significant in all regions (inner zone: p<0.05, septum and outer zone: p<0.01), the strength of the correlation, quantified by the Spearman’s correlation coefficient (ρ), differs across regions. CV in the outer zone and septum (ρ=0.85 and 0.82, respectively), has a strong correlation with gestational day, but the correlation is moderate in the inner zone (ρ=0.68). CV of the inner zone and septum correlate significantly (p<0.05), but there is no correlation between the CV of the inner zone and outer zone, or outer zone and septum.

Figure 2. CV (I) and SV (II) at different gestational stages.

A: CV and SV for the non-pregnant (NP), 15-days pregnant (PG15), 17-days pregnant (PG17), and 21-days pregnant (PG21) cervices. Violin plots show the kernel distribution of CV and SV at each gestational stage. Box plots show the distribution of CV and SV in the inner zone (IZ), outer zone (OZ), and septum (S). Statistical comparisons between the dispersion parameter (CV or SV) of different cervical regions within each stage are shown above the main figure. B: Representative quantitative results. Green schematics on the right illustrate the physical meaning of high and low values of each parameter. Larger CV and SV indicate more disorganized collagen in 2D and 3D, respectively. C: Spearman correlation coefficient quantifies the correlation between the gestational day, CV and SV in outer zone (CV_oz and SV_oz), CV and SV in inner zone (CV_iz and SV_iz), and CV and SV in septum (CV_s and SV_s). Size and color of the circles indicate the strength of correlation. The number of * indicate the significance level of the correlation, if one exists.

The region-specific changes of collagen fiber dispersion become more clear in 3D analysis in which the 3D dispersion parameter, SV, is not limited to describing the organization of the fibers in the imaging plane (Fig.2II). The Spearman correlation between gestational day and SV is statistically significant only in the outer zone (p<0.01) and septum (p<0.05), but not in the inner zone (p>0.1). The SV has a strong correlation with gestational day in the outer zone (ρ=0.80), moderate correlation in the septum (ρ=0.73), and negligible correlation in the inner zone (ρ=0.08). Interestingly, the SV in the inner zone increases with gestational day from the non-pregnant cervix to day 15 but then progressively decreases from day 15 to day 21. As with CV, the SV in the inner zone is significantly different from SV in the two other regions in all cases (p<0.05). The SV in different cervical regions do not correlate significantly (p>0.05). Collectively, our 2D (CV) and 3D (SV) results show that the increase in collagen fiber dispersion is not homogeneous across the cervical regions.

3.2. The dominant orientation of collagen fibers changes in the inner zone, but not in the other cervical regions

3D SHG images have not previously been quantitatively analyzed to distinguish different regions of the pregnant cervix. Our 3D analysis shows quantitatively that the dominant 3D orientation of collagen fibers changes during cervical remodeling. We quantified the average out-of-plane angle of collagen fibers (Φ) relative to the transverse plane (Fig.3A–C), where 0° and 90° indicate collagen fibers that are oriented parallel to the transverse and coronal planes, respectively. Like SV and CV, the evolution of Φ through gestation was region-dependent. Φ was not significantly correlated with gestational day in the outer zone (ρ=−0.146, p>0.1) or septum (ρ=−0.395, p>0.1). In the inner zone, however, we observed a strong and significant negative correlation between Φ and gestational day (ρ=−0.936, p<0.01); as pregnancy progressed, fibers in the inner zone changed from being predominantly parallel to the coronal plane to predominantly parallel to the transverse plane. Across the cervical regions, only Φ in the septum and outer zone correlate significantly (p<0.01), which is not entirely surprising based on their generally similar appearance. To our knowledge, this is the first direct report of the inner zone fiber remodeling from dominant orientation parallel to the canals to more outer zone-like, perpendicular to the canals. Thus, as pregnancy progresses, cervical collagen fibers remain mostly in-plane in the outer zone and septum, but become progressively in-plane in the inner zone despite being initially out-of-plane.

Figure 3. Φ at different gestational stages.

A: Φ for the non-pregnant (NP), 15-days pregnant (PG15), 17-days pregnant (PG17), and 21-days pregnant (PG21) cervices. Violin plots show the kernel distribution of Φ at each gestational stage. Box plots show the distribution of Φ in inner zone (IZ), outer zone (OZ), and septum (S). Comparison between the Φ of different cervical regions within each stage is shown above the main figure. B: Representative quantitative results. Green schematics to the right illustrate the physical meaning of Φ. C: Spearman correlation coefficient quantifies the correlation between the gestational day, Φ in outer zone (Φ_oz), Φ in inner zone (Φ_iz), and Φ in septum (Φ_s). Size and color of the circles both indicate the strength of correlation. The number of * indicate the significance level of the correlation, if one exists. D: Representative image of the coronal cross-section of a non-pregnant cervix, and F: A 17-days pregnant cervix. E and G show 3D quantification of selected inner zone regions from D and F, respectively.

To further verify the unexpected change of collagen fiber orientation in the inner zone, we imaged one non-pregnant cervix and one 17-days pregnant cervix in the coronal plane; fibers that are primarily parallel to the cervical canals appear in-plane when imaged in this orientation. We observed a high level of in-plane collagen fiber alignment, i.e. parallel to the canals, in the inner zone in the non-pregnant cervix, but not in the pregnant cervix (Fig.3D–G), a finding that agreed with results in transverse images. Comparison of coronal images of non-pregnant and pregnant cervices provided further evidence of significant change of collagen fiber orientation in the inner zone, from out-of-plane and parallel to the canals, to more in-plane and perpendicular to the canals.

3.3. The orientation of fibers in the inner zone influences cervical funneling

3D SHG microscopy of pregnant rat cervix revealed that the out-of-plane collagen fibers of the inner zone reorganize and become more in-plane in later stages of pregnancy. Remodeling of the inner zone region in rat cervix and its significance have not previously been described in the literature. We investigated the potential biomechanical role of fibers in this orientation in pregnancy maintenance, in particular in the case of abnormal remodeling, when fibers do not reorganize from parallel to the coronal plane to a new orientation parallel to the transverse plane. To analyze tissue-level consequences of variations in the microstructure of the inner zone, we performed a series of FE simulations to model the response of pregnant human cervical tissue to increasing intrauterine pressure, with the assumption that similar microstructural changes happen in both human and rat cervices. The structural similarities of the human and rat cervix [58] and their similar changes during the remodeling process detailed in several previous studies [33,35] support this assumption.

We generated a 2D axisymmetric model and partitioned it into five areas representing the inner and outer zones of the cervix, the uterus, and inner and outer transition zones such that there was not a sharp boundary between the cervix and uterus (Fig.S1, Supplementary). This 2D modeling approach allowed us to investigate the influence of structural parameters without confounding results with the influence of the cervical geometry and the uterine-cervix boundary. We assigned an Exponential Power Law model [52] to each cervical region (see Methods for detail on the FE and material models), using material properties from the literature [52]. In particular, we investigated the influence of the craniocaudal fibers (fibers parallel to the canals) on radial displacements of the internal os of the cervix in response to increased intrauterine pressure. To isolate this aspect of cervical remodeling, i.e., the dominant orientation of fibers in the inner zone, all of the other structural and material properties were kept constant. Our model accurately recapitulated the clinically observed dilation of the internal os that accompanies the cervical funneling, but did not adequately capture the variation of the funnel length; therefore, the dilation of the internal os is used as a measure of cervical funneling in this study. The radial displacement is compared to provide clinically-relevant information about the displacement of cervix when the microstructural parameters change.

Results from the simulation show that the internal cervical os has a ~6.5x larger radial displacement when fibers are out-of-plane (Φ=90°), which leads to a funneled shape compared to the case in which fibers are in-plane (Φ=0°). Figure 4 shows the displacement field of the cervix in response to a constant and biologically relevant intrauterine pressure for two representative cases: the collagen fibers in the inner zone oriented in-plane in 4A (Φ=0°) and out-of-plane in 4B (Φ=90°). The width of the inner zone is 50% the width of the cervical wall in both cases. Because of the difference in collagen fiber microstructure, the models respond differently to the same pressure.

Figure 4. Representative radial displacement field of the human cervix model from the finite element analysis.

A: the inner zone fibers are in-plane (Φ=0°). B: the inner zone fibers are out-of-plane (Φ=90°). The cervix funnels more in B where the inner zone collagen fibers are out-of-plane. The intrauterine pressure is 3.2 kPa for both cases, simulating the condition of cervix in pregnancy.

Figure 5A shows the radial displacement of the inner boundary of the cervix, normalized relative to the thickness of the cervix, for selected Φ. While the distal portion of the cervix does not move radially in all cases, the radial displacement of the proximal portion near the internal os, and therefore cervical funneling, increases with Φ. The fibers in the inner zone, when oriented out-of-plane (i.e., parallel to the canal), act to transfer load, thus allowing the cervix to open, or funnel, as the uterus expands. For fibers entirely parallel to the canal (Φ=90°), the radial displacement is approximately 40% the thickness of the cervix. The dependence of cervical funneling on Φ is not sensitive to the geometry of the model, boundary conditions, and material properties (Figs. S3–S6, Supplementary).

Figure 5. Influence of the inner zone out-of-plane fibers on the dilation of the 2D model of human cervix.

Normalized radial displacement of the inner boundary of the cervix is shown for selected Φ (A) and selected inner zone (IZ) widths (B). The arc length and radial displacement are normalized relative to the cervical length and the thickness of the cervix, respectively. The intrauterine pressure is 3 kPa for all cases.

The width of the inner zone also influences the radial displacement of the cervix. A wider inner zone that contains out-of-plane collagen fibers also leads to cervical funneling due to this robust, but perhaps undesirable, load transfer mechanism between the cervix and uterus. Figure 5B shows the normalized radial displacement of the inner boundary of the cervix for selected inner zone widths. The intrauterine pressure and Φ (90°) are the same in all cases.

To investigate the influence of collagen fiber dispersion on cervical funneling, we systematically changed the dominant orientation of fibers in the inner zone (0, 45, 90°) and the dispersion in the inner and outer zones (0, 0.1, 0.2) to simulate softening. Larger dispersion values indicate more disorganized fibers (range: 0-0.33). Figure 6 illustrates the combined influence of the orientation and dispersion parameters on the cervical funneling. Interestingly, the dispersion parameter has little influence on the cervical funneling; the out-of-plane fiber orientation in the inner zone has the strongest effect. Further, the influence of the dispersion parameter in the outer cervical zone is small when the inner zone fibers are oriented in-plane. Overall, softening of the inner or outer zones due to an increase in fiber dispersion, which could represent early cervical remodeling, does lead to some cervical funneling. However, the fiber orientation in the inner zone is the most important parameter for funneling.

Figure 6. Cervical funneling as a function of inner zone fiber orientation and dispersion parameters in the 2D model of human cervix.

The symbol color and size indicate the fiber orientation in the inner zone and outer zone dispersion, respectively. The intrauterine pressure is 0.7 kPa for all cases.

We also used a more anatomically accurate 3D, MRI-based cervical geometry from a pregnant woman at term to evaluate the influence of the out-of-plane fibers on cervical funneling. The solid model, provided by Fang et al. [55], is publicly available online as an STL file. We investigated whether we need a more complex and computationally expensive model to further validate our findings. The displacement fields of the models with in-plane and out-of-plane fibers are shown in (Fig. S7, Supplementary). As with the 2D case, the cervix funneled distinctly more when the inner zone fibers were oriented out-of-plane. For the purpose of our demonstration, the MRI-based model of human cervix supports our conclusions regarding the role of the out-of-plane fibers, and a more sophisticated model is not necessary for this purpose.

3.4. The cervix microstructure becomes more homogeneous as pregnancy progresses

We used the normalized standard deviation (SD), defined as the ratio of SD to the mean, to quantify the variation of structural parameters at different stages of pregnancy. All data points from each pregnancy stage, regardless of the region, contribute to the calculation of SD and mean. The normalized SD is zero when there is no variation in the data (e.g., all regions and data points within the regions are the same) but increases when the variability in data increases (e.g., the regions become more dissimilar). Figure 7 shows the variation of the normalized SD for different microstructural parameters as a function of pregnancy stage. Overall, normalized SD decreases, and thus the microstructure of the different regions converges to a more homogeneous structure during pregnancy, and the cervical regions become more similar to one another.

Figure 7.

Normalized standard deviation for different structural parameters at different stages of pregnancy

We used PCA to show how the microstructural parameters, calculated from SHG images of rat cervix, converge as pregnancy progresses. Principal component analysis is an orthogonal linear transformation that projects data from an original coordinate system onto a lower dimensional space such that the principal components are uncorrelated and the variance of the projected data is maximized [59]. For our results, PCA showed that structural parameters cluster closer to the mean as pregnancy progresses (Fig.8A), shown by the contraction of the ellipses, representing gestational day, that contain the corresponding data for all three regions. The contraction indicates a more homogeneous cervix microstructure at later stages of pregnancy.

Figure 8. Principal component analysis of the microstructural parameters.

A: graph of individuals in which individuals with similar profiles are grouped together. 95% confidence ellipse is shown for the non-pregnant (NP), 15-days pregnant (PG15), 17-days pregnant (PG17), and 21-days pregnant (PG21) cervices. B: graph of variables showing the relationship between the SHG parameters and the principal components. The angle between two vectors shows the degree of correlation; positively correlated variables have adjacent vectors, negatively correlated variables have opposite vectors, and uncorrelated variables have orthogonal vectors. Data show that CV and SV are highly correlated, but they are independent from Φ.

Principal component analysis also revealed that the CV and SV are highly correlated, but Φ is indeed an independent measure of structural organization; the CV and SV vectors are similar, and each is nearly orthogonal to the Φ vector (Fig.8B). The major axis of the confidence ellipse in Fig.8A is initially aligned with the orientation of Φ in Fig.8B, but then rotates clockwise from non-pregnant to 21-days pregnant, aligning more with the direction of SV and CV in later stages. This observation suggests that in non-pregnant/early pregnant cervices, Φ changes the most across different cervical regions, but in later stages, CV and SV change the most across the regions. Figure S8 (Supplementary) groups the data based on cervical region instead of pregnancy stage and serves to explain the variation of microstructural parameters within cervical regions.

4. Discussion

In this study, we quantified the 3D microstructural properties of rat cervix at multiple gestational stages and showed that unique, region-specific changes occur during cervical remodeling. Because of the regional changes, the initially distinct cervical regions converged to a more homogeneous microstructure with pregnancy. We used FE modeling to show that microstructural changes of the inner cervical zone could influence cervical funneling in human pregnancy, assuming similar microstructural changes take place in humans and rats. Funneling is significant because the larger the funnel becomes, the greater the risk that the fetal membranes (amnion) are exposed to normal vaginal bacteria. Bacteria and cytokines break down the membranes resulting in membrane rupture; preterm labor and birth follow. To our knowledge, we are the first investigators to quantitatively assess and report evolution of the 3D region-specific microstructural properties of the rat cervix during pregnancy and to illustrate its potential role in cervical funneling.

Characterization of tissue remodeling is usually limited to measurements from representative tissue regions or evaluation of the bulk properties [60,61]. Our results, however, suggest that the regional changes of tissue are indeed heterogeneous; in our model system, the out-of-plane orientation of collagen fibers had a strong, significant correlation with gestational day in the inner zone, but not in the other regions. Further, the cervical regions were not significantly correlated to one another when we investigated the microstructural properties; in most pairs (7 out of 9 pairs), correlations between the regions were insignificant (p>0.05). This observation indicates that the cervical regions are not evolving together, and changes of one region cannot be estimated from the changes of other regions. Region-specific or microstructure-specific mechanisms could be involved in cervical remodeling. Neglecting such regional changes may result in the loss of valuable information which could otherwise be used as an accurate measure of tissue condition. Therefore, local characterization of tissue properties is necessary for in-depth understanding of remodeling in the cervix and other tissues.

A surprising finding of this study was the structural changes of the inner zone of the rat cervix in which the fibers reorient from the coronal plane to the transverse plane. Our simulation results showed that out-of-plane fibers in the inner zone, if they were to remain intact throughout pregnancy, would lead to cervical funneling, which can, in turn, increase the risk of PTB. This observation suggests that the craniocaudal fiber remodeling in later stages of pregnancy could play a role in pregnancy maintenance that allows the cervix to stay closed while the uterine pressure accompanying growth of the fetus increases. Therefore, intact, out-of-plane fibers in the inner zone in pregnant women may signal a higher risk for PTB.

Previous OCT studies on tissues acquired by hysterectomy or cesarean hysterectomy of human uterus and cervix have shown that the width of the inner zone of the human cervix varies significantly between individuals [23], but its potential relationship with risk of preterm delivery has not been described. Using finite element simulations, we found that a wider inner zone with out-of-plane fibers could lead to cervical funneling. Therefore, we suggest that an abnormally wide inner zone early in pregnancy may predict an increased risk of cervical funneling in later stages when intrauterine pressure increases. Individuals with this condition may have higher risk of preterm delivery. More clinical investigations are required to verify the correlation between the width of the cervical inner zone and cervical funneling.

Characterization of the collagen fiber organization in the inner and outer zones and the size of these regions provide valuable information about the state of remodeling and the condition of the cervix during pregnancy. Despite its importance, there is currently no clinical method to determine cervical fiber organization in pregnant women; neither ultrasound imaging methods nor even quantitative ultrasound techniques provide a direct measure of collagen fiber organization [62]. Therefore, there is a need to incorporate new modalities capable of imaging collagen microstructure in routine cervical examinations. Second-harmonic generation microscopy, for example, is a promising technique that could have applications for in-vivo measurements beyond the laboratory, as described in previous studies [63,64]. Microstructural information, particularly if combined with patient-specific FE simulation as detailed in the current study, would improve the prediction of PTB by evaluating larger-scale changes of cervical tissue in response to pregnancy forces.

Intensity of the SHG signal was used in several previous studies as an indicator of collagen density and porosity [37,65–67]. In this work, however, we did not quantify intensity of the SHG signal, or the dark parameter, for two reasons. First, unlike comparison of different regions of the same tissue, comparison of SHG signal intensity across different tissue samples is influenced by sample-specific conditions such as use of laser power optimized for best illumination. Such differences in data acquisition may influence accuracy of comparisons across different samples. Second, the dark parameter does not provide information about underlying reasons for the weak signal; regions with a low concentration of collagen and regions with out-of-plane fibers both result in low-intensity regions in SHG images [68]. Thus, analysis of the dark parameter would not provide information about the microstructural changes that take place during tissue remodeling. It is worth emphasizing that analysis of the intensity of 2D images as a measure of collagen density could be misleading in some cases. Two-dimensional images may appear dark, signaling a low density of collagen, when the fibers are simply oriented outside the imaging plane. Accurate quantification of such regions is possible by imaging the tissue in multiple planes using 3D image acquisition and quantification.

There were several limitations to this study. We only quantified three structural parameters, CV, SV, and Φ. Other aspects of tissue microstructure could also change during cervical remodeling, and those properties could also influence the overall function of the tissue. Further, the finite element simulation did not model the contact between the fetal membrane and uterus and was not able to capture the variation of the funnel length. More realistic finite element simulation may provide more insight into important structural features of the cervix that change during pregnancy. Critically, the results of this study should be evaluated clinically in pregnant women to confirm that similar microstructural changes and displacements take place in human cervix. Currently, however, there is not a clinical modality that will give collagen organization information like that shown here.

5. Conclusions

We probed the 3D microstructural properties of the rat cervix at multiple gestational stages and demonstrated that cervical remodeling does not occur uniformly across cervical regions. The changes of the inner cervical zone, where collagen fibers orient progressively more in-plane with pregnancy, were quite distinct from the other regions. These changes proved to have a significant influence on the dilation of the internal cervical os, which is related to cervical funneling and the risk of PTB. Collectively, our results provide a new explanation for a possible origin of cervical funneling and illustrate how microscale changes of tissue can control tissue-level mechanical behavior.

Statement of significance.

Cervical funneling, or dilation of the internal cervical os, is highly associated with increased risk of preterm birth. This study explores the 3D microstructural changes of the rat cervix during pregnancy and illustrates how these changes influence cervical funneling, assuming similar evolution in rats and humans. Quantitative imaging showed that microstructural remodeling during pregnancy is nonuniform across cervical regions and that initially distinct regions become more similar. We report, for the first time, that remodeling of the inner cervical zone can influence the dilation of the internal cervical os and allow the cervix to stay closed despite increased intrauterine pressure. Our results suggest a possible relationship between the microstructural changes of this zone and cervical funneling, motivating further clinical investigations.

Data availability

The data that support the findings and plots of this study are available from the corresponding author upon reasonable request.