Detecting benign uterine tumors by autofluorescence lifetime imaging microscopy through adjacent healthy cervical tissues

Maojia Huang*, Zixiao Zhang*, Xinyi Wang*, Yonghui Xie, Yiyan Fei*, Jiong Ma*, Jing Wang, Li Chen, Lan Mi*||†† and Yulan Wang¶**†† *Department of Optical Science and Engineering Shanghai Engineering Research Center of Ultra-precision Optical Manufacturing Fudan University, Shanghai 200433, P. R. China Department of Pathology, The Central Hospital of Wuhan Tongji Medical College, Huazhong University of Science and Technology Wuhan 430014, P. R. China State Key Laboratory of High Field Laser Physics Shanghai Institute of Optics and Fine Mechanics Chinese Academy of Sciences, Shanghai 201800, P. R. China School of Arts and Sciences, MCPHS University Boston, MA 02115, U. S. A. ¶Department of Gynecology, The Central Hospital of Wuhan Tongji Medical College Huazhong University of Science and Technology Wuhan 430014, P. R. China ||lanmi@fudan.edu.cn **wyl407@163.com


Introduction
Optical imaging techniques especially°uorescence lifetime imaging microscopy (FLIM) are promising for the detection of malignancies [1][2][3][4][5][6] Di®erentiating malignancies from normal tissues by auto-°u orescence can rely on the biochemical and tissue morphological changes. The traditional gold standard technique for tissue characterization is haematoxylin and eosin (H & E) stained histopathology, based on morphological changes. However, biochemical change occurs when the cellular metabolism becomes abnormal in the case of rapid cell division, which occurs before morphological changes become apparent. 7,8 That means optical techniques have the great potential to detect malignancies or precancer much earlier than traditional histopathology.
It is well known that the endogenous°uorophores in cells and tissues include reduced nicotinamide adenine dinucleotide (phosphate) (NAD (P)H),°avin adenine dinucleotide (FAD), collagen, and elastin. 9 The structural proteins, mainly collagen and elastin, can show the extracellular matrix of tissues re°ecting the tissue morphology and structure. 10,11 While NAD(P)H and FAD are known as classic molecules involving in oxidative phosphorylation and glycolysis, [12][13][14][15] which can be used to describe the cellular metabolic environment in tissues. Based on FLIM, di®erent cancers including cervical cancer, 4 lung cancer, 6 oral cancer, 16 breast cancer, 17 were studied on the auto°uorescence for early detection. Nevertheless, little attention was paid on benign tumors so far.
Uterine leiomyomas (¯broids) are common gynecologic benign tumors in women, whose rate were reported more than 50-80% of reproductive-aged women. 18 Adenomyosis is another common benign gynecological disease. The estimates of its incidence vary widely and were reported from 5% to 70%, 19 with the mean frequency of adenomyosis at hysterectomy given as approximately 20-30%. 20 The con¯rmed diagnosis of these common diseases for women currently requires minimally invasive surgery or hysterectomy to obtain the uterine tissues. 21 However, minimally invasive surgery is generally not suitable for patients with adenomyosis. Furthermore, hysterectomy is contraindicated for women with a desire to preserve their fertility. For this reason, new technologies are needed to aid the detection of benign tumors in uterus. If this could be achieved at the very early stage of the benign tumor development, drug therapy can be prescribed and administered continuously as an alternative to surgery.
Since NAD(P)H and FAD are important coenzymes in cellular energy metabolism that can emit°uorescence, the abnormal metabolism can be monitored by FLIM. Though the benign tumor develops in uterus, the cellular metabolic status at the adjacent site of cervix may be a®ected and revealed by FLIM.
In this study, we proposed a novel method for detecting benign uterine tumors such as leiomyomas and adenomyosis at the adjacent site of healthy cervix through¯xed cervical tissue samples. It is very promising to apply FLIM method to identify benign tumors in uterus by means of cervix biopsies. It suggested an approach to avoid surgery for benign tumor detection. Wuhan, and all the involved patients provided informed consent to participants. The histologic diagnoses were provided by Department of Pathology, the Central Hospital of Wuhan. The samples comprised 17 patients undergoing hysterectomy, including 8 women with uterine leiomyomas and 9 women with adenomyosis or coexisting with leiomyomas. The control (normal) group included 10 women who underwent biopsies with a histologic diagnosis of healthy uterus and cervix. The average age of all the patients was 50: 8 AE 8:6 years old, ranging between 35 and 71 years old. The participants had received at least one health examination before the gynecological surgery or biopsy, including blood pressure (BP), thyroid hormones (THs) and thyroid-stimulating hormone (TSH) tests, blood glucose (Glu), body mass index (BMI). Their personal medical histories were also recorded. Table 1 lists the ages, chronic diseases that a®ect metabolism, and medical history of patients.

Materials and
The cervical tissue samples embedded in para±n were obtained after the surgeries or biopsies with the standard procedures, 22 by pathologists from the Department of Pathology. One unstained tissue slice of 4 m thick from each patient was cut o® and placed on a glass slide for FLIM measurements. After FLIM imaging, all the tissue slices were stained by H & E and examined by pathologists.

Experimental imaging setup
The°uorescence lifetime imaging for cervical tissue samples were acquired by a time-correlated single  To get the photons from the two PMTs correctly separated into two channels, the delay of the system was set to 20 ns for the two-channel system optimization. A°uorescence spectrum of a cervical tissue sample was shown in Fig. S1 as an example, which was excited by a 405 nm continuous-wave (CW) laser. It can be seen that the°uorescence of FAD is about 6 times higher than that of NAD(P) H. In this case, the bleed-through of NAD(P)H for the FAD detection has a little impact. Therefore, a more suitable¯lter or excitation laser should be used for the detection of FAD signal if samples have higher NAD(P)H contributions.
Each FLIM image of 256 Â 256 pixels was acquired in 60-120 s and the area with the size of about 280 m Â 280 m was imaged only once to avoid photobleaching. At least six di®erent areas were imaged for each slide. It should be noted that the measurements were applied on the epithelium part of tissue samples, while the epithelium length in each slice is di®erent. Some slices with longer epithelia could be measured by both single and twochannel TCSPC FLIM, but those with shorter Fig. 1. Schematic of the two-channel TCSPC FLIM system for detecting benign uterine tumors though adjacent healthy cervical tissues. epithelia could be measure by either single or two-channel system. As listed in Table 1, about half of patients (9 of 17) undergoing hysterectomy with uterine leiomyomas or adenomyosis were measured by both single and two-channel systems. However, the tissue samples obtained from the normal group of patients who underwent biopsies were fairly small. Only 1/10 had enough length of epithelia to be detected by both single and two-channel systems.

Data analysis
Time-decay data of each pixel in FLIM images were was¯tted with multi-exponential decay models using the commercial SPCImage software (Becker & Hickl). The mask de¯nition was applied for a region of interest (ROI) selection on the super¯cial and midzone layers in each FLIM image. The data is calculated from the pixels inside the ROI.
When the FLIM images were collected by the single-channel TCSPC system, triple-exponential function was used for¯tting. The mean°uorescence lifetime of each pixel is calculated as t m ¼ a 1 t 1 þ a 2 t 2 þ a 3 t 3 , where t 1; 2; 3 is the lifetime of the°uorescent component, a 1; 2; 3 is the corresponding contribution of the exponential component (a 1 þ a 2 þ a 3 ¼ 1). The t m distribution curve of the 256 Â 256 pixels in each FLIM image was obtained by the SPCImage software, and the peak of the distribution curve was recorded.
When the FLIM images were collected by the two-channel TCSPC system, double-exponential decay analysis was applied to the FLIM images of NAD(P)H or FAD as previously reported. 26 The mean°uorescence lifetime of each pixel is calculated as t m ¼ a 1 t 1 þ a 2 t 2 . For NAD(P)H, the°uorescence lifetime of free NAD(P)H solution was reported between 0.3 and 0.7 ns. [27][28][29] In this study, free NADH solution was measured and it was found that the°uorescence lifetime of free NADH is 0.58 ns (see Fig. S2). Therefore, the value 0.5 ns was set as the short-lifetime component t 1 of NAD(P)H. Then, the long lifetime component t 2 (bound-NAD(P)H), and their relative fractional contributions (a 1 and a 2 ) were estimated by double-exponential¯tting for NAD(P)H analysis. FAD FLIM images were analysed similarly. Free FAD solution was¯rst measured to obtain t 2 of FAD as 2.9 ns, and then the double-exponential¯tting was used to estimate t 1 of FAD, and their relative fractional contributions (a 1 and a 2 ). Based on the analysis of NAD(P)H and FAD FLIM images, the t 2 of NAD(P)H and t 1 of FAD distribution curves of each FLIM image was obtained and the curve peak of each image was analysed. All the goodness-of-¯t 2 values were below 1.4 indicating good¯ts.

FLIM images of cervical tissue
The cervical tissue samples were¯rst detected by the single-channel TCSPC system with a¯lter of 430 nm long-pass. As an example, Fig. 2 showed the auto°uorescence FLIM images of an unstained slice and H & E images of the stained slice. The images exhibited the typical structure of normal cervix, which was covered by squamous epithelial cells, containing super¯cial, midzone and basal layers. The white dash line in Fig. 2 indicates the boundary of the basement membrane with columnar cells. Figure 2(b) displayed a capillary in the epithelium of cervix, which was composed of vascular endothelial cells. It can be seen that the FLIM images present the cellular morphology features and the characteristics of epithelium structure clearly as the traditional H & E staining method.
As the lifetime bar shows, the red-orange color represented short-auto°uorescence lifetime, and the blue color represented long lifetime. In Fig. 3, the epithelial structure and the basement exhibited di®erent characteristics. For a cervix tissue slide of the normal group, the°uorescence lifetime of super¯cial layer (Area 1 in Fig. 3(a)) was fairly short around 0.93 ns, while the lifetimes increased gradually in deeper layers to around 1.73 ns in Area 3. For a cervix tissue slide of uterine leiomyomas patients, the lifetime of super¯cial layer (Area 1 in Fig. 3(b)) was about 1.92 ns and decreased to 1.38 ns at the boundary of the basal layer (Area 2 in Fig. 3(b)), and that of basement across the boundary (Area 3 in Fig. 3(b)) was up to 2.03 ns. For the uterine adenomyosis patients, there was no signi¯cant change at di®erent layers of the epithelium (between Area 1 and Area 2 in Fig. 3(c)). The lifetime increased over 0.3 ns in the basement (Area 3 in Fig. 3(c)) comparing with in the epithelium. It should be noted that all the images in Fig. 3 were from healthy cervixes, which had similar structures and clear distinct layers. However, the auto-°u orescence lifetime values showed di®erent patterns for the normal group and benign tumor groups.
The mean°uorescence lifetime distributions of FLIM images were analysed with the triple-exponential decay model and averaged with at least six FLIM images for each patient. Twenty-two patients were divided into three groups, normal (n = 5), uterine leiomyomas (n = 8), and uterine adenomyosis (n = 9). The scatter points corresponding to the averaged lifetime for each patient were shown in Fig. 4. As shown in Fig. 4(a), the°uorescence lifetime is linearly related to patient age with a goodness of¯t R 2 ¼ 0:999. It should be noted that the a green scatter point of Patient No. 2 was excluded for¯tting. The cervical tissue of this patient (No. 2) was collected in a biopsy and measured in Nov 2017 in this work. However, the patient who underwent a operative hysteroscopy was found multiple endometrial polyps in Jan 2019. Because this data is quite di®erent from other data, we assume that she might have endometrial polyps at the time of FLIM measurement, but it was not found in the biopsy examination. A linearly¯tted line with R 2 ¼ 0:411 was also shown in Fig. 4(b). It can be found that the average°uorescence lifetime values of normal (black squares) and uterine leiomyomas (red circles) groups decreased with age, while those of uterine adenomyosis (blue triangles) seemed to have no correlation with age. Similar age dependence were reported for NAD(P)H and FAD redox ratio of normal cervical tissues based on spectroscopical analysis by our group 7 and for NAD(P)H concentration in breast tissues by Gupta et al. 30 As our group reported, the age dependence was only observed in the normal cervix group between patients aged 45 years and >45 years, whereas those of cervical intraepithelial neoplasia (CIN) or cervical cancer groups were indistinguishable between 45 and >45 years. In this study, the group of adenomyosis patients revealed similar feature as CIN and cervical cancer groups reported previously. 7 Therefore, the result may imply that uterine adenomyosis growth is more invasive compared with leiomyomas, which is consistent with clinical manifestations.

NAD(P)H and FAD FLIM images of cervical tissue
Considering the FLIM images (Figs. 2-4) were excited by 405 nm laser and detected by the singlechannel TCSPC system with a 430 nm long-pass lter, the sources of°uorescence may include various endogenous°uorophores, such as NAD(P)H, FAD, collagen, elastin, and porphyrins. It is complicated to analyze the metabolism change only based on the auto°uorescence with a 430 nm long-pass¯lter. For this reason, FLIM of the two coenzymes, NAD(P)H and FAD, were simultaneously recorded by the two-channel TCSPC system with lters of 417-477 nm band-pass and 508 nm longpass for further study.

Statistical analysis of NAD(P)H and FAD
For example, based on the result in Fig. 5, the peak of t 2 of NAD(P)H curve (a-ii) was 1.51 ns, and that of t 1 of FAD (b-ii) was 0.90 ns. At least six pairs of peak values of t 2 of NAD(P)H and t 1 of FAD were obtained for each patient, and then the average values and standard deviation of t 2 of NAD(P)H versus t 1 of FAD were obtained and shown in Fig. 6. A total of¯fteen patients were analysed, including 6 normal, 4 uterine leiomyomas and 5 uterine  adenomyosis, whose average age was 53:9 AE 8:9 years old. The patients with age < 40 years were excluded owing to the limited number of young patients with adenomyosis or leiomyomas. In addition, the patients with age < 40 years may have di®erent metabolic status. The 15 patients were marked their numbers, chronic diseases and medical histories in Fig. 6. Their ages and other major medical informations were also listed in Table 1. It was found that the healthy uterine group (n ¼ 6) and the group of benign uterine tumors (n ¼ 9) were in two di®erent areas (Fig. 6). All of the benign uterine tumor patients were in the lower left. Most of the healthy uterine group (5 of 6) were in the upper right. The separating line (y ¼ À5:70x þ 7:51, green dash line in Fig. 6) was performed using the linear discriminant analysis with the accuracy of 92.1%. It should be specially noted that Patient No. 5 is in the healthy uterine group, who has a long surgery history. This patient underwent modi¯ed radical mastectomy for breast cancer, partial colectomy for colon adenomas, cholecystectomy, and lumpectomy for breast cancer metastasis during the past ten years. It is reasonable to hypothesize that the metabolic status of Patient No. 5 with a malignant tumor history is extraordinarily di®erent from others of the healthy uterine group. Therefore, the result suggests that FLIM might be useful for the discrimination of healthy uterine and benign uterine tumor by detecting the   6. Di®erentiating benign uterine tumors from healthy uterine group by studying the cervix tissues. The plot t 2 of NAD(P)H versus t 1 of FAD was obtained from 15 patients, whose ages, chronic diseases and medical histories were in Table 1. BC: breast cancer; CA: colon adenomas; BCM: breast cancer metastasis; DM: diabetes mellitus; HT: hyperthyroidism; HT-H: hypothyroidism; HTN: hypertension; OBS: obesity; OWT: overweight. °uorescence lifetime of NAD(P)H and FAD from the adjacent cervical tissues.

E®ect of chronic diseases on metabolism
It is well known that there are some chronic diseases such as diabetes, hyperthyroidism, hypertension, and obesity, associating with metabolism problems or a®ecting metabolism level. 31,32 Patients Nos. 13 and 18 highlighted in Fig. 6 showed much lower t 2 of NAD(P)H than others. Patient No. 13 was diagnosed with diabetes, hyperthyroidism, and hypertension. Patient No. 18 has hypothyroidism and obesity. For a comparison, Patient No. 19 has only hypertension, whose data showed little di®erence than others. It may indicate that there is a correlation between the chronic diseases and metabolically abnormal environment, leading to lower t 2 of NAD(P)H. For Patient No. 6, who is an overweight patient, her data was in the area of the lower left, but showed larger standard deviation on t 1 of FAD than others. Moreover, Patient No. 18 with hypothyroidism and obesity showed much larger standard deviation on t 1 of FAD. The larger variability of one patient may indicate the variation of cellular metabolism and microenvironment became larger with weight gain progression. Further investigation is needed with a larger number of cases.

Conclusion
NAD(P)H and FAD are important coenzymes in cellular energy metabolism. The FLIM measurements of NAD(P)H and FAD in the adjacent healthy cervical tissue may be useful for detecting the benign uterine tumors such as leiomyomas and adenomyosis. This method proposed a novel strategy for detecting uterine leiomyomas and adenomyosis to avoid hysterectomy by measuring the adjacent healthy cervical tissues. It was also found that ages, diabetes, hyperthyroidism, and obesity could a®ect metabolic level and environment, as well as a malignant tumor history.