Individual-oocyte transcriptomic analysis shows that genotoxic chemotherapy depletes human primordial follicle reserve in vivo by triggering proapoptotic pathways without growth activation

Gonadotoxic chemotherapeutics, such as cyclophosphamide, can cause early menopause and infertility in women. Earlier histological studies showed ovarian reserve depletion via severe DNA damage and apoptosis, but others suggested activation of PI3K/PTEN/Akt pathway and follicle ‘burn-out’ as a cause. Using a human ovarian xenograft model, we performed single-cell RNA-sequencing on laser-captured individual primordial follicle oocytes 12 h after a single cyclophosphamide injection to determine the mechanisms of acute follicle loss after gonadotoxic chemotherapy. RNA-sequencing showed 190 differentially expressed genes between the cyclophosphamide- and vehicle-exposed oocytes. Ingenuity Pathway Analysis predicted a significant decrease in the expression of anti-apoptotic pro-Akt PECAM1 (p = 2.13E-09), IKBKE (p = 0.0001), and ANGPT1 (p = 0.003), and reduced activation of PI3K/PTEN/Akt after cyclophosphamide. The qRT-PCR and immunostaining confirmed that in primordial follicle oocytes, cyclophosphamide did not change the expressions of Akt (p = 0.9), rpS6 (p = 0.3), Foxo3a (p = 0.12) and anti-apoptotic Bcl2 (p = 0.17), nor affect their phosphorylation status. There was significantly increased DNA damage by γH2AX (p = 0.0002) and apoptosis by active-caspase-3 (p = 0.0001) staining in the primordial follicles and no change in the growing follicles 12 h after chemotherapy. These data support that the mechanism of acute follicle loss by cyclophosphamide is via apoptosis, rather than growth activation of primordial follicle oocytes in the human ovary.

Ovarian xenografts. Ovarian cortical pieces from four cadaveric organ donors aged ≤ 32 years were cryopreserved according to our established protocol 9,10 . Tissue pieces 2-3 mm 3 (2-3 × 1 × 1 mm) were xenografted subcutaneously to SCID mice, as described previously 4 . Briefly, after animals were anesthetized, a small incision was made on the dorsal midline of the animal. Human ovarian cortex fragments from four different donors were transplanted to each animal (n = 4 tissue pieces/animal). The grafts were distributed in a way that a sample from each different donor was present in each condition. After allowing xenografts to fully vascularize for 10 days, the mice were given a single intraperitoneal (i.p.) dose of cyclophosphamide (75 mg/kg) or the vehicle. The tissues were then recovered 12 h later and prepared for either histology and immunohistochemistry or embedded in a medium for frozen sectioning and isolation of primordial follicle oocytes by laser capture dissection microscopy.
All methods were carried out in accordance with relevant guidelines and regulations consistent with the Declaration of Helsinki. The utility of human ovarian tissue was exempted by the Yale Institutional Review Board as the tissues are from deceased, non-living cadavers. Research on deceased organ donor cadaver tissue is not considered human subjects research and is exempt from IRB oversight. However, the LiveonNY scientific board reviewed and approved our study protocol and consent was obtained from the relatives of deceased organ donors to donate ovarian tissue to research, along with other organs for transplant procedures.
Immunohistochemistry. Ovarian tissues were fixed in 4% paraformaldehyde (PFA), embedded in paraffin, and serially sectioned at 5-micron intervals. Sections were stained with Hematoxylin and Eosin (H&E) for differential follicle counts, as previously described 4 . Every fifth section of an entire xenograft was chosen for staining (7-8 stained slides from a total of 30-35 slides per xenograft) and evaluated by observers who were blinded to the intervention and each other's findings. To avoid over-counting, follicles were recorded only when the nucleus was visible. Follicles were classified based on our previously established criteria 11 . A primordial follicle was defined as an oocyte surrounded by a single layer of flattened, squamous follicular cells. A primary follicle was defined as an oocyte surrounded by a single layer of cuboidal granulosa cells. A follicle was classified atretic based on morphological criteria, including nuclear pyknosis, cytoplasm contraction, and nuclear pyknosis of granulosa cells and dissociation of the granulosa cells from the basal membrane 12,13 .
Randomly selected sections coming from ovarian tissue of least 3 different individuals were used for immunohistochemistry for Anti-active Caspase-3 (AC-3, 1:1000, AF-835, R&D Systems), γH2AX (1:250 IHC-00059, Bethyl Laboratories), rabbit phospho-Akt (S473, 1:400, #4060S, Cell Signaling Technology, Denver, MA), rabbit phospho-rpS6 (S235-236, 1:100; #2211S, Cell Signaling Technology, Denver, MA), and phospho-FOXO3A (S253, 1 µg/ml, # ab47285, Abcam, Cambridge, MA, USA) 3 . Images were captured on an Olympus IX73 microscope. For double immunofluorescent staining, the slides were deparaffinized, subjected to heat-induced antigen retrieval with citrate buffer and subsequently blocked with 1% BSA and 5% normal goat serum. The mix of primary antibodies BAD (1:250, A302-384A, Bethyl Laboratories) and Bcl2 (1:200, sc-509, Santa Cruz Biotechnology) was used. The secondary antibodies were conjugated with Alexa 488 (1:500, A11001, Invitrogen) and Alexa 594 (1:500, A11012, Invitrogen). The slides were counterstained with DAPI (1 µg/ml, D3571, Invitrogen), and mounted with ProLong Gold (P36930, Life Technologies). The imaging and intensity calculations were done on Nikon 90i Eclipse microscope. By overlapping both BAD and Bcl2 staining color channels, and selecting the area around primordial follicles, we calculated the mean pixel intensities. We compared the intensity means between primordial follicle oocytes of vehicle and chemo groups after subtracting the background, which was set based on the negative control, and by setting a constant threshold for both channels across the images. Stringent experimental conditions ensured that the threshold values did not change significantly from day to day of image acquisition. For the calculation, we used 3 slides of each xenografted tissue, and 'n' represents the number www.nature.com/scientificreports/ of different tissue pieces. The analysis was done using ImageJ software. Additional representative images were acquired on a Leica SP8 Gated STED 3X Super Resolution confocal microscope using 60 × oil immersion lens.
Laser-capture microdissection (LCM) and isolation of single primordial follicle oocytes. After harvesting ovarian xenografts, tissue blocks were immediately prepared in optimum cutting temperature (OCT) compound and stored at − 80 °C until cryostat sectioning (− 20 °C). These were later sectioned at 8-micron thickness and stained with H&E according to the manufacturer's protocol. Based on the morphological appearance, i.e., follicle with a single layer of flat granulosa cells around with positively stained oocyte inside, single primordial follicle oocytes were individually isolated using an LCM excluding any pre-granulosa cell from the dissection (see Supplementary Video S1). LCM was performed on the Leica LMD7000 microscope, under 40 × magnification. To each Eppendorf tube, filled with cell lysis buffer (RNase inhibitor in 0.2% (vol/vol) Triton X-100 solution), the individual primordial oocyte was collected from 2-3 adjacent tissue sections to ensure that one full cell was included (Leica Microsystems, Buffalo Grove, IL, USA).
RNA-seq analysis on laser captured single primordial follicle oocytes. The primordial follicle oocytes were processed for sequencing with modifications of a previously established protocol on non-germline single cells for LCM material 14 . Samples were sequenced to depths of up to 44 million read pairs, 75 nt length reads per sample using the Illumina Rapid v2 kit (75 cycles) on an Illumina HiSeq2500 Sequencing System. Image analysis, base calling, and generation of sequence reads were produced using the HiSeq Control Software v2.0 (NCS) and Real-Time Analysis Software (RTA). Data were converted to FASTQ files using the bcl2fastq2 v1.8.4 software (Illumina Inc.). The reads were trimmed for quality. Those were then aligned to the human reference genome (hg38 gencode for humans) using HiSAT2 15 . The alignments were processed using Ballgown (free online software) 15 , and per-gene counts were obtained. The raw counts were processed using DESeq2 and R. Statistics. For RNA-seq analysis, we determined statistical significance for differential gene expression based on the q-value (or the False Discovery Rate), with stringent cut-offs of 0.05 or less. This ensures that the Type II error is kept as low as possible, given the constraints of sample size. Our observations of gene transcript abundances by RNA-seq from a relatively low sample-size are justified as we observe similar trends while interrogating the samples using other more direct biological and biochemical approaches such as qRT-PCR analysis and immunohistochemistry. Other data from chemotherapy-and vehicle-treated groups were compared using Student's two-tailed t-test. P values less than 0.05 were considered significant. For statistical analysis we used GraphPad Prism 6 (GraphPad Software, San Diego, CA).

Cyclophosphamide treatment does not increase primordial follicle growth initiation in vivo;
it induces DNA DSBs and apoptotic death. First, to morphologically determine if cyclophosphamide exposure increases the entry of primordial follicles into the growth pool, we calculated the ratio of primary to primordial follicles (py/pd = follicle growth initiation index) in xenografts 17 . Based on the growth initiation ratio, we are only able to give an estimate, and not a definite answer of the follicle fate since we use only a small fragment of the cortex and of the unequal distribution of the follicles in each tissue. For H&E and AC-3 staining, we analyzed a total of 70 follicles (51 primordial and 19 primaries) from 3 slides of 8 different tissue pieces for the vehicle group and 202 follicles (177 primordial and 25 primaries) from 3 slides of 10 pieces of cortex from the chemo-treated group. We found no significant difference in the py/pd ratio between the vehicle and the cyclophosphamide -treated group (34.62 ± 7.94, vs. 19.97 ± 6.79; p = 0.17) (Fig. 1A). Because atretic follicles may not initiate growth, we also analyzed the ratio of non-atretic py/pd follicles. We, again, did not observe a significant difference in follicle growth initiation rate between the vehicle-and cyclophosphamide-treated groups (34.59 ± 9.75 vs. 18.58 ± 10.21; p = 0.08) (Fig. 1B). Next, we studied whether cyclophosphamide induces apoptotic oocyte death using AC-3 immunostaining. We found a significant increase in the percentage of apoptotic primordial (65.03% ± 3.55% vs. 28.27% ± 5.77%; p < 0.0001 cyclophosphamide vs vehicle) and primary follicles (63.67% ± 9.89%, n = 8 vs. 28 Individual-oocyte RNA-seq analysis of pathways activated in primordial follicles in acute response to chemotherapy in vivo. We performed high throughput RNA-seq on non-atretic, individual primordial follicle oocytes from cyclophosphamide-and vehicle-treated xenografts to have a broader view of the pathways that are altered in primordial follicles in acute response to chemotherapy. For each condition, we used two individual primordial follicles from each of the two different donor ovaries. On average, 29.8 mln reads were obtained per sample with 74% alignment to the human genome. To verify that the obtained material represents oocyte transcriptome, the expression of oocyte-specific genes, such as ZP3, KIT and DDX4 was confirmed in all the samples. We found that 190 genes were differentially expressed between the chemo-and vehicle-treated groups (fold change ≥ 2, p < 0.05) ( Fig. 2A  Interleukin-8 signaling (p = 0.03) were suppressed. We also found significant changes in Axonal Guidance Signaling (p = 0.0003), Gap junction signaling (p = 0.01), and Focal Adhesion Kinase (FAK) Signaling (p = 0.04), but no predictions could be made as to the direction of change. The Phospholipase C Signaling, Ephrin Receptor Signaling, and IL8 signaling pathways are known to be associated with inhibition of apoptosis. We used IPA to generate the network connecting the genes present in these pathways with the apoptosis function (p = 5.37E −19) and to predict if apoptosis is favored based on the expression status of these genes in our data. The IPA suggested activation of apoptotic processes in the chemotherapy-treated samples ( Supplementary Fig. S1). Further, we used IPA to find the overlap between the differentially expressed genes after chemotherapy exposure and those genes in the IPA knowledge base associated with the functions of apoptosis and activation of primordial follicles. We found in our data set that PTEN, Akt-1 Foxo3, BAD, ANGPT1, Pde and BTG2 overlapped with apoptotic function (p = 3.58E−18), while PTEN and KITLG overlapped with the activation function of ovarian follicles (p = 7.90E−07). The analysis of our dataset showed that the growth activation of the primordial follicle is inhibited while apoptosis is triggered in acute response to chemotherapy in vivo (Fig. 2C).  (Fig. 3D-F). These findings suggest that cyclophosphamide treatment does not result in the activation of follicle growth or anti-apoptotic pathways but triggers an acute apoptotic response.
qRT-PCR and immunohistochemical confirmation of chemotherapy-induced acute pathway changes in human primordial follicles. To further explore and confirm the expression patterns of the PI3K/PTEN/Akt pathway members in primordial follicle oocytes in acute response to chemotherapy, we performed qRT-PCR analysis on LCM, individual primordial follicle oocytes. Based on the relative gene expression analysis, we found no change in the expressions of Akt-1 (n vehicle = 5 and n chemo = 6; p = 0.93), rpS6 (n vehicle = 5 and n chemo = 6; p = 0.32) and Foxo3a expression (n vehicle = 5 and n chemo = 6; p = 0.12) 12 h after cyclophosphamide exposure compared to the vehicle (Fig. 4A). To further confirm our findings, we investigated the changes in the immunohistochemical expression of the key PI3K/PTEN/Akt pathway proteins in response to chemotherapy (Fig. 4B). We calculated the percentage of positive to the total number of follicles from 3 slides per xenografted tissue, 'N' indicating the number of primordial follicles analyzed, 'n' indicating the number of xenografts from which the follicles originated. We observed no significant differences between the vehicle and cyclophosphamide treatment groups in the expressions of p(phospho)-Akt (7.29 ± 0.22, N = 58, n = 8 vs. 11.59 ± 0.8, N = 149, n = 6; p = 0.52), p-Foxo3a (12.03 ± 1.48, N = 38, n = 9 vs. 18.85 ± 1.43, N = 162, n = 8; p = 0.63), and p-rpS6 (6.06 ± 0.14, N = 38, n = 11 vs.12.55 ± 0.94, N = 114, n = 7; p = 0.4). These results further confirm that chemotherapy does not activate the PI3K/PTEN/Akt pathway in primordial follicles within 12 h of exposure.

Chemotherapy activates pro-apoptotic BAD-Bcl2 action. Bcl2 (B-cell lymphoma 2) is a known
anti-apoptotic gene 23 . It is localized to the outer membrane of the mitochondria, and it plays an important role in promoting cell survival via inhibiting the actions of pro-apoptotic genes such as the BAD. By qRT-PCR analysis of laser-captured primordial follicle oocyte we observed a trend towards declining expression of Bcl2 (n vehicle = 5 and n chemo = 6; p = 0.17), but together with BAD (n vehicle = 5 and n chemo = 6; p = 0.93), it did not reach statistical significance in the cyclophosphamide -treated xenografts vs. the vehicles (Fig. 5A,B). Certain stress conditions,  www.nature.com/scientificreports/ such as oxidative stress and DNA damage, initiate the Bcl-2 protein family-related apoptotic response [24][25][26] . BAD is activated by dephosphorylation and dimerizes with BCL2, preventing it from its anti-apoptotic function, and promote apoptosis 27,28 . Therefore, we studied the BAD-Bcl-2 colocalization in response to chemotherapy. We showed a significantly increased colocalization of BAD and BCL2 protein staining in cyclophosphamidetreated samples (N = 18, n = 6) compared with the vehicle (N = 20, n = 7; p = 0.004) (Fig. 5C,D), which may suggest increased protein colocalization and potential interactions. 'N' indicating the number of primordial follicles analyzed, 'n' indicating the number of xenografts from which the follicles originated. These results collectively indicate that within 12 h, chemotherapy exposure results in increased liability to apoptotic death in primordial follicle oocytes, possibly via the BAD-Bcl2 pathway.

Discussion
Maintenance of primordial follicle reserve is crucial for the reproductive life span of a female individual. This reserve incurs a massive loss when a patient undergoes chemotherapy treatment, later presenting itself as early menopause and infertility. In this study, we demonstrated the molecular mechanisms of chemotherapy-induced acute damage to ovarian primordial follicles using our unique in vivo human ovarian xenograft model and singlecell transcriptomic approaches. Because our earlier work showed that human primordial follicle apoptosis peaks www.nature.com/scientificreports/ 12 h post-exposure to chemotherapy 5 , and as our focus was to determine the mechanisms of acute follicle loss after cancer treatments, we limited our experiments to 12 h time point. We provided multiple lines of evidence converging that cyclophosphamide induces primordial follicle death by prompt activation of apoptotic death pathways, not the PI3K/PTEN/Akt pathway. While we showed that within 12 h of treatment, chemotherapy induces DNA DSBs in human oocytes, we found no evidence that it results in increased growth activation of primordial follicles. Likewise, our previous immunohistochemical analyses in rodents and human ovarian xenograft models showed that both cyclophosphamide and doxorubicin induce DSBs and apoptotic oocyte death in primordial follicles 3 . In this unique in vivo human ovarian xenograft model, we first showed that chemotherapy exposure did not acutely result in the change of total and non-atretic primary vs. primordial follicle ratio rather, it induced DNA DSBs and AC-3 expression in primordial follicles. This morphological evidence supports that chemotherapy induces primordial follicle apoptotic death rather than growth initiation (Supplementary Fig. S2).
The morphological data was then backed by LCM-based individual-oocyte RNA-seq and IPA analysis, which we used to interrogate the key pathways that are activated in primordial follicles in response to chemotherapy. The LCM-based cell isolation method enables us to precisely capture individual primordial follicle oocytes from ovarian tissue where they are identified based on their morphology without the need of unique genetic markers (Supplementary Video S1). This initial screening directed us towards pathways related to apoptotic processes rather than those that are involved in follicle growth activation as culprits in acute chemotherapy-induced ovarian reserve loss. Considering that the number of the sequenced primordial oocyte follicles was relatively small, we then confirmed these findings by qRT-PCR and immunohistochemistry in a larger number of samples (n = 5-6 and n = 8-11, where 'n' is the number of single primordial follicle oocytes and number of xenografted ovarian tissues belonging to different donors, respectively).
The transcriptomic analysis with our LCM-based individual-oocyte RNA-sequencing approach indicated suppression of several signaling pathways such as Phospholipase-C (PLC), Ephrin (Eph) Receptor, and IL-8 that are anti-apoptotic and involved with Bad-Bcl2-mediated apoptosis [29][30][31] . Activation of IKBKE induces NF-κB nuclear accumulation and DNA-binding activity by phosphorylation, which then increases the transcription of BCL2 32 . Although we did not find a significant decrease in the expression of pro-survival gene BCL-2 and pro-apoptotic BAD, we observed an increased colocalization of BAD with BCL2, which indicates an increased tendency for apoptotic death in human primordial follicle oocytes. We, on the other hand, did not find an increase in the phosphorylation of the key PI3K/PTEN/Akt members in acute response to chemotherapy. Though the follicle growth activation via PI3K/PTEN/Akt pathway is primarily regulated by protein phosphorylation, a previous study suggested that the inducible ablation of Foxo3a was associated with primordial follicle depletion through growth activation 33 . Therefore, we additionally studied transcriptional changes in individual primordial follicle oocytes by qRT-PCR and also did not find a change in response to chemotherapy exposure. These converging findings support the common theme that chemotherapy induces human primordial follicle death via apoptotic pathways rather than massive follicle growth activation and resulting 'burn-out' in the acute phase. While our data cannot rule out the activation of the primordial follicles at the later time point, it confirms apoptosis as the acute response of the oocyte to chemotherapy.
In agreement with our findings in humans, other studies in rodents showed that alkylating agents busulfan and cyclophosphamide induce apoptosis and deplete primordial follicle reserve 8,[34][35][36][37] . Studies in mouse oocytes also showed that cyclophosphamide induces a reduction in mitochondrial transmembrane potential and accumulation of cytochrome-c in the cytosol, leading to activation of the caspase family and apoptosis 38 . A plethora of factors such as DNA damage, energy stress, loss of growth factor signaling and hypoxia can trigger apoptosis by activation of BCl2 pathway proteins 39 . In our previous in vitro and xenografting studies, we showed that a topoisomerase inhibitor, cancer drug doxorubicin induces DNA DSBs and apoptosis in primordial follicles 3 . In this study, cyclophosphamide treatment likewise increased DNA DSBs, as shown by increased expression of γH2AX staining in the primordial follicles. Increased DNA damage in primordial follicles is a likely stress signal for the recruitment of the pro-apoptotic BCL2 that causes the activation of the apoptotic cascade in the primordial follicles.
Our individual-oocyte transcriptomic analyses also showed significant downregulation of platelet endothelial cell adhesion molecule-1 (PECAM), and Angiopoietin-1(ANGPT1) (Fig. 3D-F). Previous studies have shown that Jurkat cells with a 70% reduction of PECAM-1 expression by siRNA-silencing were significantly more sensitive to chemotherapy-induced apoptosis 40 . Neuronal cell death was caused by the downregulation of Ang-1 (a.k.a. ANGPT1, followed by Akt pathway inhibition 41 . It is also shown that Ang-1 induces phosphorylation of Akt, which is associated with the up-regulation of the apoptosis inhibitor survivin 22 . Inhibition of ANGPT1 in rat follicles increased the levels of AC-3 and decreased Akt phosphorylation 41 . The latter concurs with our finding of increased AC-3 expression after chemo exposure in primordial follicles and brings up the interesting possibility that chemo exposure may indeed result in the suppression of Akt functions, which further favors apoptosis. Interestingly, in addition to PECAM and ANGPT1, IKBKE is also a positive regulator of Akt activation 19,21,22 , which we found to have decreased in expression in primordial follicles after chemotherapy. In summary, our novel in vivo single-cell human transcriptomics data show that by suppression of antiapoptotic mechanisms, gonadotoxic chemotherapy exposure induces a pro-apoptotic state in primordial follicles in the acute phase (Fig. 6). Rather than PI3K/PTEN/Akt pathway activation resulting in massive follicle growth and 'burn-out' , we found contrary evidence that chemotherapy may favor inactivation of the pathway. Nonetheless, our study only focused on the acute effects of chemotherapy, and we chose the 12 h timepoint based on our earlier xenograft work 5 . It is likely that the apoptotic process was well underway by the 12th hour time point, and future studies may be needed to dissect the timeline of earlier events. It is also possible that further follicle loss occurs at later time points as a result of additional mechanisms such as the stromal/microvascular damage and follicle activation 8 and the potential delayed mechanisms of chemotherapy-induced ovarian reserve loss should Scientific Reports | (2021) 11:407 | https://doi.org/10.1038/s41598-020-79643-x www.nature.com/scientificreports/ be studied. However, given the magnitude of acute damage from DNA damage and apoptotic death, we propose that the research for the development of targeted medical gonadal fertility preservation treatments should primarily focus on preventing DNA damage and/or chemo-induced pro-apoptotic state in primordial follicles.
Received: 25 June 2020; Accepted: 9 November 2020 Figure 6. Mechanisms of acute chemotherapy-induced damage to ovarian reserve in human. Chemotherapy exposure induces apoptotic death in primordial follicles by direct and indirect mechanisms, and results in the massive depletion of human ovarian reserve. Directly, gonadotoxic chemotherapy agents such as cyclophosphamide induce double-strand DNA breaks in human primordial follicle oocytes which results in the activation of apoptotic cascades and increased co-localization of BAD-BCL2. Indirectly, same chemotherapy exposure also leads to suppression of anti-apoptotic pathways such as the Ephrin Receptor Signaling, Phospholipase C Signaling, and Interleukin-8 Signaling. Likewise, indirect suppression of pro-Akt pathways such as PECAM-1 (Platelet/Endothelial Cell Adhesion Molecule-1), IKBKE (inhibitor of nuclear factor kappa-B kinase subunit epsilon) and ANGPT1 (Angiopoietin-1) likely leads to the inhibition of the prosurvival functions of the Akt pathway. The sum of the latter two indirect mechanisms is increased sensitivity to apoptotic, gonadotoxic agents. While this scheme explains the mechanisms of acute primordial follicle losses after chemotherapy, there could be other direct and indirect mechanisms responsible for delayed losses, which are not studied here. For more detailed discussion and an expanded figure, please refer to Szymanska et al., 2020.