Peritoneal Milky Spots Serve as a Hypoxic Niche and Favor Gastric Cancer Stem/Progenitor Cell Peritoneal Dissemination Through Hypoxia-Inducible Factor 1α

Peritoneal dissemination is the most common cause of death in gastric cancer patients. The hypoxic microenvironment plays a major role in controlling the tumor stem cell phenotype and is associated with patients' prognosis through hypoxia-inducible factor-1α (HIF-1α), a key transcriptional factor that responds to hypoxic stimuli. During the peritoneal dissemination process, gastric cancer stem/progenitor cells (GCSPCs) are thought to enter into and maintained in peritoneal milky spots (PMSs), which have hypoxic microenvironments. However, the mechanism through which the hypoxic environment of PMSs regulated GCSPC maintenance is still poorly understood. Here, we investigated whether hypoxic PMSs were an ideal cancer stem cell niche suitable for GCSPC engraftment. We also evaluated the mechanisms through which the HIF-1α-mediated hypoxic microenvironment regulated GCSPC fate. We observed a positive correlation between HIF-1α expression and gastric cancer peritoneal dissemination (GCPD) in gastric cancer patients. Furthermore, the GCSPC population expanded in primary gastric cancer cells under hypoxic condition in vitro, and hypoxic GCSPCs showed enhanced self-renewal ability, but reduced differentiation capacity, mediated by HIF-1α. In an animal model, GCSPCs preferentially resided in the hypoxic zone of PMSs; moreover, when the hypoxic microenvironment in PMSs was destroyed, GCPD was significantly alleviated. In conclusion, our results demonstrated that PMSs served as a hypoxic niche and favored GCSPCs peritoneal dissemination through HIF-1α both in vitro and in vivo. These results provided new insights into the GCPD process and may lead to advancements in the clinical treatment of gastric cancer. Stem Cells 2014;32:3062–3074


INTRODUCTION
Gastric cancer is one of the most common malignant tumors in the gastrointestinal system and is associated with a pessimistic prognosis [1][2][3]. Peritoneal dissemination frequently occurs in late-stage gastric cancer and is the most serious problem that prevents the long-term survival of gastric cancer patients [4,5]. Peritoneal milky spots (PMSs) are omentum-associated lymphoid tissues that have generally been recognized as the site of origin of immature PMS macrophages [6,7]. Free tumor cells in the peritoneal cavity have been shown to preferentially engraft and colonize in PMSs, resulting in peritoneal dissemination [8][9][10]. The unique anatomical structure and vascular distribution of PMSs provide a hypoxic microenvironment that is suitable for tumor cell residence [11]. It was presumed that, during gastric cancer peritoneal dissemination (GCPD), PMSs could partly eliminate mature/senescent gastric cancer cells (GCCs) by nonspecific cytotoxic effects; however, PMSs could not eliminate gastric cancer stem/ progenitor cells (GCSPCs) [12,13]. GCSPCs are relatively quiescence and exhibit antiapoptotic abilities in PMS hypoxic microenvironment, which could partly explain why GCSPCs can survival after traditional intraperitoneal chemotherapy and are capable of rapidly regenerating tumors, thereby leading to GCPD. Conversely, PMS macrophages can be remodeled by tumor cells, forming tumor-associated macrophages with an alternative active phenotype, which in turn support tumor cell maintenance [14].
The identification of tumor stem cell locations and regulatory mechanisms that control the maintenance of tumor stem cells is a vital a Department of Surgical Oncology and b Department of Breast Surgery, The First prerequisite to successful antitumor strategies [15]. The hypoxic microenvironment has an established role in the regulation of stem cell stemness, which can drive tumor progression by triggering a set of adaptive transcriptional responses that regulate tumor stem cell differentiation and self-renewal and ultimately promote a more aggressive tumor phenotype [16,17]. These cellular responses are primarily controlled by the transcription factor system of hypoxia-inducible factors (HIFs) [18][19][20][21]. HIF-1 is a heterodimeric transcription factor that consists of two subunits. The HIF-1b subunit is constitutively expressed, while the HIF-1a subunit is regulated by oxygen levels. It is stable under hypoxic conditions, but is rapidly degraded under normoxic conditions [22]. After stabilization or activation, HIF-1 translocates to the nucleus, where it induces the transcription of numerous downstream target genes via their hypoxia-response elements [23]. To date, the hypoxia response has been shown to control the GCC phenotype and to be associated with prognosis in gastric cancer patients [24,25]. However, direct evidence is still needed to prove GCSPCs preferred reside in hypoxic PMSs, and the function of the hypoxic microenvironment and HIF regulation in GCSPC maintenance have not been fully elucidated.
In this study, we isolated GCSPCs from gastric cancer patients who suffered peritoneal dissemination using the side population approach. Using GCSPC-related protein markers and a pimonidazole label, we demonstrated that tumor stem cells were highly enriched in hypoxic regions of PMSs by a time-dependent GCPD mouse model. Moreover, we showed that hypoxia was involved in the regulation of GCSPC stemness through HIF-1a. These findings were further confirmed by the observation that GCPD was alleviated when the hypoxic microenvironment was destroyed in a mouse model. In summary, we identified the hypoxic niche as a critical regulator in GCSPC maintenance and provided new insights that may lead to successful targeting of GCSPCs in clinical treatment of gastric cancer.

Tissue Samples
Tumor specimens were obtained from 175 patients with gastric cancer who underwent surgery at the Department of Surgical Oncology, First Affiliated Hospital of China Medical University from 2003 to 2005. All patients underwent gastrectomy, and their clinical and pathological data were available. All surgical specimens were examined by experienced pathologists, and the distal resected margin was tumor-free. The study protocol was reviewed and approved by the Ethics Committee of China Medical University and all patients provided written informed consent to participate in the study.

Lentivirus Production and Infection of GCCs
The shRNA sequences directed against HIF-1a (shHIF) were shHIF1: AACCGGTTGAATCTTCAGATAT, shHIF2: AACCTCAGTGT GGGTATAAGA, and shHIF3: AAGACCGTATGGAAGACATTA, and the scramble sequence was CCGGGGGTCTGTATAGGTGGAGAC. Lentivirus construction was carried out by Genepharma (Shanghai, China). Primary GCCs were infected, and stable transfectants were selected in puromycin for 7 days. After this period, cells were expanded and exposed to normoxic or hypoxic conditions for 24 hours to test HIF-1a downregulation. The shHIF1 construct was the most efficient for downregulation of HIF-1a expression in GCCs based on Western blot analysis.

Boyden Chamber Migration Assay
Boyden chambers (BD) with 8-mm pore size polystyrene filter inserts for 24-well plates were used according to the manufacturer's instructions. Briefly, 2 3 10 4 GCCs were seeded into the upper compartment of each chamber in 300 ml DMEM with 10% FBS. The chambers were placed into wells containing 750 ml of complete medium. The migration chambers were incubated for 24 hours in normoxic or hypoxic conditions at 37 C. Following incubation, the inserts were fixed and stained, and the number of migrating cells was counted as described. Two independent experiments were performed in duplicate. Images were collected and quantified using Image-Pro Discovery software (80i, Nikon, Japan).

Side Population Analysis and Sorting
Sorting the side population of GCCs was described previously [27]. Briefly, GCCs were harvested and resuspended at 1 3 10 6 cells per milliliter in prewarmed 37 C DMEM/F-12 with 1% FBS. The cells were then labeled with Hoechst 33342 (Sigma-Aldrich, Saint Louis, MO) at a concentration of 5 mg/ ml. The labeled cells were incubated in the dark for 75 minutes in a 37 C water bath with intermittent mixing, either alone or with 75 mmol/l verapamil (Sigma-Aldrich). The cells were resuspended in ice-cold PBS containing 1% FBS after staining and were maintained at 4 C until flow cytometry analysis. Stained cells were analyzed using a flow cytometer (FACS Aria II, BD).

Clonogenic Assays
Cells were plated at 100 cells/well in six-well culture plates for 14 days with DMEM/F-12 supplemented with 10% FBS, and colonies were fixed with 4% paraformaldehyde and stained with 1% crystal violet. Images were collected using a digital microscope (80i, Nikon). The number of colonies formed (>2 mm diameter) was counted manually by three independent researchers.

Peritoneal Dissemination Model and Establishment of the Peritoneum Hypoxic Region
Eight-to ten-week-old male BALB/c nude mice (weighing 18-20 g each) were obtained from Vital River (Beijing, China). All animal protocols received prior approval from the China Medical University Animal Ethics Committee, and all experiments were performed in accordance with relevant guidelines and regulations.
To investigate the peritoneal hypoxic region and GCSPC distribution, a group of 15 mice were intraperitoneally (i.p.) injected with 1 3 10 4 GC1 sp s to establish a time-dependent GCPD model. Peritoneal tumor-bearing mice were sacrificed on days 7, 14, and 21 (five mice at each time point). The peritoneal hypoxic region was assessed by pimonidazole incorporation, as described previously [29]. Briefly, 60 mg/kg pimonidazole hydrochloride (Chemicon) was injected intravenously (i.v.) into tumorbearing mice. Two hours after injection, mice were sacrificed by cervical dislocation, and the peritoneal greater omentum was fixed and sectioned. The peritoneal hypoxic region was assayed by immunofluorescence confocal staining with a rabbit polyclonal anti-pimonidazole antibody (Chemicon).

PMS Macrophage Depletion and Tumor Xenografts
For peritoneal depletion of macrophages, mice were treated with i.p. injections of clodronate liposomes (Encapsula Nano Sciences) at 25 mg/kg every 3 days, as described previously [30]. After 9 days (three injections), mice were sacrificed by cervical dislocation. The greater omentum was excised, fixed, and sectioned. PMS macrophage quantity and PMS area were assayed by immunohistochemistry. Major axis (L) and minor axis (S) of PMS were measured, the area of milky spots was calculated by the formula: area (mm 2 ) 5 p 3 L 3 S. For tumor xenograft, 1 3 10 4 GC1 sp s or GC1HIF-1a D sp s were i.p. injected into normal (np) or PMS macrophage-depleted (dp) mice. Ten mice were included in each of the four groups. Mice were sacrificed by cervical dislocation 30 days after inoculation. Ten minutes before sacrifice, mice were anesthetized via inhalation of isoflurane to reduce the suffering. GCPD status was then estimated by measuring the volume of ascites and counting the number of metastatic nodules in the peritoneum.

Statistics
All statistical analyses were performed using SPSS 16.0 software. Overall survival rates were determined using Kaplan-Meier curves, with the event being defined as death related to cancer. The log-rank test was used to identify differences between the survival curves of different patient groups. In univariate analysis, continuous data obtained from two groups were analyzed by two-tailed Student's t tests, continuous data from three or more separate groups were analyzed by analysis of variance, and categorical data were analyzed by twotailed v [2] tests. All data were expressed as means 6 SDs. Statistical significance was set at an alpha value of p < .05.

Hypoxia-Inducible Proteins Were Overexpressed in Gastric Cancer and Correlated with Stem Cell-Related Protein Expression and GCPD
In order to investigate the correlation between the expression of hypoxia-inducible proteins and the stem cell-related proteins Oct4 and Nestin, we measured HIF-1a, HIF-2a, Oct4, and  2a protein were found to have elevated Oct4 expression but no relation with Nestin expression (p < .001 and p 5 .175, respectively). Moreover, the expression of HIF-2a protein was shown to have no significantly relation with gastric cancer metastasis patterns (p 5 .824 for hematogenesis dissemination and p 5 .377 for GCPD, respectively). Calculation of the survival times of the 175 patients using the Kaplan-Meier method revealed that patients who harbored HIF-1a-or HIF-2a-positive tumors both had a shorter survival than patients who harbored HIF-1a-or HIF-2a-negative tumors (Fig. 1E, 1F, p 5 .003 and p 5 .017, respectively). Patients' survival times were also determined according to stem cell-related protein expression, as shown in Figure 1G, 1H, Oct4-and Nestinpositive expression were both negative prognostic factor in patients with gastric cancer (p 5 .025 and p 5 .006, respectively).

Hypoxia Induced Epithelial-Mesenchymal Transition in GCCs and Enhanced the GCSPC Ratio in GCCs Through HIF-1a
In order to investigate the role of the hypoxic microenvironment and HIF-1a during GCPD, we performed primary cultures of GCCs from seven patients who exhibited GCPD. As shown in Figure 2A, GC1 and GC2 cells showed better responses to hypoxia, as measured by upregulation of HIF-1a expression. However, as the HIF-2a expression was measured, GC1 and GC2 cells exhibited no HIF-2a expression, only two in seven primary cultured GCCs showed weak HIF-2a expression. Next, we stably transfected GC1 and GC2 cells with shHIF; no elevated HIF-1a expression was observed in either GC1HIF-1a D or GC2HIF-1a D following hypoxic stimulation, indicating that HIF-1a was a key regulator of the response to hypoxic stimuli in GC1 and GC2 cells (Fig. 2B). To investigate the roles of HIF-1a and hypoxia in metastatic ability, migration ability was assayed using Boyden chamber assays under both hypoxic and normoxic conditions. In all cases, GC1 and GC2 cells showed significantly higher migratory capacity in hypoxic conditions than in normoxic conditions (2.8-and 2.4-fold increases, respectively; both p < .05, Fig. 2C). In contrast, GC1HIF-1a D and GC2HIF-1a D showed no significant differences in migrated cell numbers under both hypoxic and normoxic conditions. Because of the observed increase in migration ability, E-cadherin and a-SMA expression were also assayed to investigate whether GC1 and GC2 cells undergo epithelialmesenchymal transition (EMT) under hypoxic conditions. As shown in Figure 2D, GC1 and GC2 cells exhibited downregulation of the epithelial marker E-cadherin and upregulation of the mesenchymal marker a-SMA, which are considered specific features of the EMT, compared to those of GC1HIF-1a D and GC2HIF-1a D cells under hypoxic conditions. Because the EMT has been shown to endow tumor cells stem-like properties [31,32], we next assayed the surface expression of CD44 and lgr5, which have been shown to be specific markers of GCSPCs [33][34][35]. As shown in Figure 2E, GC1 and GC2 cells showed elevated lgr5 and CD44 expression under hypoxic conditions compared with those under normoxic conditions (8.7-and 5.1-fold increases for lgr5, both p < .05; 1.7-and 7.1-fold increases for CD44, p 5 .24 and p < .05, respectively). Moreover, GC1HIF-1a D and GC2HIF-1a D cells showed no significant differences in lgr5 or CD44 expression under both normoxic and hypoxic conditions. The side population (SP) of GCCs has been shown to be rich in GCSPCs, which have selfrenewal and pluripotent differentiation abilities, as previously described [27,36]. Therefore, we measured the verapamilsensitive SP ratio in GC1 (GC1 sp ), GC2 (GC2 sp ), GC1HIF-1a D (GC1HIF-1a D sp ), and GC2HIF-1a D (GC2HIF-1a D sp ) which were cultured under hypoxic or normoxic conditions for at least 7 days. Under normoxic condition, GC1 and GC1HIF-1a D cells were both rich in verapamil-sensitive SPs (6.4% and 3.5%, respectively), after 7 days hypoxic culture, SPs showed a 3.5and 1.6-fold increases (for GC1, SP ratio was 6.4% 6 0.54% in normoxia and 22.46% 6 3.57% in hypoxia, p < .05, Fig. 2F). However, very few SP cells (both <0.5%, Supporting Information Fig. S1A) were isolated from GC2 or GC2HIF-1a D cells, despite the observation that these cells exhibited high levels of CD44 and lgr5 expression.

Hypoxia Enhanced GCSPC Self-Renew, But Reduced GCSPC Differentiation
We isolated the verapamil-sensitive SPs from GC1 and GC1HIF-1a D . Clone formation ability was assayed in GC1 sp and GC1HIF-1a D sp cells under hypoxia and normoxia. All SP cells were shown to be heterogeneous [37,38], containing both stem/progenitor and mature/senescent cells. The variation tendency of stem/progenitor cells can be evaluated by colony formation assays because stem/progenitor cells are colonyforming cells. As shown in Figure 3A, GC1 sp cells formed more clones under hypoxic conditions when serum was added than those under normoxic conditions (31 6 3.6 vs. 13.7 6 2.1, p < .05). However, GC1HIF-1a D sp cells showed no differences in clone numbers under both hypoxic and normoxic conditions (5.3 6 0.5 vs. 5.1 6 2.1). Serial passaging of tumor spheres was performed to assay the self-renewal ability of GCSPCs, and GC1 sp cells showed incremental tumor sphere formation from the first to the third generation under hypoxic conditions (Fig. 3B, p < .05). As expected, tumor sphere formation ability was stable in GC1 sp cells under normoxic conditions. Tumor sphere formation ability was also assayed in GC1HIF-1a D sp cells, but no significant differences were found under both hypoxic and normoxic condition for all three generations (Fig. 3B). Clonogenic and tumorsphere formation assays were also carried out in GC2 cells. Consistent with GC1, hypoxia endowed GC2 an increased self-renewal ability through HIF-1a (Supporting Information Fig. S1B, S1C). To further confirm the presence of GCSPC in GC2, a gradient of 1 3 10 4 /10 5 /10 6 GC2 or GC2HIF-1a D was injected i.p. to the nude BALB/C mice. GC2 could generate metastatic nodules at a low GC2 gradient (Supporting Information Table S1). The expression levels of Oct4 and Nestin were assayed in single tumor spheres to verify the self-renewal ability of cells within the spheres using immunofluorescence staining and confocal imaging. Third-generation tumor spheres arising from GC1 sp cells cultured under hypoxic conditions showed greater numbers of Oct4-and Nestin-expressing cells compared with that of the other three groups. HIF-1a and Oct4 or Nestin were coexpressed in a subpopulation cells under hypoxic conditions (HIF-1a and Oct4 coexpressed in 55.66% 6 9.21% hypoxic cultured GC1 spheroid cells, HIF-1a and Nestin coexpressed in 45.41% 6 6.37% hypoxic cultured GC1 spheroid cells), these cells were considered GCSPCs (Fig. 3C). These results indicated that hypoxic environment endowed GCSPC stable and even growing self-renewal ability. Pluripotent differentiation  Side population of GC1 (GC1 sp ) and GC1HIF-1a (GC1HIF-1a D sp ) were plated at 100 cells/well in six-well culture plates for 14 days with Dulbecco's modified Eagle's medium (DMEM)/F-12 supplemented with 10% fetal bovine serum and cultured under hypoxic or normoxic conditions, colonies were fixed and stained. The number of colonies formed (magnification 340, >2 mm diameter) was counted manually by three independent researchers (number of colons were expressed as mean 6 SD, n 5 6, *, p < .05). (B): GC1 sp and GC1HIF-1a D sp were diluted in GCSPC sphere medium (5,000 cells per milliliter) and plated at 500 ml per well in ultra-low attachment 24-well plates in normoxic or hypoxic conditions. The medium consisted of serum-free DMEM/F-12, 13 B27 supplement, 13 N2 supplement, 50 ng/ml epidermal growth factor, 100 ng/ml basic fibroblast growth factor, 10 nM gastrin, and 100 ng/ml noggin. Cells were fed 50 ml GCSPC sphere medium every other day for 7 days. Secondary and tertiary generations of tumor spheres were cultured in the same conditions as the primary generation (magnification 360, numbers of tumorsphere were expressed as mean 6 SD, n 5 3, *, p < .05). (C): Spheroids of GC1 sp and GC1HIF-1a D sp were fixed and sectioned, HIF-1a and stem-related protein Oct4 and Nestin expression were analyzed by immunofluorescence staining (DAPI showed blue staining, Oct4 and Nestin showed green staining, and HIF-1a showed red staining, magnification 3600, *, p < .05 compared with GC1 sp cultured under normoxia and **, p < .05 compared with GC1HIF-1a D sp cultured under hypoxia). (D): Spheroids of GC1 sp and GC1HIF-1a D sp were fixed and sectioned, HIF-1a and differentiation-related protein Mucin 5ac and Mucin 6 expression were analyzed by immunofluorescence staining (DAPI showed blue staining, Mucin 5ac and Mucin 6 showed green staining, and HIF-1a showed red staining, magnification 3600, *, p < .05 compared with GC1 sp cultured under normoxia and **, p < .05 compared with GC1HIF-1a D sp cultured under hypoxia). Abbreviation: HIF-1a, hypoxia-inducible factor-1a. potential is another feature of cancer stem cells [39]. Therefore, we also evaluated the effects of hypoxic conditions on the expression of mucin5ac and mucin 6, specific markers of GCSPCs for foveolar and pyloric gland cell differentiation [40], in GCSPCs. As shown in Figure 3D, GC1 sp cells cultured under hypoxic conditions showed the lowest levels of mucin5ac and mucin6 expression (22.4% 6 7.14% for mucin5ac, 15.71% 6 6.28% for mucin6), indicating that these cells were immature/progenitor cells. Importantly, in the central region of the tumor sphere, no HIF-1a-expressing cells coexpressed mucin5ac or mucin6, suggesting that GCSPCs maintained an undifferentiated status.

PMSs Served as a Hypoxic Niche During GCPD
Hypoxic PMSs have been suggested to be the origin of GCPD [14]. We positioned the hypoxia region of PMS in untreated mice by confocal immunofluorescence staining. As shown in Figure 4A, hypoxic region located in the bottom of normal mice PMSs where newly formed blood and lymph capillaries were assembled [6]. Next, we established a group of peritoneal tumor-bearing mice that received i.p. injections of 1 3 10 4 GC1 sp cells in order to reveal the relationships between hypoxic regions and PMS distribution during GCPD. We discovered GCCs preferred anchorage-dependent growth within PMSs during GCPD. The areas of PMSs were enlarged after injected into normal (np) or PMS macrophagedepleted (dp) mice. Ten mice were included in each of the four groups. Mice were sacrificed by cervical dislocation 30 days after inoculation. GCPD status was then estimated by measuring the volume of ascites and counting the number of metastatic nodules in the peritoneum (red arrows showed metastatic nodules). Abbreviation: HIF-1a, hypoxia-inducible factor-1a. exposure to GCCs, and the margin of milky spots was blurred as GCC began clonal growth (Fig. 4B). Next, we investigated the positional relationships between hypoxic regions, PMS macrophages, and GCCs. We found that there was a band-like hypoxic region within the margins of PMSs. Most GCCs resided in the hypoxic region, while PMS macrophages were located in the center of the PMSs, in the vicinity of the hypoxic region (Fig. 4C). Regions of hypoxia are vital feature of the cancer stem cell niche (CSCN), and PMSs have also been speculated to represent a favorable microenvironment for GCSPC engraftment [11,12]. Therefore, we wanted to investigate whether the localization of stem cell-related protein-expressing GCCs was in hypoxic regions. As shown in Figure 4D, the stem cell-related proteins OCT4 and Nestin were highly expressed in the hypoxic region, as compared to that in nonhypoxic regions.

Block the Regulation of Hypoxia Microenvironment to GCSPC Alleviated GCPD in a Tumor Xenograft Model
PMSs are suitable for GCSPC engraftment, and GCSPCs resided in hypoxic PMSs expressed higher levels of stem cellrelated proteins. We depleted PMS macrophages, which were considered to be the main cellular component of PMSs. IHC staining of CD68 showed significant reductions in both PMS macrophage numbers and the PMS area, which indicated a successful reduction and damage to PMSs (Fig. 5A). To exclude unintended effects of clodronate liposomes to GCCs, GCCs growth and invasion ability were examined when exposed to clodronate liposomes directly under both normoxia and hypoxia (Supporting Information Fig. S2). Normal dose (20 ng/ml) clodronate liposomes did not have significant effect to the growth and invasion ability of GCCs. To demonstrate whether damage PMS hypoxic microenvironment or blockage of HIF-1a transduction could efficiently reduce GCPD, 1 3 10 4 GC1 sp or GC1HIF-1a D sp cells were injected i.p. into np or dp mice. As shown in Figure 5B, the most severe GCPD occurred in the GC1 sp np group. In contrast, mice in the GC1 sp dp group exhibited reduced GCPD, confirming that PMSs served as a hypoxic niche and favored GCSPC self-renewal and outgrowth. In the GC1HIF-1a D sp np group, GCPD was also alleviated compared that in the GC1 sp np group, which indicated that HIF-1a was a key regulator of the GCSPC phenotype in the hypoxic niche. As expected, the GC1HIF-1a D sp dp group showed the fewest number of mice suffered with GCPD ( Table 2). All these results confirmed that PMSs served as a hypoxic niche and favored GCSPC engraftment and self-renewal, thus inducing GCPD through the regulation of HIF-1a.

DISCUSSION
Gastric cancers are characterized by a hypoxic microenvironment, which correlates with tumor aggressiveness and poor outcomes. Overactivity of hypoxia-inducible proteins, sensor of low-oxygen conditions, is implicated in gastric cancer progression and metastasis [25,41,42]. In our study, patients with HIF-1a-or HIF-2a-positive expression both had shorter survival times compared to patients with negative expression. However, increased incidence of GCPD was only discovered in HIF-1a-positive expression patients, indicating that HIF-1a is a predominant hypoxia sensor during GCPD process ( Table 1). The hypoxic microenvironment has been established its vital role in cancer stem cell regulation and maintenance [43,44]. Thus, the expression of the stem cell-related proteins Oct4 and Nestin was also observed in gastric cancer patients. Patients with positive HIF-1a expression were also found to have elevated Oct4 and Nestin expression ( Table 1), indicating that hypoxia-regulated GCSPC maintenance may be involved in the GCPD process.
HIF-1a and HIF-2a are master regulator known to regulate various aspects of the cellular response to hypoxic stress [22][23][24][25]. In this study, HIF-1a and HIF-2a expression were examined in GCCs primary cultured from seven GCPD patients. Strong HIF-1a expression was discovered in GC1 and GC2 cells; however, no remarkable HIF-2a was examined in all seven GCCs (Fig. 2A). Along with the results from gastric cancer patients tissue samples, we concluded that, HIF-1a and HIF-2a may both important regulator of GCCs behavior under hypoxia, HIF-1a seemed to be a predominant hypoxia regulator in milky spots hypoxic microenvironment during GCPD.
GC1 and GC2 showed enhanced expression of the GCSPCrelated markers lgr5 and CD44 under a hypoxic microenvironment, indicating the presence of an elevated GCSPC ratio in the GCC population. To further investigate the effects of the hypoxic microenvironment on GC1 self-renewal and pluripotent differentiation ability, clonogenic and tumorsphere formation assays were carried out in vitro. We discovered that GC1 exhibited an enhanced self-renewal ability and was prone to maintain an undifferentiated status under hypoxic stress through HIF-1a (Fig.  3C, 3D). Even if no SPs were discovered in GC2, clonogenic and tumorsphere formation assays were also executed, because not all cancer stem cells must necessarily possess a SP phenotype. Consistent with GC1, hypoxia endowed GC2 an increased selfrenewal ability through HIF-1a (Supporting Information Fig. S1B, S1C). In conclusion, our results demonstrated that the hypoxic microenvironment determined GCSPC fate and contributed to GCSPC survival and maintenance.
According to the "seed and soil" theory, metastases only occurs when tumor cells encounter a favorable microenvironment in which they can survive and proliferation [45]. To trace GCSPCs during GCPD, we established a time-dependent mouse peritoneal dissemination model (Fig. 4B). We found that GCCs preferred anchorage-dependent growth within the hypoxic regions of PMSs in a time-dependent manner, which provided direct evidence that PMSs were an ideal niche whose hypoxic microenvironments favored GCPD. The anatomical environment includes cellular and acellular components that provide a highly specialized microenvironment for stem cell maintenance; this is defined as the CSCN [46]. Elucidation of the characteristics of the CSCN and regulators of CSCN biology has been a subject of intense research. Initial reports have primarily focused on tumor vasculature, and a perivascular CSCN was observed in a glioma model [47]. Recently, hypoxia, a common characteristic of solid malignant tumors, has been proposed as another feature of the CSCN [48,49]. In our study, we demonstrated that GCSPCs were located in particular microenvironments within the PMS, which was consistent with the CSCN feature described above. Specifically, we showed that GCSPCs were located within the hypoxic margins of PMSs. Expression of the stem cell-related proteins Oct4 and Nestin was also higher in hypoxic regions than in nonhypoxic regions (Fig. 4C, 4D). To fully state PMSs are hypoxic, we also conducted a confocal immunofluorescence staining in normal mice to show the hypoxic region. As shown in Figure 4A, hypoxic region located in the bottom of normal mice PMSs. When the GCSPC entered into PMSs, hypoxic region was enlarged and turn to a band-like outline surround the PMSs (Fig.  4C). Together, these findings suggested that GCSPCs in the peritoneal cavity promoted self-growth by entering pre-existing hypoxic niches in PMSs, which could support the maintenance of GCSPCs and eventually cause GCPD.
To further confirm that the hypoxic microenvironment in PMSs was the key regulator responsible for GCSPC maintenance, we selectively destroyed PMSs by macrophage depletion or blocked the hypoxic response by knockdown of HIF-1a expression in GCSPCs. Mice in the GC1 sp dp group exhibited reduced GCPD, which confirmed that PMSs served as a hypoxic niche and favored GCSPC self-renewal and outgrowth. In the GC1HIF-1a D sp np group, GCPD was also reduced compared to that in the GC1 sp np group, indicating that HIF-1a was a key regulator of the cellular response to hypoxic stress. Consistent with this, very slight GCPD was observed in the GC1HIF-1a D sp dp group.

CONCLUSIONS
In summary, our results suggested that the hypoxic microenvironment in PMSs regulated GCSPC proliferation and maintenance through HIF-1a both in vitro and in vivo. These results provided new insights into the mechanisms of GCPD, which may lead to the development of novel therapies for gastric cancer in clinical practice.