一句话亮点

GSTT1不是单纯的代谢酶，它是胰腺癌转移灶里干性状态的"放大器"——它不负责启动干性，但能把CD133阳性的细胞"盘活"成能长成漂亮肿瘤球的干性细胞，同时意外地让这群细胞对FGFR抑制剂更敏感。

背景/痛点

胰腺导管腺癌（PDA）这玩意儿五年生存率只有8%左右，主要死因是转移和耐药。大家早就知道肿瘤里有一小撮"干性"细胞（CSCs）是罪魁祸首——它们能启动肿瘤、抵抗放化疗、导致复发。但问题是，在已经形成的转移灶里，这群干性细胞到底怎么维持？靠什么信号活着？没人说得清。

作者团队之前挖出来一个叫GSTT1（谷胱甘肽S-转移酶theta 1）的分子，发现它在转移灶里高表达的那群细胞长得慢、有EMT特征、而且特别能转移[11]。但GSTT1跟干性有没有关系？它怎么干活的？完全不知道。这篇就是顺着这条线往下挖。

推理链分步拆解

第一步：先问"GSTT1高的转移细胞是不是更有干性？"

他们手里有一个好东西——之前构建的GSTT1-mCherry报告系统[11]，内源Gstt1启动子驱动mCherry表达。从KPC小鼠的肺转移灶里拿到的PDAC细胞，FACS分出mCherry⁺和mCherry⁻两群，放到肿瘤球培养条件下（无血清、悬浮、加EGF/FGF，这是经典的CSC富集方法）。

结果有意思：mCherry⁺那群长出来的球又多又大，而且RNA-seq一看，干性标志物（Aldh1a1、Prom1、Lgr5）蹭蹭往上涨，分化相关的通路往下掉。但注意一个细节：GSTT1低的球长到第10天，有一部分又把GSTT1捡回来了——说明这个干性状态不是固定死的，是动态的、可塑的。

@方法论点评：他们用"分选→成球培养→再测表达"这步非常关键，直接排除了"GSTT1高只是静态标记"的可能性，证明了GSTT1高状态在干性条件下是可以被诱导富集的。RNA-seq里Prom1冒出来是第一个线索，把他们引向CD133。

第二步：“那GSTT1和CD133在人源细胞里怎么配合的？”

他们先查了公共数据库——PROM1高表达的患者总生存和复发后生存都差（Figure 2A），这跟既往报道对得上。然后扫了一堆人源胰腺癌转移细胞系，发现有的两个都高（SU8686、CFPAC1、Capan1、JOPACA1），有的一高一低（SUIT2：CD133⁺GSTT1⁻），有的两个都没有（ASPC1、HS766T、PACADD135、KP4）。

成球实验：CD133⁺GSTT1⁺的细胞球多、球大、结构紧凑漂亮；CD133⁺GSTT1⁻的SUIT2能启动成球（单细胞也能长出东西），但长出来的是松散的不规则聚集体；CD133⁻的KP4虽然启动能力差，但一旦长起来球特别大——说明启动（initiation）和扩增（expansion）是两码事。

最漂亮的实验：把GSTT1强行塞进SUIT2（本来CD133⁺GSTT1⁻），结果CD133蛋白水平也上来了，球的数量增加了，形态也变好了。反过来，用shRNA敲掉GSTT1，CD133蛋白掉下来，球也没了。

@方法论点评：这个"回补"实验（gain-of-function）加上"敲低"实验（loss-of-function）构成了因果验证的闭环。尤其值得注意——GSTT1敲低后PROM1 mRNA没变，但蛋白掉了，暗示GSTT1是在蛋白水平上调控CD133，不是转录水平。他们后面用MG132（蛋白酶体抑制剂）能rescure CD133蛋白，进一步指向了蛋白稳定性调控。

Fig. 2：PROM1 is Co-expressed with GSTT1 in a Subset of Metastatic Human Pancreatic Cancer Cell Lines and is Associated with High Tumor Sphere Formation Capacity. (A) Kaplan–Meier analysis of PROM1 expression in the KM Plotter online database (https://kmplot.com), selecting the pancreatic ductal adenocarcinoma cohort (N = 177). Patients were stratified into high- and low-expression groups using the mean gene expression cutoff. Overall survival (OS) and relapse-free survival (RFS) were analyzed using the Mantel–Haenszel log-rank test, and hazard ratios (HRs) with 95% confidence intervals were calculated by KM Plotter. (B) Western blot analysis of CD133 and GSTT1 protein levels in whole-cell lysates from a panel of human metastatic pancreatic cancer cell lines, N = 9. Data are representative of at least three independent experiments. (C) Schematic of tumor sphere culture conditions for the pancreatic cancer cell lines shown in (B). Created with BioRender.com. (D) Representative bright-field images of tumor spheres at day 14 from cell lines plated at 1 × 103 cells per well and stratified by CD133 and GSTT1 expression. Images are representative of N = 3 independent experiments. Scale bar, 50 μm. (E) Quantification of average tumor sphere number per well (y-axis) and average tumor sphere size (x-axis) at day 14 across N = 9 cell lines. Cell lines are color-coded based on PROM1 expression. (F) Same analysis as in (E), with cell lines color-coded based on GSTT1 expression. PROM1 and GSTT1 mRNA expression values represent data from Supplemental Fig. 3A–B. Data represent N = 3 independent experiments with three technical replicates each. (G) SUIT2 cells expressing inducible ipCW control or ipCW-GSTT1 were treated with doxycycline for 48 h and analyzed by Western blot using the indicated antibodies. (H) Representative bright-field images of tumor spheres derived from SUIT2 conditions in (G) at day 5 post-seeding. Scale bar, 200 μm. (I) Quantification of tumor sphere growth from (H). Data represent N = 3 independent experiments with three technical replicates each, error bars indicate mean ± s.e.m. (J) Schematic model summarizing findings from tumor sphere initiation and expansion assays, illustrating the hierarchical roles of CD133 and GSTT1. Created with BioRender.com. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 2. PROM1 is Co-expressed with GSTT1 in a Subset of Metastatic Human Pancreatic Cancer Cell Lines and is Associated with High Tumor Sphere Formation Capacity. (A) Kaplan–Meier analysis of PROM1 expression in the KM Plotter online database (https://kmplot.com), selecting the pancreatic ductal adenocarcinoma cohort (N = 177). Patients were stratified into high- and low-expression groups using the mean gene expression cutoff. Overall survival (OS) and relapse-free survival (RFS) were analyzed using the Mantel–Haenszel log-rank test, and hazard ratios (HRs) with 95% confidence intervals were calculated by KM Plotter. (B) Western blot analysis of CD133 and GSTT1 protein levels in whole-cell lysates from a panel of human metastatic pancreatic cancer cell lines, N = 9. Data are representative of at least three independent experiments. (C) Schematic of tumor sphere culture conditions for the pancreatic cancer cell lines shown in (B). Created with BioRender.com. (D) Representative bright-field images of tumor spheres at day 14 from cell lines plated at 1 × 103 cells per well and stratified by CD133 and GSTT1 expression. Images are representative of N = 3 independent experiments. Scale bar, 50 μm. (E) Quantification of average tumor sphere number per well (y-axis) and average tumor sphere size (x-axis) at day 14 across N = 9 cell lines. Cell lines are color-coded based on PROM1 expression. (F) Same analysis as in (E), with cell lines color-coded based on GSTT1 expression. PROM1 and GSTT1 mRNA expression values represent data from Supplemental Fig. 3A–B. Data represent N = 3 independent experiments with three technical replicates each. (G) SUIT2 cells expressing inducible ipCW control or ipCW-GSTT1 were treated with doxycycline for 48 h and analyzed by Western blot using the indicated antibodies. (H) Representative bright-field images of tumor spheres derived from SUIT2 conditions in (G) at day 5 post-seeding. Scale bar, 200 μm. (I) Quantification of tumor sphere growth from (H). Data represent N = 3 independent experiments with three technical replicates each, error bars indicate mean ± s.e.m. (J) Schematic model summarizing findings from tumor sphere initiation and expansion assays, illustrating the hierarchical roles of CD133 and GSTT1. Created with BioRender.com. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)（图注取自PDF文本层，来源：Cancer Letters, 2026）

第三步：“那么问题来了——有没有药能靶向这群CD133⁺GSTT1⁺的细胞？”

传统化疗杀不死CSCs，他们已经知道了。所以他们换了个思路——用公共药敏数据库（GDSC）筛，按PROM1表达分层找敏感性化合物。结果第一名是BIBF-1120（Nintedanib，多靶点RTK抑制剂，靶VEGFR/PDGFR/FGFR）。

二维培养下，BIBF-1120对所有细胞系效果差不多，没有选择性。但一换到肿瘤球培养条件下，CD133⁺GSTT1⁺的细胞对药物敏感得不行，CD133⁻的基本没反应。这个差异在专门的FGFR抑制剂BGJ-398（Infigratinib）上也能复现——而且SUIT2（CD133⁺GSTT1⁻）只在更高浓度才有反应，暗示GSTT1也在贡献敏感性。

他们把GSTT1强行塞进ASPC1或HS766T（CD133⁻GSTT1⁻），这些细胞在肿瘤球条件下就对FGFR抑制剂敏感了——虽然它们并没有真正长出漂亮的球。这就很有意思了：GSTT1过表达足以赋予药物敏感性，但不足以赋予完整的干性功能。

@方法论点评：这里对比二维vs三维培养的差异是教科书级别的提醒——CSC的生物学和药理学特性在常规培养条件下完全被掩盖了。另外，“GSTT1过表达能敏化但不一定能成球"这个分离现象，提示"干性功能"和"药物敏感性"可能是不完全重叠的两个维度。

Fig. 3：GSTT1 potentiates Nintedanib sensitivity in CD133þ metastatic pancreatic tumor spheres. (A) Unbiased analysis of the Sanger GDSC1 drug sensitivity screen (CCLE), stratified by PROM1 expression, identified BIBF-1120 as a candidate compound associated with increased sensitivity in PROM1High pancreatic cancer cell lines. Drug–gene associations are shown as −log10 P-values, with a correlation coefficient approaching 1. (B,C) GSTT1LowCD133Low and GSTT1HighCD133High

Fig. 3. GSTT1 potentiates Nintedanib sensitivity in CD133þ metastatic pancreatic tumor spheres. (A) Unbiased analysis of the Sanger GDSC1 drug sensitivity screen (CCLE), stratified by PROM1 expression, identified BIBF-1120 as a candidate compound associated with increased sensitivity in PROM1High pancreatic cancer cell lines. Drug–gene associations are shown as −log10 P-values, with a correlation coefficient approaching 1. (B,C) GSTT1LowCD133Low and GSTT1HighCD133High（图注取自PDF文本层，来源：Cancer Letters, 2026）

第四步：“那维持这个CD133⁺GSTT1⁺状态的信号从哪里来？”

他们观察到肿瘤球培养条件下，CD133和GSTT1的表达比二维培养高很多（Figure 4B）。肿瘤球培养基里有EGF和FGF——那是不是FGF信号在搞事情？

扫了一圈FGFR家族成员，发现FGFR3蛋白水平跟GSTT1和CD133的表达高度相关（19个细胞系，Pearson r = 0.73）。更重要的是，磷酸化的FGFR3（激活状态）只在CD133⁺GSTT1⁺的细胞里能检测到——SUIT2虽然有FGFR3蛋白，但没有磷酸化，正好对应它缺乏GSTT1、成球能力差。

在病人组织芯片（TMA）上，GSTT1和p-FGFR3在CK19⁺的肿瘤上皮细胞里共表达，相关性r = 0.84，p = 0.001——说明这个信号轴在真实病人里也是活跃的。

然后他们做了三件事验证因果：（1）用CRISPR敲掉FGFR3，球没了，CD133和GSTT1都掉了；（2）用FGFR抑制剂处理，同样效果；（3）用STAT3抑制剂WP1066处理，完美复刻了上述表型——说明FGFR3→STAT3是维持这个干性程序的关键通路。

@方法论点评：这里"蛋白水平相关性 > RNA水平相关性"的观察很有深意——他们特意拿出来说，是因为单细胞测序里FGFR3⁺PROM1⁺双阳性细胞只占约1.9%，但蛋白水平上p-FGFR3跟GSTT1/CD133的关联非常强。这说明信号通路的激活状态比转录本丰度更能反映功能状态，这点在做转化研究时容易被忽视。

![Fig. 4：FGF–FGFR3–STAT3 signaling sustains the GSTT1–CD133 stem-like program in a subset of metastatic pancreatic cancer cells. (A) Schematic depicting differences between 2D attachment culture and low-attachment tumor sphere conditions. Created with BioRender. (B) Western blot showing CD133 and GSTT1 protein levels in whole-cell lysates from CFPAC1 and SU8686 metastatic cell lines grown under 2D attachment or low-attachment tumor sphere conditions for 7 days. (C) Western blot of CD133, GSTT1, FGFR1, FGFR3, and phospho-FGFR3 (Y724) in a panel of human metastatic pancreatic cancer cell lines (N = 9). (D) Western blot quantification of GSTT1 versus FGFR3 (left) and GSTT1 versus FGFR1 (right) across N = 19 pancreatic cancer cell lines. Heatmap coloring indicates CD133 protein levels. All protein values are normalized to tubulin or GAPDH loading controls. Quantification represents N = 3 independent experiments. (E) Principal component analysis (PCA) of N = 19 pancreatic cancer cell lines using normalized protein expression of GSTT1, CD133, and FGFR3, showing clustering patterns based on these markers. Color coding indicates GSTT1 levels. Created with Clustvis 37）. (F) Representative multiplex immunofluorescence images of human pancreatic cancer tissue microarray (TMA) cores demonstrating heterogeneous GSTT1 and phospho-FGFR3 (Y724) expression within CK19-positive tumor epithelial cells. Repre­ sentative tumors exhibiting low, intermediate, and high GSTT1/pFGFR3 expression are shown. Scale bar = 100 μm. (G) Quantification of GSTT1 and phospho-FGFR3 fluorescence intensity within CK19-positive tumor epithelial regions across 15 evaluable human pancreatic cancer TMA cores. Each data point represents one tumor core. Pearson correlation analysis was used to assess the relationship between GSTT1 and phospho-FGFR3 expression. (H) Western blot of FGFR3, GSTT1, CD133, and GAPDH in JOPACA1 cells expressing either non-targeting control (NTC) or CRISPR-mediated FGFR3 knockout (KO) using three pooled guides. (I) Brightfield images of tumor sphere growth in NTC or FGFR3 KO JOPACA1 cells 10 days post-plating. (J,K) Quantification of the number of tumor spheres per well (J) and tumor sphere size (diameter, μm) (K) for each condition and cell line. Each data point represents a single sphere across N = 3 independent experiments in JOPACA1 and SU8686 NTC or FGFR3 KO cells. Data represent N = 3 replicates from N = 3 independent experiments per cell line. Statistical significance determined by two-sided t-test with Welch’s correction (P < 0.001, P < 0.0001). (L) Schematic depicting the JAK/STAT inhibitor WP1066 and the experimental hypothesis. Created with BioR ender.com. (M) Western blot of indicated proteins in SU8686 cells treated with DMSO or 1 μM WP1066 for 48 h. (N) Brightfield images of tumor sphere growth in SU8686 cells treated with DMSO or 1 μM WP1066 10 days post-plating. (O) Quantification of the number of tumor spheres per well in SU8686 cells. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Statistical significance determined by two-sided t-test with Welch’s correction (P < 0.0001). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4. FGF–FGFR3–STAT3 signaling sustains the GSTT1–CD133 stem-like program in a subset of metastatic pancreatic cancer cells. (A) Schematic depicting differences between 2D attachment culture and low-attachment tumor sphere conditions. Created with BioRender. (B) Western blot showing CD133 and GSTT1 protein levels in whole-cell lysates from CFPAC1 and SU8686 metastatic cell lines grown under 2D attachment or low-attachment tumor sphere conditions for 7 days. (C) Western blot of CD133, GSTT1, FGFR1, FGFR3, and phospho-FGFR3 (Y724) in a panel of human metastatic pancreatic cancer cell lines (N = 9). (D) Western blot quantification of GSTT1 versus FGFR3 (left) and GSTT1 versus FGFR1 (right) across N = 19 pancreatic cancer cell lines. Heatmap coloring indicates CD133 protein levels. All protein values are normalized to tubulin or GAPDH loading controls. Quantification represents N = 3 independent experiments. (E) Principal component analysis (PCA) of N = 19 pancreatic cancer cell lines using normalized protein expression of GSTT1, CD133, and FGFR3, showing clustering patterns based on these markers. Color coding indicates GSTT1 levels. Created with Clustvis [37]. (F) Representative multiplex immunofluorescence images of human pancreatic cancer tissue microarray (TMA) cores demonstrating heterogeneous GSTT1 and phospho-FGFR3 (Y724) expression within CK19-positive tumor epithelial cells. Repre­ sentative tumors exhibiting low, intermediate, and high GSTT1/pFGFR3 expression are shown. Scale bar = 100 μm. (G) Quantification of GSTT1 and phospho-FGFR3 fluorescence intensity within CK19-positive tumor epithelial regions across 15 evaluable human pancreatic cancer TMA cores. Each data point represents one tumor core. Pearson correlation analysis was used to assess the relationship between GSTT1 and phospho-FGFR3 expression. (H) Western blot of FGFR3, GSTT1, CD133, and GAPDH in JOPACA1 cells expressing either non-targeting control (NTC) or CRISPR-mediated FGFR3 knockout (KO) using three pooled guides. (I) Brightfield images of tumor sphere growth in NTC or FGFR3 KO JOPACA1 cells 10 days post-plating. (J,K) Quantification of the number of tumor spheres per well (J) and tumor sphere size (diameter, μm) (K) for each condition and cell line. Each data point represents a single sphere across N = 3 independent experiments in JOPACA1 and SU8686 NTC or FGFR3 KO cells. Data represent N = 3 replicates from N = 3 independent experiments per cell line. Statistical significance determined by two-sided t-test with Welch’s correction (P < 0.001, P < 0.0001). (L) Schematic depicting the JAK/STAT inhibitor WP1066 and the experimental hypothesis. Created with BioR ender.com. (M) Western blot of indicated proteins in SU8686 cells treated with DMSO or 1 μM WP1066 for 48 h. (N) Brightfield images of tumor sphere growth in SU8686 cells treated with DMSO or 1 μM WP1066 10 days post-plating. (O) Quantification of the number of tumor spheres per well in SU8686 cells. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Statistical significance determined by two-sided t-test with Welch’s correction (P < 0.0001). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)（图注取自PDF文本层，来源：Cancer Letters, 2026）

第五步：“GSTT1是上游还是下游？”

如果把前面的线索串起来——FGFR3敲掉后GSTT1和CD133都掉；但反过来，GSTT1过表达能增加FGFR3磷酸化、增加CD133，在SUIT2里还能增加成球——那GSTT1到底是上游还是下游？

答案是两者之间是正向反馈。他们给出的模型是：FGF→FGFR3→STAT3建立了一个"干性信号框架”（给了CD133表达的基础），而GSTT1在这个框架下被诱导表达，然后GSTT1反过来稳定CD133蛋白、进一步放大FGFR3信号，形成一个自我强化的回路。

但他们有一个实验做了没得到阳性结果——试图在CD133⁻的细胞里单靠GSTT1过表达诱导出完整的干性功能，没成功。所以结论很克制：GSTT1是"放大器"，不是"启动子"。必须有CD133作为基础（哪怕只是低水平），GSTT1才能发挥作用。

@方法论点评：这种"干性功能需要多个因子协同"的结论，比"找到了唯一master regulator"更符合肿瘤异质性的现实。他们最后在6个病人来源的类器官（PDO）里验证了——GSTT1和PROM1共表达高的类器官球大、对Nintedanib敏感——算是把细胞系里的发现拉回到了临床相关模型上。

Fig. 5：GSTT1 sustains CD133 Expression, Tumor Sphere Formation and Sensitivity to FGFR Inhibitors. (A) Whole-cell lysates from SU8686, CFPAC1 and CAPAN1 (GSTT1HighCD133High) cells expressing either NT Control vector or two independent GSTT1 shRNAs were subjected to western blotting using indicated antibodies. (B) Bright-field (BF) representative images of Day 7 tumor spheres from each condition. Scale bar, 100 μm. (C) Bar plot depicting the quantification of the number of tumor spheres per well for each condition. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s. d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.01, P < 0.005). (D) Violin plot depicting tumor sphere size (diameter, μm) for each condition and cell line. Each data point represents a single sphere across N = 3 independent experiments. Two-sided t-test with Welch’s correction was used to determine statistical significance between groups, (P < 0.01, P < 0.0001). (E) Flow cytometry analysis of CD133+ cells based on surface marker expression in SU8686 tumor spheres expressing either NT Control vector or two independent GSTT1 shRNAs. Data are presented as geometric mean (GeoMean), error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.05, P < 0.01). (F,G) Flow cytometry analysis of CD44+CD133+ cell population based on surface marker expression in SU8686 tumor spheres expressing either NT Control vector or two independent GSTT1 shRNAs. Data represented as CD44+CD133+ population relative to the live tumor sphere population. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.05). (H) SU8686 (GSTT1HighCD133High) cells expressing either NT Control vector or two individual GSTT1 shRNAs were cultured as tumor spheres for 5 days and subsequently subjected to a dose response of BIBF-1120 (Nintedanib) for 48 h. Results are shown as % tumor sphere growth relative to each DMSO treated control for each condition. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between NT Control and each shRNA at both 2.5 μM and 1 μM dose (P < 0.01, P < 0.005).

Fig. 5. GSTT1 sustains CD133 Expression, Tumor Sphere Formation and Sensitivity to FGFR Inhibitors. (A) Whole-cell lysates from SU8686, CFPAC1 and CAPAN1 (GSTT1HighCD133High) cells expressing either NT Control vector or two independent GSTT1 shRNAs were subjected to western blotting using indicated antibodies. (B) Bright-field (BF) representative images of Day 7 tumor spheres from each condition. Scale bar, 100 μm. (C) Bar plot depicting the quantification of the number of tumor spheres per well for each condition. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s. d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.01, P < 0.005). (D) Violin plot depicting tumor sphere size (diameter, μm) for each condition and cell line. Each data point represents a single sphere across N = 3 independent experiments. Two-sided t-test with Welch’s correction was used to determine statistical significance between groups, (P < 0.01, P < 0.0001). (E) Flow cytometry analysis of CD133+ cells based on surface marker expression in SU8686 tumor spheres expressing either NT Control vector or two independent GSTT1 shRNAs. Data are presented as geometric mean (GeoMean), error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.05, P < 0.01). (F,G) Flow cytometry analysis of CD44+CD133+ cell population based on surface marker expression in SU8686 tumor spheres expressing either NT Control vector or two independent GSTT1 shRNAs. Data represented as CD44+CD133+ population relative to the live tumor sphere population. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between groups (P < 0.05). (H) SU8686 (GSTT1HighCD133High) cells expressing either NT Control vector or two individual GSTT1 shRNAs were cultured as tumor spheres for 5 days and subsequently subjected to a dose response of BIBF-1120 (Nintedanib) for 48 h. Results are shown as % tumor sphere growth relative to each DMSO treated control for each condition. Data represent N = 3 replicates from N = 3 independent experiments, error bars indicate standard deviation (s.d). Two-sided t-test with Welch’s correction was used to determine statistical significance between NT Control and each shRNA at both 2.5 μM and 1 μM dose (P < 0.01, P < 0.005).（图注取自PDF文本层，来源：Cancer Letters, 2026）

Fig. 6：Heterogeneous GSTT1 and PROM1 Expression in Patient-Derived PDA Organoids Predict Tumor Sphere Formation and Nintedanib Sensitivity. (A) Schematic illustrating the workflow for culturing PDA patient-derived organoids under tumor sphere conditions, followed by qPCR analysis and tumor sphere growth assays. Created with BioRender.com. (B) qRT-PCR Expression of GSTT1 and PROM1 in a panel of PDA patient-derived organoid-derived tumor spheres (PDA Org1-6). Data are represented as mean s.e.m. Data represents 3 independent experiments. (C) Representative brightfield images of tumor sphere growth in PDA organoids. Scale bar, 100 μm. (D) ImageJ quantification of the average number of tumor spheres per well. Data are represented as mean s.d. Data represent 3 independent experiments. (E) Quantification of tumor sphere size (diameter, μm). Each data point represents a single sphere across three independent experiments. (F) Dos\u00ad e–response of organoid-derived tumor spheres to BIBF-1120 (Nintedanib). Results are shown as percentage of DMSO control for each organoid line. Data represent N = 2 independent experiments, with n = 3 replicates each, error bars indicate standard deviation (s.d). (G) Representative bright-field images showing tumor sphere growth in PDA organoids treated with 10 μM BIBF-1120. Scale bar: 100 μm. (H) LogIC50 values for organoid-derived tumor spheres treated with BIBF-1120. (I) Schematic model depicting overall findings. Created with BioRender.

Fig. 6. Heterogeneous GSTT1 and PROM1 Expression in Patient-Derived PDA Organoids Predict Tumor Sphere Formation and Nintedanib Sensitivity. (A) Schematic illustrating the workflow for culturing PDA patient-derived organoids under tumor sphere conditions, followed by qPCR analysis and tumor sphere growth assays. Created with BioRender.com. (B) qRT-PCR Expression of GSTT1 and PROM1 in a panel of PDA patient-derived organoid-derived tumor spheres (PDA_Org1-6). Data are represented as mean s.e.m. Data represents 3 independent experiments. (C) Representative brightfield images of tumor sphere growth in PDA organoids. Scale bar, 100 μm. (D) ImageJ quantification of the average number of tumor spheres per well. Data are represented as mean s.d. Data represent 3 independent experiments. (E) Quantification of tumor sphere size (diameter, μm). Each data point represents a single sphere across three independent experiments. (F) Dos­ e–response of organoid-derived tumor spheres to BIBF-1120 (Nintedanib). Results are shown as percentage of DMSO control for each organoid line. Data represent N = 2 independent experiments, with n = 3 replicates each, error bars indicate standard deviation (s.d). (G) Representative bright-field images showing tumor sphere growth in PDA organoids treated with 10 μM BIBF-1120. Scale bar: 100 μm. (H) LogIC50 values for organoid-derived tumor spheres treated with BIBF-1120. (I) Schematic model depicting overall findings. Created with BioRender.（图注取自PDF文本层，来源：Cancer Letters, 2026）

核心结论

GSTT1和CD133共定义了一个干性分层体系：CD133负责"启动"（initiation），GSTT1负责"扩增和组织化"（expansion + organization）。只有两者都高的细胞才有最完整的干性功能。 FGFR3-STAT3是维持这个状态的上游信号轴，而GSTT1作为下游的"放大器"，通过稳定CD133蛋白来维持这个回路的运转。 高GSTT1/CD133的干性细胞对FGFR抑制剂（Nintedanib、Infigratinib）敏感，但这种敏感只在肿瘤球/干性富集条件下才显现，常规二维培养看不出来。 GSTT1本身就能赋予FGFR抑制剂敏感性，即使在没有完整干性功能的情况下——提示"敏感性"和"干性功能"可能是GSTT1的两个可分离的效应。

对耐药/DTP/PGCC 的启示

干性可塑性与DTP的关系：GSTT1低细胞在干性培养条件下能重新上调GSTT1（Figure 1E），这说明"干性"不是固定的细胞身份，而是一种可逆的状态。药物耐受持久细胞（DTP）很可能就是处于这种可塑状态的细胞——它们在药物压力下可能通过上调GSTT1/CD133轴进入干性状态存活下来，停药后又可能"回到"非干性状态。这提示靶向干性状态的药物可能需要联合"锁定"可塑性的策略，否则细胞会"逃"出来。 三维培养作为DTP药物筛选模型：常规二维培养下FGFR抑制剂没有选择性，但肿瘤球条件下差异显著——这个现象在DTP/PGCC研究里有直接借鉴意义：许多"耐药机制"可能只在三维结构或干性富集条件下才暴露出来。用常规贴壁培养筛选抗DTP药物，可能漏掉大量真实有效的化合物。 CD133作为生物标志物的再思考：SUIT2（CD133⁺GSTT1⁻）能启动但不能扩增，KP4（CD133⁻GSTT1⁻）启动差但扩增强——说明单靠CD133一个标志物无法准确界定干性功能状态。对于临床上用CD133预测预后或指导治疗，可能需要联合GSTT1或其他标志物（如CD44）才能更准确。组合标志物（GSTT1+CD133）可能比单一标志物更有预测价值——无论是预测复发还是预测FGFR抑制剂的疗效。 从"静态标记"到"动态程序"的思维转换：GSTT1对CD133的调控发生在蛋白稳定性层面而非转录层面（PROM1 mRNA不变），这在药物耐受研究中是一个常见但容易被忽略的现象——耐药/DTP状态下很多蛋白水平的变化并不伴随转录变化。在研究DTP时，除了转录组，蛋白水平的动态调控（降解、修饰、定位）同样需要关注。

局限

GSTT1调控CD133蛋白稳定性的具体机制尚未完全阐明：他们用免疫共沉淀没看到CD133的直接谷胱甘肽修饰（Figure S7B），用MG132证明涉及蛋白酶体途径，但中间的具体分子连接（E3连接酶？去泛素化酶？）还是黑箱状态。 体内验证不够充分：除了TMA的共表达相关性，没有在体内移植瘤模型里验证FGFR抑制剂对CD133⁺GSTT1⁺细胞的靶向效果和疗效。 病人类器官样本量偏小（N=6），且只做了qPCR和药敏，没有在蛋白水平验证CD133和GSTT1的共表达。 GSTT1的"经典功能"（谷胱甘肽代谢/解毒）与"新功能"（调控CD133稳定性）之间的关系没有厘清——是酶活性依赖的还是非酶活性的？文中没有做酶活突变体的回补实验。

来源

期刊：Cancer Letters，2026年。DOI: 10.1016/j.canlet.2026.218733