Metronomic chemotherapy offsets HIFα induction upon maximum-tolerated dose in metastatic cancers

Abstract

Conventional maximum-tolerated dose (MTD) chemotherapy relies on periodic, massive cancer cell ablation events followed by treatment-free intermissions, stereotypically resulting in resistance, relapse, and mortality. Furthermore, MTD chemotherapy can promote metastatic dissemination via activation of a transcriptional program dependent on hypoxia-inducible factor (HIF)-1α and (HIF)-2α (hereafter referred to as HIFα). Instead, frequent low-dose metronomic (LDM) chemotherapy displays less adverse effects while preserving significant pre-clinical anticancer activity. Consequently, we hereby compared the effect of MTD or LDM chemotherapy upon HIFα in models of advanced, metastatic colon and breast cancer. Our results revealed that LDM chemotherapy could offset paralog-specific, MTD-dependent HIFα induction in colon cancers disseminating to the liver and lungs, while limiting HIFα and hypoxia in breast cancer lung metastases. Moreover, we assessed the translational significance of HIFα activity in colorectal and breast TCGA/microarray data, by developing two compact, 11-gene transcriptomic signatures allowing the stratification/identification of patients likely to benefit from LDM and/or HIFα-targeting therapies. Altogether, these results suggest LDM chemotherapy as a potential maintenance strategy to stave off HIFα induction within the intra-metastatic tumor microenvironment.

Keywords: HIF-1; breast cancer; colon cancer; hypoxia; low-dose metronomic.

Figure 1. LDM chemotherapy selectively offsets HIF‐1α levels in experimental colon cancers

1. A

   HIF‐1α levels in HT29 primary tumors. Left: vehicle‐treated controls. Middle/left: Doublet LDM cyclophosphamide + capecitabine (LDM^CTX + LDM^CPB). Middle/right: LDM cyclophosphamide + MTD capecitabine (LDM^CTX + MTD^CPB). Inset, high‐magnification image of the region marked with a green asterisk (*). Right: Quantification of the effect of monotherapies or doublet LDM/MTD regimens on HIF‐1α⁺ area. F _5,12 = 8.791 and P = 0.001 for overall treatment by Brown–Forsythe ANOVA; *P = 0.046 versus vehicle; ^# P < 0.0001 versus LDM^CTX + MTD^CPB by Benjamini, Krieger, and Yekutieli post hoc test.

2. B

   HIF‐2α levels in HT29 primary tumors. Left: Example of immunostaining in vehicle‐treated controls. Inset, high‐magnification image of the region marked with a green asterisk (*). Right: Quantification of the effect of monotherapies or doublet LDM/MTD regimens on HIF‐2α⁺ area. F _5,18 = 1.215 and P = 0.3424 (not significant) for overall treatment by Brown–Forsythe ANOVA.

3. C

   CA9 levels in HT29 primary tumors. Left: Example of immunostaining in vehicle‐treated controls. Inset, high‐magnification image of the region marked with a green asterisk (*). Right: Quantification of the effect of monotherapies or doublet LDM/MTD regimens on CA9⁺ area. F _5,11 = 2.466 and P = 0.0961 (not significant) for overall treatment by Brown–Forsythe ANOVA.

4. D

   Correlation between HIF‐1α levels and proliferation indexes in HT29 primary tumors. Left and middle/left: HIF‐1α and Ki67 expression in vehicle‐treated controls; consecutive sections are shown. Middle/right: HIF‐1α⁺ versus Ki67⁺ proliferation index scatterplot. Each point represents median values for HIF‐1α⁺ tumors. Regression line (red) and 95% CI (shaded blue area) are indicated. F _1,27 = 34.45 and P = 10⁻⁴; slope ≠ 0 by F‐test. Right: Machine‐learning quantification of the effect of monotherapies or doublet LDM/MTD regimens upon Ki67⁺ proliferative index. Median indexes per tumor are indicated. F _5,12 = 1.609 and P = 0.2312 (not significant) for overall treatment by Brown–Forsythe ANOVA.

Data information: Violin plots present 50^th (red line), 25^th and 75^th percentiles (blue line); numbers in brackets indicate number of tumors per group. L, low‐dose metronomic; M, maximum‐tolerated dose; r, correlation coefficient. Low power magnification images of all experimental conditions can be found in Appendix Fig S1. Blue frames in D indicate consecutive sections stained for HIF‐1α and Ki67.

Figure 2. LDM chemotherapy selectively offsets HIF‐1α levels in colon cancer metastases to the liver

1. A

   Effect of LDM and MTD chemotherapy on HT29 liver metastatic nodule size. Left: Histogram of pooled cross‐sectional metastatic diameter (Ø). Middle: Metastatic diameter (Ø) by chemotherapy regimen. A decreasing linear trend for median metastatic diameter was observed (left to right, slope = −143.2, P = 0.043 by post‐test for trend). Right: Dichotomized metastatic size at median diameter of vehicle‐treated tumors. Liver nodule size was classified as small (blue) or large (red). Individual nodule counts per group and size category (small/large) are plotted on the right ordinate. χ² (df = 5) = 6.80; P = 0.2361 (not significant) for overall effects on size; χ² (df = 1) = 4.71; P = 0.0299 (LDM^CTX + LDM^CPB versus vehicle).

2. B

   HIF‐1α levels in HT29 liver metastatic nodules. Left: Vehicle‐treated controls. Middle/left: Doublet LDM cyclophosphamide + capecitabine (LDM^CTX + LDM^CPB). Middle/right: LDM cyclophosphamide + MTD capecitabine (LDM^CTX + MTD^CPB). Right: Automatic quantification of the effect of monotherapies or doublet regimens on HIF‐1α⁺ areas in individual metastatic nodules. F _2,13 = 4.796 and P = 0.028 for overall treatment by Brown–Forsythe ANOVA; *P = 0.038 LDM^CTX + MTD^CPB versus vehicle or P = 0.0123 LDM^CTX + LDM^CPB versus vehicle; ^# P = 0.0098 LDM^CTX + LDM^CPB versus LDM^CTX + MTD^CPB by Benjamini, Krieger, and Yekutieli post hoc test.

3. C

   Correlation between metastatic and peri‐metastatic HIF‐1α levels in the liver. Left: Nodule immunostaining and parenchymal HIF‐1α expression. Each point represents median values per nodule. Regression line (blue) and 95% CI (shaded blue area) are indicated. F _1,57 = 150.7 and P < 0.0001; slope ≠ 0 by F‐test. Right: Correlation between intra‐metastatic and peri‐metastatic parenchymatous ring in the liver. Each point represents median values per nodule. Regression line (blue) and 95% CI (shaded blue area) are indicated. F _1,54 = 55.2 and P < 0.0001; slope ≠ 0 by F‐test.

4. D

   Correlation between metastatic and peri‐metastatic HIF‐2α levels in the liver. Left: Nodule immunostaining and parenchymal HIF‐2α expression. Right: Correlation between intra‐metastatic and peri‐metastatic parenchymatous ring in the liver. Each point represents median values for each nodule. Regression line (blue) and 95% CI (shaded blue area) are indicated. F _1,54 = 55.2 and P < 0.0001; slope ≠ 0 by F‐test. n, number of nodules.

Data information: Violin plots present 50^th (red line), 25^th, and 75^th percentiles (blue line); numbers in brackets indicate number of nodules (n) or animals (mice). L, low‐dose metronomic; M, maximum‐tolerated dose; ns, not significant; Met, metastasis; r, correlation coefficient. Dashed green lines encircle the histological limit between metastatic nodules and their surrounding normal liver parenchyma. Insets show high‐magnification images of regions marked with asterisks (*). Blue frames in C and D indicate consecutive sections from the same liver metastasis, stained for HIF‐1α or HIF‐2α, respectively.

Figure EV1. LDM chemotherapy offsets HIF‐2α levels in colon cancer metastases to the lung

1. A

   Effect of LDM and MTD chemotherapy on HT29 lung metastatic nodule size. Left: Histogram of cross‐sectional metastatic diameter (Ø). Middle: Metastatic diameter (Ø) by chemotherapy regimen. F _5,43 = 9.066 and P < 0.0001 for overall treatment by Brown–Forsythe ANOVA; *P < 0.05 versus vehicle; ^# P < 0.01 versus LDM^CTX + MTD^CPB by Benjamini, Krieger, and Yekutieli post hoc test. Right: Dichotomized metastatic size at median diameter of vehicle‐treated tumors. Lung nodule size was classified as small (blue) or large (red). Individual nodule counts per group and size category (small/large) are plotted on the right ordinate. χ² (df = 5) = 16.44, P = 0.0057 for overall effects on size; χ² (df = 1) = 4.10, P = 0.043 LDM^CPB versus vehicle; χ² (df = 2) = 7.00, P = 0.028 LDM^CTX versus vehicle.

2. B

   HIF‐2α levels in HT29 lung metastatic nodules. Left: vehicle‐treated controls. Middle/left: LDM doublet cyclophosphamide + capecitabine (LDM^CTX + LDM^CPB). Middle/right: LDM cyclophosphamide + MTD capecitabine (LDM^CTX + MTD^CPB). Right: Automatic quantification of the effect of monotherapies or doublet regimens on HIF‐2α⁺ areas in individual metastatic nodules. F _5,64 = 12.38 and P < 0.0001 for overall treatment by Brown–Forsythe ANOVA; *P < 0.01 versus vehicle; ^# P < 0.05 versus LDM^CTX + MTD^CPB by Benjamini, Krieger, and Yekutieli post hoc test.

3. C

   Effect of lung metastatic diameter on HIF‐2α levels. Individual nodule diameter and HIF‐2α⁺ area (logarithm). Regression line (red) and 95% CI (shaded red area) are shown. F _1,84 = 40.18 and P < 0.0001; slope ≠ 0 by F‐test.

4. D

   HIF‐1α levels in HT29 lung metastatic nodules. Left: Example of HIF‐1α expression in a para‐bronchiolar nodule. RB, respiratory bronchiole showing lysed intralumenal erythrocytes. Low power magnification image can be found in Appendix Fig S5B. Right: Automatic quantification of the effect of monotherapies or doublet regimens upon intra‐metastatic HIF‐1α⁺ area. F _5,14 = 1.670 and P = 0.2053 (not significant) for overall treatment by Brown–Forsythe ANOVA.

Data information: Violin plots present 50^th (blue line), 25^th, and 75^th percentiles (red lines); numbers in brackets indicate number of nodules (n) or animals (mice). L, low‐dose metronomic; M, maximum‐tolerated dose; Met, metastasis; r, correlation coefficient. Dashed green lines encircle the histological limit between metastatic nodules and surrounding normal lung parenchymae. Insets, high‐magnification images of regions marked with asterisks (* in all microphotographs). Low power magnification images of all experimental conditions of B can be found in Appendix Fig S5A.

Figure 3. LDM chemotherapy offsets HIF‐1α levels in breast cancer metastases to the lung

1. A

   EMT6‐CDDP lung metastatic nodule size. Left: Histogram of cross‐sectional metastatic diameter (Ø). Middle: Metastatic diameter (Ø) by chemotherapy regimen. F _5,99 = 0.9310 and P = 0.4644 (not significant) for overall treatment by Brown–Forsythe ANOVA. Right: Dichotomized metastatic size at median diameter of vehicle‐treated tumors. Lung nodule size was classified as small (blue) or large (red). Individual nodule counts per group and size category (small/large) are plotted on the right ordinate. χ² (df = 5) = 2.631; P = 0.7567 (not significant) for overall effects on size.

2. B

   Hypoxic fraction in lung metastatic nodules. Pimonidazole adduct immunoreactivity was automatically quantified and expressed as fractional positive areas per nodule (rightmost panel). F _5,110 = 7.544 and P < 0.0001 for overall treatment by Brown–Forsythe ANOVA; *P < 0.01 versus vehicle; ^# P = 0.0111 doublet LDM cyclophosphamide + capecitabine versus LDM cyclophosphamide + MTD capecitabine by Benjamini, Krieger, and Yekutieli post hoc test.

3. C

   HIF‐1α levels in lung metastatic nodules. Automatic quantification of HIF‐1α levels measured as fractional positive areas per nodule (rightmost panel). F _5,114 = 5.325 and P = 0.0004 for overall treatment by Brown–Forsythe ANOVA; *P < 0.01 versus vehicle by Benjamini, Krieger, and Yekutieli post hoc test.

4. D

   Microvessel density in lung metastatic nodules. Automatic quantification of CD31 fractional areas per nodule (rightmost panel). F _5,124 = 7.531 and P < 0.0001 for overall treatment by Brown–Forsythe ANOVA; *P < 0.01 (LDM capecitabine, cyclophosphamide, or doublet capecitabine + cyclophosphamide versus vehicle), ^# P = 0.0417 (doublet LDM capecitabine + cyclophosphamide versus doublet LDM cyclophosphamide + MTD capecitabine) by Benjamini, Krieger, and Yekutieli post hoc test.

Data information: Violin plots present 50^th (blue line), 25^th, and 75^th percentiles (red line); numbers in brackets indicate number of nodules (n) or animals (mice) per group. CPB, capecitabine; CTX, cyclophosphamide; L, low‐dose metronomic; M, maximum‐tolerated dose. Inset: high‐magnification image of the region marked with an asterisk (*; C rightmost panel). Green, blue, and red frames in B, C, and D indicate consecutive sections from the same lung metastatic nodule, stained for pimonidazole (hypoxia), HIF‐1α, or CD31 (microvascular density).

Figure EV2. Relationship among hypoxia, HIF‐1α, and microvessel density in breast cancer metastases to the lung

1. A

   Intra‐metastatic hypoxia (pimonidazole) and microvessel density (CD31). Left and middle: Consecutive sections showing pimonidazole (left) and CD31 (middle) immunoreactivities. Right: Correlation between intra‐metastatic hypoxia (pimonidazole) and microvessel density (CD31). Points represent median values per nodule. F _1,165 = 192.2 and P < 0.0001, slope ≠ 0 by F‐test.

2. B

   Intra‐metastatic hypoxia (pimonidazole) and HIF‐1α levels. Left and middle: Consecutive sections showing pimonidazole (left) and HIF‐1α (middle) immunoreactivities. Right: Correlation between intra‐metastatic hypoxia (pimonidazole) and HIF‐1α. Points represent median values per nodule. F _1,143 = 14.3 and P = 0.0002, slope ≠ 0 by F‐test.

3. C

   HIF‐1α levels and microvessel density (CD31). Left and middle: Consecutive sections showing HIF‐1α (left) and CD31 (middle) immunoreactivities. Right: Correlation between HIF‐1α and CD31. Points represent median values per nodule. F _1,142 = 8.313 and P = 0.0045, slope ≠ 0 by F‐test.

Data information: Pearson regression line (red) and 95% CI (shaded red area) are indicated. Immunoreactivities are expressed as fractions of nodular areas. Met, metastasis; r, correlation coefficient. Dashed green lines demarcate corresponding intra‐metastatic areas of immunoreactivity among pimonidazole, CD31, and HIF‐1α in consecutive histological sections from example specimens (blue frames/arrows).

Figure 4. HIFα transcriptional activation in human colon and breast cancers

1. A

   Correlation matrix among statistically over‐represented and biologically validated HIFα targets in TCGA colon or breast cancer data. Spearman (ρ) coefficients were encoded in a pseudo‐color scale wherein flattening of ellipses denotes increasing |ρ| values; blank cells indicate non‐significant correlation pairs after Bonferroni post hoc comparisons. Transcripts are annotated using official gene symbols.

2. B

   HIFα‐inducible colon cancer signature (HIFi‐CCS). Left and middle/left: List of HIFα targets and overall survival in TCGA (left) or aggregated microarray data (middle/left) from colon adenocarcinomas. Cases are classified as low (blue) or high (red) HIFi‐CCS according to their relationship to the median. Kaplan–Meier analysis followed by log‐rank test. Middle/right: HIFi‐CCS and lymph node (LN) invasion status. Data are standardized as z‐scores. t ₁₉₄ = 4.399 and P < 0.0001 by Student t‐test with Welch's correction. Right: HIFi‐CCS is associated with accelerated distant metastasis. Cases are classified as low (blue) or high (red) HIFi‐CCS according to their relationship to the median. Kaplan–Meier analysis followed by log‐rank test.

3. C

   HIFα‐inducible breast cancer signature (HIFi‐BCS). Left and middle/left: List of HIFα targets and overall survival in METABRIC (left) or recurrence‐free survival in microarray (middle/left) from breast adenocarcinomas. Cases are classified as low (blue) or high (red) HIFi‐BCS according to their relationship to the median. Kaplan–Meier analysis followed by log‐rank test. Middle/right: HIFi‐BCS and estrogen receptor (ER) status. Data are standardized as z‐scores. t ₁₁₂ = 3.789 and P = 0.0002 by Student t‐test with Welch's correction. Right: HIFi‐BCS is associated with worsened distant metastasis‐free survival. Cases are classified as low (blue) or high (red) HIFi‐BCS whenever they are below or above the distribution median. Kaplan–Meier analysis followed by log‐rank test.

4. D

   Median RNAseq expression of individual HIFα targets as split by median HIFi‐CCS (left) or HIFi‐BCS (right). Transcripts up/downregulated by ± 1.5‐fold are shown in blue or yellow, respectively. Error bars indicate 95% CI of the median. Log₂ scale; n = 209 (colon, TCGA) or n = 609 (breast, METABRIC).

Data information: Gene expression omnibus accession numbers are indicated; HR, hazard ratios, brackets indicate 95% CIs; DMFS, distant metastasis‐free survival; OS, overall survival; RFS, recurrence‐free survival; n, number of patients. Violin plots present 50^th (thick line), 25^th, and 75^th percentiles (thin lines) Blue, low signature index (below median); red, high signature index (above median). Source data are available online for this figure.