HMGA1 acts as an epigenetic gatekeeper of ASCL2 and Wnt signaling during colon tumorigenesis

Exquisite control of transcriptional networks that regulate plasticity and other stem cell properties allow for tissue specification during embryogenesis and tissue maintenance after birth (1–9). Often referred to as “stemness” networks, genes involved in self-renewal and plasticity are largely silenced in differentiated tissues, although a subset remain active in adult stem cells where they contribute to tissue regeneration during homeostatic conditions or following injury and other stressors (3–6). By contrast, neoplastic cells frequently corrupt these same transcriptional networks to foster aberrant growth and differentiation (1–7, 9, 10). Moreover, tumor progression is associated with increased expression of genes controlling stem cell properties, which may foster the emergence of highly plastic tumor cells capable of metastatic progression, immune evasion, and therapy resistance (1–7, 9, 10). While mechanisms responsible for cell state during tissue regeneration and tumorigenesis remain incompletely understood, it is clear that chromatin reorganization and epigenetic alterations contribute to plasticity, self renewal, and other stem cell properties (4, 11).

As a highly regenerative tissue and frequent site for cancer, the colon epithelium provides a unique paradigm to study plasticity and cell state during tumorigenesis. Colon epithelium comprises an intricately organized hierarchy of epithelial cells maintained by proliferative stem cells that reside at the base of crypts (3, 4, 6–8). Moreover, it is among the most regenerative tissues of the body, renewing itself every 3–5 days to maintain nutrient absorption essential for life and provide a protective barrier from gut pathogens and toxins. Stem cells at the crypt base in colon and small intestinal epithelium are marked by the serpentine coreceptor for Wnt signals, leucine-rich, repeat-containing G-protein–coupled receptor 5 (LGR5) (7, 8). Recent studies in murine small intestine also identified distinct populations of upper crypt cells, marked by fibroblast growth factor binding protein 1 (FGFBP1) or LGR4, that regenerate all lineages, including LGR5+ cells, the latter of which requires the Wnt agonist, R-spondin (12–14). In both small intestine and colon, a Wnt gradient maintains LGR5+ cells by repressing differentiation at the base where Wnt levels are highest, while allowing cells to differentiate as they move up the crypt with decreasing Wnt. Thus, tightly regulated Wnt signaling is fundamental to epithelial regeneration in the gut.

Not surprisingly, mutations that activate Wnt signals are common in colon adenomas and adenocarcinomas (15–22). Inactivating mutations in the gene encoding the Adenomatosis Polyposis Coli (APC) tumor suppressor protein, first described in the familial adenomatosis polyposis (FAP) syndrome, are the most common genetic lesions found in colon adenomas and carcinomas (15–22). APC normally restrains Wnt function by maintaining β-catenin in the cytoplasm, thereby preventing β-catenin entry into the nucleus to activate Wnt target genes together with the TCF-4 transcription factor. Genomic studies established a model whereby colon carcinomas develop from polyps harboring APC mutations after the stepwise accumulation of mutations that inactivate additional tumor suppressor genes and/or activate protooncogenes (15–16). Neoplastic polyps are thought to arise in colon stem cells expressing LGR5, although targeting mutated LGR5+ cells in therapy has not been feasible (7, 8). Importantly, colon cancer is the third leading cause of cancer-related deaths in the US, and the incidence is rising globally, particularly in younger individuals (23–26). Thus, studies are warranted to decipher changes in cell state occurring during colon tumorigenesis to identify mechanisms that could be targeted to intercept the transition from mutant cells and localized tumors to advanced disease.

High Mobility Group A (HMGA1) proteins are architectural transcription factors that bind AT-rich sequences where they modulate chromatin structure and gene expression (4, 27–55). The HMGA1 gene is highly expressed during embryogenesis (4, 9, 34, 37) and in adult stem cells (4, 29, 30, 47), but is silenced in most differentiated cells. HMGA1 becomes reexpressed in aggressive cancer cells and high levels portend adverse clinical outcomes (28–38, 40–43, 56–64). In colon cancer, HMGA1 is among the genes most highly overexpressed compared with nonmalignant colon epithelium (4, 30, 57, 62). While mechanisms upregulating HMGA1 in cancer are incompletely understood, increasing evidence suggests that diverse oncogenic pathways, including growth factors (65, 66), mutations — such as mutant Apc (67, 68), KRAS (28, 36, 69), or mutant JAK2 (29) — and oncogenic transcription factors, like cMYC (70) or cJUN (71, 72), converge on HMGA1 to induce its expression in distinct settings. In transgenic mouse models, Hmga1 overexpression leads to tumorigenesis (35, 41, 42, 58, 64). For example, transgenic mice overexpressing Hmga1 in lymphoid cells develop clonal expansion with evolution to leukemia by upregulating transcriptional networks active in proliferating stem cells, poorly differentiated cancer cells, and inflammation (37, 38, 41, 58, 64). In experimental models of pancreatic cancer and myeloproliferative neoplasms, HMGA1 activates gene networks in a cell-intrinsic fashion to drive aberrant proliferation and differentiation, while inducing signals within the tumor microenvironment that promote fibrosis, culminating in tumor progression (28, 29). HMGA1 also upregulates genes involved in an epithelial-to-mesenchymal transition in colon cancer cell lines (35). In small intestinal stem cells, HMGA1 induces Sox9 and Wnt signals from the stromal and epithelial niches to maintain the stem cell compartment and Paneth cell niche during epithelial regeneration (4, 30). However, HMGA1 function in colon stem and progenitor cells during tumorigenesis was previously unknown.

Here, we uncover a previously unknown role for HMGA1 in modulating transcriptional networks to expand LGR5+ stem cells and Paneth-like cells during tumorigenesis driven by Apc deficiency. Strikingly, loss of just a single Hmga1 allele disrupts tumorigenesis while prolonging survival in two different models of colon tumorigenesis with APC inactivation, including mice with biallelic deletion of colonic epithelial Apc (CDX2P-CreER^T2Apc^fl/fl) (73, 74) and mice harboring monoallelic mutant Apc (Apc^Min/– or Min mice) colonized with the inflammatory human symbiote, enterotoxigenic Bacteroides fragilis (ETBF) (75–79). Single cell RNA sequencing (scRNA-seq) in Apc-deficient crypt epithelium reveals that HMGA1 maintains colon crypt cells in a stem and Paneth-like cell state while depleting differentiated enterocytes. Integration of transcriptomic analyses with assays of chromatin accessibility demonstrate that HMGA1 activates Wnt signals by “opening” chromatin at gene loci governing Wnt signaling, including the stem cell regulator, Achaete-Scute Family BHLH Transcription Factor 2 (Ascl2), in addition to Wnt agonist receptors (Lgr5, Lrp5) and downstream effectors. We focus on the gene encoding ASCL2 as a master regulator of cell fate in the small intestine, although its role in the colon was previously unknown. In human colon cancer cells, HMGA1 directly induces ASCL2 by recruiting activating histone marks. Further, silencing HMGA1 disrupts oncogenic properties (proliferation and clonogenicity), while reexpression of ASCL2 partially rescues oncogenic phenotypes in HMGA1-depleted human colon cancer cells. Most importantly, both HMGA1 and ASCL2 are coexpressed and upregulated in human colon cancer. Surprisingly, HMGA1 depletion has minimal effects on colon epithelial regeneration under homeostatic conditions. Our results establish HMGA1 as an epigenetic gatekeeper of ASCL2 and Wnt signals in colon stem cells during tumorigenesis, but not steady state homeostasis, highlighting HMGA1 pathways as promising therapeutic targets for colon carcinogenesis.

Loss of just a single Hmga1 allele is sufficient to decrease tumorigenesis and prolong survival in mice with colon tumors driven by biallelic Apc inactivation. Because previous studies from our group and others showed that HMGA1 is highly overexpressed in human colon cancer (30, 35, 62) and required for oncogenic properties in colon cancer cell lines (35), we sought to assess its role in colon tumorigenesis in vivo. Since APC is the most commonly mutated gene in human colon cancer (15), we examined CDX2P-CreER^T2Apc^fl/fl mice, an established model of colon tumorigenesis caused by inducible, biallelic loss of Apc within colon epithelium (73, 74). CDX2P-CreER^T2Apc^fl/fl mice were crossed to mice with global Hmga1 deficiency (heterozygous or homozygous). Importantly, mice with heterozygous Hmga1 have normal development and lifespans, whereas those with homozygous deficiency have partial embryonic lethality and develop premature aging phenotypes (kyphosis, bone loss, greying, and shortened lifespans) beginning after 10–12 months of age (29, 56). As expected, CDX2P-CreER^T2Apc^fl/fl mice with Hmga1 heterozygous or homozygous deficiency have lower Hmga1 gene expression and protein levels in colon epithelium (Supplemental Figure 1, A and B; supplemental material available online with this article; https://doi.org/10.1172/JCI184442DS1). Following induction of Cre recombinase-mediated Apc deletion by tamoxifen (TAM), CDX2P-CreER^T2Apc^fl/fl mice with intact Hmga1 alleles develop epithelial hyperplasia in the cecum, proximal, and midcolon regions (Figure 1A) associated with weight loss by 4 weeks (Figure 1B) and decreased survival (median survival 43 days after TAM; n = 16) (Figure 1C). Strikingly, loss of just a single Hmga1 allele in CDX2P-CreER^T2Apc^fl/fl mice mitigates weight loss while prolonging survival (median survival 61.5 days; P < 0.0001, n = 12; Figure 1, A–C). Survival is prolonged further (median survival 78 days; P < 0.0001, n = 7) (Figure 1C) in mice with Hmga1 homozygous deficiency, indicating that Hmga1 gene dosage modulates tumor progression in this model.

Figure 1

Loss of a single Hmga1 allele mitigates colon tumorigenesis and prolongs survival in CDX2P-CreER^T2/Apc^fl/fl mice. (A) Representative images (H&E) of proximal colon in CDX2P-CreER^T2/Apc^fl/fl mice with Hmga1 intact (Hmga1^+/+ top), heterozygous deletion (Hmga1^+/–, middle), or homozygous deletion (Hmga1^–/–, bottom) at survival endpoint necropsy (top: day 35 after TAM; middle: day 57; bottom: day 81). Scale bars: 250 μm. (B) Relative weight changes in CDX2P-CreER^T2/Apc^fl/fl models after TAM. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, 1-way ANOVA). (C) Kaplan-Meier plot showing survival in CDX2P-CreER^T2/Apc^fl/fl mice with Hmga1^+/+, Hmga1^+/–, or Hmga1^+/+. (****P < 0.0001; Mantel-Cox test). (D) Relative colon weight to body weight in CDX2P-CreER^T2/Apc^fl/fl mice with Hmga1^+/+, Hmga1^+/–, or Hmga1^–/– (*P < 0.05; Hmga1^+/+ versus Hmga1^+/–, **P < 0.01, Hmga1^+/+ versus Hmga1^–/–; Tukey’s multiple comparison test following significance by 1-way ANOVA). (E) Proximal colon crypt depth in CDX2P-CreER^T2/Apc^fl/fl models (*P < 0.05, ***P < 0.01, ****P < 0.0001; 1-way ANOVA with Tukey’s multiple comparison test). Each shape (circle, square, triangle) corresponds to a different mouse (n = 2–3/genotype). The solid shapes show the mean from each mouse; the open, smaller shapes represent individual measurements/crypt (range = 9–13 crypts/mouse) at × 20 magnification.

To determine more precisely how HMGA1 modulates tumorigenesis in CDX2P-CreER^T2Apc^fl/fl mice, we compared colon weights as a surrogate for tumor burden since extensive, contiguous tumors in the proximal colon precludes precise enumeration. Both the absolute and relative colon weight (% colon weight/body weight) increase in CDX2P-CreER^T2Apc^fl/fl mice with intact Hmga1 compared with those with heterozygous or homozygous Hmga1 deficiency (Figure 1D and Supplemental Figure 1C). At 21 days following Apc inactivation, colon epithelium becomes thickened and dysplastic with extensive adenomatous changes and increased crypt depth in CDX2P-CreER^T2Apc^fl/fl mice with intact Hmga1 (Figure 1, A and E, and Figure 2, A and B). Intranuclear HMGA1 is prominent throughout the crypts up to the luminal enterocytes in CDX2P-CreER^T2Apc^fl/fl mice with intact Hmga1 by IHC. By contrast, intranuclear HMGA1 is normally restricted to the crypt bases in WT mice lacking Apc-deficiency (Supplemental Figure 1D). Since APC restrains Wnt signals by sequestering β-catenin within inhibitory, cytoplasmic complexes, we compared β-catenin levels and localization in nuclei and cytoplasm in the Apc-deficient models. Following Apc inactivation (day 21), both nuclear and cytoplasmic β-catenin levels increase in colon epithelial cells of mice with intact HMGA1 compared with those with heterozygous or homozygous Hmga1 deficiency, paralleling the distribution of intranuclear HMGA1 (Figure 2, A and B, and Supplemental Figure 1E). Cells staining positive for the proliferation marker Ki67 are also increased in CDX2P-CreER^T2Apc^fl/fl mice with intact HMGA1 compared with those with HMGA1 deficiency early in tumorigenesis, although they predominate at the crypt bases (Figure 2, A and B). In addition, HMGA1 increases in the adenomatous epithelium compared with nontumor, midcolon epithelium in mice with intact HMGA1 (Figure 3, A and B, and Supplemental Figure 1, A and B). Intriguingly, HMGA1 also increases in tumors from mice with Apc inactivation and Hmga1 heterozygous deficiency compared with adjacent nontumor colon epithelium (Figure 3, A and B, and Supplemental Figure 1, A and B). These findings indicate that loss of just a single Hmga1 allele in the setting to Apc inactivation decreases hyperproliferation, β-catenin levels, and tumorigenesis.

Figure 2

Hmga1 haploinsufficiency decreases β-catenin and Ki67 in CDX2P-CreER^T2/Apc^fl/fl mice. (A) Representative IHC images of nuclear HMGA1 (top), β-catenin (middle), and Ki67 (bottom) in CDX2P-CreER^T2/Apc^fl/fl models at 3 weeks after TAM. Scale bar: 200 μm. (B) Quantitative comparisons of IHC images (*P < 0.05, **P < 0.01; Tukey’s multiple comparison test following significance by 1-way ANOVA). Each shape (circle, square, triangle) corresponds to a different mouse [top bar graph (n = 2–3/genotype), middle bar graph (n = 3/genotype), bottom bar graph (n = 3/genotype)]. The solid shapes show the mean from each mouse; the open, smaller shapes represent individual values/field (range = 8–24 fields/mouse) at × 20 magnification.

Figure 3

Hmga1 deficiency decreases colon tumorigenesis. (A) Representative IHC images of nuclear HMGA1 (top), β-catenin (middle), and Ki67 (bottom) in CDX2P-CreER^T2/Apc^fl/fl models at 3 weeks after TAM. Scale bars: 1,000 μm (top panel); 200 μm (lower panel). (B) Quantitative comparisons of IHC images (*P < 0.05, ***P < 0.001, ****P < 0.0001; Tukey’s multiple comparison test following significance by 1-way ANOVA). Each shape (circle, square, triangle, hexagon) corresponds to a different mouse (n = 3–4/genotype). The solid shapes show the mean values from each tumor; the open, smaller shapes represent individual values/field (range = 6–17 fields/tumor from 1–5 tumors/mouse) at × 20 magnification.

HMGA1 hemizygous deficiency mitigates colon tumorigenesis induced by ETBF in Min mice. Next, we investigated HMGA1 function in the multiple intestinal neoplasia (Min) model, which harbors a heterozygous Apc loss-of-function mutation (Apc^Min/+ or Min^+/–) and develops distal colon tumors following inoculation with the human symbiotic bacterium ETBF. This model recapitulates salient features of human colon tumors with respect to the mutational status, location in the distal colon, and histopathology (75–79). Furthermore, ETBF colonization is common in colon cancer (up to 90%) and epidemiologic studies suggest that it increases the risk of carcinogenesis (75–85). Following inoculation with ETBF at 5–6 weeks of age, Min mice with intact Hmga1 exhibit poor weight gain and robust distal colon tumorigenesis by 11–12 weeks with a median survival of 17 weeks; by contrast, Min mice with global Hmga1 hemizygosity gain more weight, develop fewer tumors, and exhibit prolonged survival (Figure 4, A–D). Histologic examination shows hyperproliferative colon epithelium and adenomatosis with increased crypt depth in the distal colons of Min mice with intact HMGA1 compared with those with HMGA1 deficiency (Figure 4, E and F). As expected, HMGA1 mRNA and protein levels by IHC are increased in mice with intact HMGA1 compared with Hmga1 haploinsufficient Min mice (Figure 4, E and F, Figure 5, A–C, and Supplemental Figure 2A). Similarly, the proportion of cells staining positive for intranuclear β-catenin and cytoplasmic β-catenin are greater in colon epithelium and tumors of Apc^Min/+ mice with intact HMGA1, although Ki67 was unchanged in mice with intact or haploinsufficient HMGA1 (Figure 4, E and F, and Supplemental Figure 2B). Intriguingly, HMGA1 protein staining is similar in tumors from Apc^Min/+ mice with intact HMGA1 and HMGA1 haploinsufficiency, suggesting that mice with HMGA1 haploinsufficiency can upregulate the intact Hmga1 allele to increase HMGA1 levels within their tumors (Figure 5, A–C). The small intestinal tumor burden is also greater in Apc^Min/+ mice with intact HMGA1 (Supplemental Figure 2C), indicating that HMGA1 contributes to tumorigenesis in both the colon and small intestine of Min mice.

Figure 4

Hmga1 haploinsufficiency disrupts colon tumorigenesis induced by ETBF in APC^Min/+ mice. (A) Body weights at necropsy after ETBF in Apc^Min/+ mice with intact Hmga1 or heterozygous Hmga1 (*P < 0.05, **P < 0.01, ***P < 0.001; student’s t test). (B) Representative images of methylene-blue stained colons to visualize tumors in Apc^Min/+/Hmga1^+/+ mouse (top) compared with Apc^Min/+/Hmga1^+/– mouse (bottom) at 11–12 weeks after ETBF. (C) Normalized tumor numbers in Apc^Min/+ models (*P < 0.05; Mann-Whitney test). (D) Kaplan-Meier plot showing survival in in Apc^Min/+ mice with intact Hmga1 or heterozygous (*P < 0.05; Mantel-Cox test). (E) Representative images (H&E left; IHC right; Scale bars: 200 μm) for HMGA1 (second column), β-catenin (third column), and Ki67 (right) in distal colon of Apc^Min/+ models at 11–12 weeks after ETBF. (F) Comparison of crypt depths (**P < 0.01) and IHC for nuclear HMGA1 (**P < 0.01), nuclear β-catenin (****P < 0.0001) and Ki-67(P = 0.16, unpaired student’s t test for each comparison) in Apc^Min/+ models. For crypt depth (left), each shape (circle, square, triangle, hexagon) corresponds to a different mouse (n = 3–4/genotype). The solid shapes show the mean from each mouse; the open, smaller shapes represent individual measurements/crypt (range = 9–16 crypts/mouse). For the IHC comparisons, each shape (circle, square, triangle, hexagon) corresponds to a different mouse (n = 3–4/genotype), the solid shapes show the mean value from each mouse; the open, smaller shapes represent individual values/field (range=9-19 fields/mouse) at x20 magnification.

Figure 5

Hmga1 haploinsufficiency decreases β-catenin and tumorigenesis induced by ETBF in APC^Min/+ mice. (A) Representative images (H&E) of distal colon tumors in Apc^Min/+ models at 11–12 weeks after ETBF. (B) Representative images (IHC) of distal tumors in Apc^Min/+ models for HMGA1 and β-catenin at 11–12 weeks after ETBF. Scale bars: 200 μm. (C) Quantitative IHC comparisons of distal tumors in Apc^Min/+ models for HMGA1 (P = 0.09) and β-catenin (****P < 0.0001; unpaired student’s t test for both). Each shape (circle, square, triangle, hexagon) corresponds to a different mouse (n = 3–4/genotype). The solid shapes show the mean from each mouse; the open, smaller shapes represent individual values/field (range = 2–4 tumor fields/mouse) at × 20 magnification.

Loss of a single Hmga1 allele within the colon epithelium is sufficient to reduce colon tumorigenesis induced by ETBF in Min mice. To determine whether HMGA1 deficiency within colon epithelium is sufficient to mitigate tumorigenesis in the Min-ETBF model, we generated Min mice with Hmga1 deficiency (hetero- and homozygous genetic deletion) restricted to colon and small intestinal epithelium by crossing Apc^Min/+ mice with Hmga1^fl/fl mice on a Vil-cre background. Notably, Min mice with tissue-specific HMGA1 deficiency developed fewer colon tumors and decreased crypt depth compared with Min mice with intact HMGA1 (Figure 6, A–E). Surprisingly, Min mice with Hmga1 haploinsufficiency had a similar decrease in colon tumor number as Min mice with homozygous Hmga1 loss, suggesting that a relatively modest decrease in HMGA1 within the epithelial compartment alone is sufficient to mitigate tumorigenesis. Small intestinal tumors also decrease modestly in this model, but only with homozygous loss of Hmga1 (Supplemental Figure 3A). By contrast, tissue-specific biallelic loss of Hmga1 in colon crypts from WT mice lacking Apc mutation show no significant changes in crypt depth, suggesting that HMGA1 deficiency under steady state, homeostatic conditions (no ETBF colonization) is not deleterious to colon epithelial regeneration (Supplemental Figure 3B). Together, these findings demonstrate that HMGA1 within the crypt epithelium drives tumorigenesis, and, moreover, tissue-specific, Hmga1 haploinsufficiency is sufficient to impair colon tumor formation driven by mutant Apc and inflammatory ETBF, highlighting HMGA1 as a promising potential therapeutic target.

Figure 6

Loss of Hmga1 allele within colon epithelium decreases colon tumorigenesis induced by ETBF in Apc^Min/+ mice. (A) Representative images of methylene-blue–stained colons of Apc^Min/+ mice (top) compared with Apc^Min with tissue-specific heterozygous Hmga1 deletion (middle) and tissue-specific homozygous Hmga1 deletion (bottom) at 11–12 weeks after ETBF. (B) Relative tumor numbers (%) in Apc^Min mice with intact Hmga1, tissue-specific heterozygous Hmga1 deletion, or tissue-specific homozygous Hmga1 deletion from 3 separate experiments; tumor numbers in control were assigned a value of 100 (*P < 0.05; Mann-Whitney test for both comparisons). (C) Representative images (H&E) of distal colon of Apc^Min/+ with or without tissue-specific Hmga1 deficiency models. Scale bars: 100 μm. (D) Distal colon crypt depths in Apc^Min/+ mice with or without tissue-specific Hmga1 deficiency. (**P < 0.01, ****P < 0.0001; Tukey’s multiple comparisons test following significance by 1-way ANOVA). Each shape (circle, square, triangle, hexagon) corresponds to a different mouse (n = 3–4/genotype). The solid shapes show the mean value from each mouse; the open, smaller shapes represent individual measurements/crypt (range = 9–19 crypts/mouse) at × 20 magnification. (E) Representative images (H&E) of distal colon tumors of Apc^Min/+ with or without tissue-specific Hmga1 deficiency.

HMGA1 expands colon stem cells and Paneth-like cells while depleting more differentiated enterocytes in Apc-deficient colon crypts. To investigate molecular mechanisms underlying HMGA1 in Apc-deficient colon crypts, we performed scRNA-seq in proximal colon crypt cells from CDX2P-CreER^T2Apc^fl/fl mice with intact Hmga1 compared with those with heterozygous or homozygous Hmga1 deficiency. We examined transcriptomes at early stages in tumorigenesis (21 days following Apc inactivation via TAM) to identify mechanisms involved in tumor initiation. Single-cell transcriptomes depicted by uniform manifold approximation and projection (UMAP) reveal differences in overall distribution in cells with or without HMGA1 (Figure 7A). Unsupervised hierarchical clustering of transcripts (via Seurat) revealed 12 clusters (Figure 7B) from which cell identities were imputed using established markers (Supplemental Table 1). Of these clusters, five are comprised of epithelial crypt cells (denoted epithelial island) based on expression of the colon epithelial cell adhesion marker gene (Epcam), colon stem and progenitor cell genes (Lgr5, Sox9, Ctnnb1), and proximity by UMAP (Figure 7C). The remaining clusters are comprised of immune cells (Figure 7B). Within the immune cell islands, we identified Cd4⁺ and Cd8⁺ T cells with smaller populations of B cells, macrophages, and other myeloid lineages (macrophage-like, mast cells, and neutrophils) (Supplemental Figure 4A).

Figure 7

HMGA1 expands colon stem cells and Paneth-like cells while depleting more differentiated cells in Apc-deficient colon crypts. (A) UMAPs from scRNA-seq of crypt cells from CDX2P-CreER^T2 Apc^fl/fl mice with Hmga1^+/+ or Hmga1^–/–; shown together (left) or separately to highlight differences (center and right). (B) UMAP from scRNA-seq by cluster. Three distinct islands capture epithelial cell types (red circle), T cells (blue circle), and other immune cells (yellow). Imputed cell identities are designated by separate colors. TA, transit amplifying cells; EC, enterocytes. (C) Epcam, Lgr5, and other Wnt genes (Sox9, Ctnnb1) are enriched in the epithelial island. Single cell transcripts from both genotypes are shown. (D) Relative proportion of cell types in crypt cells by genotype (bar graph, left; Table, right). (Association between cell and HMGA1 status was evaluated by χ² test for each cell type versus all others).

To dissect HMGA1-dependent changes in the cell of origin for colon tumors, we focused on the epithelial island. With intact HMGA1, the LGR5⁺ stem cell population, defined by high levels of Lgr5, Msi1, Bmi1, and other stem cell transcripts (Supplemental Table 1), comprise the majority of cells (40.1%) within CDX2P-CreER^T2Apc^fl/fl colon crypts (Figure 7D). Transit amplifying (TA) cells are the next most abundant population, constituting 26% of epithelial crypt cells, whereas Paneth-like cells, based on Paneth cell markers (Lyz1, Mmp7, Sox9, Retnlb, Chil3, Reg3g, and Deta; Supplemental Table 1), comprise 20% of crypt cells. Intriguingly, while Paneth cells are not present in normal colon epithelium, Paneth cell “metaplasia” has been reported in proximal colon epithelium in adenomas, adenocarcinoma, and inflammatory bowel disease, and ectopic Paneth cells were observed in colon epithelium of Apc-deficient mouse models (73, 74, 85, 86). The terminally differentiated enterocyte (EC) and goblet cell clusters comprise the least frequent crypt cell types in this model (9.7% and 4.1%, respectively). Strikingly, HMGA1 deficiency decreases the proportion of stem and Paneth-like cell clusters by about 50% (P < 0.0001) together with a concurrent expansion in the proportion of differentiated ECs (from 9.7% to 28.9%; P < 0.0001), TA cells (from 26% to 34.9%; P < 0.0001), and goblet cells (from 4.1% to 8%; P < 0.0001) within the crypt epithelium. In Apc-deficient crypts with Hmga1 heterozygous deficiency, the changes in most clusters are intermediate between crypts with intact or homozygous deficiency of Hmga1 (Supplemental Figure 4, B and C). Together, these results indicate that intact HMGA1 is required to maintain the colon stem and Paneth-like cells in the setting of Apc deficiency while depleting more differentiated cells (ECs and goblet cells).

Within the immune cell islands, Cd4⁺ and Cd8⁺ T cells increase in frequency in the HMGA1-deficient crypt cells (Supplemental Figure 4), the latter of which could reflect an increase in tumor-infiltrating T lymphocytes. Both Cd4⁺ and Cd8⁺ T cells with Hmga1 genetic deletion also exhibit a shift on UMAP, indicating that HMGA1 loss within these T cell populations alters their underlying transcriptomes (Supplemental Figure 4A).

Trajectory and cell state analyses show that HMGA1 maintains an earlier cell state in Apc-deficient crypt cells. To delineate HMGA1 function in differentiation dynamics in Apc-deficient colon crypt epithelium, we performed pseudotime trajectory analyses, assigning time = 0 to the most dedifferentiated, stem cell cluster (via Seurat, Monocle 2). Intact HMGA1 results in a greater proportion of cells at earlier stages in development (time = 0; undifferentiated stage) whereas HMGA1 deficiency leads to more cells in later stages (time = 12; more differentiated stage) within the differentiation trajectory (Figure 8A). Next, we applied cell state analysis (Seurat, Monocle 2) as a more static assessment of differentiation status of each cell along the trajectory in Apc-deficient epithelial crypt cells (Figure 8B). Cell states (defined by the top 200 most differentially expressed genes within 5 groups with distinct transcriptomes) were assigned to individual cells along the trajectory. Similar to our cluster analysis, Apc-deficient crypt cells with intact Hmga1 include a greater proportion of cells in an undifferentiated stem cell state (state 0) or Paneth-like state (state 1) compared with those with HMGA1 deficiency, which skews development to later, more differentiated cell states (states 3–4) (Figure 8B) or ECs (Figure 8C).

Figure 8

Hmga1 deficiency alters cell state, decreasing stem and Paneth-like cell populations while expanding more differentiated cell populations in Apc deficient crypt cells. (A) Pseudotime trajectory analysis estimated from scRNA-seq of CDX2P-CreER^T2 Apc^fl/fl crypt cells from the epithelial island with Hmga1^+/+ or Hmga1^–/–. HMGA1 deficient cells are more prominent in later stages of pseudotime (indicated by black ovals) compared with time 0 cells. (B) Cell states defined by the top 200 most differentially expressed genes on the trajectories from pseudotime analysis were assigned 0–4 and indicated by color on a trajectory plot (left) or bar graph (right). Note the skewing to cell states 3 and 4 in HMGA1 deficient cells. (C) Stem cells and enterocytes (ECs) imputed from scRNA-seq are shown on the trajectories to highlight the major differences between CDX2P-CreER^T2 Apc^fl/fl cells with intact HMGA1 or HMGA1 deficiency. HMGA1 deficient cells have increased ECs (blue) with decreased stem cells (violet). Bar graphs show relative cell frequencies (right); the top graphs show only stem and ECs, the bottom includes all cells with grey depicting cells that are not stem cells nor ECs.

Single cell transcriptomes suggest that HMGA1 accelerates proliferation by inducing gene networks involved in cell cycle progression. Next, we inferred cell cycle status of each cell in the epithelial cluster from scRNA-seq (Seurat; standard settings). Transcriptomic changes suggest that intact HMGA1 in Apc-deficient crypt epithelial cells function by increasing proliferation, as evidenced by decreases in the proportion of cells in G0/G1 concurrent with increases in the proportion reaching G2/M; the proportion of S phase cells were similar in Apc-deficient crypt cells with or without HMGA1 deficiency (Supplemental Figure 5). The proliferation marker gene encoding Ki67 is among the most upregulated genes of all G2/M genes. Collectively, our single-cell transcriptomes, together with increases in crypt depth, Ki67 protein staining, and tumorigenesis in mice with colon epithelial Apc inactivation and intact HMGA1 (Figure 1) are consistent with a model whereby HMGA1 increases proliferation restricted to cells at the earliest developmental stages, leading to expansion in stem and Paneth-like cells at the expense of more differentiated cells within the crypt epithelium.

HMGA1 activates gene networks within crypt epithelial cells involved in IFN signaling, inflammation, DNA repair, proliferation, and Wnt signaling. To elucidate mechanisms underlying HMGA1 in Apc-deficient crypt cells, we performed gene set enrichment analysis (GSEA; MSigDB) with Hallmark and Curated gene sets (87, 88). GSEA with transcripts from all clusters (epithelial and immune cells) reveal that HMGA1 upregulates heterogenous pathways, including those associated with metabolism (oxidative phosphorylation and glycolysis), proliferation (MYC Targets V1 and MYC Targets V2), and inflammation (IFN-α Response and IFN-γ Response) (Supplemental Figure 6). HMGA1 also activates multiple WNT networks, including Wnt Pathway requiring MYC, Degradation of the β-catenin Destruction Complex, TCF Dependent Signaling in Response to WNT, and APC Targets (Table 1). By contrast, transcriptional networks repressed by HMGA1 include allograft rejection, IL2-STAT5 signaling, and mitotic spindle genes (Hallmark). Notably, repression in allograft rejection and IL2-STAT5 gene networks have been implicated in immune escape and decreases in cytotoxic tumor infiltrating lymphocytes (Supplemental Figure 6) (89). The heterogeneity in these pathways is consistent with the diverse cell populations (epithelial and immune) within the crypts.

Table 1

Curated WNT pathways upregulated by HMGA1

To focus our analysis on the tumor-initiating cells, we performed GSEA exclusively on transcripts from the crypt epithelial island. Further, this island is comprised of the majority of cells from the crypt isolates and cell numbers are sufficient for pathway analyses. In the remaining immune islands, cell numbers were insufficient for further GSEA. Strikingly, HMGA1 activates transcriptional networks involved in inflammation, including IFN-α and IFN-γ response genes and proliferation (MYC targets V1) within the epithelial island (Figure 9A). DNA repair genes are also induced, which is a frequent transcriptional response when quiescent stem cells are triggered to cycle and proliferate (Figure 9A) (90–92). Among the IFN networks, multiple IFN-induced genes that mediate inflammatory signals are upregulated by HMGA1, including IFN-induced transmembrane proteins 1, 2, 3 (Ifitm 1–3), IFN stimulated gene 15 (Isg15), Stat1, Stat2, and cytokines (Ccl5, Cxcl9/10) (Figure 9B). Wnt pathway genes are also prominent among the networks activated by HMGA1 within the epithelial crypt cells (Figure 9A and Table 1). By contrast, HMGA1 represses gene networks involved in fatty acid metabolism and adipogenesis, metabolic pathways used extensively by differentiated ECs in intestinal epithelium (Figure 9A) (93). HMGA1 also represses genes controlling protein secretion (Figure 9A), an important cellular function of differentiated ECs, which secrete digestive enzymes (93). Together, our single-cell transcriptomes and pathway analysis demonstrate that HMGA1 expands the stem and Paneth-like cells while driving transcriptional networks involved in proliferation, inflammation, Wnt signaling, and DNA repair. Conversely, HMGA1 restrains differentiation and represses metabolic gene networks active in differentiated ECs.

Figure 9

HMGA1 activates gene networks within the crypt epithelial island involved in IFN signaling, inflammation, DNA repair, proliferation, and Wnt signaling. (A) GSEA analysis (left) of single cell transcripts from the epithelial island reveals that HMGA1 activates pathways involved in inflammation (IFN-α, IFN-γ), DNA repair, and proliferation (MYC) while repressing pathways active in differentiated ECs (fatty acid metabolism, protein secretion); FDR ≤ 0.25. Enrichment plots (right) show HMGA1 networks in more detail, including genes involved in inflammation (IFN-α), DNA repair, and Wnt signaling. Normalized enrichment score (NES) and normalized P values are indicated. (B) IFN-inducible genes that mediate inflammatory signals, including IFN-induced transmembrane 1, 2, 3, (Ifitm1, 2, 3) genes, IFN stimulated gene 15 (Isg15), Stat1, Stat2, and cytokines (Ccl5, Cxcl9, Cxcl10) are activated by HMGA1. Dot plots depict gene expression (–0.4 to +0.4) and the proportion of cells (25%–75%) expressing each transcript within the epithelial island.

HMGA1 enhances chromatin accessibility at gene loci involved in proliferation, DNA repair, inflammation, and Wnt signaling. Because HMGA1 is an architectural transcription factor that modulates chromatin structure, we performed assays to detect accessible chromatin mediated by HMGA1 via assays of transposase-accessible chromatin sequencing (ATAC-seq) in Apc-deficient crypt cells with or without HMGA1 deficiency (94). Notably, overall chromatin accessibility is enhanced in Apc-deficient crypt cells with intact HMGA1 compared with crypts lacking HMGA1 (Figure 10, A–C). HMGA1 results in both more peaks and longer stretches of accessible chromatin overall (Figure 10, A–C). Focusing on promoter regions (up to –3 kb from the transcription start sites), we also identified more peaks and longer stretches of open chromatin within these regulatory regions with intact HMGA1 (Figure 10, B and C). Similar to our scRNA-seq results, gene networks associated with HMGA1-mediated accessible chromatin included pathways involved in proliferation (MYC Targets V1, E2F Targets, G2M Checkpoint genes) and inflammation (IFN-γ, TNF-α signaling via NF-κB (Figure 10, D and E, and Supplemental Figure 7). Accessible chromatin is also enriched at Wnt signaling gene networks (Table 1). Intersecting pathways identified by both ATAC-seq and scRNA-seq (epithelial island) revealed that HMGA1 increased chromatin accessibility and expression of genes involved in proliferation (MYC Targets V1), DNA repair, inflammation (IFN-γ response genes), and Wnt signaling (Figure 10E and Table 1).

Figure 10

HMGA1 enhances chromatin accessibility at gene loci involved in proliferation, DNA repair, and inflammation. (A) HMGA1 increases chromatin accessibility in crypt cell nuclei globally in CDX2P-CreER^T2 Apc^fl/fl mice. (B) HMGA1 enhances chromatin accessibility in promoter regions ranging from 0 to –3 kb upstream of the transcription start sites shown by average peak lengths. (****P < 0.0001; student’s t test). (C) HMGA1 enhances chromatin accessibility in promoter regions ranging from 0 to –3 kb upstream of the transcription start sites shown by number of significantly expanded peaks. (P < 0.0001; χ²). (D) HMGA1 enhances chromatin accessibility in gene sets involved in proliferation, inflammation, and metabolism. (E) GSEA pathways identified by intersecting ATAC-seq and scRNA-seq pathways with associated P values.

HMGA1 amplifies expression of Wnt pathway genes in Apc-deficient colon crypts. Given the fundamental role for Wnt signaling in colon tumorigenesis and HMGA1-dependent upregulation of Wnt genes and β-catenin levels in our tumor models, we further examined the relationship between HMGA1 and Wnt genes. We focused on canonical Wnt pathway genes, including Wnt effectors (Ctnnb1, Tcf4, Axin2, Cd44, Ets2, Ephb2, Ascl2, cMyc, Prom1, Sox9) and Wnt receptors (Lgr5, Lrp5, Lrp6, Fzd5, Fzd7). Remarkably, all Wnt effector genes were upregulated at the level of single cells in the setting of intact HMGA1 and Apc deficiency (Figure 11A). Of the Wnt receptors, both Lgr5 and Lrp5 transcripts are upregulated in crypt cells with intact HMGA1. We also found significant positive correlations between Hmga1 and multiple Wnt effectors (Figure 11B) with the strongest correlations (r > 0.68; P < 0.05) for Ascl2, Axin2, Tcf4, Ctnnb1, Ephb2, and cMyc. To determine whether HMGA1 enhances chromatin accessibility at promoter regions for these genes, we examined our ATAC-seq results, which revealed increased chromatin accessibility at promoter regions for Ascl2, Tcf4, Prom1, Lgr5, and Lrp5 (Supplemental Figure 8). Intriguingly, HMGA1 is also associated with accessible chromatin at the Hmga1 promoter (Supplemental Figure 8), suggesting that high levels of HMGA1 induce its own expression in Apc-deficient crypts by opening chromatin at its promoter region. Together, these results demonstrate that HMGA1 enhances chromatin accessibility to activate Wnt agonist receptor signaling and Wnt effector genes. In small intestinal epithelium, ASCL2 activates Wnt genes (95–98), and upregulation in Ascl2 could trigger a feed-forward loop whereby HMGA1 activates Ascl2, which, in turn, amplifies Wnt gene expression in colon epithelium with Apc inactivation.

Figure 11

HMGA1 amplifies Wnt genes in Apc-deficient colon crypts, and these HMGA1-Wnt pathways are activated in human colon cancer. (A) Dot plot of Wnt effector and receptor gene expression in crypt cells of CDX2P-CreER^T2 Apc^fl/fl mice with Hmga1^+/+ versus Hmga1^–/–, demonstrating that HMGA1 activates all Wnt effectors and many Wnt receptor genes. (B) Hmga1 is positively and strongly correlated with Wnt genes, including Wnt effector genes (Ascl2, Axin2, Tcf4, Ctnnb1, Myc, Ephb2, Sox9) and Wnt receptor genes (Lgr5, Lrp5, Fzd7) (Spearman’s rank correlation test). (C) Heatmap showing HMGA1 and WNT genes. HMGA1 correlates positively with ASCL2 and MYC (log scale) in human colon cancer. (D) HMGA1 and WNT genes in nonmalignant colon epithelium (n = 41) and human colorectal adenocarcinoma (n = 286) from TCGA (student’s t test). ^†HMGA1 and SOC9 expression were previously reported in ref. 30. TPM, transcripts per million. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

HMGA1 and ASCL2 are upregulated and coexpressed in human colorectal cancer. To determine which HMGA1 pathways are relevant to human colon tumorigenesis, we queried the Cancer Genome Atlas (TCGA) for expression of HMGA1 and Wnt genes (Figure 11, C and D). As we previously reported, HMGA1 and SOX9 are upregulated in colon cancer compared with nonmalignant epithelium (4, 30, 62). Strikingly, most of the Wnt genes upregulated by HMGA1 in our murine model are also upregulated in human colon cancer, including the WNT effectors, ASCL2, AXIN2, CTNNB1, MYC, EPHB2, CD44, and ETS2 and the WNT receptors, LGR5, LRP5, and LRP6 (Figure 11, C and D). Further, both ASCL2 and cMYC are upregulated and positively correlated with HMGA1, suggesting that HMGA1 may directly induce their expression in human colon tumorigenesis (Figure 11C).

HMGA1 upregulates ASCL2 and promotes oncogenic properties in human colon cancer cells. The ASCL2 transcription factor is critical to cell fate in the small intestine (95–98) although its role in the colon has not been studied in detail. Because our results strongly link HMGA1 to ASCL2 in colon tumorigenesis in humans and mice, we tested whether HMGA1 directly activates ASCL2 expression in human colon cancer cells. Silencing HMGA1 in 2 human colon cancer cell lines (SW620, SW480) by lentiviral-mediated delivery of short hairpin RNA (shRNA) or CRISPR/Cas9 represses ASCL2, demonstrating that ASCL2 expression depends on HMGA1 (Figure 12A and Supplemental Figure 9A). Next, we tested whether HMGA1 deficiency affects oncogenic properties in these cells. We previously reported that HMGA1 knockdown decreases clonogenicity in SW480 cells using plasmid-mediated gene silencing (35). Here, we found that HMGA1 silencing (via shRNA or CRISPR) disrupts proliferation and clonogenicity similarly in SW620 and SW480 cells (Figure 12, A and B, and Supplemental Figure 9, A and B), demonstrating that these in vitro, oncogenic phenotypes depend on high levels of HMGA1. To determine whether ASCL2 restoration will rescue these phenotypes, we reexpressed ASCL2 in both colon cancer cell lines (SW620 and SW480) with HMGA1 silencing (Supplemental Figure 9C). Restoration in ASCL2 levels results in a partial rescue of proliferation and full rescue of clonogenicity in both cell lines, suggesting that ASCL2 mediates some, but not all, effects of HMGA1 in these colon cancer cells (Supplemental Figure 9, C and D).

Figure 12

HMGA1 induces ASCL2 by directly binding to the promoter and recruiting activating histone marks. (A) Silencing HMGA1 represses ASCL2 and decreases proliferation in SW620 and SW480 cells. Control cells were transduced with empty lentiviral vector versus shRNA targeting HMGA1. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; student’s t test). (B) Silencing HMGA1 decreases clonogenicity in SW620 and SW480 (**P < 0.01, ****P < 0.0001; student’s t test). (C) Predicted HMGA1 binding sites 1–5 and region 6–7 in the ASCL2 promoter region shown with activating histone marks from SW620 and SW480 (GSE106921). (D) ChIP assay results at sites 1–5 and region 6–7 in SW620 cells from one representative biological replicate for HMGA1 and activating histone marks (H3K4me3,and H3K27ac). (E) ChIP assay results at sites 1–5 and region 6–7 in SW620 cells from one representative biological replicate for H3 and the repressive histone (H3K27me3). (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; student’s t test following significance by ANOVA).

HMGA1 directly induces ASCL2 by binding to its promoter and recruiting activating histone marks in human colon cancer cells. To ascertain whether HMGA1 binds directly to the ASCL2 promoter to activate its expression, we used an in silico prediction algorithm (TRAP) (99), which identified 7 potential HMGA1 binding sites within the ASCL2 promoter-enhancer region (labeled 1, 2, 3, 4, 5, 6, and 7; Supplemental Figure 10A). Because sites 6 and 7 are within 10 base pairs of each other, they could not be resolved by ChIP-PCR, and were therefore denoted region 6–7. Intriguingly, all of these sites are positioned near the homologous regions of HMGA1-dependent accessible chromatin in the mouse Ascl2 promoter (Figure 12C). By chromatin immunoprecipitation-PCR (ChIP-PCR), we assessed HMGA1 chromatin occupancy at these sites in SW620 cells, as these cells have higher levels of both HMGA1 and ASCL2 compared with SW480 cells. There was robust HMGA1 occupancy throughout the ASCL2 promoter compared with the IgG antibody as a negative control (Supplemental Figure 10, A and B). Next, we compared HMGA1 chromatin binding in SW620 cells with or without HMGA1 silencing, which showed enrichment for HMGA1 occupancy at the same regions in control cells and depletion of HMGA1 with HMGA1 silencing, validating the specificity of our HMGA1 antibody (Figure 12D and Supplemental Figure 10C). Because HMGA1 recruits active histone marks to upregulate developmental genes in other settings (28, 29), we tested whether HMGA1 binding associates with activating histones, including histone H3 lysine 4 trimethylation (H3K4me3) and histone H3 lysine 27 acetylation (H3K27Ac), which mark active promoters and enhancers, respectively. These marks also associate with HMGA1 chromatin binding in other tumor settings (28, 29). We found enrichment for both activating histone marks (H3K4me3, H3K27Ac) in the regions of HMGA1 binding in SW620 and SW480 cells from public databases (GSE10692) (Figure 12, C and D). By ChIP-PCR, we found that both H3K4me3 and H3K27Ac bind to the ASCL2 promoter, with greatest enrichment near the HMGA1 binding site number 1 (located near the transcription start site [TSS]) and these active marks are depleted with HMGA1 silencing (Figure 12D and Supplemental Figure 10C). By contrast, there was no change with HMGA1 silencing in the positive control, histone H3, which is a ubiquitous histone that does not modulate gene expression (Figure 12E and Supplemental Figure 10C). We also tested whether HMGA1 depletion enables repressive histones to bind to the ASCL2 promoter as a mechanism of downregulating ASCL2 with HMGA1 silencing. Because the repressive histone 3 lysine 27 trimethyl (H3K27me3) was identified in a colon cancer cell line (HCT116) from a public database (GSE171817), we assessed its binding relative to that of HMGA1. In control SW620 cells with high HMGA1, there was minimal binding of the repressive mark, H3K27me3; however, HMGA1 depletion results in modest, yet significant increases in the H3K27me3 repressive mark at the ASCL2 promoter enhancer region (Figure 12E and Supplemental Figure 10D).

To determine if HMGA1 activates the ASCL2 promoter, we cloned the human ASCL2 promoter sequence (–2.5 kb from the TSS) upstream of the luciferase reporter gene and transfected this construct into colon cancer cell lines (SW620, SW480) (Supplemental Figure 11A). In both cell lines with abundant levels of HMGA1, the ASCL2 promoter is induced compared with control vector lacking the ASCL2 promoter sequence (Supplemental Figure 11B). By contrast, HMGA1 silencing decreases ASCL2 promoter activity, consistent with HMGA1-dependent activation of the ASCL2 promoter (Supplemental Figure 11B). Together, our results support a model whereby HMGA1, present in high levels, binds to the ASCL2 promoter, enhances chromatin accessibility, and recruits activating histones to induce ASCL2 and downstream Wnt genes, thereby driving tumorigenesis in the setting of Apc deficiency. Moreover, our findings further highlight HMGA1 as a promising potential therapeutic target, particularly since loss of HMGA1 in colon epithelium has only subtle effects on epithelial regeneration under homeostatic conditions.

Changes in nuclear structure and function are required for normal development, tissue regeneration, and tumorigenesis (4, 11). While underlying mechanisms remain incompletely understood, chromatin state has emerged as a fundamental player required for diverse cell fate decisions in tumor biology. Embryonic stem cells and tissue-specific, adult stem cells have large nuclei harboring “open” accessible chromatin, which is thought to endow these cells with developmental potency or the capacity to differentiate into diverse progeny with distinct functions (4, 11). Similarly, nuclei in aggressive cancer cells are often enlarged and irregular (4, 11), whereas nuclear compaction accompanies differentiation in normal tissues (4, 11). Though somatic mutations accumulate in adult stem cells over time, particularly in highly proliferative tissues, such as the colon crypts, most mutated cells do not evolve into tumors. Thus, changes in chromatin structure and cell state provide a plausible requisite for tumor development. Indeed, pathologists distinguish cancer cells from nonmalignant cells primarily by alterations in nuclear architecture. These observations suggest that understanding mechanisms underlying chromatin structure and cell state during tumor evolution could reveal strategies to intercept the transition from early neoplasia to invasive cancers. Colon tumorigenesis offers a unique opportunity to study cell state, adult stem cells, and tumorigenesis given the hierarchical organization of stem and progenitors within the crypts along with evidence implicating mutated LGR5+ colon stem cells as a tumor initiating cell (100). Moreover, the incidence of colon cancer is increasing globally, particularly in younger individuals, highlighting the significance of this work (24–26).

Here, we discover that HMGA1 acts as an epigenetic regulator that imposes a stem-like chromatin state within Apc-deficient crypt epithelial cells. HMGA1 enhances chromatin accessibility at key loci, leading to activation of gene networks involved in Wnt signaling, proliferation, and inflammation early in tumorigenesis. As an architectural transcription factor, HMGA1 binds to DNA and recruits histones and other chromatin complexes to modulate gene expression, rather than acting on its own. While HMGA1 drives clonal expansion, aberrant differentiation, and transformation in diverse settings, its function in colon tumorigenesis has not been studied in detail despite the fact that it is among the most overexpressed genes in colon cancer compared with nonmalignant epithelium (30, 62). We found that Hmga1 haploinsufficiency dampens colon tumor development and prolongs survival in 2 models. Importantly, mice with Hmga1 heterozygosity (and WT Apc) have normal development and lifespans (28, 29, 56). By contrast, Apc deletion together with intact HMGA1 results in increasing HMGA1 and nuclear β-catenin protein levels, not only at their normal location at the crypt base, but throughout the crypt extending toward the luminal epithelium. Intriguingly, HMGA1 protein levels increase within the colon tumors compared with nontumor colon epithelium, even in the setting of Hmga1 haploinsufficiency. Precisely how this occurs will require further investigation, although these results underscore the importance of HMGA1 in tumorigenesis in the CDX2P-CreER^T2Apc^fl/fl. While prior studies show that HMGA2 is overexpressed in colon cancer and Hmga2 drives tumorigenesis in mouse models with Let-7 deficiency (101), we focus on HMGA1 since transcripts are approximately 100-fold higher than HMGA2 in colon cancer datasets (TCGA) and in many other human tumors (28–30).

In the Min model following inoculation with ETBF, Hmga1 haploinsufficiency globally or within the colon epithelium is sufficient to decrease tumorigenesis. Surprisingly, complete loss of Hmga1 (homozygous deficiency) from colon epithelium decreases tumor incidence similar to that of haploinsufficiency. This was unexpected, since global deletion of Hmga1 in the biallelic Apc-deficient model led to the greatest impact on tumor development and survival. However, in models of pancreatic tumorigenesis (28), we found a similar relationship whereby tissue-specific loss of just one Hmga1 allele was sufficient to dampen tumorigenesis and prolong survival, akin to results with tissue-specific loss of both Hmga1 alleles. Intriguingly, crypt depth is similar in the colon with intact Hmga1 or complete loss of Hmga1 in the WT Apc epithelial compartment. Based on our transcriptomic data showing that HMGA1 deficiency fosters differentiation of stem cells to enterocytes, we surmise that the crypt depth is maintained in the Hmga1 deficient model lacking Apc mutation through skewing of more quiescent stem cells towards proliferating enterocytes. Unfortunately, there are no pharmacologic inhibitors to directly disrupt HMGA1 function in the clinics, although modulating HMGA1 function or levels by approximately 50% is likely to be a more feasible therapeutic goal than a more comprehensive disruption of its function.

To identify potential therapies to disrupt HMGA1 function, we focused on the epigenetic landscape downstream of HMGA1. HMGA1 enhances chromatin accessibility globally, in addition to “opening” regulatory regions of the genome important for activation of proliferation, inflammation, and Wnt signaling genes, including Ascl2. ASCL2 functions as a master regulator of stemness in small intestinal epithelium where it activates its own expression and that of downstream Wnt effector and receptor genes (95–97), and in esophageal cancer (98), although its role in colon epithelium had not been studied in detail. A recent study also links ASCL2 expression to early-onset colorectal cancer in a Japanese cohort (102).

We also found that HMGA1 binds directly to the ASCL2 promoter region and recruits activating histones (H3K4me3, H3K27Ac) to upregulate its expression. Indeed, Hmga1 and Ascl2 are the most tightly coregulated genes in murine crypt epithelium (Figure 7). Hmga1 is also coregulated with other Wnt effectors (Axin2, Tcf4, Ctnnb1, cMyc, Sox9) and Wnt agonist receptors (Lgr5, Lrp6, Fzd7, and Ephb2). In addition, HMGA1 enhances chromatin accessibility at promoter regions for Tcf4, Prom1, Lgr5, and Lrp5, suggesting that it may directly induce these Wnt genes or modify chromatin to facilitate their expression. In small intestinal stem cells and Caco2 cells (a human colon cancer cell line), HMGA1 directly induces SOX9, and upregulation of SOX9 by HMGA1 could enhance expansion in Paneth-like cells, since Paneth cell differentiation in small intestine depends on SOX9 (30). While classical Paneth cells are not present in the colon, Paneth cell metaplasia occurs in the proximal colon in Apc mice (73, 74) and in humans with inflammatory bowel disease (103), although their role in tumorigenesis is not yet clear. HMGA1 also correlates with MYC in human colon cancer and prior work in other settings shows that MYC directly induces HMGA1 (70). HMGA1 also binds to the MYC promoter to induce its expression in embryonic stem cells (34). In colon crypt cells, HMGA1 enhances chromatin accessibility at the Hmga1 promoter, suggesting that HMGA1 induces its own expression, a feature common to many stemness transcriptional regulators, such as ASCL2 (95). Given the link between ASCL2 and early onset colon cancer (102), further studies to explore HMGA1 and ASCL2 are warranted. Restoration of ASCL2 only partially rescues proliferation in colon cancer cell lines with HMGA1 silencing, indicating that HMGA1 regulates additional networks during colon tumorigenesis.

We also identified inflammatory and proliferative networks that are induced by HMGA1 and associated with HMGA1-dependent accessible chromatin. Both IFN-α and γ signaling networks are upregulated in crypt epithelial cells by HMGA1, leading to activation in IFN-stimulated genes and inflammatory networks, including signal transduction and activator of transcription 1 and 2 (Stat1/2), and chemokine genes encoding C-X-X motif chemokine ligands 9 and 10 (CXCL9/10). Importantly, inflammatory cytokines and/or their receptors are often amenable to pharmacologic blockade. Our scRNA-seq results show that tumor-infiltrating T-lymphocytes increase in the setting of HMGA1 deficiency (Figure 7), suggesting that HMGA1 in colon tumor cells may foster an immunologically “cold” tumor microenvironment to facilitate tumor progression. Alternatively, the changes in T cell number could reflect altered transcriptomes from HMGA1 deficiency and associated changes in cell behavior, including proliferation and motility. Given the immune pathways identified from colon epithelial cells, studies focusing on HMGA1 inflammatory networks and immune escape are warranted and could identify new therapeutic strategies.

In summary, we discovered that HMGA1 acts as a molecular key that “opens” chromatin to activate transcriptional networks that maintain a stem and Paneth-like cell state early in colon tumorigenesis. Within crypt epithelial cells, HMGA1 enhances chromatin accessibility to activate the Ascl2 master regulator gene, additional Wnt genes, and inflammatory networks in murine models with Apc inactivation. Further, in human colon cancer, HMGA1 and ASCL2 are coexpressed and upregulated along with downstream Wnt pathway genes. Together, our results establish HMGA1 as an epigenetic gatekeeper of ASCL2 and Wnt signals, inflammation, and a stem-like state in colon cells with APC inactivation, highlighting HMGA1 as a promising potential therapeutic target in colon cancer.

Sex as a biologic variable. All studies were carried out on male and female mouse populations and similar findings were observed for both sexes.

Detailed methods, statistical analyses, and reagents are provided in the supplemental material section, including culture medium, primers, antibodies, and in silico approaches (Supplemental Table 5). Sequencing data were deposited into the Gene Expression Omnibus (GSE) with accession numbers GSE279070 (scRNA-seq) and GSE278871 (ATAC-seq).

Animal models. CDX2P-CreER^T2Apc^fl/fl (73, 74) or Apc^Min/– (Min mice) (75–79) mice were previously described. The CDX2P-CreER^T2Apc^fl/fl were generated and provided in house at the University of Michigan (73, 74). The Apc^Min/– (Min mice) were originally obtained from Bert Vogelstein at Johns Hopkins University who developed this model (104). Both were crossed to mice with global deficiency of one or both Hmga1 alleles (all on C57Bl6 backgrounds) (28, 29). Tissue-specific Hmga1-deficient models were generated by crossing to mice with floxed Hmga1 alleles. Additional details are provided in the supplement (Supplemental Data Set 1).

Statistics. To compare continuous variables across 2 groups, statistical significance was determined using a 2-tailed student’s t test when normally distributed (ascertained by Ryan-Joyner and D’Agostino-Pearson tests). If not normal, the Mann-Whitney test was used. To compare more than 2 groups, we used a 1-way ANOVA with Dunnett’s or Turkey’s multiple comparisons (Prism 10, GraphPad Software) after which 2 groups were compared via 2-tailed student’s t test if normally distributed or Mann-Whitney if not. For categorical data, association with condition was evaluated by Fisher’s exact test. We compared survival analyses under the assumption of Cox proportional hazards using the log-rank test. P < 0.05 was considered significant. All code for the scRNA-seq analysis was performed using Seurat at the indicated resolutions; code will be made available from the corresponding author upon request.

Study approvals. All mouse studies were approved by the Johns Hopkins University Institutional Animal Care and Use Committee (IACUC).

Data availability. As above, metadata are available in the NCBI GEO database (access numbers: scRNA-seq: GSE279070 and ATAC-seq: GSE278871); the remaining data are provided in the Supporting Data Values file.