Hsp90-stabilized MIF supports tumor progression via macrophage recruitment and angiogenesis in colorectal cancer

Macrophage migration inhibitory factor (MIF) is an upstream regulator of innate immunity, but its expression is increased in some cancers via stabilization with HSP90-associated chaperones. Here, we show that MIF stabilization is tumor-specific in an acute colitis-associated colorectal cancer (CRC) mouse model, leading to tumor-specific functions and selective therapeutic vulnerabilities. Therefore, we demonstrate that a Mif deletion reduced CRC tumor growth. Further, we define a dual role for MIF in CRC tumor progression. Mif deletion protects mice from inflammation-associated tumor initiation, confirming the action of MIF on host inflammatory pathways; however, macrophage recruitment, neoangiogenesis, and proliferative responses are reduced in Mif-deficient tumors once the tumors are established. Thus, during neoplastic transformation, the function of MIF switches from a proinflammatory cytokine to an angiogenesis promoting factor within our experimental model. Mechanistically, Mif-containing tumor cells regulate angiogenic gene expression via a MIF/CD74/MAPK axis in vitro. Clinical correlation studies of CRC patients show the shortest overall survival for patients with high MIF levels in combination with CD74 expression. Pharmacological inhibition of HSP90 to reduce MIF levels decreased tumor growth in vivo, and selectively reduced the growth of organoids derived from murine and human tumors without affecting organoids derived from healthy epithelial cells. Therefore, novel, clinically relevant Hsp90 inhibitors provide therapeutic selectivity by interfering with tumorigenic MIF in tumor epithelial cells but not in normal cells. Furthermore, Mif-depleted colonic tumor organoids showed growth defects compared to wild-type organoids and were less susceptible toward HSP90 inhibitor treatment. Our data support that tumor-specific stabilization of MIF promotes CRC progression and allows MIF to become a potential and selective therapeutic target in CRC.


Introduction
Macrophage migration inhibitory factor (MIF), which was originally discovered as a secreted proinflammatory cytokine with a central role in immune and inflammatory responses, has also been identified as a tumor promoter 1,2 .
Colorectal cancer (CRC) patients also present elevated MIF levels, which are associated with a worse prognosis 12,15,[19][20][21][22] . Among cancers, CRC has the third highest incidence 23 . Previous in vitro studies in human CRC cells showed that MIF increases proliferation, angiogenesis, and migration 12,24,25 . Functionally, MIF can bind to its main receptor CD74 to activate p38, MAPKs, or PI3K/ AKT, which induces the expression of angiogenic factors 4,12,24,[26][27][28] . Furthermore, MIF regulates therapeutic resistance via regulation of STAT3, MAPKs, AMPK, or hypoxia-dependent mechanisms [28][29][30][31] . Other studies using CT26 allograft models support that MIF promotes CRC progression 12,24 . In vivo, it has been shown that MIF stimulates the early stages of small intestinal adenomas in Apc min mice 27 . Although all these studies showed a positive correlation between aberrant MIF function and CRC growth, an in vivo model of causative and severe CRC that mimics the human CRC was not available.
In our study, we investigated whether MIF promotes tumor growth in an autochthonous colorectal azoxymethane (AOM)/dextran sodium sulfate (DSS) mouse model and whether MIF can serve as a potential drug target. Because of the tumor-specific Hsp90-mediated stabilization of MIF, this protein could be selectively targeted in CRC. Our data suggest that MIF increases CRC growth and supports tumor-specific macrophage recruitment, tumor cell proliferation, and neoangiogenesis without affecting overall inflammation in established tumors.
Strikingly, a recent study in a mouse model of chronic colitis-dependent CRC reported a tumor-protective role for MIF 32 . This phenomenon was not observed in neither the Apc min mouse model 27 nor several other in vivo cancer studies, including myc-induced lymphoma, chronic lymphocytic leukemia, breast, prostate, bladder, and skin cancer 3,4,11,[33][34][35][36][37][38] . An important difference between the previous work and our study is that we used a mouse model of acute colitis-associated CRC, which is more similar to human sporadic CRC 39 . Importantly, in our sporadic CRC model, MIF as a tumor-promoting factor is selectively targetable in tumor cells by inhibiting Hsp90, supporting a strong rationale for MIF as a potential therapeutic target in sporadic CRC.

MIF supports tumor growth in a mouse model of CRC
Given the importance of MIF in cancer and to determine whether MIF supports CRC tumorigenesis, we used the severe CRC AOM/DSS mouse model, which includes one phase of acute colitis (Fig. 1A). After a recovery phase, mice exclusively develop tumors within 12 weeks in the large intestine 40   mice showed a reduction in the tumor burden (Fig. 1B). Quantification of colonic tumors by a scoring system 41,42 revealed a reduction in tumor multiplicity in Mif −/− mice (Fig. 1C). Moreover, at 12 weeks post-AOM, during which the CRC tumors are well established, MIF deficiency decreased tumor burden and numbers (Fig. 1D, E). In summary, MIF supports tumor growth in an acute colitisassociated CRC mouse model.

MIF levels are elevated in CRC cells
During tumorigenesis, MIF protein levels are increased 12,27 . Our data confirm elevated MIF levels in cancer cells from CRC patients (Figs. 2A-C, S1A). Compared to the moderate increase in MIF mRNA levels (Figs. 2A, S1A), MIF protein levels were strongly increased in tumors from patients (Fig. 2B, C). Similar to the patient tumors, established AOM/DSS-induced tumors confirmed tumor-specific elevation of MIF expression (Fig. 2D). Intriguingly, epithelial cancer cells express high levels of MIF compared to those in the normal surrounding epithelium (Fig. 2D), indicating that the major source of MIF is tumor epithelial cells. Measurement of MIF expression in murine tumor lysates indicated increased MIF expression in tumors compared to normal colonic tissue (Figs. 2E, F and S1B, C). Taken together, these results confirm an enhanced tumorspecific increase in MIF occurrence within the epithelial tumor compartment.
A Mif deletion protects mice from inflammation-associated cancer initiation As a proinflammatory cytokine, MIF regulates immune responses and is suggested to be a link between inflammation and cancer 1,2 . Therefore, we hypothesized that the  loss of Mif expression protects mice during the colitisassociated phase of tumor initiation. Indeed, during the recovery phase, colonic tissue damage and epithelial cell loss, as reflected by the inflammatory score, were increased in Mif +/+ mice compared to Mif −/− mice and were accompanied by increased immune cell infiltration (Fig. 3A, B). To further examine the inflammatory response, we histologically analyzed the immune cell composition within the tumor microenvironment. Infiltrates from the colonic tissue of Mif +/+ mice had higher percentages of CD3-positive (T-lymphocyte marker), MPO-positive (neutrophil/granulocyte marker), and FoxP3-positive (regulatory T-cell marker) cells than did colonic tissue from Mif −/− mice (Figs. 3C, S2A). Immune infiltrates and the inflammatory score showed a positive correlation ( Figure S2B). Interestingly, CD68-positive (monocyte/macrophage marker) cell infiltration was unchanged between the two mice groups (Fig. 3C, S2A). Similar to the changes in the inflammatory cell composition, the expression of inflammation-associated cytokines was downregulated in Mif −/− tissues during the recovery period, confirming a reduction in inflammation in the absence of Mif (Fig. 3D, S2C). Consistent with the protective effect of Mif deletion during recovery, Mif −/− mice showed a reduced overall inflammatory response under DSS administration ( Figures S2D-G). Furthermore, since MIF inhibits p53 activity 11,43 , we pursued whether MIF interferes with the DNA damage response and apoptosis in response to AOM treatment. Surprisingly, neither the levels of phosphorylated histone H2A.X Mmp9 (a DNA damage marker) nor the expression of p53 target genes (Mdm2, Cdkn1a, Ccnd1, Gadd45a, and Bax) was altered in colonic tissues in Mif −/− mice compared to those in Mif +/+ mice, suggesting that MIF failed to regulate an AOM-induced p53-dependent response in colonic epithelia ( Figures S2H, I).
Overall, a Mif deficiency protected mice during the early phases of inflammation in the AOM/DSS model and demonstrated that during colitis-associated tumor initiation, MIF acts as a proinflammatory cytokine.

MIF supports CRC development via tumor-specific macrophage recruitment and angiogenesis without affecting overall inflammation
To determine whether MIF also acts as an inflammatory cytokine to support established tumors, we analyzed the expression of inflammatory markers at 12 weeks post-AOM. Interestingly, immunohistological staining ( Figure S3A) and their corresponding quantifications ( Fig. 4A) did not show any differences in the extent of infiltrating lymphocytes, regulatory T-cells or neutrophils/granulocytes within established tumors. In line with these findings, an assessment of inflammatory cytokines from tumor lysates failed to show major differences between Mif-expressing and Mif-deficient tumors, although all cytokines were upregulated in tumor samples ('T') compared to normal epithelium samples from untreated animals ('N') ( Figure S3B). However, interestingly, CD68-positive macrophage/ monocyte infiltration was decreased in Mif −/− tumors compared to Mif +/+ tumors (Fig. 4B, upper panel), supporting the function of MIF as a chemokine to mediate macrophage recruitment 3,4,44 . To clarify whether elevated MIF expression in tumor cells mediates tumor-specific macrophage recruitment, we monitored the adjacent epithelium ( Fig. 4B, lower panel). Indeed, macrophages specifically infiltrated tumors, suggesting that MIF regulates the chemotaxis of tumor-associated macrophages to promote CRC tumorigenesis. Tumorassociated macrophages are known to secrete tumorpromoting cytokines during cancer progression to stimulate tumor cell proliferation and angiogenesis 45,46 . Indeed, levels of Vegfa, an angiogenic cytokine known to be secreted by macrophages 46,47 , are reduced in Mif −/− tumors (Figs. 4C, F, G, S3C). Moreover, in established CRC tumors, Mif +/+ mice showed stronger vessel formation, as indicated by CD31-positive staining, compared to Mif −/− mice (Fig. 4D, E). Immunoblots confirmed increased activation of proangiogenic factors such as p38 and ERK in Mif-containing samples (Fig. 4F), an effect described previously 3,10,12,27,48 . MIF also affected tumor cell proliferation in AOM/DSS-induced tumors (Fig. 4H), which might explain the smaller tumors observed in Mif −/− mice (Fig. 1D, E).
Given that MIF also intrinsically regulates apoptosis via p53, e.g., in HER2-positive breast cancer or macrophages 11,51 , we clarified whether the loss of MIF expression also activates p53 target genes in AOM/DSS tumors. In our CRC mouse model, Mif deficiency did not upregulate the expression of p53 target genes involved in apoptosis (e.g., Bax, Bcl2l1, Bcl2, and Mcl1) ( Figure S3E). TUNEL staining confirmed the lack of altered apoptosis in Mif −/− tumors ( Figure S3F). However, the expression of the cell cycle inhibitor p21/Cdkn1a was upregulated in Mif −/− tumors ( Figure S3E), supporting the diminished proliferation upon MIF loss (Fig. 4H).
To assess whether angiogenesis and proliferation are affected during the recovery phase, we evaluated Vegfa, CD31, and Ki67 expression in colonic tissues at 8 days post-DSS (Figures S3G-K) and found that neither vessel formation and Vegfa expression nor proliferation was dependent on the presence of MIF during colonic tissue recovery.
Albeit our data confirmed that MIF supports inflammatory processes during colitis-associated tumor-initiating phases, we identified that in established tumors, MIF contributes to tumor-specific macrophage recruitment, tumor cell proliferation, and vessel formation without affecting overall inflammatory responses. Whether these infiltrated macrophages release proangiogenic cytokines 45,47 or whether MIF regulates angiogenic pathways in tumor cells themselves 52 must be further elucidated.
The CD74-MIF receptor complex facilitates the expression of proangiogenic factors in human CRC cells MIF functions through CD74/CD44 and/or CXCR2/4 receptor complexes in proliferation, angiogenesis, and with its chemokine-like properties in monocyte and leukocyte recruitment 3,8,[53][54][55] . The CD74 receptor is the main MIF receptor 53,56 . Since the expression of Vegfa is downregulated in Mif-deficient tumors (Figs. 4C, S3C), we examined whether tumor cells themselves are able to express angiogenic genes via MIF binding to CD74 to activate MAP kinases to induce VEGF and IL8 expression 12,24,[26][27][28] . First, we used the CD74-expressing ( recombinant MIF (rhMIF) in DLD-1 cells to mimic MIF secretion, also failed to activate ERK or angiogenic gene expression (Fig. 5D, E). Importantly, supplementation of both, MIF by rhMIF and CD74 by plasmid-based ectopic expression, lead to ERK activation and increased VEGFA and CXCL8/IL8 expression confirming that concomitant CD74 and secreted MIF are necessary for expression of angiogenic markers (Fig. 5F, G). To further investigate the MIF-CD74 axis, we performed clinical correlation studies based on MIF and CD74 expression levels of human CRC patients (Fig. 5H, I, J). Interestingly, simultaneous high levels of MIF and CD74 showed a trend for patient shortest survival (53.1 months) compared to stabilized MIF alone (71.4 months) (Fig. 5I). In contrast, CD74 status in patients with low MIF levels did not impact overall survival (Fig. 5J). These findings underline the importance of MIF in cancer and support that MIF acts via CD74 in CRC.

MIF-driven CRC is vulnerable to Hsp90 inhibition
Next, we asked whether constitutive MIF stabilization in CRC cells creates vulnerabilities that can be therapeutically targeted. Since MIF is stabilized by Hsp90 11 , we used the pharmacological Hsp90 inhibitor 17AAG. When Mif +/+ and Mif −/− mice reached a defined tumor burden, they were treated with 17AAG (Fig. 6A). Hsp90 inhibition reduced MIF protein levels in AOM/DSS tumors (Fig. 6B) and showed a trend for decreased tumor burden in Mif +/+ mice (Fig. 6C-E). Differences were not statistically significant but showed a trend in Mif +/+ mice (Fig. 6D, E, left panels). By contrast, Hsp90 inhibition in Mif −/− mice failed to achieve tumor reduction (Fig. 6D, E,  right panels).
To further support MIF as tumor-relevant Hsp90 client in CRC progression, we used genetically deleted MIF tumor organoid cultures. We observed a decreased growth in Mif-depleted organoids (Fig. 6F, G), further confirming, that MIF loss reduces tumor cell proliferation (Fig. 4H). Whether these growth defects arise from intracellular MIF functions and/or an MIF-CD74 axis remains elusive. Moreover, Vegfa expression was reduced in those organoids (Fig. 6G). To further support MIF as a tumor-relevant Hsp90 substrate in CRC, we analyzed these Mif-depleted organoids after treatment with 17AAG (Fig. 6G, H). Indeed, a Mif depletion led to a decreased susceptibility toward 17AAG treatment compared to Mifproficient organoids (Fig. 6H). Furthermore, apoptotic markers such as cleaved caspase-3 and Parp were only upregulated after 17AAG treatment in Mif-proficient organoids, but not in Mif-deficient organoids (Fig. 6I).
These data support a relevant point: Hsp90 inhibition seems to stronger target CRC tumors with elevated MIF, although the HSP90 system stabilizes innumerable oncogenes. These findings support that MIF is a tumorrelevant Hsp90 substrate in CRC.

MIF is a selective therapeutic target of Hsp90 inhibition in CRC-derived organoids
To exploit further therapeutically targeting of stabilized MIF, we administered Hsp90 inhibitors to healthy epithelial/mucosal-derived and tumor-derived murine colonic organoids from the same AOM/DSS-induced mice (i.e., matched pairs). Since organoids derived from mice with a 129S1/SvImJ background failed to grow in vitro in our laboratory ( Figure S5A), we used C57BL/6 mice. Observation of the organoid morphology and the subsequent quantifications showed higher levels of cell death after 17AAG in tumor-derived organoids, compared to the epithelial-derived organoids (Fig. 7A). Immunoblot analysis confirmed strong reduction of Mif levels especially after treatment with 500 nM 17AAG (Fig. 7B). This prompted us to test Ganetespib and Onalespib, two clinically relevant second-generation HSP90 inhibitors that have been extensively tested in clinical trials and have a suitable toxicity profile [58][59][60][61] . Both inhibitors induced cell death to a far lesser extent in normal epithelial-derived organoids than in tumor-derived organoids (Fig. 7C) and showed promising specificity toward tumor organoids. Although Mif protein was degraded by Hsp90 inhibition in normal and tumorderived organoids treated with either inhibitor (Fig. 7D), only tumor-derived organoids were morphologically disrupted upon Hsp90 inhibition (Fig. 7C), indicating that MIF plays a tumorigenic role rather than an essential function in normal epithelial cells. Importantly, and in line with our findings, we confirmed the enhanced Mif levels in tumorderived organoids (Fig. 7B, D). Furthermore, in MIFexpressing patient-derived CRC organoids, Ganetespib markedly increased organoid death compared to that observed in the control organoids (Fig. 7E).
Therapeutic selectivity toward tumor cells plays an important role in therapy implementation. To further test whether Hsp90 inhibitors affect healthy tissues, we used organoids derived from the murine small intestine. Upon implementing the same treatment scheme as that used for colonic tumor-derived organoids, we discovered that 17AAG, Ganetespib, and Onalespib only exerted minor or no effects on the small intestine-derived organoids ( Figure  S5B). Indeed, Ganetespib failed to significantly degrade Mif protein in those organoids ( Figure S5C), while another Hsp90 substrate, Stat3, was degraded. This confirms the selectivity of Hsp90 inhibition toward stabilized MIF in tumors. Even though immuno blot analysis also showed reduction of Mif levels after 17AAG treatment ( Figure S5C), again depletion of unstabilized Mif in small intestinal organoids, did not impact morphology or survival of organoids ( Figure S5B) as observed for normal colonic epithelia-derived organoids (Fig. 7A, B).
Thus, our findings highlight that MIF degradation via Hsp90 inhibition is a promising mechanism in CRC therapy. MIF acts not only as a critical driver in CRC but also as a selective target for Hsp90 inhibition in tumors.

Discussion
We used the immune-competent AOM/DSS mouse model, which mimics CRC progression in humans, to exploit the therapeutic potential of MIF. We demonstrated that MIF is specifically elevated in tumor cells and drives tumor growth in this acute colitis-associated ('sporadic') CRC model. Thus, in established tumors, stabilized MIF preferentially supports tumor-specific macrophage infiltration, vessel formation, and tumor cell proliferation.
Concomitantly, we also showed within this model that MIF regulates overall inflammatory signatures but especially during tumor initiation. Compared with Mif +/+ mice, Mif −/− mice were protected against acute colitisassociated tumor initiation (Fig. 3), confirming the general function of MIF as a proinflammatory cytokine 3,10 . By contrast, established Mif-deficient tumors did not show reductions in overall inflammation (Fig. 4); rather, only tumor-associated macrophages significantly infiltrated Mif +/+ tumors. Thus, MIF seems to lose its overall proinflammatory function once CRC tumors are established. Proliferation, vessel formation and angiogenic cytokine expression were reduced in Mif-deficient tumors, an effect described previously 3,10,27,48 .
Studies showing that human tumor cells themselves are able to activate MAPK-mediated IL8 and VEGF expression by binding of MIF to its main receptor CD74 7,12,27,28,35,53 were confirmed within this study in human CRC cells (Fig. 5)  macrophages to the tumor (which consequently secrete angiogenic factors) (Fig. 4B); and provide an autocrine MIF-CD74 interaction to induce the MAPK-VEGF axis ( Fig. 5B and F), albeit we have not specifically tested it in this study. However, reduced expression of VEGF in Mifdeficient organoids (Fig. 6G), support the idea, that tumor cells themselves contribute to VEGF expression. Nevertheless, MIF is known to act as chemokine on tumorspecific macrophage recruitment and/or macrophage polarization, and macrophages are known to secrete angiogenic factors, further promoting CRC tumorigenesis 44,55 . In sum, tumor cells and tumor-associated macrophages might contribute to angiogenic factor expression but stabilized MIF in epithelial tumor cells provides the prerequisite for both scenarios.
To further test whether tumor epithelial cells with elevated MIF expression provide dual control over tumor growth, additional experimental models with inducible, tissue-specific Mif deletions are required. In principle, reduced chemotaxis of Mif −/− macrophages 62 or Mif-depleted fibroblasts within the tumor stroma 63 might also contribute to tumor reduction in The co-expression of MIF and CD74 seems to be important in tumorigenesis (Fig. 5), and either MIF or CD74 alone might not be strong tumor biomarkers. Our CRC patient study (Fig. 5H-J) as well as patient studies of lung cancer and colon carcinomatosis indicate that MIF/ CD74 co-expression corresponds to an even worse prognosis 28,64 . Moreover, a recent mouse study revealed a strong upregulation of CD74 during colonic inflammation, promoting mucosal healing, and epithelial tissue recovery by enhanced cell proliferation 65 . While this study confirms the importance of a MIF/CD74 co-existence in proliferation, it also clarified that a CD74 deficiency alone massively increases overall inflammation with a reduced recovery rate 65 . In contrast, MIF deletion or ablation alone protects against inflammation, demonstrated in experimental models of gastrointestinal inflammation [66][67][68] . Why a CD74 single deletion intensifies inflammation remains speculative 65,69 . One explanation might be altered macrophage recruitment. MIF −/− macrophages exhibited reduced overall chemotaxis compared to wild-type macrophages, whereas CD74 −/− macrophages showed random chemokinesis 62 , leading to an accelerated inflammatory response. Moreover, receptors often co-regulate each other, and after CD74 loss, MIF might increase its affinity to CXCR2 and/or CXCR4 receptors driving inflammation instead of proliferation and angiogenesis 54,70 . Dual roles for ligand-receptor complexes are becoming increasingly evident in the context of active inflammation and mucosal recovery 69 . In sum, MIF/CD74 co-expression might be the major predictor for tumor growth in CRC.
MIF is mainly stabilized in tumors but not stromal or inflammatory cells (Fig. 2). MIF stabilization occurs via binding to Hsp90 16 , which offers therapeutic approaches to target cancer cells via Hsp90 inhibition. We showed for the first time that clinically relevant Hsp90 inhibitors decreased MIF levels in CRC and subsequently reduced tumor growth (Figs. 6 and 7). Given the plethora of known Hsp90stabilized oncogenes 18 , it is interesting to see that Hsp90mediated stabilization of MIF is critical for the survival of Mif-proficient murine colonic tumor-derived organoids (Fig.  6H). MIF reduced tumor-derived organoids show a reduced antitumor response to Hsp90 interference compared to that in Mif-proficient organoids, indicating that MIF is an important Hsp90-stabilized protein in CRC. Moreover, Hsp90 interference provides therapeutic selectivity toward tumor cells (Fig. 7). Since Hsp90 inhibitors exhibit fundamental differences in action 71 , we focused on newly developed inhibitors such as Ganetespib and Onalespib.
In summary, since MIF stabilization is a crucial event, specifically in tumor cells, Hsp90 inhibition provides a potential approach to target MIF function in CRC. These findings support the tumor-promoting role of MIF in CRC and highlight the necessity to better understand the underlying MIF-induced tumorigenic mechanisms in CRC.

Patient samples
Clinical samples (protein samples, RNA samples, PFAfixed paraffin-embedded sections, patient tissue for cultivation) were provided by the Department of General, Visceral and Pediatric Surgery of the University Medical Center Göttingen (UMG, Germany).

Mouse models and genotyping
Mouse experiments were approved by state (Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit, LAVES, Germany) and institutional (Göttingen University Medical Center) committees, which ensured that all experiments conformed to the relevant regulatory standards. Constitutive Mif knockout in the 129S1/SvImJ background has been described in detail in ref. 72 . DNA isolation and genotyping were performed using DirectPCR lysis Reagent and One-Taq®Quick-Load® MasterMix. Genotyping primers are specified in Supplemental Table 1. Mif flox/flox mice in C57BL/6NCrl background were described in detail 72,73 ; and were used for the development of murine organoids. In brief, to remove floxed MIF alleles from colonic epithelial tissue, we crossed Mif fl/fl mice with villinCreERT2harboring mice to generate Mif fl/fl ;villincreERT2 transgenic mice. Mice were housed and handled under pathogen-free barrier conditions.

Murine CRC induction, colonoscopy, and treatment
For experiments, randomly selected 10-week-old female and male mice (>20 g) were used. For the induction of colorectal tumors, mice were treated with a single intraperitoneal injection of 10 mg/kg azoxymethane (AOM, Sigma-Aldrich) in 0.9% sodium chloride. After 1 week of rest, 1.5% (129S1/SvImJ background) or 2% (C57BL/6 background) DSS (MP-Biomedicals) was added to the drinking water for 6 consecutive days. Throughout the AOM/DSS phase, the body weights of the mice were continuously measured.
Five weeks after AOM induction, tumor development was monitored weekly by conducting a colonoscopy (Karl Storz GmbH) on anesthetized mice (1.5-2% isoflurane inhalation). Tumor sizes were determined according to the method described by Becker and Neurath 41 based on the colonic luminal perimeter as follows: S1 = just detectable, S2 = 1/8 of the lumen, S3 = 1/4 of the lumen, S4 = 1/2 of the lumen, and S5 > 1/2 of the lumen. For analysis of established tumors, we chose an endpoint study design, terminating the experiment at 12 weeks after AOM treatment to avoid losing mice to extraneous reasons such as intestinal obstruction and anal prolapse.
For pharmacological Hsp90 inhibitor analysis, tumors were visualized and validated by colonoscopy. Reaching a defined tumor burden, at least one S2/3 tumor and at least three S1/2 tumors when scored by colonoscopy, mice were treated with 17-allylamino, 17-demethoxygeldanamycin (17AAG, provided by the National Cancer Institute, NCI). Therefore, 17AAG was pre-dissolved in DMSO and further diluted in 10% DMSO/18% Kolliphor® RH40/3.6% Dextrose (Sigma-Aldrich) in H 2 O. 60 mg/kg of 17AAG or vehicle were given by intraperitoneal injection for 5 days per week for 3 consecutive weeks. During 17AAG treatment, tumors were weekly visualized and monitored by colonoscopy.
At endpoints, mice were euthanized and the entire large intestine was harvested, longitudinally opened, and displayed. Tumor numbers were counted and sizes were measured with an electronic caliper. For subsequent analysis, single tumor biopsies were taken. Each large intestine was 'swiss rolled', fixed in 3.7% paraformaldehyde/PBS, processed for embedding and bisected. Both halves were placed into a mold for paraffin embedding.
For Mif depletion in vivo, AOM/DSS-treated Mif fl/fl mice were used for Tamoxifen (TAM, Sigma-Aldrich) or respective vehicle control (oil). Reaching a defined tumor burden, at least one S2/3 tumor and at least three S1/2 tumors when scored by colonoscopy, mice were treated for 5 consecutive days, followed by 2 days of rest and another 3 days TAM/oil treatment. Twelve days after TAM-end, mice were dissected, and organoids were prepared (see section above).
All animal experiments were carried out in full agreement with the guidelines outlined above.

Human cell cultures, treatment, and transfection
The human CRC cell line DLD-1 was cultured in RPMI 1640 medium, whereas HCT116 CRC cells were cultured in McCoy's 5A modified medium. Media were supplemented with 10% FBS, Penicillin-Streptomycin, and Lglutamine (RPMI 1640). Cell lines were cultured at 37°C and 5% CO 2 in a humidified atmosphere and were regularly tested for Mycoplasma contamination.
Knockdown of MIF or CD74 was achieved by siRNA transfection using Lipofectamine™3000 reagent according to the manufacturer's instructions. All siRNAs were purchased from Ambion and used according their guidelines; the sequences are listed in Supplemental Table 2. CD74 overexpression in DLD-1 cells was performed using Lipo-fectamine™3000 transfection reagent. In brief, 24 h after cell seeding, DLD-1 cells were cotransfected with 0.5 μg of GFP-containing plasmid and 1.5 μg of either pcDNA3.1-CD74 expression plasmid 56 or the corresponding pcDNA3.1/V5-His-TOPO control plasmid. Forty-eight hours post-transfection, cells were treated with recombinant human MIF as indicated.

Preparation and cultivation of colonic and small intestinal organoids
Tumor-harboring large intestines of C57BL/6 mice were harvested. Three to four tumors per mouse and in parallel, parts of the normal epithelium were biopsied from the same mouse allowing generation of matched organoid pairs. Normal epithelial tissue was cut, washed, and incubated in 4 mM EDTA/PBS for 30 min on ice. Pieces were thoroughly, mechanically dissociated in PBS. Tumor samples were digested with 2 mg/mL Collagenase type-I solution at 37°C for 30 min. Normal crypts and tumor fragments were filtered using cell strainers. Cell pellets were washed, resuspended in cold Matrigel, and droplet-plated allowing Matrigel polymerization at 37°C for 30 min. Organoids-containing domes were covered with organoid culture medium, cultivated at humidified 37°C with 5% CO 2 . Medium was exchanged every 2 to 3 days. Organoids splitting was performed once a week. For passaging, organoids were manually disrupted in icecold PBS, and cultured as described above.

Organoid treatments and morphological quantification
Experiment with matched pairs of normal epitheliaderived and tumor-derived colonic organoids, murine small intestinal organoids, and human organoids were performed between passage 2-7. HSP90 inhibitors 17AAG (National Cancer Institute, NCI), Onalespib and Ganetespib (Synta Pharmaceuticals) were dissolved in DMSO and used as indicated. For quantification of treatment response, light microscopy images of ≥5 Matrigel domes were taken from each condition. The amount of images was dependent on size and culture density as indicated. Based on morphology, dead organoids were defined as organoids with a partial or complete loss of outer epithelial barrier leading to disruption into clumps of dead cells or separation of dead cells 74 . The percentage of dead organoids was calculated relative to the total amount of organoids per image. For dead organoid quantification and measurement of organoid diameter ImageJ was used. For analysis of organoid lysates, Matrigel domes were disrupted using ice-cold PBS. Suspension was centrifuged and organoid-containing pellets were further washed and incubated with Cell Recovery solution (Corning) for complete removal of Matrigel. Organoids were resuspended in standard RIPA buffer for protein lysates and in TRIZOL for RNA isolation.
Hämalaun & Eosin G stained sections were used to determine the inflammatory score. The inflammatory score is based on morphological changes (grade of damage) of the tissue due to immune cell infiltration and epithelial layer disruption. Grade 0 = factor 0, no infiltration of immune cells, normal distribution of epithelia and amount of goblet cells; grade 1 = factor 1, minor infiltration of immune cells, epithelia is still intact and minor changes in goblet cell number; grade 2 = factor 2, moderate infiltration of inflammatory cells, epithelia is partly damaged and reduced number of goblet cells; grade 3 = factor 3, massive infiltration of immune cells, complete disruption/loss of epithelia and loss of goblet cells. For calculation, amount of tissue in percentage with respective grade of tissue damage was multiplied with the corresponding factor (factor 0-3). The obtained percentages were summed, resulting in a value for the inflammatory score (minimum 0-maximum 300) for each mouse. To ensure unbiased quantification, the inflammatory score was individually determined by one scientist and one pathologist.
TUNEL staining was used to detect DNA-strand breaks occurring during apoptotic cell death in established tumors. TUNEL reaction mix (Sigma-Aldrich) consists of TUNEL enzyme solution and TUNEL label mix. The assay was performed according to manufactures guidelines. DAPI served as counterstaining, slides were mounted with Fluorescent Mounting Medium (DakoCytomation).

Quantitative immunohistochemistry on colon cancer patient samples
Section of a tissue micro array (TMA) for primary colonic tumors was kindly provided by the Department of Pathology of the University Medical Center Göttingen (UMG, Germany). According to described standard protocols for immunohistochemistry (see above), TMAs were stained for MIF (Sigma-Aldrich, HPA003868) and CD74 (Sigma-Aldrich, HPA010592). For CD74 staining tumors with more than 10% strongly positive stained cells or more than 40% overall stained cells with lower intensity were graded as high (CD74 high ). For MIF staining, biopsis with high intensity in more than 70% of cells were graded as MIF high . Biopsis with moderate or low intensity were graded as MIF low .

Quantification and statistical analysis
Statistics of each experiment such as number of animals, number of tumors, biological replicates, technical replicates, precision measures (mean and ±SD), and the statistical tests used for significance are provided in the figures and figure legends.
Densitometric measurements for quantification of immunoblot bands were done with the gel analysis software Image Lab™ (BioRad) and normalized to loading controls.
Pearson correlation factor R was used for analysis of immunohistochemical correlation studies on CRC tissue. GraphPad Prism was used for analysis of Kaplan-Meier survival curves with the Log-rank (Mantel-Cox) test.

Ethics approval
Patient samples (protein samples, RNA samples, PFA-fixed paraffin-embedded sections, patient tissue for cultivation) were provided by the Department of General, Visceral and Pediatric Surgery of the University Medical Center Göttingen (UMG, Germany) with approval from the ethics committee of UMG for the collection of CRC samples (approval numbers 9/8/08 and 25/3/17).