17-AAG: A Promising Cancer Drug’s Rise and Fall

17-AAG (tanespimycin) was one of the first drugs designed to shut down a cancer cell’s protein-maintenance machinery, and for a stretch in the mid-2000s it looked like it might change how several hard-to-treat cancers were managed. In a phase II trial of HER2-positive breast cancer patients who had stopped responding to trastuzumab, the drug combined with trastuzumab produced a median overall survival of 17 months and a clinical benefit rate of 59%. Then, in July 2008, its developer pulled the plug. The story of 17-AAG is a case study in how a genuinely clever idea can collide with the messy realities of drug formulation, liver toxicity, built-in resistance pathways, and corporate decision-making.

A Drug Born From Soil Bacteria

17-AAG is a semi-synthetic derivative of geldanamycin, a natural compound first isolated from the soil bacterium Streptomyces hygroscopicus. Geldanamycin turned out to be a potent inhibitor of the chaperone protein Hsp90, and early lab work showed it could stop cancer cells from growing. But geldanamycin itself was too toxic for human use, causing severe liver damage in animal models and behaving unpredictably in the bloodstream.¹ Chemists modified the molecule by swapping one chemical group for an allylamino side chain, creating 17-allylamino-17-demethoxygeldanamycin, or 17-AAG. The modification preserved the cancer-fighting activity while taming some of the parent compound’s worst qualities. It was enough to get the drug into human trials, which geldanamycin never managed.

How It Attacks Cancer Cells

Hsp90 is a chaperone protein, meaning its job is to fold, stabilize, and protect other proteins inside the cell. Cancer cells lean on Hsp90 more heavily than normal cells because many of the mutated or overexpressed proteins driving tumor growth are unstable and need constant chaperoning to stay functional. 17-AAG binds to Hsp90 and blocks this chaperoning activity, which sends those unstable client proteins to the cell’s garbage disposal system for degradation.²

The list of Hsp90 client proteins reads like a who’s who of cancer biology. It includes Akt (a survival signal), Raf-1 (a growth signal), HER2 (an amplified receptor in certain breast cancers), mutant p53, and CDK4 (a cell-cycle driver). When 17-AAG knocks out Hsp90, these proteins lose their chaperone support and get degraded, hitting the cancer cell from multiple angles at once.³ This multi-target approach was one of the main reasons researchers found 17-AAG so exciting. Rather than blocking one growth signal that cancer might reroute around, the drug threatened to collapse the entire support network that kept oncogenic proteins alive.

The Enzyme That Made or Broke the Drug

One of the more surprising findings about 17-AAG was that it worked better in some cancer cells than others, and not because of anything related to Hsp90 itself. The difference came down to an enzyme called NQO1 (also known as DT-diaphorase). NQO1 converts 17-AAG into a reduced form called 17-AAGH2, a hydroquinone that turned out to be a more potent Hsp90 inhibitor than the parent drug.⁴ In cancer cells engineered to express high levels of NQO1, sensitivity to 17-AAG jumped by roughly 12-fold compared to cells lacking the enzyme. The mechanism was confirmed by blocking NQO1 with a specific inhibitor called ES936, which wiped out the extra sensitivity entirely.

Earlier work had already hinted at this relationship. In two independent panels of human tumor cell lines, higher NQO1 expression consistently tracked with greater growth inhibition by 17-AAG. When the active NQO1 gene was put into a colon cancer cell line that naturally lacked it, sensitivity to 17-AAG jumped 32-fold.⁵ The drug still worked through the same Hsp90 inhibition mechanism in cells with low NQO1, just much less efficiently. This NQO1 dependency would later become both a potential biomarker opportunity and a source of resistance problems.

First Steps Into Patients

Several phase I trials in the early-to-mid 2000s established that 17-AAG could reach therapeutic concentrations in human blood without killing the patient. One trial showed that plasma levels exceeded those needed for activity in preclinical models, and biopsies confirmed that the drug was doing what it was supposed to: Akt and Raf-1 levels dropped in tumor tissue.⁶ A separate phase I study pushed dosing up to 450 mg/m² per week and showed that drug exposure increased in a roughly linear fashion across the dose range. Tumor biopsies at the higher doses showed Raf-1 inhibition in four of six patients, CDK4 depletion in eight of nine, and HSP70 induction in eight of nine, clear pharmacological proof that 17-AAG was hitting its target.⁷

But even in these early safety-focused trials, trouble signs emerged. Drug exposure varied substantially from patient to patient, with coefficient of variation values ranging from about 17% to as high as 55% at certain dose levels.⁸ That kind of unpredictability makes dosing a gamble: some patients get too little drug, others get too much. And the dose-limiting toxicities were concerning, as the next section makes clear.

The Toxicity Problem

Liver toxicity was 17-AAG’s constant companion. Across multiple phase I studies, reversible elevations in liver enzymes cropped up in roughly a third of patients.⁹ At higher doses, grade 3 liver damage (meaning serious enough to require medical intervention) appeared. One trial reported grade 3 liver enzyme elevations at a dose of just 56 mg/m², forcing expansion of the dosing cohort.¹⁰ A twice-weekly dosing schedule trial hit dose-limiting toxicity that included grade 3 dehydration and grade 4 liver enzyme spikes.¹¹

Beyond the liver, common side effects included fatigue, nausea, vomiting, diarrhea, and loss of appetite. These are familiar complaints with many cancer drugs, but they stacked on top of the hepatotoxicity concerns to narrow the therapeutic window considerably. And then there was the formulation headache: 17-AAG is poorly soluble in water, so it had to be dissolved in Cremophor EL or ethanol for intravenous delivery.¹² Cremophor EL is the same solvent that caused hypersensitivity reactions with early paclitaxel formulations, adding yet another layer of risk. The drug itself had a limited shelf life in solution, making logistics difficult for clinical sites.

Where It Actually Worked

Despite these problems, 17-AAG produced genuinely encouraging results in two cancers. The strongest signal came in HER2-positive breast cancer. In a phase II trial, patients whose tumors had progressed on trastuzumab received 17-AAG (tanespimycin) plus trastuzumab. The overall response rate was 22%, with a clinical benefit rate of 59%. Median progression-free survival reached six months, and median overall survival was 17 months.¹³ For patients who had already failed on trastuzumab, those numbers were meaningful. HER2 is an Hsp90 client protein, so the rationale was straightforward: knock out the chaperone that keeps HER2 stable, and the receptor degrades even in cells that have found ways to resist trastuzumab alone.

In multiple myeloma, a phase I trial of tanespimycin combined with bortezomib showed responses in patients who had already been heavily pretreated. Among patients who had never received bortezomib, about 71% responded. Even among patients whose disease was refractory to bortezomib, about a third responded.¹⁴ A follow-up phase II study confirmed antitumor activity, though it was closed early for what the investigators described as “resource-based reasons” rather than scientific ones.¹⁵ That phrasing hints at what was really happening behind the scenes.

Why Development Was Halted

In July 2008, Bristol-Myers Squibb, which had acquired the rights to tanespimycin through its purchase of Kosan Biosciences, stopped all development of the drug. The decision came after the promising breast cancer data but before pivotal phase III trials could be launched. The official explanation was never fully detailed, but analysis of the situation points to a convergence of factors. The drug’s formulation difficulties, its hepatotoxicity, its narrow therapeutic window, and looming patent expiry concerns all played a role.¹⁶

From a purely scientific standpoint, 17-AAG was arguably still a viable candidate. The breast cancer results were among the best data generated by any Hsp90 inhibitor, before or since. But drug development decisions are not made purely on science. If the patent clock was running out and a phase III trial would take years to complete, the business case evaporates regardless of how well the drug performs. The formulation issues also meant that manufacturing at commercial scale would have been expensive and logistically challenging. Taken together, the economics simply did not work, even though the biology was sound.

A Built-In Escape Hatch for Cancer Cells

Alongside the practical and commercial problems, 17-AAG faced a fundamental biological obstacle. When the drug blocks Hsp90, cancer cells respond by activating a transcription factor called HSF1 (heat shock factor 1), which cranks up production of other protective proteins, especially Hsp70. Hsp70 is itself a pro-survival chaperone, and elevated levels of it can partially compensate for the loss of Hsp90 function.¹⁷ In other words, 17-AAG triggers the very defense mechanism that cancer cells need to survive the attack.

The scale of this response is dramatic. In one experimental system, 17-AAG treatment produced a 23-fold increase in heat shock reporter expression within 60 hours, with corresponding increases in Hsp70 protein levels detectable by imaging.¹⁸ Work in glioblastoma cells confirmed the same pattern: 17-AAG successfully degraded the target (the EGFR client protein lost stability as expected), but simultaneously drove a significant increase in Hsp70 levels that undermined the drug’s effectiveness.¹⁹ Silencing HSF1 or blocking it pharmacologically lowered the threshold at which 17-AAG triggered cancer cell death, suggesting that the heat shock response was genuinely propping up cells that would otherwise have died. This feedback loop is not unique to 17-AAG; it plagues the entire class of N-terminal Hsp90 inhibitors and remains one of the central puzzles in the field.

Acquired Resistance Through NQO1 Loss

The NQO1 connection created a second, distinct resistance pathway. Recall that cells with high NQO1 levels convert 17-AAG into a more potent form. Researchers found that when glioblastoma cells were continuously exposed to 17-AAG, resistant clones emerged that had dramatically reduced NQO1 expression. In some lines this happened through reduced gene transcription, and in others through selection of cells carrying an inactive NQO1 gene variant (the NQO1*2 polymorphism). The same pattern showed up in melanoma cells that had developed acquired resistance.²⁰ When the NQO1 inhibitor ES936 was applied, the sensitivity difference between parental and resistant cell lines vanished, confirming that NQO1 loss was the key driver.

This finding had a troubling clinical implication. If tumors could escape 17-AAG simply by downregulating one metabolic enzyme, then even patients who initially responded might develop resistance relatively quickly. The NQO1*2 polymorphism is common in certain populations, meaning a substantial fraction of patients might have been poor candidates for the drug from the start. But there was a flip side too: if NQO1 status could be measured before treatment, it could serve as a biomarker for patient selection, concentrating the drug’s use where it was most likely to work.

NQO1 as a Biomarker for Smarter Treatment

The idea of using NQO1 levels to pick the right patients gained traction in research labs even as clinical development stalled. A study in esophageal squamous cell carcinoma found that NQO1 status could identify tumors likely to respond well to 17-AAG at lower doses, potentially reducing toxicity while preserving efficacy.²¹ However, the picture was not perfectly clean. Comparative work in pancreatic and colorectal cancer cell lines showed that while 17-AAG sensitivity correlated with NQO1 protein levels and enzyme activity in general, some NQO1-lacking cells were still sensitive to the drug.²² That complicates any simple rule of “NQO1-high, treat; NQO1-low, don’t.” The enzyme clearly matters, but it is not the whole story.

Had 17-AAG’s development continued, a biomarker-guided trial design might have improved response rates and reduced the number of patients exposed to futile toxicity. This is one of the sharper what-ifs in the drug’s history. The era when 17-AAG was being tested clinically slightly predated the routine use of companion diagnostics in oncology trials, and one wonders whether a more modern trial design might have changed the outcome.

Radiosensitization and Combination Possibilities

Beyond its use as a standalone Hsp90 inhibitor, 17-AAG showed potential as a radiosensitizer, meaning it made cancer cells more vulnerable to radiation. In human lung cancer cells, pretreatment with 17-AAG significantly enhanced the killing effect of both conventional X-ray radiation and heavy-ion irradiation. The mechanism appeared to involve suppression of the homologous recombination repair pathway, which cells use to fix double-strand DNA breaks after radiation damage.²³ By blocking the DNA repair machinery, 17-AAG made radiation-induced breaks more likely to be lethal.

This radiosensitization work remained preclinical and was never tested in patients, but it illustrated a broader principle: Hsp90 inhibitors might be most useful not as monotherapy but in combination with other treatments that create specific types of cellular stress. The breast cancer trial that combined 17-AAG with trastuzumab was working along similar lines, attacking the tumor from two directions at once. More recent nanoparticle-based delivery research has explored whether packaging 17-AAG differently could enhance its synergy with trastuzumab even further, with one study reporting an 85% reduction in cell viability with the combination, compared to roughly 50% or 65% for either agent alone.²⁴

What Came After 17-AAG

The failure of 17-AAG did not kill interest in Hsp90 inhibition. If anything, it sharpened the field’s focus on what needed to be fixed. A wave of second-generation inhibitors entered clinical testing, split into two categories. One group, including 17-DMAG and IPI-504 (retaspimycin), kept the geldanamycin backbone but tried to improve on 17-AAG’s pharmacology. The other consisted of entirely synthetic molecules unrelated to geldanamycin, such as BIIB021, SNX-5422, STA-9090 (ganetespib), and NVP-AUY922 (luminespib). The synthetic compounds were particularly appealing because some could be given orally and showed less liver toxicity, which had been the ceiling that kept geldanamycin-based agents from reaching higher, more effective doses.²⁵

Yet none of these successors has reached regulatory approval either. The heat shock response problem and the question of how to select patients remain largely unsolved. Newer research has shifted toward more fundamental redesigns of the Hsp90 inhibition strategy. One approach targets the C-terminal end of Hsp90 rather than the N-terminal ATP-binding pocket where 17-AAG and its relatives bind, which appears to avoid triggering the protective heat shock response altogether.²⁶ Another strategy pursues inhibitors that are selective for specific Hsp90 isoforms rather than blocking all forms of the protein indiscriminately, in hopes of reducing off-target effects while maintaining potency against cancer-relevant pathways.²⁷

Whether these newer approaches will finally deliver an approved Hsp90 inhibitor remains an open question. The field has been humbled enough times that predictions feel risky. But the underlying logic of attacking the chaperone network that cancer cells depend on is still compelling, and 17-AAG, for all its flaws, was the drug that proved the concept could work in real patients. The breast cancer data in particular remains a kind of North Star for the field: evidence that Hsp90 inhibition can produce meaningful clinical benefit when paired with the right combination partner in the right patient population.