ELRIG 2025: Basal-like tumour states explain why AKT inhibitors fail to hold triple negative breast cancer in check

Research presented by Dr Erik Sahai has shown that early, reversible diversity in tumour cell state can blunt the effect of protein kinase B inhibition in triple negative breast cancer. Patient-derived organoids, high-content three-dimensional microscopy and single-cell sequencing have revealed a basal-like programme that weakens the link between phosphoinositide 3-kinase pathway blockade and growth arrest

Dr Erik Sahai described his research programme that aims to explain why targeted cancer therapies so often deliver only partial or transient benefits. He framed the challenge as therapy resistance in its broadest sense in that even when a drug successfully blocks its intended molecular target, tumours could continue to grow because that blockade fails to translate into the desired outcome. This disconnect has remained a central obstacle to durable responses in precision oncology.

“Even when we hit the pathway [that] we think matters, the tumour sometimes behaves as if it does not care,” said Sahai, who leads the Sahai Tumour Cell Biology Laboratory at the Francis Crick Institute, introducing his work on inhibitors of protein kinase B, commonly known as AKT.

AKT is a central node in the phosphoinositide 3-kinase (PI3K) signalling network which many cancers activate through PI3K mutations, aberrant upstream receptor signalling, or loss of negative regulators such as phosphatase and tensin homologue. Although AKT inhibition has shown clinical activity in some settings, Sahai stressed that triple negative breast cancer remains associated with a poor clinical prognosis and outcomes with limited therapeutic options, despite frequent alterations affecting the PI3K–AKT pathway.

The central question examined in his work concerned heterogeneity in the earliest response to therapy. When all tumour cells received the same drug, did they respond uniformly, or could early diversity in behaviour seed later resistance? To address this, Sahai’s group used patient-derived material, beginning with a patient-derived xenograft and then generating matched organoids. Previous work had already revealed heterogeneity in cell state through single-cell RNA sequencing, while lineage barcoding had shown that individual cells could move through multiple states across several weeks. These observations supported a view of tumours as plastic systems rather than fixed mosaics of genetically distinct subclones.

When the team treated organoids with different therapies and tracked them by brightfield time-lapse microscopy, they observed a clear contrast. Paclitaxel produced a relatively uniform response across organoids, consistent with behaviour seen in xenografts. By contrast, an AKT inhibitor generated divergent outcomes within the same field of view. Some organoids continued to grow, others slowed markedly, and others underwent apoptosis. Quantitative analysis confirmed that responses to AKT inhibition varied widely, even under identical conditions.

Sahai outlined two possible explanations. In one model, a rare, stable genetic subclone would carry intrinsic resistance and expand under drug selection. In the alternative model, resistance would arise from a non-genetic and reversible cell state. To distinguish these possibilities, the group derived sublines and assessed their drug responses. If genetic resistance dominated, some sublines would show near-complete resistance while others would remain sensitive. Instead, every subline displayed a similar spectrum of responses. Moreover, when cells were selected for growth under AKT inhibition and the drug was then withdrawn, resistance diminished. This reversibility argued against hard-wired genetic resistance and supported a state-based explanation.

To characterise this resistant state in more detail, Sahai described a high-content imaging workflow based on dual-view oblique plane microscopy, which provides light-sheet-like three-dimensional imaging through a single objective. This approach allowed the team to image many organoids in multiwell plates over time. The platform was combined with genetically encoded biosensors. One reporter provided a cell-by-cell readout of AKT activity via nucleocytoplasmic localisation, while a second ratiometric sensor reported metabolic state as a proxy for glycolytic and biosynthetic activity. Image analysis segmented nuclei and estimated cytoplasmic signal, enabling per-cell measurements within each organoid and generating a high-dimensional dataset linking morphology, signalling, metabolism and growth.

Using these data, the group compared organoids in three conditions: before treatment, within 24 hours of AKT inhibition, and several days later. Most organoids experienced robust AKT inhibition within the first day, yet the degree of early inhibition did not predict longer-term growth behaviour. The key difference emerged between 24 and 96 hours after treatment. During this period, organoids diverged in their AKT dynamics. Some rapidly restored AKT activity, and this rebound correlated with higher metabolic activity and sustained proliferation. Heterogeneity therefore arose less from incomplete target engagement at the outset and more from differences in how quickly signalling and growth became reconnected.

To link these behaviours to molecular programmes, the team performed single-cell RNA sequencing on organoids treated with AKT inhibition. Clustering analyses identified subsets of cells that failed to exit the cell cycle despite drug exposure. These resistant clusters expressed markers of a basal-like mammary epithelial state, including SRY-box transcription factor 2 and keratin 14 (KRT14). Sahai placed this finding in the context of normal mammary biology, in which luminal cells support milk production while basal myoepithelial cells provide contractile and structural functions. The resistant state resembled the basal programme which suggested that resistance exploited pre-existing tissue biology rather than arising de novo in response to therapy.

The link was reinforced by bulk sequencing after long-term AKT inhibition which again revealed basal-associated markers. For functional isolation, the team used integrin alpha 6 (ITGA6) as a practical – if imperfect – cell-surface marker. Cells with high ITGA6 expression generated a greater fraction of organoids that continued to grow under AKT inhibition than ITGA6-low cells, supporting the conclusion that a pre-existing subpopulation carried disproportionate resistance capacity.

Sahai then asked whether this principle extended beyond a single model. He described a re-analysis of published datasets comparing breast cancer models with basal or luminal character. In luminal models, suppression of AKT activity aligned closely with reduced downstream growth signalling. In basal models, AKT activity still declined, yet downstream growth outputs fell far less. This pattern mirrored the organoid data and suggested that, in basal-like states, growth is less tightly coupled to AKT signalling.

To examine causality, the group asked whether enforcing a basal-like programme could induce resistance. Transcription factor enrichment analyses implicated tumour protein 63 (TP63), a regulator previously linked to basal identity. Overexpression of a TP63 isoform increased KRT14 expression and reduced sensitivity to AKT inhibition, while a DNA-binding mutant produced a weaker effect. Sahai suggested this was evidence that entry into a basal-like state could suffice to generate a therapy-persistent phenotype.

Finally, the group explored therapeutic implications. Using the organoid platform, they screened compounds for selective activity against AKT-resistant organoids and identified sensitivities to inhibitors of Src-family signalling and to an inhibitor of monocarboxylate transporter 1, a lactate transporter. Sahai cautioned that the basal-like state did not confer blanket resistance but instead altered dependencies, which opened the possibility of rational drug combinations.

Returning to the xenograft setting, the team found that baseline tumours contained a subset of KRT14-high cells with a TP63-associated resistance signature. Consistent with predictions based on state plasticity, AKT inhibition produced only a short-lived response in vivo, with tumour growth resuming within around three weeks and increased KRT14 staining apparent after treatment.

Sahai concluded that high-content organoid imaging, integrated with single-cell sequencing and lineage tracing, can reveal early sources of heterogeneity that can later be associated with treatment failure. In triple negative breast cancer, resistance to AKT inhibition has appeared to arise from a reversible basal-like state that weakens the coupling between AKT signalling and growth, a state rooted in normal mammary epithelial biology and amenable to targeted combination strategies.