The high frequency of PI3K pathway alterations in cancer has motivated numerous efforts to develop drugs targeting this network. Although many potent and selective inhibitors have been developed and evaluated in preclinical models, their progress to clinical approval has been limited. Here we discuss the pressing need to develop improved biomarker strategies to guide patient selection and improve assessment of patient responses to PI3K pathway inhibitors to address unresolved issues surrounding the efficacy and tolerability of these compounds in patients with cancer.

The PI3K/AKT pathway is a highly conserved signaling network that responds to extracellular stimuli to promote many cellular processes including growth and survival. Aberrant activation of this pathway is frequently observed across cancer types, most commonly through oncogenic mutation or amplification of PIK3CA or AKT, PTEN inactivation, or RTK overexpression. Thus, significant efforts have been dedicated to developing compounds that target this pathway for therapeutic benefit, including pan-PI3K inhibitors, isoform-specific PI3K inhibitors, AKT inhibitors, and mTOR or dual PI3K/mTOR inhibitors. Despite these efforts, to date only one PI3K alpha-selective inhibitor, alpelisib, has been FDA approved for the treatment of solid tumors. Notably, the PI3K inhibitors idelalisib, copanlisib, duvelisib, and umbralisib have been FDA approved for the treatment of various hematologic malignancies including chronic lymphocytic leukemia and follicular lymphoma (Table 1). For a more complete landscape of PI3K inhibitors in preclinical and clinical development, we refer readers to this recent review (1).

Table 1.

PI3K inhibitors that have been FDA approved or have reached late-stage clinical development.

InhibitorCompound nameSpecificityClinical statusYear FDA approved
Alpelisib BYL-719 PI3Kα-selective HR+ HER2− PIK3CA mBC in combination with fulvestrant 2019 
Copanlisib BAY 80–6946 Pan-PI3K Relapsed follicular lymphoma 2017 
Duvelisib IPI-145 PI3Kδ and PI3Kγ Relapsed follicular lymphoma, chronic lymphocytic leukemia, small lymphocytic leukemia 2018 
Idelalisib CAL-101 PI3Kδ-selective Relapsed follicular lymphoma, chronic lymphocytic leukemia, and small lymphocytic leukemia 2014 
Taselisib GDC-0032 PI3Kβ-sparing Discontinued during phase 3 – 
Umbralisib TGR-1202 PI3Kδ-selective, (also CK-1ϵ) Relapsed follicular lymphoma and marginal zone lymphoma 2021 
PI3K pathway biomarkers, their function, advantages, and disadvantages. 
Type Biomarker Function Advantages Disadvantages 
Predictive PIK3CA mutations Predict sensitivity to PI3K pathway inhibitors Predict sensitivity to alpelisib Pathway feedback and tumor heterogeneity complicate use of prediction of genotype as marker of response. Lack of standardized assays (i.e., biopsy), driver mutations, and predictive value of coexisting mutations related to resistance. 
 PTEN loss Predict sensitivity to PI3K pathway inhibitors   
 18F-FDG-PET/CT imaging Predict early metabolic response to PI3K pathway inhibitors  Not yet utilized in clinical trials; predictive value unclear 
PI3K pathway signaling AKT (pThr308, pSer473), PRAS40 (pThr246), 4EBP1 (pSer65, pThr70), and RPS6 (pSer240, pSer244) Demonstrate target engagement and pathway modulation  Targets may be regulated alternative pathways; cells sampled (skin/hair follicle biopsies) may not accurately reflect exposure or efficacy in tumor 
Metabolic response Plasma glucose Surrogate measure of pathway modulation Minimally invasive Inferior to insulin and C-peptide, as plasma glucose is subject to compensatory insulin and C-peptide release 
 Insulin Surrogate measure of pathway modulation Minimally invasive Relationship between dose and response not yet clear 
 C-peptide Surrogate measure of pathway modulation Minimally invasive Relationship between dose and response not yet clear 
Functional imaging 18F-FDG-PET Surrogate measure of pathway modulation Minimally invasive, highly quantitative, assessment of on- and off-target toxicities Relationship between dose and response not yet clear 
 18F-FLT-PET Surrogate measure of pathway modulation, proliferation Minimally invasive, highly quantitative Not yet utilized in clinical trials for PI3K pathway inhibitors 
 Magnetic resonance spectroscopy Surrogate measure of response Minimally invasive Not yet utilized in clinical trials for PI3K pathway inhibitors 
 Diffusion-weighted and dynamic contrast enhanced MRI Surrogate measure of response Minimally invasive Not yet utilized in clinical trials for PI3K pathway inhibitors 
Circulating Circulating tumor cells Surrogate marker of target engagement Minimally invasive, monitor response over time Cells in blood exposed to plasma drug concentrations, not concentrations achieved in solid tumors 
 Cell-free DNA (cfDNA) Surrogate marker of target engagement Minimally invasive Low plasma DNA levels may prevent detection of mutations and clonal evolution 
 Circulating markers of cell death Surrogate marker for cell death Minimally invasive High variability between patients 
InhibitorCompound nameSpecificityClinical statusYear FDA approved
Alpelisib BYL-719 PI3Kα-selective HR+ HER2− PIK3CA mBC in combination with fulvestrant 2019 
Copanlisib BAY 80–6946 Pan-PI3K Relapsed follicular lymphoma 2017 
Duvelisib IPI-145 PI3Kδ and PI3Kγ Relapsed follicular lymphoma, chronic lymphocytic leukemia, small lymphocytic leukemia 2018 
Idelalisib CAL-101 PI3Kδ-selective Relapsed follicular lymphoma, chronic lymphocytic leukemia, and small lymphocytic leukemia 2014 
Taselisib GDC-0032 PI3Kβ-sparing Discontinued during phase 3 – 
Umbralisib TGR-1202 PI3Kδ-selective, (also CK-1ϵ) Relapsed follicular lymphoma and marginal zone lymphoma 2021 
PI3K pathway biomarkers, their function, advantages, and disadvantages. 
Type Biomarker Function Advantages Disadvantages 
Predictive PIK3CA mutations Predict sensitivity to PI3K pathway inhibitors Predict sensitivity to alpelisib Pathway feedback and tumor heterogeneity complicate use of prediction of genotype as marker of response. Lack of standardized assays (i.e., biopsy), driver mutations, and predictive value of coexisting mutations related to resistance. 
 PTEN loss Predict sensitivity to PI3K pathway inhibitors   
 18F-FDG-PET/CT imaging Predict early metabolic response to PI3K pathway inhibitors  Not yet utilized in clinical trials; predictive value unclear 
PI3K pathway signaling AKT (pThr308, pSer473), PRAS40 (pThr246), 4EBP1 (pSer65, pThr70), and RPS6 (pSer240, pSer244) Demonstrate target engagement and pathway modulation  Targets may be regulated alternative pathways; cells sampled (skin/hair follicle biopsies) may not accurately reflect exposure or efficacy in tumor 
Metabolic response Plasma glucose Surrogate measure of pathway modulation Minimally invasive Inferior to insulin and C-peptide, as plasma glucose is subject to compensatory insulin and C-peptide release 
 Insulin Surrogate measure of pathway modulation Minimally invasive Relationship between dose and response not yet clear 
 C-peptide Surrogate measure of pathway modulation Minimally invasive Relationship between dose and response not yet clear 
Functional imaging 18F-FDG-PET Surrogate measure of pathway modulation Minimally invasive, highly quantitative, assessment of on- and off-target toxicities Relationship between dose and response not yet clear 
 18F-FLT-PET Surrogate measure of pathway modulation, proliferation Minimally invasive, highly quantitative Not yet utilized in clinical trials for PI3K pathway inhibitors 
 Magnetic resonance spectroscopy Surrogate measure of response Minimally invasive Not yet utilized in clinical trials for PI3K pathway inhibitors 
 Diffusion-weighted and dynamic contrast enhanced MRI Surrogate measure of response Minimally invasive Not yet utilized in clinical trials for PI3K pathway inhibitors 
Circulating Circulating tumor cells Surrogate marker of target engagement Minimally invasive, monitor response over time Cells in blood exposed to plasma drug concentrations, not concentrations achieved in solid tumors 
 Cell-free DNA (cfDNA) Surrogate marker of target engagement Minimally invasive Low plasma DNA levels may prevent detection of mutations and clonal evolution 
 Circulating markers of cell death Surrogate marker for cell death Minimally invasive High variability between patients 

Abbreviations: HR, hormone receptor; mBC, metastatic breast cancer.

The clinical efficacy of drugs targeting the PI3K pathway has been limited by dose-dependent on-target toxicities, development of resistance through compensatory feedback mechanisms, and suboptimal patient selection for clinical trials. Moreover, the disconnect between the efficacy of these inhibitors in preclinical studies and clinical trials underscores the need to refine our understanding of the determinants of patient response to PI3K/AKT therapies. Navigating these shortcomings will likely involve the use of molecular biomarkers to (i) guide patient selection and (ii) monitor drug response. Here, we discuss current biomarkers, challenges in their development and implementation, and their potential value in the development of PI3K pathway therapies for solid tumors.

Predictive biomarkers

Despite the frequency of PI3K pathway mutations in solid tumors, establishing predictive biomarkers based on pathway mutational status has been a challenge. Preclinical data indicate that PIK3CA and AKT are relatively weak oncogenic drivers that produce variable degrees of tumor cell addiction to PI3K/AKT signaling (2). Furthermore, hotspot mutations such as PIK3CA E545K or H1047R may differentially potentiate cancer growth, especially on distinct mutational backgrounds. It is still largely unclear whether tumors with distinct PIK3CA mutations respond differentially to different PI3K inhibitors. Recent studies have shown that double PIK3CA mutations in cis, which occur in up to 15% of patients with breast cancer, increase PI3K activity and, as a consequence, sensitivity to PI3K alpha-selective inhibitors (3). The complexities of pathway regulation and feedback mechanisms pose a challenge when determining if response (or lack of response) is correlated to mutational status, or as a result of multiple coexisting mutations. In addition, certain pathway alterations occur at low frequency (i.e., PTEN or AKT mutations) and thus their incorporation into predictive models, especially retrospective analyses, may be compromised due to low statistical power.

Despite initially contradictory preclinical and clinical evidence, the sensitivity of PIK3CA-mutant tumors to PI3K inhibitors was confirmed by the SANDPIPER and SOLAR-1 trials, the latter leading to approval of alpelisib (4). PIK3CA mutations in tissue or liquid biopsies have since been approved by the FDA as predictive biomarkers for the use of alpelisib (5). However, PIK3CA mutation had no predictive value in response to the pan-PI3K inhibitor pictilisib (OPPORTUNE) or alpelisib (NEO-ORB) in the neoadjuvant setting (5). This disconnect in predictive value of PIK3CA mutation status may be explained by the low tolerability of pan-PI3K inhibitors compared with isoform-selective PI3K inhibitors. Most of the analyses conducted to date, with the exception of the SANDPIPER and SOLAR-1 trials, which used PIK3CA mutational status in their inclusion criteria, have been retrospective, exploratory, and based on a relatively small number of patients (4). Another possibility is that efficacy of these inhibitors is only revealed in combination settings, for example, with endocrine therapy in estrogen receptor-positive tumors.

Mutational status is already used somewhat successfully in predicting combination therapies. Because PI3K pathway inhibitors display limited efficacy as monotherapies, preclinical modeling with adequate biomarker information may optimize selection for combination therapies, such as endocrine therapies, chemotherapy, as well as molecular targeted agents such as CDK4/6 inhibitors, mTOR inhibitors, or HER2 inhibitors (2, 5, 6). As immunotherapy becomes a potential avenue of treatment for patients with breast cancer, it will become important to assess the role of the tumor microenvironment in predicting response to treatments, especially given recent data showing that PI3K inhibitors act on the stroma to contribute to antitumor effects (2).

The evidence presented above emphasizes the critical importance of proper patient selection and argues for the expansion of predictive biomarker studies for future clinical trials. We propose that prospective mutational analysis will help refine patient selection and determine whether different PI3K mutations respond better to specific inhibitors. Mutational analysis of intertumoral heterogeneity, including predicted markers of resistance to PI3K pathway inhibitors (KRAS, TP53), may also help inform patient selection (2). Alternative methods for predicting response to PI3K pathway inhibitors, for example through assessment of early metabolic response by 18F-FDG-PET/CT imaging may also efficiently predict response to PI3K inhibitors (5, 7).

Response biomarkers

Inhibition of the PI3K pathway produces well-characterized signaling and metabolic changes that can be monitored as surrogate measures for pathway inhibition. Measurement of phosphoprotein biomarkers, such as AKT (pThr308, pSer473), PRAS40 (pThr246), 4EBP1 (pSer65, pThr70), and RPS6 (pSer240, pSer244), have been used to evaluate the degree of pathway inhibition and confirm target engagement in early-phase clinical trials (8). Although more direct and quantitative measurements of PI3K inhibition have been developed, such as cellular and tissue measurement of PIP3, these techniques have yet to reach the clinic (9).

Phosphoprotein biomarkers are often imperfect at gauging target engagement or predicting efficacy of inhibitors in patients, and this has proven to be a particular challenge in the PI3K and AKT inhibitor arena. Inconsistency in sample handling, especially in multicenter clinical trials, may prevent accurate assessment of phosphoproteins downstream of PI3K/AKT. Many downstream targets that are used as biomarkers, such as RPS6 and 4EBP1 can be regulated by alternative pathways and thus mask successful target engagement. The disadvantage of this approach is the imperfect ability to predict efficacy, as robust responses can be associated with incomplete inhibition of protein phosphorylation. In addition, the cells sampled to measure response, such as skin or hair follicle biopsies, often have a different compound exposure level than the tumor, leading to inaccurate assessment of exposure or efficacy. Finally, discrepancies between robust preclinical efficacy validated by downstream signaling markers and the corresponding modest inhibition observed in clinical trials highlights the challenge of using signaling markers at predicting efficacy PI3K inhibitors.

An alternative means of measuring target engagement is through circulating tumor cells (CTC), cell-free circulating tumor DNA (ctDNA), or circulating markers of apoptosis. These minimally invasive methods detect cancer cells originating from solid tumors, fragments of DNA, or markers of cell death circulating in peripheral blood. A major advantage of circulating biomarkers is their ability to rapidly demonstrate target engagement or tumor shrinkage, and their potential to follow response to inhibitors in real time. Phosphoprotein biomarkers for pathway modulation can be assessed in CTCs, though a clear correlation between these markers and clinical response have not yet been established. One major limitation of circulating biomarkers it that plasma blood concentrations may be significantly different from concentrations that tumors are exposed to.

Because of the pivotal role the PI3K pathway plays in cellular metabolism, several biomarkers to measure metabolic changes in response to PI3K pathway inhibition have been developed. Inhibition of PI3K or AKT prevents AKT-mediated plasma membrane translocation of the glucose transporter GLUT4, dampening cells' ability to uptake glucose. To compensate, the body releases insulin from the pancreas to check glucose levels. Consequently, measurement of plasma glucose, insulin, and C-peptide levels have been adopted as surrogate markers of target engagement. However, a robust understanding of the relationship between these markers and pathway inhibition remains under evaluation (8).

In addition to these plasma biomarkers, several noninvasive functional imaging technologies have emerged as novel approaches to measure both on- and off-target effects and could also be used to inform optimum dosing schedules to reduce toxicities. Metabolic imaging techniques such as PET with 2-deoxy-2-[fluorine-18]fluoro- D-glucose ([18F]-FDG-PET) or 3-deoxy-3-[18F]-fluorothymidine ([18F]-FLT-PET), magnetic resonance spectroscopy, and diffusion weighted or dynamic contrast enhanced magnetic resonance imaging are means of measuring the metabolic effects of PI3K pathway modulation. Noninvasive imaging techniques offer the advantage of not requiring sample biopsy or acquisition of tumor tissue. These highly quantitative measurements also allow for assessment of both tumor-specific and systemic metabolic effects and the assessment of both on- and off-target toxicities. Quantitative measurements are important for assessing time- and dose-dependent inhibition of downstream signaling to confirm target engagement, because the magnitude and duration of pathway inhibition is often used to predict efficacy (8). However, there is a clear need for biomarker studies to robustly associate these biomarkers to clinical response. Many of these new technologies have not yet reached PI3K clinical trials.

The use of any single biomarker cannot optimally assess clinical response. Instead, the use of a composite biomarker that monitors signaling, metabolic, and immunologic effects may allow for a more detailed evaluation of drug efficacy in patients. Expanding the use of biomarkers throughout the duration of a clinical trial may also inform our understanding of disease progression and development of resistance. Similarly, collecting biomarker data at multiple timepoints, not simply trial endpoint, will allow monitoring of treatment efficacy and disease progression. This strategy may be able to predict emergence of therapeutic resistance. This also emphasizes the need to prospectively, rather than retrospectively, assess patient mutation status. For example, the predictive value of PIK3CA mutations was only discovered with prospective mutational analysis in the SOLAR-1 and SANDPIPER trials, indicating that retrospective studies are not sufficient to identify clinical benefit for specific subpopulations of patients (5).

Finally, there is a clear need for a management strategy for clinical biomarker data. A major barrier to the development of predictive and prognostic biomarkers is the lack of a centralized set of clinical and biomarker data. In the era of big data and personalized patient care, the expansion of biomarker utility will depend on platforms to collect, manage, and share clinical data. Increasingly detailed biomarker data from genetic sequencing, flow cytometry, metabolic assays, and omics analyses have become less expensive and complex to perform. The vast amounts of data that clinical trials generate to characterize disease progression and provide evidence of drug efficacy may serve as a resource to inform future trials as well as preclinical studies. Although the collection of these data has become increasingly common, their diverse sources and formats have led to challenges integrating and centralizing information. The development of broadly accessible platforms to strategically manage molecular profiling data and clinical features represents a critical need in cancer management.

Although PIK3CA is one of the most frequently altered genes in cancer, the complexity of this signaling network has made it difficult to develop therapeutic agents that successfully target these aberrations (3). The recent success of the PI3K alpha-specific inhibitor alpelisib in combination with fulvestrant in advanced ER+ breast cancer demonstrates that life-extending therapies are within developmental capacity if the context is correctly understood. Success can be enhanced by improved translation of preclinical models, pre-selection of patients who will respond to therapies, refined monitoring of patient response, and prediction of resistance. Efforts to collect and share clinical data, understand the translatability of in vitro models, and exploring additional biomarkers of prediction and response may promote the success of PI3K pathway-targeting agents in the clinic.

A. Toker reports grants from NCI during the conduct of the study; personal fees from ASBMB, Bertis, Inc., and OncXerna outside the submitted work. E.C. Erickson reports grants from F31 CA254000 outside the submitted work.

This work was supported by grants from the NCI (R35 CA253097 to A. Toker and F31CA254000 to E.C. Erickson). The authors apologize to any colleagues whose work was not cited due to space limitations.

1.
Castel
P
,
Toska
E
,
Engelman
JA
,
Scaltriti
M
. 
The present and future of PI3K inhibitors for cancer therapy
.
Nat Cancer
2021
;
2
:
587
97
.
2.
Hanker
AB
,
Kaklamani
V
,
Arteaga
CL
. 
Challenges for the clinical development of PI3K inhibitors: strategies to improve their impact in solid tumors
.
Cancer Discov
2019
;
9
:
482
91
.
3.
Ellis
H
,
Ma
CX
. 
PI3K Inhibitors in breast cancer therapy
.
Curr Oncol Rep
2019
;
21
:
110
.
4.
Chang
DY
,
Ma
WL
,
Lu
YS
. 
Role of alpelisib in the treatment of pik3ca-mutated breast cancer: patient selection and clinical perspectives
.
Ther Clin Risk Manag
2021
;
17
:
193
207
.
5.
Brandão
M
,
Caparica
R
,
Eiger
D
,
de Azambuja
E
. 
Biomarkers of response and resistance to PI3K inhibitors in estrogen receptor-positive breast cancer patients and combination therapies involving PI3K inhibitors
.
Ann Oncol
2019
;
30
:
x27
42
.
6.
LoRusso
PM
. 
Inhibition of the PI3K/AKT/mTOR pathway in solid tumors
.
J Clin Oncol
2016
;
34
:
3803
15
.
7.
Mayer
IA
,
Abramson
VG
,
Isakoff
SJ
,
Forero
A
,
Balko
JM
,
Kuba
MG
, et al
Stand up to cancer phase Ib study of pan-phosphoinositide-3-kinase inhibitor buparlisib with letrozole in estrogen receptor-positive/human epidermal growth factor receptor 2-negative metastatic breast cancer
.
J Clin Oncol
2014
;
32
:
1202
9
.
8.
Josephs
DH
,
Sarker
D
. 
Pharmacodynamic biomarker development for PI3K pathway therapeutics
.
Transl Oncogenomics
2016
;
7
:
33
49
.
9.
Guillou
H
,
Stephens
LR
,
Hawkins
PT
. 
Quantitative measurement of phosphatidylinositol 3,4,5-trisphosphate
.
Methods Enzymol
2007
;
434
:
117
30
.