High levels of p21 correlate with a poor prognosis in TP53 wild-type NSCLC

p21 has been shown to act as both a tumour suppressor and an oncogene, depending on context [22]. Here, we wanted to investigate the potential pro-tumorigenic, or pro-survival, properties of p21 in NSCLC since a high level of p21 protein correlates with a worse prognosis in this disease (Fig. 1a) and we hypothesised that p21 may provide a fitness advantage to NSCLC cells by allowing tumour cells with replication stress that persists beyond S-phase, to enter a p21-dependent quiescent state (G0) after mitosis (Fig. 1b [8,9,10,11]).

If our hypothesis is correct, we would expect to see that p21 expression correlates with a worse prognosis only in TP53WT tumours, where cells can respond to replication stress through p53-dependent p21 expression [8], and not in TP53mutant tumours. Therefore, we analysed NSCLC cases in The Cancer Genome Atlas (TCGA), separated by TP53 status and level of CDKN1A (encoding p21) expression. Consistent with our hypothesis, a high level of CDKN1A expression only correlated with a poor prognosis in TP53WT NSCLC and not in TP53mutant tumours (Fig. 1c). We observed the same correlation when analysing protein expression data from TCGA ((quantified by Reverse Phase Protein Array); Supplementary Fig. 1).

p21 expression has been reported to be higher in tumours than in surrounding normal tissue [27]. However, p21 is frequently used as a marker of senescence in tissues and tumours. We wanted to identify if p21-expressing cells in lung tumours maintain proliferative capacity and whether some of these cells could be quiescent and not senescent. Since there are no good markers that uniquely distinguish quiescent from senescent cells, we opted to use Ki67 staining as a marker of proliferative potential. It was recently shown that Ki67 is a graded marker of cellular proliferation, that is expressed in quiescent cells but that decreases in protein level the longer cells are held in quiescence [35]. We immunostained TP53WT NSCLC tumours for p21 and Ki67 and quantified the fraction of p21+, Ki67+ and p21+/Ki67+ cells (Fig. 1d). We identified a small and variable fraction of p21+ /Ki67+ cells, indicating the possibility of p21-dependent quiescent cells in patient tumours (Fig. 1e). Cells negative for both p21 and Ki67 are most likely terminally-differentiated cells. We investigated if p21+ /Ki67+ cells had any spatial arrangement, for example, localised near vessels or at the tumour periphery. We focussed on tumour 8734 which had the largest number of p21+/Ki67+ cells and identified 18 'hotspots' of double-positive cells (see Materials and Methods). Of these 18 hotspots, 9/18 (50%) were at the tumour edge and 15/18 (83.3%) were located adjacent to vessels (identified by their morphology). This suggests that p21+/Ki67+ cells while equally likely to be spread throughout the tumour mass may be preferentially close to vessels. However, this is just one tumour and a larger cohort is needed to determine how widespread this spatial distribution is.

In summary, our analyses show that high p21 expression correlates with poor NSCLC patient prognosis specifically in TP53WT tumours, which account for ~50% of NSCLC cases. We provide evidence that p21 expressing cells in TP53WT tumours can be proliferative and are not terminally arrested in senescence.

TP53 wild-type NSCLC cells can enter a p21-dependent quiescent state

We first investigated whether proliferating NSCLC cells could enter a 'spontaneous' quiescent state and if this was p21-dependent. Spontaneous here refers to cells entering quiescence in the presence of full growth media and in the absence of contact inhibition [7]. We selected a panel of NSCLC cell lines – five TP53WT (NCI-H460, A549, NCI-H1944, NCI-H1666, NCI-H1563), three TP53mutant (NCI-H1299, CORL23, NCI-H358) and one that has a mutation causing psi-p53 to be expressed (NCI-H1650 [36]). We validated that TP53WT cells had an intact p53-p21 pathway by treating cells with the Mdm2 inhibitor, Nutlin-3, and quantifying p21 expression by single-cell imaging. All TP53WT and none of the TP53mutant cell lines induced p21 expression upon Nutlin-3 treatment (Fig. S2A). psi-p53 has been suggested to be transcriptionally inactive [37]. However, we observed slight upregulation of p21 protein expression in NCI-H1650 cells in response to Nutlin-3 and downregulation of p21 after p53 depletion, more similar to TP53WT than TP53mutant cells (Fig. S2A, B). Therefore, we consider the NCI-H1650 cell line as an intermediate between TP53WT and TP53mutant cells that retains the ability to regulate p21 in a p53-dependent manner.

We developed a high-throughput single-cell imaging assay to identify quiescent cells. We plated cells at low cell density, then added EdU, a nucleoside analogue incorporated into cells undergoing DNA replication, for the final 24 h before fixing and staining cells. In addition to EdU labelling, we co-stained cells for hyperphosphorylated pRb (P-Rb), a marker for cells that have passed the Restriction Point and committed to proliferation [38, 39] (Figs. 2a, b; S3A, B). EdU negative cells have not undergone S-phase for at least 24 h and can be said to be quiescent. Similarly, cells that are negative for hyperphosphorylated pRb can be considered quiescent. Using this assay, we identified a fraction of spontaneously quiescent cells across all cell lines tested, ranging from 1-47% of quiescent cells (EdU assay) and 10–66% quiescent cells (P-Rb assay; Fig. 2c, Supplementary Fig. 3c). The fraction of quiescent cells identified in the two assays differs due to the different assay timescales but the two measurements correlate (Supplementary Fig. 3G, left panel). Ki67 is a third marker that has been used to distinguish between quiescent and proliferating cells [40]. However, as mentioned, Ki67 is a graded marker of proliferation ([35]; Supplementary Fig. 3D) unlike the binary on/off readouts from EdU and P-Rb. We triple-stained cells with EdU, P-Rb and Ki67 and determined a Ki67 intensity cut-off to distinguish between quiescent and proliferating cells (Supplementary Fig. 3E, F). We found that the Ki67 low population was almost identical to the EdU negative population (Supplementary Fig. 3G). Therefore, since EdU and P-Rb produce definitive quiescent versus proliferating populations, subsequent assays used these markers.

Fig. 2: p53 wild-type NSCLC cells can enter a p21-dependent quiescent state.

a Schematic shows assay principle – EdU is taken up by cells during S-phase (highlighted in blue). Cells in quiescence for >24 h (G0, red circle) will not be labelled with EdU. b Sample images of A549 cells labelled with EdU to quantify the quiescent fraction. Red circles represent EdU negative, quiescent cells. EdU is in red and Hoechst in blue in merged image. Scale bar 20 μm. c Graph shows quantification of percentage of EdU negative (G0) cells across cell lines. Data are plotted as superplots, n = 3, 4 technical replicates per experiment. Grey triangle represents the mean. TP53WT cell lines are shown in red, psi-p53 mutant is shown in purple and TP53mutant cell lines in blue. d Graph shows the effect of p21 depletion by siRNA on the fraction of quiescent cells within each cell line, measured using the EdU assay. Data are plotted as super plots, n = 3, 4 technical replicates per experiment. Non-targeting control (NTC) siRNA values are shown in black, p21 siRNA in red. One-way ANOVA to test for significance, ****p < 0.0001, **p < 0.01, ns: not significant. Note that for the TP53mutant CORL23 line, significance indicates that the fraction of EdU negative cells increases after p21 siRNA. e Quantification of nuclear p21-mVenus in single NCI-H1944 cells over time. All single-cell traces are aligned to mitotic exit. Red curves represent cells that exit mitosis with high p21-mVenus levels (defined as >60 a.u. at 5 h post-mitotic exit). Percentage in red is the fraction of cells that enter a p21High state (proxy for quiescence, 24/57 cells (42.1%)). f Correlation between G1 length and p21-mVenus levels. g Left: schematic to show how the CDK2 activity sensor works [30]. Right: quantification of CDK2 activity in single A549 cells over time. All single-cell traces are aligned to mitotic exit. Red curves represent cells that exit mitosis with low CDK2 activity (defined as <0.5 at 6 h post-mitotic exit). Percentages in red are the fraction of cells in each condition that enter a CDK2Low activity state (proxy for quiescence). NTC siRNA, 14/48 cells (20.8%). p21 siRNA, 4/49 cells (8.2%). h Heatmap shows clustering of significantly differentially expressed genes between p21-High versus p21-Low cells based on RNA-seq profiling. Four biological repeats were profiled. Red represents increased expression, blue represents decreased expression. i GO analysis of genes with lower expression in p21High cells. j GO analysis of genes with higher expression in p21High cells.

To determine if entry into this spontaneous quiescent state is p21-dependent in NSCLC cells, we depleted p21 using siRNA (Supplementary Fig. 3H, I). Four of the five TP53WT cell lines and the psi-p53 cell line all showed a reduction in the quiescent fraction when p21 was depleted (Fig. 2d, Supplementary Fig. S3J). None of the three TP53mutant lines showed a reduced quiescent fraction after p21 depletion. These data show that p21-dependent quiescence exists in TP53WT NSCLC cells.

Quiescence is a reversible cell cycle arrest state and cells enter and exit quiescence on different timescales. While our fixed cell assay is useful for characterising quiescence across cell lines in a high-throughput manner, we need to quantify quiescence in a more dynamic manner. To do this, we generated fluorescent reporter TP53WT NSCLC cell lines to quantify entry and exit into quiescence. First, we generated an NCI-H1944 cell line where we labelled endogenous PCNA with mRuby to follow cell cycle dynamics and to accurately define phase transitions [29] and tagged endogenous p21 with a mVenus fluorophore [31] to follow p21 expression (Supplementary Figure S3K-N). Live cell imaging revealed that p21 expression is heterogeneous, highest in G1 cells and that high p21 expression correlates with entry into quiescence. Approximately 42% of NCI-H1944 cells exit mitosis with high p21 levels and remain in a period of quiescence (longer G1 phase) before re-entering the cell cycle (Fig. 2e, red curves), while the remaining cells have low levels of p21 and re-enter S-phase more rapidly (Fig. 2e, black curves, 2f; Supplementary Movie 1 [8, 13]). We also generated an A549 cell line expressing mRuby-PCNA and a CDK2 activity sensor [30] (Fig. 2g, Supplementary Fig. 3O). In cells that enter quiescence, we anticipate CDK2 activity to decrease after mitotic exit, as p21 levels increase [12, 30]. Indeed, we observed that approximately 21% of A549 cells downregulate CDK2 activity upon exiting mitosis and that this is p21-dependent (red curves, Fig. 2g).

We wanted to determine what, in addition to p21 expression, defined these spontaneously quiescent NSCLC cells. As well as demonstrating localisation changes throughout the cell cycle, PCNA also displays expression changes throughout the cell cycle, with G0/G1 cells displaying the lowest PCNA protein levels. We used FACS to separate p21-mVenusHigh mRuby-PCNALow (‘p21High’) quiescent cells from p21-mVenusLow mRuby-PCNAHigh (‘p21Low’) proliferating cells (Supplementary Fig. 4A) and validated the two populations by immunostaining for additional cell cycle markers (Supplementary Fig. 4B). We characterised the transcriptomes of these two populations by RNAseq (Fig. 2h; Supplementary Fig. 4C). As expected, genes involved in regulating the cell cycle were significantly reduced in p21High quiescent cells (Fig. 2i; Table S1), further validating this as a quiescent population. Genes that were significantly enriched in p21High quiescent cells included genes encoding proteins involved in cell interactions with the external environment (Fig. 2j; Table S2). These included genes associated with GO terms involving response to external stimuli (IL1B, TNFSF14, NR1H4, PHEX, MAP1LC3C, WIPI1, SCX, FOLR1, CASP1, RRAGD, TLR5, PDK4, CD68, CDKN1A, AGT, ZFYVE1, BMT2, NAMPT, GABARAPL1), in extracellular matrix and structure organisation (TGM3, ADAMTS14, ADAMTS7, IL6, SCX, COL4A3, COL17A1, LRP1, ADAMTSL4, COL5A1, MMP11, PBXIP1, MMP2, ADAMTS13, LOXL2, AGT, LUM, ADAMTS10, LAMB3, BMP1) and in organic anion transport (IL1B, KMO, SLC22A11, CEACAM1, SLC16A4, SLC6A12, SLC4A10, FOLR1, SLC26A9, CES1, SLC4A3, SLC23A1, SLC27A1, SLC17A5, AGT, ABCC6, PTGES, PSAP, CYB5R1). The enrichment of mRNAs for ADAMTS (a disintegrin and metalloproteinase with thrombospondin motifs) and MMP (matrix metalloproteinase) suggest that p21High quiescent cells may be more invasive than their proliferative counterparts.

Together, these data show that p21-dependent quiescence exists in TP53WT, but not TP53mutant, NSCLC. p21-High quiescent cells have unique transcriptional profiles suggesting they are a discrete cell state. There is considerable heterogeneity amongst cell lines in the fraction of cells in p21-dependent quiescence at any one time and within cell lines in the length of time spent in quiescence by individual cells. For the remainder of this work, we focus our efforts on trying to understand the causes and consequences of p21-dependent quiescence in NSCLC. Therefore, from here, we focus only on TP53WT NSCLC cells.

Replication stress can drive entry into p21-dependent quiescence

We, and others, have previously shown that a key cause of p21 upregulation and entry into p21-dependent quiescence is intrinsic DNA damage caused by DNA replication stress [8,9,10,11]. Therefore, we wanted to see if the spontaneously quiescent cells identified in TP53WT NSCLC had persistent DNA damage. We pulse-labelled cells with EdU 24 h prior to fixation and immunostained for p21 and 53BP1, the latter of which forms nuclear bodies around damaged DNA and is associated with persistent replication stress (Fig. 3a [41]). We identified EdU-negative or p21-High quiescent cells and quantified the fraction of these cells with at least one 53BP1 nuclear body, compared to EdU-positive and p21-Low proliferating cells (Fig. 3b). We consistently observed across TP53WT NSCLC cell lines that a higher fraction of EdU-negative and p21-High cells had at least one 53BP1 nuclear body, more than is observed in proliferating cells in the same population (Fig. 3c, Supplementary Fig. 5A). This suggests that entry into the quiescent state in TP53WT NSCLC is linked to persistent DNA damage.

Fig. 3: Replication stress can drive entry into p21-dependent quiescence.

a Images of NCI-H1563 cells labelled with EdU for 24 h and stained for p21 and 53BP1. Red circle highlights an EdU negative (quiescent) cell with a 53BP1 nuclear body and expressing p21. EdU is red, p21 is green, 53BP1 is orange and Hoechst is blue in merged image. Scale bar is 10 μm. b Graphs show the percentage of p21-High (upper panel) or EdU negative (lower panel) quiescent cells with a 53BP1 nuclear body, compared to proliferating cells. One repeat is shown and data are representative of n = 3. c Summary of percentage of quiescent cells (red) with at least one 53BP1 nuclear body, compared to proliferating cells (blue). Black line represents the mean of n = 3. Student’s t-test to test for significance, *p < 0.05, ns: not significant. d Percentage of cells arresting in quiescence after vehicle (grey dots) or low-dose (0.3 μM) aphidicolin (orange dots) treatment. Data are plotted as superplots, n = 3, 4 technical replicates per experiment. Black line is the mean of n = 3. Student’s t-test to test for significance, ****p < 0.0001, ***p < 0.001, **p < 0.01 *p < 0.05, ns: not significant. e Quantification of CDK2 activity in single A549 mRuby-PCNA CDK2L-GFP cells after DMSO or 0.3 μM aphidicolin treatment. All single-cell traces are aligned to the S-G2 transition. Blue curves represent cells that re-enter S-phase after DMSO or aphidicolin treatment, red curves represent cells that arrest in G1 and purple curves (just 1/20 cells in NTCsi + aphidicolin) represent cells arresting in G2. Percentages in red are the fraction of cells that enter an arrest state after aphidicolin treatment. NTC siRNA, 15/20 cells (75%). p21 siRNA, 3/14 cells (21.4%).

To test if replication stress can push cells into a quiescent state in our TP53WT NSCLC cell lines, we treated cells with a low dose of the DNA polymerase-α inhibitor, aphidicolin, to induce more replication stress. We verified that this dose of aphidicolin induced replication stress in these cell lines by quantifying the number of 53BP1 bodies per cell and the level of p21 expression, both of which increase after aphidicolin treatment (Supplementary Fig. 5B–E). We observed an increased fraction of cells arresting in quiescence after low-dose aphidicolin treatment (Fig. 3d). We also performed live cell imaging with our A549 mRuby-PCNA CDK2L-GFP cell line, treated with either a negative control or p21 siRNA and with or without aphidicolin treatment to enable us to monitor entry into quiescence in real-time (Fig. 3e). We observed that in the presence of p21 a higher fraction of A549 cells treated with aphidicolin enter a quiescent state post-mitosis (75% in NTCsi versus 21.4% in p21si).

Together our data suggest that replication stress correlates with a p21-dependent quiescent state and can drive entry into that state in TP53WT NSCLC cells.

Loss of p21 leads to propagation of DNA damage into S-phase, perturbation of cell cycle dynamics and spontaneous cell death

Tumour cells often exhibit high levels of DNA damage in the form of replication stress as a consequence of oncogene activation [42, 43]. Therefore, we wanted to examine the consequences of the loss of p21 on TP53WT NSCLC cells on genome stability and proliferative potential.

We depleted p21 using siRNA in mRuby-PCNA expressing A549 and NCI-H1944 cells to track cell cycle phenotypes and cell fates by timelapse imaging. After acute p21 depletion, we observed a higher rate of spontaneous cell death (Fig. 4a, b; Supplementary Fig. 6A). Cell death was not restricted to a single cell cycle phase. To test the hypothesis that p21 helps to maintain cell viability in TP53WT NSCLC cells, we used CRISPR/Cas9 to disrupt the p21 gene (CDKN1A) in our five TP53WT NSCLC cell lines to generate p21 knockout (p21KO) cell lines. We were only able to generate p21KO clonal lines in three of the five cell lines (A549 mRuby-PCNA, NCI-H1944 mRuby-PCNA and NCI-H460; Supplementary Fig. 6B, C), despite multiple targeting attempts. As expected, all p21KO cell lines had a reduced fraction of cells entering spontaneous quiescence (Supplementary Fig. 6D, E). We used timelapse imaging of p21WT and p21KO A549 and NCI-H1944 mRuby-PCNA cells to track cell cycle phenotypes and cell fates. Again, we observed increased spontaneous cell death in p21KO cell lines compared to p21WT cells (Fig. 4c). Of note, these rates of death were lower than those observed after acute depletion of p21 by siRNA (Fig. 4b) which may reflect p21KO cell lines adapting to p21 loss.

Fig. 4: Loss of p21 leads to propagation of DNA damage into S-phase, perturbation of cell cycle dynamics and spontaneous cell death.

a Cell cycle phase plots for NCI-H1944 mRuby-PCNA cells transfected with NTC or p21-targeting siRNA. 34 cells for NTCsi and 36 cells for p21si. 'D' stands for death. b Percentage of cells that die during the 72 h imaging period after acute depletion of p21 by siRNA for NCI-H1944 cells (left) and A549 cells (right). c Percentage of cells that die during the 72 h imaging period in p21KO cells in NCI-H1944 cells (left) and A549 cells (right). d Graph shows measurement of cell cycle phase lengths in A549 p21WT and p21KO cells. Each dot is the phase length in an individual cell, the red line represents the mean. Statistical significance was calculated using an unpaired t-test, between each cell cycle phase, with Welch’s correction. *p < 0.05, ****p < 0.0001. e Graph shows measurement of cell cycle phase length in NCI-H1944 p21WT and p21KO cells. Each dot is the phase length in an individual cell, the red line represents the mean. Statistical significance was calculated using an unpaired t-test, between each cell cycle phase, with Welch’s correction. ****p < 0.0001, ns = not significant. f Representative images of NCI-H1944 mRuby-PCNA p21KO cells pulse-labelled with EdU (red in merged image), immunostained for 53BP1 (green in merged image) and stained with Hoechst to label nuclei (blue in merged image). Purple-circled cells are those where 53BP1 bodies are present in S-phase. White circled cells are 53BP1 bodies present outside S-phase. Scale bar 20 μm. g Quantification of percentage of S-phase cells with at least one 53BP1 nuclear body present. Data are presented as superplots, n = 3 with 4 technical replicates. Black line represents the mean. Statistical significance was calculated using a one-way ANOVA, between each p21KO clone and the p21WT counterpart. ns = not significant, *p < 0.05, ****p < 0.0001.

We analysed cell cycle phase lengths to assess the impact of p21 loss. In A549 mRuby-PCNA cells, we observed a reduction in G1 length, largely due to the loss of cells entering the prolonged G0/G1-like state of quiescence, an increase in the length of S-phase, and a decrease in the lengths of G2 and mitosis (Fig. 4d). We observed a similar increase in S-phase length in p21KO NCI-H1944 mRuby-PCNA cells compared to p21WT cells (Fig. 4e). We hypothesised that with a reduced ability to enter quiescence, which would normally allow cells to repair any inherited DNA damage, more DNA damage may be propagated into S-phase and that this may contribute to a longer S-phase length. Therefore, we quantified the levels of DNA damage in p21WT and p21KO NSCLC cell lines. We saw an overall increase in the levels of nuclear γH2AX in p21KO versus p21WT cells, consistent with an overall increase in genomic instability in the absence of p21-dependent quiescence (Supplementary Fig. 6F). To identify damage specifically in S-phase cells, we pulse-labelled cells with EdU for 30 min before fixation and immunostained with 53BP1 (Fig. 4f). We observed an increase in the fraction of S-phase cells with DNA damage in p21KO versus p21WT cells, that was statistically significant in two out of the three p21KO lines (Fig. 4g). This suggests that more DNA damage is being propagated into S-phase cells in the absence of p21-dependent quiescence.

Our data suggest that p21-dependent quiescence is required for the maintenance of genome stability, limiting the amount of DNA damage being propagated into S-phase cells, and that p21 acts to maintain the proliferative potential of TP53WT NSCLC cells by promoting cell fitness.

Loss of p21 sensitises TP53 wild-type NSCLC cells to chemotherapy

Since quiescence can protect cancer cells from chemotherapy agents that target cycling cells, we wanted to see if loss of p21-dependent quiescence in TP53WT NSCLC cells would sensitise them to chemotherapy.

We performed dose curves in A549, NCI-H460 and NCI-H1944 cell lines to determine sensitivity to three chemotherapy agents used to treat NSCLC: cisplatin, gemcitabine and etoposide. We observed a small fraction of cells remaining in all cell lines, even at the highest doses of drug (grey boxes, Supplementary Fig. 7A). The fraction of cells surviving drug treatment was positively correlated with the fraction of quiescent cells in each cell line (Fig. 5a, Supplementary Fig. 7B). To determine if p21 had an impact on the response of these cell lines to chemotherapy, we compared survival of p21WT and p21KO cell lines treated with cisplatin, gemcitabine and etoposide. We observed that p21KO A549 and NCI-H1944 cells were significantly more sensitive to etoposide and gemcitabine than p21WT cells, while p21KO NCI-H460 cells were significantly more sensitive to cisplatin than p21WT cells (Fig. 5b, Supplementary Fig. 7C).

Fig. 5: Loss of p21 sensitises TP53 wild-type NSCLC cells to chemotherapy.

a Correlation between fraction of cells surviving etoposide treatment and quiescent fraction in that cell line. Cells were treated for three days with 5 μM etoposide. Mean ± std for each repeat plotted and n = 3 for each cell line. b p21WT and p21KO cell lines were treated for three days with 5 μM of etoposide and surviving fraction was calculated. Data are plotted as superplots, n = 3 biological repeats and four technical repeats per experiment. Black line represents the mean. One-way ANOVA was used to calculate statistical significance. ****p < 0.0001, ns = not significant. c Cell cycle phase plots for A549 mRuby-PCNA p21WT (upper panels) and p21KO (lower panels) cells treated with gemcitabine or cisplatin for 3 days. Cells were imaged for two days before drug addition (marked by vertical black lines at 2880 mins). Tables underneath show the percentage of cells that started in each fate upon drug addition (G1, S or G2) and their end fate (G1, S, G2 or death (d)). d Graphs showing percentage of cells surviving drug treatment (gemcitabine, etoposide or cisplatin) in quiescence. Cells were arrested in quiescence for one day in CDK2/4/6i prior to three days treatment with chemotherapy agent (still in the presence of CDK2/4/6i). Data are plotted as superplots, n = 3 biological replicates with three technical replicates per repeat. Black line represents the mean. Student’s t-test to test for significance. ****p < 0.0001, **p < 0.01.

To determine if it was the p21-dependent quiescent G0/G1 state that we observe in unperturbed cell populations that confers chemo-protection, or another function of p21, we used timelapse imaging. We imaged A549 and NCI-H1944 mRuby-PCNA expressing cells for two days to establish the cell cycle state before drug treatment. We then added either gemcitabine or cisplatin and continued to image for three days to track cell fates (Fig. 5c, Supplementary Fig. 7D). Cells that started in G1 upon drug treatment were equally likely to die as those cells in S or G2, suggesting residing in p21-dependent quiescence prior to drug treatment does not protect cells from chemotherapy. For gemcitabine, which targets DNA synthesis in S-phase, this is perhaps not too surprising and indeed, gemcitabine-treated cells that were in G1 at time of treatment, complete S-phase before arresting in G2 or dying. However, for cisplatin, which forms DNA adducts and blocks DNA repair, and therefore can affect cells at all cell cycle stages, we found this more surprising. Most cisplatin-treated G1 cells arrest in G2 or die (Fig. 5c; Supplementary Fig. 7D), perhaps because cells in G1 do not mount a sufficient DNA damage response to cisplatin and so progress into the cell cycle. Specifically in A549 cells, we did observe that approximately one-third of cells that were in G2 upon cisplatin addition survived in a p21-dependent G0/G1 state and that cells in any state upon either gemcitabine or cisplatin treatment could enter a p21-dependent G2 arrest (Fig. 5c).

We next asked if TP53WT NSCLC cells held in quiescence for the duration of chemotherapy treatments would be protected from chemotherapy. Addition of the CDK2/4/6 inhibitor, ebvaciclib [44], was able to arrest NCI-H1944 cells in quiescence (Supplementary Fig. 7E). Therefore, we held NCI-H1944 cells in quiescence with CDK2/4/6 inhibitor and then treated arrested cells with gemcitabine, etoposide or cisplatin for three days. While cisplatin was able to kill quiescent NSCLC cells as efficiently as it killed proliferating cells, gemcitabine and etoposide were not (Fig. 5d). These results are consistent with their mechanism of action and suggest that a sufficiently prolonged quiescence would not be sufficient to protect cells from cisplatin but may protect them from etoposide and gemcitabine.

Our data show that loss of p21 sensitises cells to chemotherapy. This is not due to a chemoprotective action of cells residing in a p21-dependent quiescent state during drug treatment as this state may be too transient to confer protection but instead is due to p21 promoting cytoprotective prolonged G1 and G2 arrests.

p21-dependent quiescence could drive tumour relapse

The prolonged p21-dependent G1 and G2 arrest states observed after chemotherapy treatment raise the question of how stable these arrest states are i.e. are these cells transiently arrested (quiescent) or terminally arrested (senescent)? These arrested cells express high levels of p21 and have enlarged nuclei, suggesting that these surviving cells could be in, or progressing towards, senescence (Fig. 6a). However, these features are shared with quiescent cells [45]. Therefore, we wanted to investigate if a population of p21-dependent quiescent cells were present post-drug treatment by determining if surviving cells can re-initiate cell proliferation, as a model of tumour relapse.

Fig. 6: p21-dependent quiescence could drive tumour relapse.

a Representative images and quantification from A549 cells showing increase in p21 levels and nuclear area three days post-treatment with vehicle or gemcitabine. Hoechst is in blue and p21 is in red in merged images. Scale bar is 100 μm. Graphs on right are single-cell quantifications of p21 nuclear intensity (upper panels) and nuclear area (lower panels) three days post-treatment with indicated drugs. Veh—vehicle, Cis—cisplatin, Eto—etoposide and Gem—gemcitabine. b Representative graph of cell growth over time of NCI-H1944 p21WT (red curves) and p21KO cells (blue curves) treated with vehicle (solid line) or gemcitabine (dashed line) taken from one field of view. Cells were treated with drug for three days before the drug was washed out (‘wash’) and replaced with normal growth media to track any cell re-growth. c Quantification of percentage of well covered by NCI-H1944 nuclei at the end of the relapse assay (28 days). Mean ± stdev of n = 3 are shown and individual wells are plotted for each technical repeat. Student’s t-test to test for significance. ****p < 0.0001, **p < 0.01, ns: not significant. d Representative graph (upper panel) and mean ± stdev of n = 3 (lower panel) of growth of cells over time after sorting G1 and G2 populations. Area was calculated from DPC images. Student’s t-tests show no significant difference in growth between G1 and G2 populations at any timepoint across n = 3. e Representative images of Hoechst-stained NCI-H1944 nuclei from vehicle- or gemcitabine-treated G1 and G2 cells allowed to recover from drug treatment for three weeks. Scale bar is 1 mm. Quantification of percentage of well covered by nuclei three weeks after sorting either G1 or G2 vehicle- or gemcitabine-treated cells and replating in drug-free media. Mean ± stdev of n = 3 are shown with 3 technical repeats. Student’s t-test to test for significance between G1 and G2 cells, ns: not significant.

We treated p21WT and p21KO A549, NCI-H1944 and NCI-H460 cells for three days with cisplatin or gemcitabine before washing out the drug and imaging long-term cell growth. After the initial wave of drug-induced cell death and several days of no proliferation, p21WT cells started to proliferate again, a phenomenon that was more rare in p21KO cells (Fig. 6b, c). This was not a widespread event but individual colonies form and can eventually recolonise the well (Fig. 6c, Supplementary Fig. 8A–C, Supplementary Movies 3–6). These data suggest that p21 can maintain a quiescent pool of cells after drug treatment that can drive population regrowth.

Since our single-cell imaging assays had identified that post-chemotherapy treatment, cells could arrest in G1 or, more frequently, G2 (Fig. 5c, Supplementary Fig. 7D), we wanted to ask if cells arresting in G1 or G2 were equally likely to re-enter the cell cycle after drug removal in p21WT cells. We also noticed that the colonies that regrew had different morphologies. Some colonies contained cells that were large and flat while other colonies contained cells that more closely resembled untreated cells (Supplementary Fig. 8D, blue bounding edge) and we wondered if these different cell morphologies represented whether cells had re-entered the cell cycle from G1 or G2. Therefore, we treated p21WT NCI-H1944 cells for three days with gemcitabine and then used FACS to sort individual G1 or G2 cells into 96-well plates (Supplementary Fig.