ABSTRACT
Cellular invasion is a key part of development, immunity and disease. Using an in vivo model of Caenorhabditis elegans anchor cell invasion, we characterize the gene regulatory network that promotes cell invasion. The anchor cell is initially specified in a stochastic cell fate decision mediated by Notch signaling. Previous research has identified four conserved transcription factors, fos-1 (Fos), egl-43 (EVI1/MEL), hlh-2 (E/Daughterless) and nhr-67 (NR2E1/TLX), that mediate anchor cell specification and/or invasive behavior. Connections between these transcription factors and the underlying cell biology that they regulate are poorly understood. Here, using genome editing and RNA interference, we examine transcription factor interactions before and after anchor cell specification. Initially, these transcription factors function independently of one another to regulate LIN-12 (Notch) activity. Following anchor cell specification, egl-43, hlh-2 and nhr-67 function largely parallel to fos-1 in a type I coherent feed-forward loop with positive feedback to promote invasion. Together, these results demonstrate that the same transcription factors can function in cell fate specification and differentiated cell behavior, and that a gene regulatory network can be rapidly assembled to reinforce a post-mitotic, pro-invasive state.
INTRODUCTION
Invasion through basement membranes (BMs) is a cellular behavior that is integral to embryo placentation, tissue patterning during embryonic development, and immune response to infection and injury (Medwig and Matus, 2017; Rowe and Weiss, 2008). Increased cellular invasiveness is also a hallmark of metastatic cancer (Hanahan and Weinberg, 2011). Previous research has identified several cell-autonomous mechanisms that are highly conserved across different contexts of BM invasion (Kelley et al., 2014; Medwig and Matus, 2017). These include localization of F-actin and upregulation of matrix metalloproteinases (MMPs) to chemically degrade the BM (Hagedorn et al., 2009, 2013; Kelley et al., 2019; Levitan and Greenwald, 1998; Morrissey et al., 2014; Sherwood et al., 2005). There is also growing evidence that cells must undergo cell cycle arrest in order to invade the BM (Kohrman and Matus, 2017; Matus et al., 2015). How these tightly coordinated programs are transcriptionally regulated is not well understood.
As many contexts of cellular invasion occur deep within tissue layers and are thus difficult to visualize, we utilize morphogenesis of the Caenorhabditis elegans uterine-vulval connection as a genetically tractable and visually amenable model for examining cell invasion in vivo (Sherwood and Sternberg, 2003). During the mid-L3 stage, a specialized uterine cell called the anchor cell (AC), invades the underlying BM, connecting the uterus to the vulval epithelium to facilitate egg laying (Fig. 1A,B) (Sherwood and Sternberg, 2003). The AC itself is specified in a cell fate decision event earlier in development, during the L2 stage, when two initially equipotent cells diverge via stochastic Notch asymmetry, giving rise to the terminally differentiated AC and a proliferative ventral uterine (VU) cell (Fig. 1A,B) (Wilkinson et al., 1994).
Prior research has identified four pro-invasive transcription factors (TFs) that function cell-autonomously to regulate AC invasion (Fig. 1A). These include the basic leucine zipper TF fos-1 (Fos), the basic helix-loop-helix TF hlh-2 (E/Daughterless), the nuclear hormone receptor nhr-67 (NR2E1/Tailless/TLX) and the zinc-finger TF egl-43 (EVI1/MEL) (Hwang et al., 2007; Matus et al., 2010; Rimann and Hajnal, 2007; Schindler and Sherwood, 2011; Sherwood et al., 2005; Verghese et al., 2011). NHR-67 functions upstream of the cyclin-dependent kinase inhibitor CKI-1 (p21/p27), to induce G1/G0 cell cycle arrest, which is necessary for AC invasion (Matus et al., 2015). Independently of NHR-67 activity, FOS-1A regulates the expression of the long isoform of egl-43 (EGL-43L) and downstream effectors including multiple MMPs (zmp-1, -3 and -6) and a cadherin (cdh-3) (Hwang et al., 2007; Kelley et al., 2019; Matus et al., 2010; Rimann and Hajnal, 2007). HLH-2 independently regulates cdh-3 along with other targets and cytoskeletal polarity (Schindler and Sherwood, 2011). Prior work has also suggested that EGL-43 and HLH-2 may regulate NHR-67 based on conserved HLH-2 binding motifs present in the nhr-67 promoter as well as perturbation experiments using nhr-67 transgenic reporters (Bodofsky et al., 2018; Verghese et al., 2011). Interestingly, three of these TFs (egl-43, hlh-2 and nhr-67) also play key roles in the specification of the AC and VU cell fates during the L2 stage (Attner et al., 2019; Hwang et al., 2007; Karp and Greenwald, 2004; Sallee et al., 2017; Verghese et al., 2011). How these four TFs function, independently and together, to regulate both specification and invasive activity of the differentiated AC is poorly understood.
Here, using new highly efficient RNA interference (RNAi) vectors (Sturm et al., 2018), we identify novel roles for egl-43 and hlh-2 in maintaining the AC in a cell cycle-arrested state. Additionally, we generated GFP knock-in alleles for each pro-invasive TF using CRISPR/Cas9-mediated genome engineering. These tools allow for quantitative measures of endogenous protein in the context of native chromatin architecture and potential distal enhancer elements that may be misrepresented using traditional transgenic reporters (Dickinson and Goldstein, 2016; Dickinson et al., 2013; Kim et al., 2014). Using these alleles and improved RNAi constructs, we report the onset of relative expression of pro-invasive TFs in the lineages leading to the AC and dissect their molecular epistatic interactions. We find that, prior to AC/VU specification, egl-43, hlh-2 and nhr-67 function independently to regulate an endogenously tagged lin-12::GFP allele. This network appears to be assembled following specification, with the addition of a fourth node, fos-1, and activation of TF interactions. Furthermore, we characterized both a cell cycle-independent circuit and a cell cycle-dependent feed-forward regulatory circuit with positive feedback that is crucial for maintaining the AC in an invasive, post-mitotic state. These findings provide new insights into how a set of reiteratively used TFs can be re-purposed from regulating cell fate specification to coordinating a differentiated cellular behavior.
RESULTS
EGL-43 or HLH-2 RNAi depletion results in extra ACs
Our initial goal was to determine interactions between the four pro-invasive TFs fos-1, egl-43, hlh-2 and nhr-67. To accomplish this, we first generated new RNAi targeting constructs in the improved RNAi vector, T444T, which includes T7 termination sequences to prevent the generation of non-specific RNA fragments from the vector backbone, increasing the efficacy of gene silencing over the original RNAi targeting vector, L4440 (Fig. S1A) (Sturm et al., 2018). As three of four pro-invasive TFs (egl-43, hlh-2 and nhr-67) also function during AC specification (Attner et al., 2019; Hwang et al., 2007; Karp and Greenwald, 2004; Sallee et al., 2017; Verghese et al., 2011), we assessed TF depletions in a genetic background in which only the AC and neighboring uterine cells are sensitive to RNAi following the AC/VU decision (Matus et al., 2010). The uterine-specific RNAi-sensitive strain was generated through tissue-specific restoration of the RDE-1 Piwi/Argonaut protein in an rde-1 mutant background using the fos-1 promoter, which is expressed specifically in the somatic gonad during the late L2/early L3 stage of larval development (Haerty et al., 2008; Hagedorn et al., 2009; Matus et al., 2010, 2015). AC invasion was quantified as the presence or absence of a BM gap, visualized by laminin::GFP, at the P6.p four-cell stage, a developmental window in which 100% of wild-type ACs are invaded (Fig. 1C, Table S1). We also assessed invasion several hours later, following vulval morphogenesis, allowing us to distinguish between delays in invasion and complete loss of invasive capacity (Fig. 1E, Fig. S1D, Table S1). Depletion of each TF resulted in invasion defects ranging from moderate penetrance (38%; fos-1) to highly penetrant (69-94%; hlh-2, egl-43 and nhr-67), when synchronized L1-stage animals were plated on RNAi bacteria and scored at the mid-L3 stage (P6.p four-cell stage) and early L4 stage (P6.p eight-cell stage) (Fig. 1C-E, Fig. S1D). Consistent with previous work, depletion of fos-1 resulted in single ACs that failed to breach the BM. Depletion of nhr-67 resulted in proliferative, non-invasive ACs at a high penetrance. Depletion of egl-43 phenocopied nhr-67, with a highly penetrant defect of multiple, non-invasive ACs. Lastly, depletion of hlh-2 resulted in pleiotropic phenotypes in the AC, from animals with zero, one, two or even, in several cases, three or more ACs (Fig. 1D, Fig. S1C). In all cases, the penetrance of AC invasion defects was significantly increased using a T444T-based RNAi targeting vector compared with L4440-based vectors (Fig. S1A). Thus, use of the improved RNAi vector revealed unreported phenotypes resulting from depletion of egl-43 and hlh-2 during invasion, phenocopying depletion of nhr-67 resulting in multiple non-invasive ACs (Matus et al., 2015).
Because of the pleiotropy associated with hlh-2 depletion, which affects AC specification and invasion (Schindler and Sherwood, 2011), we sought to rule out the possibility that the extra cdh-3-expressing ACs are due to defects in AC specification. Depletions using the traditional RNAi vector L4440 in the uterine-specific RNAi-sensitive background is sufficient to bypass AC fate specification defects that occur in genetic backgrounds in which all cells are sensitive to RNAi (Matus et al., 2010). Specifically, loss of HLH-2 prior to AC specification disrupts the lin-12(Notch)/lag-2(Delta)-dependent signaling cascade, resulting in both pre-AC/VU cells adopting a VU fate and thereby producing a zero AC phenotype. During AC specification, loss of HLH-2 results in upregulation of pro-AC fate by activating lag-2/Delta resulting in both AC/VU cells adopting an AC fate (Karp and Greenwald, 2004; Schindler and Sherwood, 2011). Notably, hlh-2 depletion using the T444T vector resulted in a greater occurrence of a zero AC phenotype compared with the L4440 hlh-2 RNAi clone (Fig. 1D, Fig. S1A). This is likely due to the improved efficiency of RNAi-mediated knockdown using the enhanced T444T vector (Martinez et al., 2019), as the time period between restoration of RDE-1 function and the timing of the AC/VU decision appears decreased. Thus, to bypass these potential specification defects due to increased RNAi penetrance for hlh-2 using the T444T vector, we depleted hlh-2 and the other TFs by RNAi using an L2-plating strategy (Schindler and Sherwood 2011), growing animals to the time of the L1/L2 molt on OP50 Escherichia coli and then transferring them to TF-RNAi plates. L2 platings resulted in a lower overall penetrance of AC invasion defects for all TF-RNAi treatments (Fig. S1B), including hlh-2 (43% at the P6.p four-cell stage; Fig. 1D) but all animals possessed an AC. Strikingly, some animals possessed multiple cdh-3-expressing ACs at the P6.p four-cell stage (Fig. 1C), and 10% of animals in the early L4 stage possessed multiple ACs (Fig. 1C-E, Fig. S1C,D). As the number of ACs increased over developmental time (Fig. 1D,E), these results suggest that the multiple ACs observed could be arising through loss of G1/G0-cell cycle arrest and nhr-67 activity (Matus et al., 2015).
Characterization of GFP-tagged alleles and RNAi penetrance
Next, we sought to examine the relationship of these four TFs in the AC prior to and during invasion by combining gene knockdown with endogenous GFP-tagged alleles. To accomplish this, we used CRISPR/Cas9 genome-editing technology to knock in a codon-optimized GFP tag into the endogenous locus of each TF (Fig. S2, Tables S4-S6) (Dickinson and Goldstein, 2016). First, we examined a minimum of 50 L1/L2-stage animals to determine the onset of expression for each TF in the somatic gonad, which is derived from two founder cells, Z1 and Z4 (Kimble and Hirsh, 1979). The two cells that can stochastically give rise to the AC are the proximal great-granddaughters of Z1 and Z4 (Z1.ppp and Z4.aaa). Examination of L1- and L2-stage somatic gonads revealed that egl-43::GFP::egl-43 turns on first, in Z1 and Z4, and remains on in their descendants through the time of AC/VU specification (Fig. 2A). As reported recently (Attner et al., 2019), GFP::hlh-2 turns on at the second division of Z1/Z4 cells in Z1.pp and Z4.aa. We also detected onset of nhr-67::GFP in these cells in some animals, though it appears that their expression is weaker in Z1.pp/Z4.aa compared with GFP::hlh-2, but becomes robust in the AC/VU cells prior to specification. Finally, we detected weak expression of GFP::fos-1 in L1 and L2 Z1/Z4 descendants that increased rapidly in the somatic gonad following AC specification (Fig. 2A). In summary, consistent with previously reported roles for egl-43, hlh-2 and nhr-67 in AC/VU specification (Hwang et al., 2007; Karp and Greenwald, 2004; Verghese et al., 2011), we detected the presence of endogenous GFP-tagged TFs in Z1/Z4 descendants.
Next, we examined the expression patterns of TF::GFP alleles following AC specification. All four TFs showed robust GFP localization in the AC nucleus prior to and during invasion (Fig. 2B). Additionally, the GFP-tagged strains had similar expression domains in the somatic gonad during the L3 larval stage as previously reported by traditional multi-copy array transgenes (Hwang et al., 2007; Rimann and Hajnal, 2007; Sherwood et al., 2005; Verghese et al., 2011). After examining expression patterns, we then synchronized TF GFP-tagged strains and collected a developmental time course quantifying the expression of GFP-tagged protein in the AC during uterine/vulval development (Fig. 2B). All GFP-tagged TFs were imaged using uniform acquisition settings on a spinning disk confocal microscope using an EM-CCD camera, allowing us to compare relative expression levels of endogenous GFP-tagged proteins in the AC. Interestingly, all four TFs followed the same general trend in expression levels, with a gradual increase in levels prior to invasion, peaking at or just before the P6.p four-cell stage when 100% of wild-type animals have generated a gap in the BM. Expression levels then decreased following invasion, during vulval morphogenesis, in the early L4 stage (P6.p eight-cell stage) (Fig. 2C,D).
To visualize invasion, we first crossed a BM reporter (laminin::GFP) and an AC marker (cdh-3>mCherry:moeABD; see Materials and Methods for transgene nomenclature) into each endogenously GFP-tagged TF strain. We then performed a series of RNAi depletion experiments targeting each TF and examining synchronized animals at the P6.p 4-cell stage. As the insertion of GFP tags into native genomic loci can potentially interfere with gene function, we examined a minimum of 50 animals treated with control (T444T) empty vector RNAi (Fig. S3). All four GFP-tagged strains showed 100% BM breach at the normal time of invasion in control animals (empty vector) (Fig. S3) and the animals all appeared superficially wild type. Additionally, all alleles generated did not appear to affect viability or fertility. Although this does not rule out other more subtle hypomorphic phenotypes, for the purpose of determining TF regulatory relationships during AC invasion, we consider all alleles generated to be equivalent to wild-type AC invasion.
As RNAi penetrance can be difficult to measure at the single cell level across a population, we next assessed the efficacy of our newly generated TF-RNAi vectors quantitatively. We examined a minimum of 50 animals following targeted TF-RNAi depletions, collecting spinning disk confocal z-stacks using the same acquisition settings as used for the developmental series (Fig. 2B) in all experiments. Following image quantification, our results were consistent with the phenotypes identified in our original uterine-specific RNAi screen (Fig. 1, Fig. S3). However, by being able to quantify depletion of endogenous GFP-tagged TF targeted by RNAi, we were able to correlate phenotype with mRNA depletion more accurately. The nhr-67 and egl-43 improved RNAi constructs strongly knocked down their GFP-tagged endogenous targets, with depletions averaging 88% and 75%, respectively (Fig. S3E). This strong knockdown of nhr-67 and egl-43 was also correlated with highly penetrant AC invasion defects (61% and 100%, respectively). The completely penetrant defect from egl-43(RNAi) segregated into 26% of animals exhibiting a single AC that failed to invade and all other cases phenocopying loss of nhr-67, with multiple cdh-3-expressing non-invasive ACs (Fig. S3E). Depletion of GFP::fos-1 by RNAi was also highly penetrant, with a mean depletion of 92% of GFP-tagged endogenous protein and 76% of treated animals with a defect in AC invasion (Fig. S3E). Although only loss of hlh-2 prior to AC specification can affect AC fate directly resulting in zero ACs (Karp and Greenwald, 2004), nhr-67 and egl-43 also regulate the AC/VU decision (Hwang et al., 2007; Verghese et al., 2011). Thus, to ensure that the invasion defects we were observing were the result of post-AC specification defects following TF knockdown, we performed L2 platings and scored for invasion. Because L2 platings result in less time (∼12 h at 25°C) exposed to TF-RNAi, we were not surprised to observe a lower penetrance of AC invasion defects compared with L1 platings targeting the other TFs (Fig. S3F). As all L2 platings phenocopied L1 platings, although at lower penetrance and with only depletion of hlh-2 at the L1 stage directly affecting AC fate, we utilized L1 platings for their increased penetrance for nhr-67, egl-43 and fos-1 for the remainder of our experiments. Together, these results suggest that following strong TF depletion, loss of either egl-43 or hlh-2 phenocopies nhr-67 depletion, generating multiple, non-invasive ACs.
Identification of a feed-forward loop controlling NHR-67 activity
Next, we examined the relationship between the four pro-invasive TFs during invasion. To accomplish this, we performed a series of RNAi depletion experiments using synchronized L1 or L2 (for hlh-2) stage animals and quantified the amount of GFP-tagged endogenous TF in the AC in animals with defects in invasion at the P6.p four-cell stage. Whereas depletion of fos-1 failed to significantly downregulate levels of nhr-67::GFP, depletion of egl-43 resulted in a strong reduction (65%) of nhr-67::GFP in the AC (Fig. 3A,B). Intriguingly, in animals with a single non-invasive AC following hlh-2(RNAi) we saw only a partial reduction (19%) of nhr-67::GFP levels, whereas in animals with multiple cdh-3-expressing cells we saw a significant reduction (49%) in nhr-67::GFP (Fig. 5A,B). These results suggest that both egl-43 and hlh-2 co-regulate nhr-67 during invasion.
We repeated these TF-RNAi molecular epistasis experiments with the remaining three TF-GFP-tagged strains and quantified depletion of GFP in animals with invasion defects. Consistent with previous studies using transcriptional and translational reporters (Hwang et al., 2007; Rimann and Hajnal, 2007), we found that depletion of fos-1 regulated levels of egl-43::GFP::egl-43 (44% depletion) (Fig. 3C,D). No other TF depletion significantly regulated the levels of egl-43 in our experiments (Fig. 3C,D). hlh-2 is predicted to regulate egl-43 activity based on the presence of two conserved E-box binding motifs in the egl-43 promoter (Hwang et al., 2007). However, we did not detect regulation of egl-43::GFP::egl-43 in either the AC (Fig. 3C,D) or neighboring VU cells (Fig. S4). Next, we examined GFP::hlh-2 levels following TF-RNAi depletions. Intriguingly, we found a similar pattern of regulation based on AC phenotype. Animals with single non-invasive ACs showed partial depletion of GFP::hlh-2 following fos-1, egl-43 or nhr-67 depletion (35%, 20% and 14%, respectively). However, animals with multiple cdh-3-expressing ACs following egl-43(RNAi) or nhr-67(RNAi) showed stronger depletion of GFP::hlh-2 (66% and 73%, respectively) (Fig. 3E,F). Finally, we examined levels of GFP::fos-1 following TF depletions. Depletion of either nhr-67 or egl-43 resulted in partial reduction of GFP::fos-1 levels in the AC (52% and 49%, respectively) (Fig. 3G,H). Together, these results suggest that egl-43, hlh-2 and nhr-67 may function together in a feed-forward regulatory loop with positive feedback to maintain the AC in a post-mitotic, pro-invasive state.
Multiple ACs derive from proliferation
As depletion of egl-43 and hlh-2 both regulate levels of nhr-67 and their depletion results in multiple ACs that fail to invade, we next wanted to assess whether their depletion was functionally phenocopying depletion of nhr-67. To confirm that the presence of multiple cells expressing the cdh-3-driven AC reporter were due to defects in proliferation, we performed static imaging using a full-length translational cdt-1>CDT-1::GFP reporter that indicates cell cycle progression (Fig. 4A). As CDT-1 must be removed from origins of replication during DNA licensing, the transgene localizes to the nucleus during G1/G0 and is largely cytosolic at the onset of S phase (Matus et al., 2014, 2015). As expected, T444T (empty vector)- and fos-1(RNAi)-treated animals consistently exhibited nuclear localization of CDT-1::GFP [Fig. 4A; 100%, T444T, n=15; 100%, fos-1(RNAi), n=19 animals examined], indicating cell cycle arrest. However, following depletion of egl-43, hlh-2 and nhr-67 we identified multiple animals (Fig. 4A) possessing non-invasive ACs lacking nuclear CDT-1::GFP, indicative of cycling ACs [67%, egl-43(RNAi), n=12/18; 58%, hlh-2(RNAi), n=11/19; 71%, nhr-67(RNAi), n=10/14 animals examined]. Together, these results strongly support our molecular epistasis data suggesting that egl-43 and hlh-2 function upstream of nhr-67 to maintain the post-mitotic state of the AC during invasion.
After examining the role of pro-invasive TFs in regulating cell cycle progression, we then sought to examine other downstream targets involved in AC invasion. Using an endogenously tagged GFP reporter for the MMP zmp-1 (Kelley et al., 2019), we assessed the ability of ACs to produce MMPs following TF(RNAi) depletion. We confirmed previously published findings (Kelley et al., 2019) that fos-1 depletion results in reduction of zmp-1 expression (Fig. 4B). Interestingly, in the cases of the other TF(RNAi) treatments, a significant reduction of zmp-1 expression was only observed when multiple cdh-3-expressing ACs were present (Fig. 4B), suggesting that cell cycle progression may downregulate production of MMPs.
EGL-43 and HLH-2 play roles in cell cycle-dependent and independent pathways
We have shown previously that the invasive activity of the AC following depletion of NHR-67 can be completely rescued by restoring the AC to a post-mitotic G1/G0 state through upregulation of the cyclin-dependent kinase inhibitor CKI-1 (p21/p27) (Matus et al., 2015). As our epistasis experiments revealed that egl-43 and hlh-2 positively co-regulate NHR-67 activity, we induced AC-specific expression of CKI-1 using a cdh-3>CKI-1::GFP integrated array and assessed invasion following RNAi depletions (Fig. 5A,B). As expected, induced expression of CKI-1::GFP in the AC completely rescued nhr-67(RNAi)-treated animals [Fig. 5A,B; 100% invaded (n=42) compared with 45% of control animals lacking induced CKI-1::GFP (n=62)]. Additionally, cdh-3>CKI-1::GFP partially rescued hlh-2 depletion in L2 RNAi-feeding experiments by restoring AC invasion in most animals examined [Fig. 5A,B; 87% invaded (n=77), compared with 56% of control animals (n=59)]. Inducing G1/G0 arrest in the AC failed to rescue fos-1(RNAi)-treated animals [Fig. 5A,B; 20% invaded (n=35), compared with 51% of control animals (n=53)]. AC-specific expression of CKI-1::GFP blocked AC proliferation following egl-43(RNAi); however, this induced arrest failed to rescue invasion [Fig. 5A,B; 10% invaded (n=31), compared with 19% of control animals (n=37)]. This result suggests that egl-43 has pro-invasive functions outside of the cell cycle-dependent pathway. In support of a role for egl-43 outside of cell cycle control, induced CKI-1::GFP restored the expression of a reporter of MMP activity (zmp-1>mCherry) in nhr-67-depleted animals, but neither fos-1 nor egl-43-depleted animals showed restoration of zmp-1 transcription (Fig. 5C). Together, these results reveal the presence of two integrated sub-circuits necessary for invasive activity, with egl-43 and hlh-2 likely having a crucial role in both circuits.
We next set out to test our putative network topology. To do this, we generated two new single copy transgenic lines expressing either BFP-tagged codon-optimized HLH-2 or the genomic region of NHR-67 under the transcriptional control of a heat shock promoter. Induced expression of HLH-2 failed to rescue AC invasion following depletion of egl-43, fos-1 or nhr-67, consistent with hlh-2 functioning upstream of nhr-67 and the cell cycle-independent roles of egl-43 and fos-1 in regulating AC invasion (Fig. S5A). Similarly, induced expression of NHR-67 did not rescue invasion in egl-43(RNAi)- and fos-1(RNAi)-treated animals (Fig. S5B). Furthermore, induced expression of NHR-67 failed to rescue AC invasion in hlh-2(RNAi)-treated animals (Fig. S5B). This result, paired with the prevalence of single non-invasive ACs (Fig. 1C) and only partial rescue from induced CKI-1 (Fig. 5A,B) suggests that hlh-2 regulates invasive targets independently of its role in regulating nhr-67 and cell cycle arrest.
As our induced TF experiments failed to rescue invasion defects, we examined an nhr-67 hypomorphic allele [nhr-67(pf88)], which we have previously shown results in an incomplete penetrance of mitotic, non-invasive ACs (Matus et al., 2015). Notably, nhr-67(RNAi) using the enhanced T444T vector did not significantly increase the invasion defect of the hypomorphic allele (Fig. S6), supporting previous results using the L4440-based nhr-67 RNAi vector (Matus et al., 2015). Thus, the nhr-67(pf88) allele may represent a putative null phenotype for AC invasion. Similar to our results with nhr-67(RNAi), depletion of hlh-2 also failed to enhance the nhr-67(pf88) AC invasion defect significantly (Fig. S6). Together, these results suggest that hlh-2 functions upstream of nhr-67 in maintaining the AC in a post-mitotic, pro-invasive state.
EGL-43 isoforms function redundantly and in an autoregulative manner
Two functional isoforms of egl-43 are encoded in the C. elegans genome (Fig. S7A). Previous research has suggested that the longer isoform functions downstream of fos-1 to modulate MMP expression and other fos-1 transcriptional targets, including cdh-3 (Hwang et al., 2007; Rimann and Hajnal, 2007). The shorter isoform has been predicted to function in Notch/Delta-mediated AC specification and later Notch/Delta-mediated patterning of the ventral uterus (Hwang et al., 2007) or as a potential competitive inhibitor for long isoform binding of downstream targets (Rimann and Hajnal, 2007). Additionally, fos-1 and egl-43 have been shown to function in an incoherent feed-forward loop with negative feedback, with fos-1 positively regulating and egl-43 negatively regulating the levels of mig-10/lamellipodin, a key adhesion protein important for stabilizing the attachment of the AC to the BM. Titrating levels of MIG-10 is crucial, as overexpression of MIG-10 results in AC invasion defects (Wang et al., 2014). Although it is readily apparent that levels of EGL-43 are also important, it is unclear from previous studies whether the two isoforms have divergent functions during invasion. Thus, we next decided to explore egl-43 isoform function.
First, we generated a knock-in allele of GFP at the egl-43 N-terminus to tag the long isoform (GFP::egl-43L) specifically. This allowed us to compare expression patterns of the long isoform to the internally GFP-tagged allele that should dually label both isoforms (Fig. S7A,B). We examined animals during uterine-vulval development and found overlapping expression patterns between both isoforms with strong expression throughout the somatic gonad (Fig. S7B). Next, we generated an egl-43L-specific improved RNAi targeting vector in T444T and examined egl-43L depletion in the uterine-specific RNAi-sensitive genetic background (Matus et al., 2010). Depletion of egl-43L resulted in a penetrant AC invasion defect, as expected, but, notably, we detected the presence of animals with both a single non-invasive AC (19%) and animals with multiple ACs (39%) (Fig. 6A,B), similar to depletion of both isoforms (Fig. 1C,E). We then assessed whether the long isoform has the same quantitative relationship to the other TFs as observed when targeting both isoforms. Indeed, egl-43(RNAi) and egl-43L(RNAi) exhibited comparable levels of depletion of egl-43::GFP::egl-43 (75%, 78%), GFP::hlh-2 (66%, 60%) and nhr-67::GFP (66%, 64%, respectively) (n≥25 animals examined per treatment) (Fig. S7C).
To characterize loss-of-function phenotypes further and examine transcriptional output of the two egl-43 isoforms, we next examined endogenous transcriptional reporters that were created as an intermediate step in generating GFP knock-in alleles using the self-excising cassette (SEC) method (Dickinson and Goldstein 2016). Specifically, prior to Cre-Lox recombination, the insertion of GFP at the N-terminus of a locus with a ∼6 kb SEC generates a transcriptional reporter at the endogenous locus and often interferes with the native transcriptional machinery at the locus where it is inserted, generating a strong loss-of-function or putative null allele (Dickinson and Goldstein, 2016). Examination of pre-floxed versions of both edited knock-in alleles of egl-43 was performed in the presence of an mT1 balancer, as homozygous animals are non-fertile. Similar to the putative null allele, egl-43(tm1802) (Rimann and Hajnal, 2007), the allele resulting from the SEC insertion of the internal GFP tag had an extremely low frequency of escapers, and we were unable to obtain L3-stage animals to examine for AC invasion defects. However, similar to our results with RNAi (Fig. 6A,B), an allele resulting from SEC insertion with the long isoform [egl-43(bmd135)] displayed a 31% AC invasion defect (n=11/36), with animals containing either single or multiple non-invading ACs (Fig. S7D,E). Finally, as previous reports based on transgenic transcriptional reporters have suggested that egl-43 may be autoregulatory (Matus et al., 2010; Rimann and Hajnal, 2007; Wang et al., 2014), we examined the mT1-balanced pre-floxed allele of the internally tagged egl-43 targeting both isoforms (egl-43::GFP^SEC^::egl-43). In support of these previous studies, we found strong evidence of autoregulation at the endogenous locus, as egl-43(RNAi) reduced the expression of GFP by 65% (n≥25 animals) (Fig. 6C,D). Taken together, these results suggest that both isoforms of egl-43 function redundantly to regulate multiple transcriptional sub-circuits that are crucial for invasion.
Activation of the pro-invasive gene regulatory network occurs post-specification
Our results here identify a putative feed-forward loop between egl-43, hlh-2 and nhr-67 to maintain levels of NHR-67, which we have previously shown to be crucial for keeping the AC in a post-mitotic, pro-invasive state (Matus et al., 2015). As these same TFs also have been shown to function during the AC/VU decision (Hwang et al., 2007; Karp and Greenwald, 2004; Verghese et al., 2011), we investigated whether the same regulatory relationships were utilized during cell fate specification and invasion. Intriguingly, our examination of TF onset establishes a hierarchy of egl-43, hlh-2 and nhr-67 reaching steady-state levels in the descendants of Z1 and Z4, although all three TFs are robustly expressed in the AC/VU cells (Fig. 2A). To examine TF regulatory relationships prior to specification, we performed L1 TF RNAi platings and determined expression levels of TF-GFP in the AC/VU cells in L2-stage animals. Although each TF RNAi robustly depleted its target GFP allele, we failed to detect any molecular epistatic interactions between TFs in the AC/VU cells (Fig. 7A,B). In support of previous research demonstrating a role for these TFs in regulating the AC/VU decision (Hwang et al., 2007; Karp and Greenwald, 2004; Verghese et al., 2011), we detected significant depletion of an endogenously GFP-tagged allele of lin-12 (Notch) (Fig. 7C,D). Together, these results suggest that, despite the re-iterative use of the same set of TFs in specification and invasion, the gene regulatory network (GRN) we characterized here is specific for the post-specification pro-invasive behavior of the AC.
DISCUSSION
Cellular invasion requires coordination of extrinsic cues from the surrounding microenvironment and orchestration of intrinsic regulatory circuits (Sherwood and Plastino, 2018). We focus here on exploring the intrinsic pathways that autonomously regulate C. elegans AC invasion. Combining improved RNAi targeting vectors with GFP-tagged alleles enabled us to correlate phenotype with depletion of the endogenous mRNAs and compare regulatory relationships in a quantitative framework. Our experiments reveal new roles for egl-43 and hlh-2 in regulating nhr-67-mediated cell cycle arrest and suggest that these TFs function in a coherent feed-forward loop with positive feedback. Additionally, we confirm previous work demonstrating that egl-43 is autoregulatory and demonstrate that it functions in both the cell cycle-dependent sub-circuit as well as in a cell cycle-independent manner, through fos-1, to orchestrate invasion (Fig. 8). Together, building upon previous work using transgenic approaches (Bodofsky et al., 2018; Hwang et al., 2007; Rimann and Hajnal, 2007; Schindler and Sherwood, 2011; Verghese et al., 2011), our results using endogenously tagged alleles provide the first description of the native regulatory relationships between the four TFs that promote invasion during C. elegans uterine-vulval attachment.
A key function of nhr-67 is to maintain the post-mitotic state of the AC (Matus et al., 2015). Here, we demonstrate that egl-43 and hlh-2 both regulate nhr-67, and that nhr-67-depletion also reduces HLH-2 levels. Network inference suggests these three TFs function in a classic type I coherent feed-forward loop regulatory circuit with positive feedback (Mangan and Alon, 2003) (Fig. 8). Our model fits with data by others showing that hlh-2 may directly bind to the nhr-67 promoter through two canonical E-box motifs found in a 164 bp window in a functional promoter element deleted in several hypomorphic alleles of nhr-67 (Bodofsky et al., 2018; Matus et al., 2015; Verghese et al., 2011). Further, deletion of these hlh-2 conserved binding sites results in loss of AC-specific GFP expression in transgenic lines (Bodofsky et al., 2018). Although others have also reported positive regulation by EGL-43 on nhr-67 activity via transgenic reporters (Bodofsky et al., 2018), this interaction, and many of the other TF interactions, may be indirect, as no known binding sites exist for EGL-43 in the nhr-67 promoter. Alternatively, EGL-43 and NHR-67 have been predicted, based on yeast two-hybrid experiments, to interact at the protein-protein level (Reece-Hoyes et al., 2013).
We have also examined the activity of the two previously identified isoforms of egl-43 (Hwang et al., 2007; Rimann and Hajnal, 2007). As both mutant analyses and RNAi depletion of the long isoform phenocopy depletion of both isoforms, our data suggest that either the long isoform of egl-43 functions redundantly with the short isoform or it functions as the dominant isoform at the intersection of multiple regulatory circuits. Of these two possible hypotheses, the latter is supported by a recent report demonstrating that the long isoform of egl-43 is a key upstream regulator of the post-mitotic state of the AC as mutation of the transcriptional start methionine of the short isoform of egl-43 has no observable AC invasion defect and an AC-specific genetic knockout of the long isoform results in mitotic, non-invasive ACs (Deng et al., 2019 preprint).
Together, our data and corroborating evidence from the literature support the existence of a coherent feed-forward circuit with positive feedback among egl-43, hlh-2 and nhr-67 in maintaining the post-mitotic state of the AC (Fig. 8). We have previously shown that nhr-67 positively regulates transcripts of cki-1 specifically in the AC (Matus et al., 2015). Thus, to test our putative regulatory circuit, we induced expression of cki-1 in the AC, which prevents the AC from inappropriately entering the cell cycle. This forced G1/G0 arrest fully rescues nhr-67 depletion (Matus et al., 2015). Induced cki-1 partially rescued hlh-2 depletion but failed to rescue egl-43 depletion. Heat shock-induced expression of nhr-67 failed to rescue either hlh-2 or egl-43, suggesting that hlh-2, like egl-43, has a cell cycle-independent function. These results suggest that egl-43 and hlh-2 function in multiple regulatory circuits. For egl-43, this is supported by recent work demonstrating a type I incoherent feed-forward loop between fos-1, egl-43L and the BM adhesion protein MIG-10/lamellipodin (Wang et al., 2014). Finally, our observation that egl-43 is autoregulatory, also shown previously by promoter>GFP fusion experiments (Matus et al., 2015; Rimann and Hajnal, 2007; Wang et al., 2014), supports a model in which egl-43 occupies a key node at the intersection of multiple pro-invasive circuits (Fig. 8).
Feed-forward regulatory loops are likely the most well-described network motif occurring in all GRNs (Cordero and Hogeweg, 2006; Davidson, 2010; Mangan and Alon, 2003). From E. coli (Shen-Orr et al., 2002) and yeast (Lee et al., 2002; Milo et al., 2002) to a myriad of examples across the Metazoa (reviewed by Davidson, 2010), feed-forward loops are thought to function as filters for transient inputs (reviewed by Alon, 2007; Hinman, 2016). The addition of positive feedback, generating a coherent feed-forward loop, provides stability to a sub-circuit and is often found in differentiation gene batteries coincident with autoregulation (Davidson, 2010). These network motifs have been described in many developmental contexts, including MyoD-driven vertebrate skeletal muscle differentiation (Penn et al., 2004), patterning of the Drosophila egg shell via the TF Broad and interactions with EGFR and Dpp signaling (Yakoby et al., 2007), Pax6-dependent regulation of Maf and crystallin expression in the mouse embryonic lens (Xie and Cvekl, 2011) and terminal selector neuronal differentiation in C. elegans and mammals (reviewed by Hobert, 2008). During C. elegans embryonic development, a series of coherent feed-forward loops utilizing SKN-1/MED-1,2 and then a suite of reiteratively used GATA factors (END-1,-3) are required to pattern endomesoderm development successfully (reviewed by Maduro, 2009). Thus, the coherent feed-forward loop identified here maintaining the post-mitotic state in the C. elegans AC likely evolved to coordinate AC cell cycle exit, both as a by-product of terminal differentiation and a morphogenetic requirement for invasive behavior (Matus et al., 2015). AC invasion is necessary for egg laying, and defects in the process results in penetrant Protruding vulva/Egg-laying defective (Pvl/Egl) phenotypes, reducing fecundity in individual animals by nearly ten-fold. Thus, redundant control of the sub-circuit regulating invasion and cell cycle arrest may have been under strong selection, providing an explanation for the regulatory relationships between egl-43, hlh-2 and nhr-67 we characterize here.
As egl-43, hlh-2 and nhr-67 reach steady-state levels prior to the time of AC/VU specification and individually have been previously shown to regulate different aspects of the stochastic cell fate decision between the AC and VU (Hwang et al., 2007; Karp and Greenwald, 2004; Verghese et al., 2011), we performed RNAi depletion experiments and examined TF::GFP levels in the AC/VU cells during the L2 stage. Surprisingly, we found no evidence of regulation between the TFs at this stage, though, as expected, all three TFs regulated levels of endogenously tagged lin-12::GFP (Notch). This suggests that the feed-forward loop functions specifically in the post-specification AC to maintain the post-mitotic, invasive state (Fig. 8). Transient TF-DNA binding and GRN rewiring can result from responses to diverse stimuli (Luscombe et al., 2004; Swift and Coruzzi, 2017). One plausible explanation for this temporal change in network topology is that cell cycle-dependent changes in chromatin remodeling and TF activity likely changes the genomic landscape between the bipotential AC/VU cells and the post-mitotic, differentiated AC (Ma et al., 2015). Additionally, there may be differentially expressed co-factor(s) functioning during specification or invasion that facilitate interactions between TFs. Reiterative use of the same TFs to program different cell behaviors has been examined during neuronal differentiation (Guillemot, 2007) and neural crest specification and migration, as the winged-helix TF FoxD3 was recently shown to regulate multiple independent chromatin-organizing circuits, functioning as a pioneer factor during zebrafish neural crest specification and a transcriptional repressor during later events, including epithelial-mesenchymal transition and migration (Lukoseviciute et al., 2018). Whether or not egl-43, nhr-67 or hlh-2 function as pioneer factors during AC/VU specification is an open question, and one that could soon be tackled through isolation by fluorescence-activated cell sorting and the application of newer next-generation sequencing tools such as ATAC-seq (assay for transposase-accessible chromatin using sequencing) or CUT&RUN (cleavage under targets and release using nuclease) (Skene and Henikoff, 2017).
We are just beginning to understand the connections between regulatory circuits and morphogenetic behaviors (Christiaen et al., 2008; Martik and McClay, 2015; Saunders and McClay, 2014). Hopefully, the ease of genome editing and protein perturbation strategies will facilitate the kind of analyses we have performed here in other metazoan systems. In summary, in this study we characterize the complex relationships between the four pro-invasive TFs that function in the AC to program invasive behavior. We show that although the TFs are reiteratively used in both cell fate specification and in a differentiated cell behavior, invasion, they only interact with each other following specification. We identify a classic type I feed-forward loop regulating mitotic exit and controlling a switch between proliferative and invasive behavior. Whether or not similar circuit architecture is utilized to regulate invasive and proliferative cell biology in other developmental invasive contexts, including mammalian trophoblast implantation and placentation (Carter et al., 2015; Red-Horse et al., 2004) and epithelial to mesenchymal transition events during gastrulation (Vega et al., 2004), is still poorly understood. Finally, as there appear to be many cancer subtypes that may switch between proliferative and invasive fates (reviewed by Kohrman and Matus, 2017), improving our understanding of the GRN architecture of invasive cells may provide new therapeutic nodes to target in reducing the lethality associated with cancer metastasis.
MATERIALS AND METHODS
C. elegans strains and culture conditions
Animals were reared under standard conditions and cultured at 25°C, with the exception of temperature-sensitive strains including those containing heat shock-inducible transgenes or the rrf-3(pk1426) allele (conferring RNAi hypersensitivity), which were maintained at 15°C and 20°C (Brenner, 1974; Simmer et al., 2002). Heat shock-inducible transgenes were activated by incubating animals on plates at 32°C for 1 h following AC specification and again just prior to AC invasion. Animals were synchronized for experiments through alkaline hypochlorite treatment of gravid adults to isolate eggs (Porta-de-la-Riva et al., 2012). In the text and figures, we designate linkage to a promoter through the use of the>symbol and fusion of a proteins by a :: annotation. The following transgenes and alleles were used in this study: qyIs102[fos-1>RDE-1;myo-2>YFP] LG I hlh-2(bmd90[hlh-2>LoxP::GFP::HLH-2]), qyIs227[cdh-3>mCherry::moeABD], bmd121[LoxP::hsp>NHR-67::2xBFP], bmd142[hsp>HLH-2::2xBFP]; LG II egl-43(bmd87[egl-43>SEC::GFP::EGL-43]), egl-43(bmd88[egl-43>LoxP::GFP::EGL-43]), egl-43(bmd136[egl-43L>LoxP::GFP::EGL-43]) rrf-3(pk1426); qyIs17 [zmp-1>mCherry] LG III unc-119(ed4), LG IV nhr-67(syb509[nhr-67>NHR-67::GFP]), qyIs10[lam-1>LAM-1::GFP] LG V fos-1(bmd138[fos-1>LoxP::GFP::FOS-1]), qyIs225[cdh-3>mCherry::moeABD], rde-1(ne219), qyIs24[cdh-31.5>mCherry::PLCδPH], qyIs266[cdh-3>CKI-1::GFP] LG X qyIs7[lam-1>LAM-1::GFP]. See Table S3 for additional details of strains used and generated in this study.
Molecular biology and microinjection
TFs were tagged at their endogenous loci using CRISPR/Cas9 genome editing via microinjection into the hermaphrodite gonad (Dickinson and Goldstein, 2016; Dickinson et al., 2013). Repair templates were generated as synthetic DNAs from either Integrated DNA Technologies (IDT) as gBlocks or Twist Biosciences as DNA fragments and cloned into ccdB compatible sites in pDD282 by New England Biolabs Gibson assembly (Dickinson et al., 2015). Homology arms ranged from 690 to 1200 bp (see Tables S4-S6 for additional details). sgRNAs were constructed by EcoRV and NheI digestion of the plasmid pDD122. A 230 bp amplicon was generated replacing the sgRNA targeting sequence from pDD122 with a new sgRNA, and NEB Gibson assembly was used to generate new sgRNA plasmids (see Tables S4-S6 for additional details). Hermaphrodite adults were co-injected with guide plasmid (50 ng/μl), repair plasmid (50 ng/μl) and an extrachromosomal array marker (pCFJ90, 2.5 ng/μl), and incubated at 25°C for several days before carrying out screening and floxing protocols associated with the SEC system (Dickinson et al., 2015).
RNA interference
An RNAi library of the pro-invasive TFs was constructed by cloning 950-1000 bp of synthetic DNA (663 bp for the egl-43 long-specific isoform) based on cDNA sequences available on WormBase (www.wormbase.org) into the highly efficient T444T RNAi vector (Grove et al., 2018; Sturm et al., 2018). Synthetic DNAs were generated by IDT as gBlocks or Twist Biosciences as DNA fragments and cloned into restriction-digested T444T using NEB Gibson Assembly (see Tables S4, S5 and S7 for additional details). For most experiments, synchronized L1-stage animals were directly exposed to RNAi through feeding with bacteria expressing dsRNA (Conte et al., 2015). Because early hlh-2 RNAi treatment perturbs AC specification, animals were initially placed on empty vector RNAi plates and then transferred to hlh-2 RNAi plates following AC specification, approximately 12 h later at 25°C or 24 h later at 15°C (Schindler and Sherwood, 2011).
Live-cell imaging
All micrographs included in this manuscript were collected on a Hamamatsu Orca EM-CCD camera mounted on an upright Zeiss AxioImager A2 with a Borealis-modified CSU10 Yokagawa spinning disk scan head (Nobska Imaging) using 488 nm and 561 nm Vortran lasers in a VersaLase merge and a Plan-Apochromat 100×/1.4 (NA) Oil DIC objective. MetaMorph software (Molecular Devices) was used for microscopy automation. Several experiments were scored using epifluorescence visualized on a Zeiss Axiocam MRM camera, also mounted on an upright Zeiss AxioImager A2 and a Plan-Apochromat 100×/1.4 (NA) Oil DIC objective. Animals were mounted into a drop of M9 on a 5% Noble agar pad containing approximately 10 mM sodium azide anesthetic and topped with a coverslip.
Scoring of AC invasion
AC invasion was scored at the P6.p four-cell stage, when 100% of wild-type animals exhibit a breached BM (Sherwood and Sternberg, 2003). Presence of green fluorescence under the AC, in strains with the laminin::GFP transgene, or presence of a phase dense line, in strains without the transgene, was used to assess invasion. Wild-type invasion is defined as a breach as wide as the basolateral surface of the AC, whereas partial invasion indicates the presence of a breach smaller than the footprint of the AC (Sherwood and Sternberg, 2003). Raw scoring data is available in Tables S1 and S2.
Image quantification and statistical analyses
Images were processed using Fiji/ImageJ (v.1.52q) (Schindelin et al., 2012). Expression levels of TFs were measured by quantifying the mean gray value of AC nuclei, defined as somatic gonad cells strongly expressing the cdh-3>mCherry::moeABD transgene, subtracting the mean gray value of a background region of an equal area to account for EM-CCD camera noise, which we use as a proxy for background fluorescence, as measurements of the mean gray value of unlabeled ACs approximately correspond to camera noise (means=2514.2 and 2243.5, respectively). Levels of zmp-1>mCherry were quantified in control and cdh-3>CKI-1::GFP animals by measuring the mean gray value of the entire AC, selected either by a hand-drawn region of interest or using the threshold and wand tools in Fiji/ImageJ. For characterization of downstream targets (Fig. 4) and molecular epistasis experiments (Fig. 3, Figs S3 and S7) at the L3 stage, only TF-RNAi animals exhibiting defects in invasion were included in the analysis. Data was normalized to negative control (empty vector) values for the plots in Figs 3-7 and Figs S5-S7, and to both negative control and positive control values for determining the interaction strengths represented in Fig. 8. Images were overlaid and figures were assembled using Adobe Photoshop (v. 20.0.6) and Illustrator CS (v. 10.14), respectively. Statistical analyses and plotting of data were conducted using RStudio (v. 1.1.463). Individual data points for each experiment were collected over multiple days. Statistical significance was determined using either a two-tailed Student's t-test or Fisher's exact probability test. Figure legends specify when each test was used and the P-value cut-off.
Acknowledgements
We are grateful to R. Adikes, B. Kinney, A. Kohrman, M. Martinez and B. Martin for comments on the manuscript. We also thank I. Greenwald for providing the endogenously tagged lin-12>LIN-12::GFP strain used in Fig. 7C. Some strains were provided by the Caenorhabditis Genetics Center, which is funded by the NIH Office of Research Infrastructure Programs [P40 OD010440].
Footnotes
Author contributions
Conceptualization: T.N.M., D.Q.M.; Methodology: T.N.M., D.Q.M.; Validation: T.N.M., D.Q.M.; Formal analysis: T.N.M., J.J.S., N.J.P., D.Q.M.; Investigation: T.N.M., J.J.S., N.J.P., D.Q.M.; Resources: T.N.M., J.J.S., N.J.P., S.T., W.Z., D.Q.M.; Data curation: T.N.M., J.J.S., D.Q.M.; Writing - original draft: T.N.M., D.Q.M.; Writing - review & editing: T.N.M., J.J.S., N.J.P., D.Q.M.; Visualization: T.N.M., J.J.S., N.J.P., D.Q.M.; Supervision: W.Z., D.Q.M.; Project administration: D.Q.M.; Funding acquisition: D.Q.M.
Funding
This work was funded by the National Institutes of Health (NIH) National Cancer Institute [5R00CA154870-05 to D.Q.M.] and National Institute of General Medical Sciences (NIGMS) [1R01GM121597-01 to D.Q.M.]. D.Q.M. is also a Damon Runyon-Rachleff Innovator supported (in part) by the Damon Runyon Cancer Research Foundation [DRR-47-17]. T.N.M.-K. is supported by the NIH Eunice Kennedy Shriver National Institute of Child Health and Human Development [F31HD100091-01]. J.J.S. is supported by NIGMS [3R01GM121597-02S1] and N.J.P. is supported by the American Cancer Society [132969-PF-18-226-01-CSM]. Deposited in PMC for release after 12 months.
References
Competing interests
The authors declare no competing or financial interests.