A positive feedback loop controls Toxoplasma chronic differentiation

Successful infection strategies must balance pathogen amplification and persistence. In the obligate intracellular parasite Toxoplasma gondii this is accomplished through differentiation into dedicated cyst-forming chronic stages that avoid clearance by the host immune system. The transcription factor BFD1 is both necessary and sufficient for stage conversion; however, its regulation is not understood. In this study we examine five factors that are transcriptionally activated by BFD1. One of these is a cytosolic RNA-binding protein of the CCCH-type zinc-finger family, which we name bradyzoite formation deficient 2 (BFD2). Parasites lacking BFD2 fail to induce BFD1 and are consequently unable to fully differentiate in culture or in mice. BFD2 interacts with the BFD1 transcript under stress, and deletion of BFD2 reduces BFD1 protein levels but not messenger RNA abundance. The reciprocal effects on BFD2 transcription and BFD1 translation outline a positive feedback loop that enforces the chronic-stage gene-expression programme. Thus, our findings help explain how parasites both initiate and commit to chronic differentiation. This work provides new mechanistic insight into the regulation of T. gondii persistence, and can be exploited in the design of strategies to prevent and treat these key reservoirs of human infection. Productive infection strategies require trade-offs between pathogen amplification and persistence. Although replication increases the likelihood of transmission, the resultant tissue damage drives host immunity to promote pathogen clearance. At the same time, uncontrolled proliferation may be fatal to the host, altogether eliminating

Successful infection strategies must balance pathogen amplification and persistence. In the obligate intracellular parasite Toxoplasma gondii this is accomplished through differentiation into dedicated cyst-forming chronic stages that avoid clearance by the host immune system. The transcription factor BFD1 is both necessary and sufficient for stage conversion; however, its regulation is not understood. In this study we examine five factors that are transcriptionally activated by BFD1. One of these is a cytosolic RNA-binding protein of the CCCH-type zinc-finger family, which we name bradyzoite formation deficient 2 (BFD2). Parasites lacking BFD2 fail to induce BFD1 and are consequently unable to fully differentiate in culture or in mice. BFD2 interacts with the BFD1 transcript under stress, and deletion of BFD2 reduces BFD1 protein levels but not messenger RNA abundance. The reciprocal effects on BFD2 transcription and BFD1 translation outline a positive feedback loop that enforces the chronic-stage gene-expression programme. Thus, our findings help explain how parasites both initiate and commit to chronic differentiation. This work provides new mechanistic insight into the regulation of T. gondii persistence, and can be exploited in the design of strategies to prevent and treat these key reservoirs of human infection.
Productive infection strategies require trade-offs between pathogen amplification and persistence. Although replication increases the likelihood of transmission, the resultant tissue damage drives host immunity to promote pathogen clearance. At the same time, uncontrolled proliferation may be fatal to the host, altogether eliminating the reservoir of infection 1 . To balance these outcomes, many protozoan parasites have developed slow-growing chronic stages that can withstand unfavourable growth conditions and exhibit reduced immunogenicity 2 . An equilibrium between proliferation and latency can thereby promote the wellbeing of the host and ensure parasite transmission.
For apicomplexans of the Sarcocystidae family, persistence is achieved through the formation of large intracellular structures (tissue cysts) containing chronic stages known as bradyzoites. Whereas infected animals can survive long-term harbouring of these cysts, active replication of the proliferative forms (tachyzoites) results in severe, potentially deadly disease. The most widespread of these infections is caused by Toxoplasma gondii, which chronically infects up to one quarter of humans worldwide 3 . During acute infection T. gondii tachyzoites disseminate throughout the body and cause pathology through host-cell lysis. A portion of tachyzoites give rise to bradyzoites, which encyst in brain and muscle tissue 4 . Most infections are controlled by the host immune response; however, reactivation of latent cysts can occur during periods of weakened immune function [5][6][7][8] . In addition, there are no therapies available at present that target the chronic stages 9 , making it impossible to eradicate T. gondii from the host. Article https://doi.org/10.1038/s41564-023-01358-2 of indole-3-acetic acid (IAA; Fig. 1b) 37,38 . BFD1 was included as a control for its impact on differentiation.
We first examined the expression and localization of each factor. Following 48 h of alkaline stress, we observed staining for BFD1 and four of the five candidates (Fig. 1c), with AP2IB-1 remaining below the limit of detection (not pictured). TGME49_311100 was uniquely expressed in acute stages but upregulated in response to stress. The three predicted RNA-binding proteins localized to the cytosol, whereas BFD1 and AP2IX-9 were localized to the nucleus, consistent with previous characterization 23,35 . All five detectable factors were depleted following treatment with IAA (Fig. 1c).
We examined the impact of each factor on stress-induced differentiation by quantifying cyst-wall staining with Dolichos biflorus lectin (DBL) using automated image analysis (Fig. 1d). Consistent with published results, loss of BFD1 blocked cyst formation under alkaline stress 23 , whereas AP2IX-9 had no impact 33,35 . TGME49_224630, TGME49_253790 and AP2IB-1 were similarly dispensable; however, parasites expressing the conditional allele of TGME49_311100 showed a marked decrease in DBL-positive vacuoles under both vehicle and IAA treatment. Given the insensitivity of this effect to IAA, we conclude that TGME49_311100 is functionally inhibited by carboxy (C)-terminal tagging. Furthermore, our results implicate TGME49_311100 in the process of differentiation, leading us to name it bradyzoite formation deficient 2 (BFD2).

BFD1 and BFD2 share a common transcriptional signature
Given that DBL staining only describes one aspect of differentiation, we performed RNA sequencing (RNA-seq) on all knockdown strains following 96 h of stress. Only two strains exhibited obvious signatures of transcriptional reprogramming (Fig. 1e). AP2IX-9 and AP2IB-1 in particular had no effect when depleted from the chronic stages (Extended Data Fig. 2), despite evidence for the former as an early antagonist of chronic-stage gene expression 33,35 . Although we did identify a small regulon for TGME49_224630 (Supplementary Table 1), the affected transcripts are generally unresponsive to BFD1 (ref. 23). Thus, this regulon probably represents a response to alkaline stress rather than a bona fide component of the core bradyzoite programme. By contrast, BFD1 knockdown had the expected impact on the chronic-stage transcriptome, which inversely mirrored the effects of conditionally upregulating BFD1 under standard conditions (Fig. 1f). Strikingly, a near-identical response was observed in parasites depleted of BFD2 ( Fig. 1f,g), suggesting that the two factors function closely within the differentiation pathway to reshape gene expression.
Bradyzoite formation deficient 1 (BFD1), a Myb-like transcription factor, was discovered by screening approximately 200 T. gondii genes with putative nucleic acid-binding domains 23 . Despite being transcribed throughout the T. gondii asexual cycle, BFD1 protein is not detected in unstressed parasites, implying that its expression is translationally controlled. Parasites lacking BFD1 grow normally as tachyzoites but enter a non-differentiated G1-like state under alkaline stress. Exhibiting the behaviour of a master regulator, conditional BFD1 expression is sufficient to induce differentiation in the absence of stress and closely recapitulates the transcriptional changes observed in naturally derived bradyzoites. Yet it is unclear how BFD1 intersects with other chronic-stage regulators. We previously showed that a number of genes containing putative RNA-and DNA-binding domains are targets of BFD1, suggesting a cascade of downstream differentiation factors. In the present study we set out to unravel the regulatory network extending from BFD1 by characterizing these effectors. In doing so, we uncover a positive feedback loop acting on BFD1 to reinforce commitment to chronic differentiation.

Analysis of BFD1-regulated gene-expression factors
Previous datasets revealed five putative nucleic acid-binding proteins that are both upregulated during differentiation and bound at their promoters by BFD1, making them probable secondary effectors of the bradyzoite programme (Fig. 1a). Two of these-TGME49_306620 (also known as AP2IX-9) and TGME49_208020 (also known as AP2IB-1)-belong to the well-characterized ApiAP2 family of apicomplexan transcription factors. TGME49_311100, TGME49_224630 and TGME49_253790 encode CCCH-type zinc-finger domains involved in RNA binding. Whereas AP2IX-9 was described previously as a regulator of early bradyzoite development 33,35 , the other four genes were functionally uncharacterized. We therefore generated conditional-knockdown strains for each candidate to assess its role in differentiation. In a parental line expressing the heterologous TIR1 auxin receptor, each candidate was endogenously tagged with mNeonGreen (mNG) and the minimal auxin-inducible degron (mAID; Extended Data Fig. 1), allowing both detection and depletion of the gene product following the addition   23 , with the five genes encoding predicted nucleic acid-binding domains indicated. b, Generation of conditionalknockdown strains. Tagging with mNG-mAID targets proteins for degradation following IAA treatment. Additional strain construction details are provided in Extended Data Fig. 8. CDS, coding sequence. c, Knockdown strains after 48 h in the presence or absence of alkaline stress treated with IAA or vehicle. Parasites are labelled by GAP45, DNA with Hoechst and tagged genes by immunostaining for mNG. For clarity, the mNG signal is shown below the corresponding merged image. The assay was performed twice with similar results. d, Differentiation of knockdown strains after 48 h of alkaline stress, assessed by cyst-wall staining (DBL). Mean ± s.d. plotted for n = 3 biological replicates, with a minimum of 175 vacuoles counted per sample. ****P < 0.0001 and NS, not significant; one-way analysis of variance (ANOVA) with Dunnett's correction performed on the percentage of DBL high vacuoles. e-g, Effects of candidate knockdown on the chronic-stage transcriptome. Data reflect changes after 96 h of alkaline stress in the presence of IAA for n = 2 independent infections. e, Genes significantly affected by depletion of each factor (adjusted P < 0.05, Wald test in DESeq2) are highlighted and enumerated in the respective plot. FPKM, fragments per kilobase of exon per million mapped fragments. Complete lists of the affected genes are provided in Supplementary Table 1. f, For the BFD1-and BFD2knockdown strains, significantly affected genes are coloured according to their log 2 (fold change) value during conditional BFD1 expression 23 . g, A comparison of the transcriptional effects of BFD1 versus BFD2 depletion is shown, highlighting genes meeting the significance threshold in both or either sample independently. Pearson's correlation (r) analysis was performed on the union of statistically regulated points. Article https://doi.org/10.1038/s41564-023-01358-2 endogenously regulated BFD1 (ref. 23). After 48 h of alkaline stress, DBL staining for Δbfd2 recapitulated our observations during knockdown; however, whereas deletion of BFD1 completely blocked cyst-wall formation, a portion of Δbfd2 vacuoles stained faintly with DBL (Fig. 2b,c).
Both functional inhibition (Extended Data Fig. 3b) and deletion of BFD2 (Fig. 2d,e) caused a reduction in plaque size, which could not be attributed to a change in host cell invasion (Fig. 2f), replication rate ( Fig. 2g) or sensitivity to extracellular stress (Fig. 2h). Revisiting the results of our previous loss-of-function screens, BFD2 seems to be dispensable under standard growth conditions in the RH strain 39 ; however, in the ME49 screen that identified BFD1, four of the five guide RNAs (gRNAs) targeting BFD2 decreased in abundance over serial passaging in standard media 23 (Extended Data Fig. 4) and were further depleted in differentiating populations. The apparent cost to ME49 fitness led to BFD2 being excluded as a potential differentiation factor in our previous work.  To better understand the role of BFD2 throughout the asexual cycle of T. gondii, we next performed RNA-seq on parental and Δbfd2 parasites under alkaline-stress or unstressed conditions. Reassuringly, Δbfd2 and the non-functional knockdown strain showed strong agreement under stress (Fig. 2i). Specifically, 753 genes were dysregulated in the absence of BFD2, reflecting a clear BFD1-responsive signature (Fig. 2j  (left)). The partial differentiation of Δbfd2 parasites was recapitulated in their transcriptional response to stress, which mirrored that of the parental strain but with a marked decrease in magnitude (Fig. 2k). By contrast, we observed only minor differences in the unstressed samples ( Fig. 2j(right)), which were also limited to BFD1-regulated genes. This is consistent with an absence of spontaneously differentiating parasites in the Δbfd2 population, which normally form at low frequency in wild-type ME49 strains during routine culturing in fibroblasts (note the DBL-positive vacuoles in unstressed samples, Fig. 2c). Proteome profiling further confirmed reduced levels for several canonical bradyzoite proteins in unstressed Δbfd2 (Extended Data Fig. 3c), none of which seemed to be fitness-conferring based on previous genetic screens 39 .
Although alkaline stress is a potent trigger of stage conversion, we sought to validate our findings in a more physiologically relevant   14,40,41 , and neurons support high rates of spontaneous differentiation in vitro 42,43 . Thus, we compared the transcriptomes of wild-type, Δbfd1 and Δbfd2 parasites-this time generated in the Pru genetic background-cultured for 48 h in mouse primary neurons. As predicted, both PruΔbfd1 and PruΔbfd2 showed a distorted transcriptional response consistent with a block in the BFD1 programme (Extended Data Fig. 3d). Curiously, whereas stress-induced mutants had shown an inability to both fully induce and repress distinct gene sets ( Fig. 1f and 2j), in neurons, loss of either factor primarily affected genes transcriptionally activated by BFD1. This could reflect a strain-specific effect or a difference in the parasite response to distinct differentiation stimuli. Nonetheless, as observed under stress (Fig. 1g), knockout of BFD1 and BFD2 had similar effects on the transcriptome, with the loss of BFD1 causing changes of a greater magnitude (Extended Data Fig. 3e). These results further highlight the critical role of BFD2 in enforcing the BFD1 programme.

Parasites lacking BFD2 are unable to generate cysts in mice
We examined how BFD2 impacts virulence and cyst formation in mice. Female CD-1 mice were intraperitoneally (i.p.) injected with 100 tachyzoites of either the parental or Δbfd2 strain (Fig. 3a). Despite the reduced fitness of Δbfd2 in vitro, weight loss and mortality were equivalent between the two cohorts (Fig. 3b,c), indicating that BFD2 is dispensable for the acute symptoms of T. gondii infection. The cyst burden of the brains of all mice was assessed at the time of euthanasia, based on coincidence of DBL with a general parasite marker (CDPK1; Fig. 3d). After 45 d, all surviving animals infected with the parental strain harboured hundreds of cysts; none were detected for Δbfd2 (Fig. 3e). A similar result was observed in moribund animals, wherein cysts were first detected 23 d post injection with the parental strain and never with the mutant, thereby eliminating the possibility that Δbfd2 cysts form and are cleared early in infection. Intriguingly, we did detect Δbfd2 parasites in low numbers in the brains of infected (45 d) mice, and both parental-and Δbfd2-infected animals exhibited pathology consistent with persistent immune activation (Fig. 3f). Nonetheless, our results suggest that BFD2 is dispensable for the acute infection in mice yet necessary to produce chronic-stage cysts.

A feedback loop between BFD1 and BFD2 drives differentiation
BFD2 is induced by conditional BFD1 expression and its promoter shows evidence of BFD1 binding-characteristics that should place it  Article https://doi.org/10.1038/s41564-023-01358-2 downstream in the regulatory hierarchy. However, given the analogous effects of the two factors on differentiation, we sought to test this relationship directly. Stress-responsive changes in BFD2 transcript levels were quantified in the presence (Δbfd1::BFD1-TY; wild-type) and absence of BFD1 (Δbfd1) as well as for parasites expressing a non-functional version lacking DNA-binding motifs (Δbfd1::BFD1 ΔMYB -TY) 23 . Whereas BFD2 was upregulated in the wild-type parasites (Fig. 4a), this effect was almost completely lost in Δbfd1 and the non-functional complement, confirming that BFD2 is indeed induced by BFD1.
To assess the stress dependency of this relationship, we investigated whether loss of BFD2 could hamper differentiation induced by conditional BFD1 expression. We previously generated a regulatable version of BFD1 with an appended destabilization domain (DD; DD-BFD1-TY; Extended Data Fig. 5a) that is constitutively degraded unless stabilized by the small molecule Shield-1. In the presence of Shield-1, BFD1 accumulates, causing uniform differentiation irrespective of stress 23 . In this genetic background, we knocked out BFD2 and assessed differentiation by DBL staining following 48 h of Shield-1 or vehicle treatment. Over 94% of the Shield-1-treated DD-BFD1-TY vacuoles were DBL-positive (Fig. 4b,c), consistent with published results. Although we still observed a weak DBL signal for DD-BFD1-TYΔbfd2, the addition of Shield-1 could not overcome the loss of BFD2-reminiscent of the incomplete differentiation observed for BFD2-knockouts under stress (Fig. 2c). This result was also mirrored at the transcriptome level (Fig. 4d). Thus, BFD2 seems to facilitate the BFD1 programme, regardless of whether it is triggered by stress or targeted induction of BFD1 alone.
Positive feedback is a genetic feature commonly associated with master regulators 44 . Despite being activated downstream of BFD1, we therefore considered that BFD2 may also be required for robust BFD1 induction. To test this, we initially compared conditional BFD1 expression in the presence or absence of BFD2. Loss of BFD2 had no effect on the abundance of DD-BFD1-TY messenger RNA (Fig. 4e)the Shield-1-induced decrease in transcription can be ascribed to the TUB1 promoter driving the transgene 23 . However, immunostaining revealed significantly lower DD-BFD1-TY protein levels in the nuclei of the BFD2 knockouts (Fig. 4c,f). Encouraged by this result, we further examined endogenous BFD1 protein accumulation in alkaline-stressed wild-type, Δbfd2 and Δbfd1::BFD1 ΔMYB -TY parasites, with the hypothesis that if BFD1 and BFD2 indeed promote one another's expression, full induction of BFD1 should depend on its own activity. BFD1 was robustly detected in parasites harbouring functional copies of both factors (Δbfd1::BFD1-TY; wild-type) after 48 h of stress and continued to accumulate with time ( Fig. 4g). This effect was diminished in Δbfd2, with BFD1 levels reaching only approximately 7% of those measured for the wild-type parasites after 72 h. The non-functional BFD1 complement strain Δbfd1::BFD1 ΔMYB -TY-which cannot robustly induce BFD2 transcription-exhibited an intermediate level of BFD1 staining. Thus, we conclude that basal levels of BFD2 facilitate stress-responsive expression of BFD1. This in turn drives BFD2 upregulation, closing the feedback loop (Fig. 4h). In addition, given that the levels of BFD1 mRNA were unaffected by the presence of BFD2 (Fig. 4a,e), regulation must occur either via BFD1 translation or protein turnover.

RNA binding is necessary for the function of BFD2
Given that C-terminal fusions had rendered the protein inactive ( Fig. 1d), we sought to verify the localization of BFD2. We complemented Δbfd2 by introducing a complementary DNA copy, which had been tagged at the amino (N) terminus, back into the endogenous locus (HA-BFD2; Figs. 5a and Extended Data Fig. 6a). The HA-BFD2 parasites displayed normal plaque formation (Fig. 5b) and complete rescue of DBL staining under stress (Fig. 5c,d). By contrast, a version lacking the CCCH zinc fingers (HA-BFD2 ΔZF ) could not rescue either phenotype, confirming the importance of these domains for BFD2 function. Both HA-BFD2 and HA-BFD2 ΔZF were detectable above background in unstressed parasites and localized to the cytosol, irrespective of the culture conditions (Fig. 5c). In addition, both were upregulated after 48 h of stress; however, the magnitude of this change was approximately 30% lower in the zinc-finger mutant (Fig. 5e), providing further evidence of positive feedback, given that parasites cannot fully induce BFD1 or maximally upregulate BFD2 without BFD2 activity.

Overexpression of BFD2 triggers differentiation
In keeping with our model of BFD2 as a positive regulator of BFD1, we investigated whether its conditional overexpression is similarly sufficient for differentiation. In the Δbfd1::BFD1-TY background, we modified the native BFD2 locus to generate a regulatable allele, replacing the endogenous promoter with that of α-tubulin while appending an N-terminal DD domain to the coding sequence (DD-HA-BFD2; Fig. 5f and Extended Data Fig. 6b). Treatment with the stabilizing Shield-1 ligand led to robust cytosolic accumulation of DD-HA-BFD2 and 95% of vacuoles had developed DBL positivity above the vehicle-treated control by 48 h (Fig. 5g,h). Importantly, BFD1-TY also accumulated in the nuclei of parasites treated with Shield-1. Thus, overexpression of BFD2 is indeed sufficient for differentiation, probably through the induction of BFD1.

BFD2 binds to a cohort of transcripts that includes BFD1
We next investigated whether BFD2 binds to BFD1 mRNA. HA-BFD2 immunoprecipitation (IP) was performed on alkaline-stressed as well as unstressed samples to measure the BFD1 transcript abundance using quantitative PCR with reverse transcription (RT-qPCR). HA-BFD2 ΔZF samples were processed in parallel to control for non-specific interactions. Compared with total RNA, BFD1 was clearly enriched in the IPs from stressed HA-BFD2 parasites (Fig. 6a, WT). No enrichment was detected for HA-BFD2 ΔZF (Fig. 6a, ΔZF), confirming the specificity of the interaction. Although BFD1 was also detected at low levels in the IP from unstressed cultures, this did not exceed background measured from HA-BFD2 ΔZF under the same condition. This disparity could reflect a stress-dependent interaction between BFD2 and BFD1. However, two stage-independent transcripts (ASP5 and GCN5B) were also unexpectedly modestly enriched by IP under stress, possibly indicating a general capacity for BFD2 to bind RNA rather than preferential capture of certain transcripts.
To better describe the specificity of BFD2 binding, we subjected pulldowns from stressed HA-BFD2 parasites to next-generation sequencing ( Fig. 6b-d). After excluding messages below a minimum threshold of detection, an enrichment ratio was calculated for each transcript based on the transcripts per million (TPM) representation in the IP versus input samples (TPM IP /TPM input , Fig. 6b). The relative mRNA abundance did not correlate with enrichment after pulldown (Extended Data Fig. 7a), indicative of specific capture of BFD2 targets rather than background from non-specific binding. ENO1 and BAG1-two of the most abundant mRNAs in alkaline-stressed parasites-were not enriched in the immunoprecipitated material (Fig. 6c,d). Consistent with our results from RT-qPCR, both ASP5 and GCN5B seemed to be modestly enriched by IP (approximately twofold; Fig. 6c,d). A number of developmentally regulated genes, including LDH2, CST1 and BFD2, also showed varying degrees of enrichment; however, few exhibited the same magnitude of enrichment as BFD1, prompting us to seek a more rigorous approach to objectively stratify these interactions. Using Gaussian mixture modelling, we fitted a combination of two Gaussian curves to the log-transformed enrichment ratios (Fig. 6b). A statistical cutoff was then set based on the logarithm of the odds (LOD) of each gene in the highly enriched (green) distribution. Based on this conservative metric, we defined a subset of 375 transcripts showing the strongest evidence of BFD2 binding (LOD > 0; Supplementary Table 7). BFD1 fell comfortably within this group, ranking in the top 0.5% of enriched messages. Intriguingly, the majority of the cohort also displayed a stress-independent pattern of transcription, reminiscent of the expression pattern of BFD1 (Extended Data Fig. 7b). Given the apparent Article https://doi.org/10.1038/s41564-023-01358-2 role of BFD2 as a translational activator, this observation could point to BFD1 constituting part of a larger translational regulon. Moreover, it further emphasized the absence of confounding effects from differential transcription in the pulldown. Thus, our data provide strong support for the model of BFD2 binding to the BFD1 transcript under stress.

Discussion
Encystation is a key feature of T. gondii persistence. We recently identified BFD1 as the master regulator of this process 23 . Now we expand this model with the discovery of BFD2, a second factor indispensable for the chronic stage. We show that basal levels of BFD2 are required for   Nuclear DD-BFD1-TY (relative fluorescence)

DD-BFD1-TY
Shield-1:    Previous studies described RNA-and DNA-binding proteins that influence rates of stage conversion [28][29][30][31][32][33][34][35][36] ; however, their relationship to BFD1 is unknown. Our own work identified several direct targets of BFD1 with nucleic acid-binding domains 23 , suggesting a hierarchy of differentiation-promoting factors. It was therefore unexpected that of the five candidates examined, few exhibited obvious transcriptional signatures. In particular, AP2IX-9 was previously shown to inhibit the expression of bradyzoite-specific mRNAs 33,35,36 , yet we observed no consequence from its depletion under alkaline stress. In this case, the function of AP2IX-9 may be overshadowed by other factors based on its transient expression, despite still being detectable in stressed parasites at 96 h (ref. 35). Similarly, AP2IB-1 was previously implicated as transcriptionally upregulated in chronic forms; however, our inability to detect the tagged protein suggests that it may not be translated at the time points sampled. Subtle changes were detected for the other RNA-binding proteins, although our results suggest that TGME49_224630 may be associated with alkaline-stress responses outside the BFD1 programme. By far, the most obvious transcriptional signature was observed for BFD2.
BFD2 belongs to a widespread class of RNA-binding proteins defined by the presence of one or more CX 7-8 CX 4-5 CX 3 H (CCCH) zinc-finger motifs 45 . These domains are found in proteins affecting virtually all stages of RNA metabolism [46][47][48] ; however, few CCCH family members have been functionally characterized in apicomplexans. Two recent studies in Plasmodium species implicated ZFP4 and Pb103 in sexual reproduction due to gametocyte-specific effects 49,50 , yet the mechanism of regulation mediated by either factor was not definitive. Intriguingly, a CCCH protein in the non-apicomplexan parasite Trypanosoma brucei was found to promote differentiation through association with polyribosomes 51,52 -consistent with positive translational control, as described for BFD2. At least 33 other genes in the T. gondii genome encode putative CCCH motifs (VEuPathDB.org). Given their importance across diverse parasite phyla, we suspect that further exploration of this protein family will expose new factors involved in developmental gene expression.
Our data implicate BFD2 as a core part of the T. gondii chronic differentiation programme. Although BFD2 orthologues are found in other closely related cyst-forming apicomplexans, such as Neospora and Hammondia, the factor seems to be restricted to the Sarcocystadae family (VEuPathDB.org). BFD1 displays a similar pattern of conservation, suggesting that the relationship between the two factors may be maintained in closely related species. Notably, although dogma suggests that an inability to produce cysts should render Δbfd2 incapable of long-term persistence, we could still detect parasites in brain sections from mice up to 45 d post infection. This unexpected observation finds support in a parallel study, which showed that BFD1 and BFD2 (here, referred to as ROCY1) knockouts are capable of reinitiating acute infection as late as 5 months post inoculation, following immunosuppression 53 . Definitive characterization of these non-encysted forms will require further investigation and may expose novel mechanisms of T. gondii stress tolerance and persistence unrelated to bradyzoite conversion.
Although not essential for tachyzoite growth, we observed a modest fitness cost to BFD2 deletion in cultured ME49 parasites. The replication rates, invasion efficiency, sensitivity to extracellular stress and acute-stage virulence were puzzlingly comparable between the wild-type and Δbfd2 strains. Transcriptional differences in the knockout could be attributed to a lack of spontaneous differentiation and at the protein level we detected no deficiencies in known fitness-conferring proteins-although we acknowledge that the relevant players could be particularly low abundant and thereby missed by mass spectrometry. Given that BFD1 deletion did not elicit a similar reduction in fitness 23 , we hypothesize that minor defects in the Δbfd2 strain-subtle enough to be overlooked in cellular assays-ultimately result in reduced growth. Whereas loss of BFD2 in our previous ME49 screens also reflected compromised fitness 23 , deletion in Pru (type II) or RH (type I) Toxoplasma strains seems to be innocuous 39,54 . This suggests that strain-specific differences might account for the deficiencies of the Δbfd2 tachyzoites.
BFD2 binding to BFD1 mRNA seems to be necessary for its translation. Given that BFD2 is expressed throughout the asexual cycle, it    Table 6. b, Log-transformed enrichment ratios for all transcripts detected in alkaline-stressed HA-BFD2 parasites. Using mixture modelling, the data were fitted with a combination of two Gaussian curves, one corresponding to a broad range of moderately or non-interacting RNAs (black line) and the other representing our conservative cutoff for enrichment (green line). c, Relative abundance of mRNAs following IP of HA-BFD2 compared with the input transcriptome. Enrichment was defined based on the LOD ratio between the Gaussian functions plotted in b. Transcripts with a LOD > 0 (that is, having a greater likelihood of falling within the upper distribution) were considered highly enriched (green). Grey points represent all other RNAs that did not meet this statistical cutoff. d, Representative transcripts reflecting varying degrees of enrichment following IP of HA-BFD2. Profiles display the total read count at each nucleotide position in either input or IP-enriched samples, with paired tracks plotted on the same scale (denoted in parentheses) for a given gene. remains an open question how its stress-dependent effects are conferred. The observation that BFD1 is enriched by BFD2 pulldown under alkaline-stress (but not unstressed) conditions seems to implicate differential binding as the event licensing the translation of BFD2 targets. However, given that BFD2 expression is substantially higher in differentiating parasites, differences in the amount of BFD2 captured by IP prevent us from unambiguously attributing increased association with BFD1 to changes in binding affinity. Alternatively, BFD2 may bind mRNA constitutively, such that its translational influence is regulated through other dynamic interactions. While this manuscript was under revision, BFD2 was found among several targets of a conserved serine/ threonine phosphatase, protein phosphatase 2A (PP2A) 54 . Mutants of PP2A exhibited BFD2 hyperphosphorylation, which coincided with reduced DBL staining and incomplete induction of bradyzoite genes under stress-reminiscent of the partial differentiation of BFD2 knockouts. Although the molecular consequences of these and other post-translational modifications on BFD2 are yet to be tested, this result could support a model wherein stress-dependent dephosphorylation of BFD2 enables the recruitment of factors necessary to translate its mRNA clients.
Conditional overexpression of BFD2 is sufficient to drive stage conversion irrespective of stress. This relationship does not preclude the function of BFD1 as master regulator of the chronic stage, given that our collective data suggest that differentiation depends on the transcriptional changes induced by BFD1. However, the regulatory machinery involved seems to be responsive to the concentration of BFD2 and can be overwhelmed stoichiometrically. Based on this observation, we propose that once the amount of BFD2 exceeds a certain threshold (for example, via transcriptional activation by BFD1), maintenance of the bradyzoite state becomes less dependent on the signals that initiated it. This phenomenon, known as hysteresis, is well-described in other developmental programmes, such as cell-type determination in multicellular organisms and commitment to cell division [55][56][57][58][59] . In the context of the T. gondii chronic stage, such a circuit could help maintain parasites in a differentiated state in the presence of variable stimuli.
Feedback loops endow genetic circuits with several notable features. The discovery of a positive feedback loop acting on BFD1 therefore has fundamental implications for the properties of T. gondii latency. In addition to signal amplification, these circuits frequently exhibit bistability. That is, whereas hysteresis renders a genetic system less susceptible to noise, bistability confers switch-like binary control 55 . It was previously suggested that acute-to-chronic-stage conversion in T. gondii involves the production of slow-growing intermediates, termed pre-bradyzoites 33,35,36 . This idea is not inconsistent with a binary model, given that bistable systems can encode more than two cell states due to non-equilibrium 60 . Nevertheless, our observation that Δbfd2 parasites partially induce the chronic-stage programme but cannot produce long-lived cysts in mice suggests that robust binary commitment is important. Perhaps more intriguingly, bistable systems exhibiting hysteresis are capable of 'remembering' a stimulus long after it has been removed 55 . Network theory suggests that a double-positive loop such as we describe for BFD1 and BFD2 should lock into a self-perpetuating steady state 55 . This raises the question of how the bradyzoite programme is overturned during reactivation. It is possible that the absence of stress may be sensed directly by the circuit. Alternatively, there may be additional reactivation factors that antagonize BFD1-BFD2 feedback. Other groups have proposed that differentiation requires a balance of regulators, including proteins that repress the transcription of bradyzoite-specific genes 33,35,36 . Whether these or other factors directly inhibit BFD1 or BFD2 is yet to be examined.
In summary, our work shows that BFD2 is a major determinant of T. gondii persistence due to its involvement in a positive feedback loop with the master regulator BFD1. This discovery opens up exciting new questions surrounding the regulation of the BFD1 programme, particularly as it relates to the maintenance and reactivation of chronic stages. Further exploration of the BFD1-BFD2 circuit will probably uncover heuristics that govern timing and commitment in differentiation. Together, these advances will aid in the design of therapeutic interventions to prevent or reverse chronic stages, bringing us closer to a radical cure against toxoplasmosis.

Parasite and host cell culture
T. gondii parasites were maintained in human foreskin fibroblasts (HFFs) at 37 °C under 5% CO 2 in standard medium consisting of DMEM (GIBCO) supplemented with 3% heat-inactivated fetal bovine serum (FBS), 2 mM l-glutamine (GIBCO) and 10 μg ml −1 gentamicin (Thermo Fisher). Routine passaging was performed by scraping and transfer to fresh HFFs. For experimental infections, parasites were released from host cells by extrusion through a 27-gauge needle (that is, syringe release) and allowed to invade in standard medium containing 10% FBS. The alkaline-stress medium used for differentiation consisted of RPMI 1640 supplemented with 1% FBS, 10 μg ml −1 gentamicin and 50 mM HEPES, adjusted to pH 8.1 with 10 N NaOH. IAA was used at 50 μM in standard medium and 500 μM under alkaline stress. Shield-1 was used at 3 μM in standard medium 61 .

T. gondii strain generation
Unless otherwise stated (that is, PruΔbfd2), strains were generated in the ME49 genetic background. Cloning was performed using Q5 polymerase (NEB) and NEBuilder HiFi DNA assembly master mix. Oligonucleotides were ordered from IDT (full list in Supplementary Table  9). When needed, the following concentrations were used for selection: pyrimethamine, 3 μM; chloramphenicol, 20 μM; mycophenolic acid, 25 μg ml −1 ; and xanthine, 50 μg ml −1 . Subcloning was performed in 96-well plates by sorting or plating at limiting dilution, with strains verified by at least two rounds of genotyping PCR and sequencing.

TIR1-expressing ME49.
To generate a TIR1-expressing ME49 strain (ME49/TIR1), the heterologous TIR1 auxin receptor was introduced into a defined neutral locus on chromosome VI 62 of ME49Δku80Δhxgprt parasites 23 . Integration was aided by a previously validated gRNA-Cas9-expression plasmid targeting the neutral locus (GenBank accession no. MN019116). A repair template encoding TIR1 driven by the α-tubulin promoter (pTUB1-TIR1) and CAT expression cassette was amplified from Addgene plasmid no. 87258 using the oligonucleotides P1 and P2, which include regions of homology to the integration site. Transfectants were selected for 2 weeks in chloramphenicol before subcloning, with single-plaque wells screened using P3 and P4.
mNG-mAID-tagged conditional-knockdown strains. In ME49/TIR1, genes were endogenously tagged with mNG-mAID using a previously described high-throughput (HiT) strategy 38 . Briefly, cutting units specific to each candidate were purchased as IDT gBlocks (P5-P10) and assembled with the empty mNG-mAID HiT vector (GenBank accession no. OM640005; Extended Data Fig. 8a). The constructs were BsaI-linearized and co-transfected with pSS014 Cas9-expression plasmid (GenBank accession no. OM640002) and the transfectants were Article https://doi.org/10.1038/s41564-023-01358-2 selected in pyrimethamine. Single-plaque wells of each strain were genotyped using a common strategy (P11-P23; Extended Data Fig. 1) using PCR to verify both integration of mNG-mAID and reciprocal loss of the wild-type untagged allele. Sequencing further verified maintenance of the reading frame across junctions where recombination occurred.

BFD2-knockout strains.
Oligonucleotides encoding gRNAs targeting the 5′ (P24 and P25) and 3′ (P26 and P27) ends of the BFD2 coding sequence were assembled with the pU6-Universal Cas9-expression plasmid (GenBank accession no. OM640003) and sequenced using P28. A repair template consisting of pTUB1-tdTomato was amplified using P29 and P30, which encode 40 base pairs (bp) of homology to regions immediately up-and downstream of the gRNA target sites. The gRNAs and the repair construct (25 μg each) were transfected into ME49ΔKu80Δbfd1::BFD1-TY and ME49ΔKu80Δbfd1/ HXGPRT::pTUB1-DD-BFD1-TY parasites (described in ref. 23) to generate Δbfd2 and DD-BFD1-TYΔbfd2, respectively. After 2-3 d, knockouts were sorted by FACS directly into 96-well plates and single-plaque wells were screened for integration of tdTomato (P31 and P32) and reciprocal loss of BFD2 (P31 and P33). For PruΔbfd2, the same gRNAs were used to target BFD2 in Pru-OVA-tdTomato 63 parasites; however, BFD2 was replaced with an HXGPRT cassette to allow selection in mycophenolic acid and xanthine.
Deletion of BFD1. PruΔbfd1 was generated with published reagents 23 , using the same FACS-based strategy described in the previous section. Briefly, gRNA-Cas9-expression plasmids targeting either end of BFD1 were transfected into Pru-OVA-tdTomato parasites along with a repair template encoding pTUB1-mNG flanked by 40 bp of homology adjacent to the cut sites. The transfectants were sorted for mNG expression and subcloned, and a lineage lacking BFD1 was verified by PCR and sequencing. However, despite being mNG-positive, pTUB1-mNG was not found within the BFD1 locus, indicating that it had randomly integrated into the parasite genome.
HA-BFD2 WT and HA-BFD2 ΔZF -complemented strains. HA-tagged BFD2-either full-length or lacking the zinc-finger domains-was reintroduced into ME49ΔKu80Δbfd1::BFD1-TYΔbfd2 at the endogenous locus, allowing FACS enrichment of complemented parasites based on the loss of tdTomato. Guide RNAs targeting the 5′ and 3′ ends of the fluorescence cassette (P34 and P35, and P36 and P37, respectively) were annealed and assembled with the pU6-Universal plasmid. For the full-length repair template, BFD2 was amplified from cDNA (P38 and P39) and assembled with PCR fragments spanning approximately 500 nucleotides immediately up-(P40 and P41) and downstream (P42 and P43) of the coding sequence. The HA epitope was introduced at the junction between the upstream amplicon and BFD2 via overhangs in the corresponding reverse and forward primers (that is, P41 and P38). The ΔZF version was generated similarly but with BFD2 amplified as two separate fragments (P38 and P44, and P45 and P39) to exclude the zinc-finger domains. The annealed products were then used as templates for PCR to generate approximately 15 μg of linear repair construct (P46 and P47), which was co-transfected with 20 μg of each gRNA plasmid into Δbfd2. After an initial round of FACS to enrich for low/non-fluorescent cells, complemented mutants were sorted directly into 96-well plates for subcloning. Single-plaque wells were screened by PCR for reintroduction of tagged BFD2 (P31 and P32) and loss of tdTomato (P31 and P23).
Conditional overexpression of BFD2. The DD-HA-BFD2 strain with conditional expression of BFD2 was generated using a derivation of the HiT vector strategy 38 , modified for manipulating the N termini of targeted genes. A cutting unit specific to BFD2 was purchased as an IDT gBlock (P48) and Gibson-assembled with the empty pTUB1-DD HiT vector (GenBank accession number pending; Extended Data Fig. 8b). The resultant plasmid was sequence-verified with P49, then BsaI-linearized and transfected into ME49ΔKu80Δbfd1::BFD1-TY parasites with 50 μg of pSS014 Cas9-expression plasmid. Mutants were selected for 1 week in pyrimethamine before subcloning. Single-plaque wells were PCR-screened for successful integration of the construct (P51 and P32) and reciprocal loss of the wild-type allele (P50 and P32).

Criteria for identifying putative effectors of BFD1
Based on previous BFD1 CUT&RUN chromatin profiling 23 , we generated a list of genes that seemed to be direct targets of BFD1 (that is, promoters encompassing BFD1 peaks, called by MACS2). Of these, eight encoded putative RNA-or DNA-binding domains based on publicly available domain annotations and gene ontology. A final requirement of significant transcriptional induction (adjusted P < 0.05; calculated by DESeq2) in response to either alkaline stress or conditional BFD1 expression further narrowed the list to the five genes examined in this study.

Plate-based differentiation screen
Parasites were inoculated onto HFFs in black-wall 96-well microplates for 4 h in standard media, then washed twice with PBS and switched to alkaline medium containing 500 μl IAA or an equivalent volume of PBS (vehicle). After 48 h, the plates were fixed and permeabilized as described earlier. The wells were treated sequentially with 50 μl of primary (guinea pig anti-CDPK1) and secondary (Alexa 594-conjugated goat anti-guinea pig, DBL-488 and Hoechst 33258) staining solutions for 1 h each, with three PBS washes in between. Fluorescence images were collected at ×20 magnification using a Cytation3 cell imaging multi-mode plate reader (Agilent Technologies).
For blinded scoring of DBL staining, a mask was created for each image in Fiji (v. 2.1.0/1.53c) based on CDPK1 to delineate individual vacuoles, regardless of the differentiation status. This was applied to the corresponding DBL scan from each set, creating a second mask to enable visualization of the isolated DBL-staining profile for each vacuole. Using the Interactive Learning and Segmentation Toolkit (ilastik v. 1.3.3) 66 , vacuoles were categorized as negative, intermediate or high based on the intensity and variance of the DBL signal.

RNA-seq analysis of conditional-knockdown strains
Syringe-released parasites were allowed to invade HFFs in 15-cm dishes under standard conditions for 4 h; the monolayers were then rinsed Article https://doi.org/10.1038/s41564-023-01358-2 with PBS and switched to alkaline-stress media containing either 500 μM IAA or PBS (vehicle). After 96 h, the parasites were harvested by sequential extrusion through 27-and 30-gauge needles, followed by passage through a 5-μm filter, pelleted by centrifugation (1,200g for 10 min) and flash-frozen in TRIzol (Thermo Fisher). Total RNA was isolated using a Zymo Direct-zol RNA miniprep kit and the RNA quality was assessed using a BioAnalyzer or Fragment analyser. All samples were split in two at the time of RNA isolation and treated as technical replicates in downstream processing. Libraries were generated using a Swift rapid RNA library kit for stranded RNA-seq (Swift Biosciences) and subjected to 150 × 150 paired-end sequencing on the NovaSeq 6000 platform (Illumina). Reads were trimmed using Cutadapt v.3.6 and aligned to the ME49 genome (ToxoDB v.56 assembly) using STAR v.2.7.0a. Differential expression analysis was done using the DESeq2 package (v. 1.12.3) in R. An adjusted P value of 0.05 was used as a statistical cutoff for differential expression.

RNA-seq analysis of BFD2 deletion
In HFFs. HFF monolayers in 15-cm dishes were infected with syringe-released parasites for 4 h, rinsed with PBS and either returned to standard medium or treated with one of two differentiation triggers. For Δbfd2 and its parental strain (Δbfd1::BFD1-TY; Fig. 2i-k), stage conversion was induced by alkaline stress. For DD-BFD1-TY and DD-BFD1-TYΔbfd2 (Fig. 4d), parasites received standard media supplemented with 3 μM Shield-1 or the equivalent volume of 100% ethanol (vehicle). The parasites were cultured for an additional 48 h before being released from the host cells by scraping, syringe release and filtering, as described earlier, followed by collection by centrifugation (1,000g for 10 min) and flash-freezing in liquid nitrogen. Total RNA was isolated using a Qiagen RNeasy plus micro kit following the protocol for animal and human cells; the cell pellets were split in two and treated as technical replicates for the downstream processing. Library preparation, sequencing and analysis were performed with the same pipeline as that applied to the conditional-knockdown strains.
In mouse primary neurons. Primary mouse neurons were generated as described previously 67 , with all procedures approved by the University of Arizona's Institutional Animal Care and Use Committee (12-391). Briefly, fetal pups were isolated from pregnant Cre reporter mice 68 ( Jackson Laboratories, cat. no. 007906) and dissected to recover the cerebral hemispheres, regardless of sex. Cortical and hippocampal neurons were recovered using established protocols 69 and seeded onto poly-l-lysine-coated six-well plates containing DMEM medium supplemented with 0.6% d-glucose (Sigma), l-glutamine (Thermo Fisher) and 5% FBS. After 4 h, the medium was changed to complete Neurobasal media (Thermo Fisher) supplemented with B27 (Thermo Fisher), l-glutamine and penicillin-streptomycin antibiotic cocktail. On day 4, 5 μM cytosine arabinoside was added to the neurons to stop glial proliferation. On day 10, the neurons were infected (MOI of about 2) with parental, PruΔbfd1 or PruΔbfd2 parasites and left for an additional 48 h in culture before being dissociated with accutase (5 min at 37 °C), pelleted by centrifugation and flash-frozen in liquid nitrogen. RNA was then isolated and analysed identically to the workflow described for HFFs above.

Plaque assays
HFFs in six-well plates were inoculated with 1 × 10 3 parasites and left undisturbed in standard media containing 10% heat-inactivated fetal bovine serum. After 14-16 d (indicated in the respective figures), the plates were rinsed with PBS, fixed with 100% ethanol for 10 min and dried. The following day the wells were stained for 5 min with crystal violet solution (12.5 g crystal violet, 125 ml of 100% ethanol and 500 ml of 1% ammonium oxalate), rinsed with water and imaged. Plaques were manually delineated and their areas quantified using Fiji 70 .

Host cell invasion assays
Assays were performed in glass-bottomed 24-well plates seeded with HFFs. Immediately before infection, the wells were rinsed with PBS and filled with 500 μl of pre-warmed invasion medium (DMEM without sodium bicarbonate supplemented with 20 mM HEPES and 1% heat-inactivated fetal bovine serum, and adjusted to pH 7.4). Syringe-released parasites (7 × 10 5 ) were added to each well, settled onto the monolayer by centrifugation (290g for 5 min) and then allowed to invade at 37 °C for 20 min before fixing in 4% formaldehyde for 10 min. Blocking solution was added to the wells, which were incubated for 10 min (see the 'Immunofluorescence assays' section) and stained for SAG1 for 30 min. After permeabilizing with 0.25% Triton X-100 for 8 min, the wells were blocked a second time and stained for DNA and CDPK1, allowing discrimination between intracellular (CDPK1 + SAG1 − ) and extracellular (CDPK1 + SAG1 + ) parasites. Biological replicates were performed on different days, with technical replicates being individually infected wells. At least five fields were imaged per well using a Nikon Eclipse Ti microscope and counts were averaged to create an overall measure of invasion, expressed as the number of intracellular parasites per host-cell nucleus.

Intracellular replication assays
Δbfd1::BFD1-TY and Δbfd2 parasites were infected onto coverslips seeded with HHFs in standard media and allowed to grow for 24 h. The samples were fixed, permeabilized, and stained for DNA and CDPK1 as described in the 'Immunofluorescence assays' section. The number of parasites per vacuole was calculated from a minimum of 100 vacuoles per strain.

Extracellular stress assays
Single-cell suspensions of extracellular Δbfd1::BFD1-TY and Δbfd2 parasites were generated by scraping, extruding cultures through 27-gauge needles and passing through 5-μm filters. The suspension (50 μl) was inoculated onto HFFs on coverslips in standard media, and the remaining volume was maintained at 37 °C. At intermittent time points thereafter, additional coverslips were inoculated with the suspensions at an equivalent MOI. After 24 h, all coverslips were fixed and stained for DNA and CDPK1 as described earlier in the 'Immunofluorescence assays' section. Infectivity at each time point was quantified as the number of vacuoles containing two or more parasites normalized to the number of host nuclei per field. Within strains, data were normalized to the infectivity at t = 0.

Quantitative analysis of immunofluorescence assays
All analyses were performed using Fiji 70 . For DBL quantification, individual vacuoles were manually defined as regions of interest (ROIs) based on a CDPK1 counterstain. The ROIs were saved using the ROI Manager function, applied to the corresponding DBL channel and their pixel intensity (Integrated Density) was quantified. As a measure of background, the pixel intensity over uninfected regions of the monolayer was quantified. Corrected total fluorescence (CTF) for individual ROIs was calculated using the following function: CTF = [integrated density − (ROI area × mean background)] / (ROI area). Cytosolic HA-BFD2 accumulation was determined similarly, with individual parasites delineated based on the HA signal itself. For BFD1-TY, parasite nuclei were defined as ROIs based on Hoechst staining and the background measured from non-nuclear regions of intravacuolar parasites.

Acute-stage quantitative proteomics
Sample harvest and processing. Δbfd1::BFD1-TY and Δbfd2 parasites were cultured in HFFs in 15-cm dishes under standard conditions for 48 h, and then harvested by scraping, extrusion through a 27-gauge needle and passage through 5-μm filters. Samples were pelleted by centrifugation at 1,200g for 10 min and resuspended in lysis buffer (10% SDS, 100 mM TEAB pH 7.5, 2 mM MgCl 2 and 2×HALT protease inhibitor) Article https://doi.org/10.1038/s41564-023-01358-2 for storage at −80 °C until further steps. Proteins were prepared for quantitative mass spectrometry exactly as described previously 71 , using a modified version of the S-trap protocol (Protifi), followed by lyophilization, labelling with TMT reagents (Thermo Fisher) and fractionation.
Mass spectrometry acquisition. Samples were analysed using an Orbitrap Eclipse Tribrid mass spectrometer equipped with a FAIMS Pro source 72 connected to a Vanquish HPLC chromatography system, using 0.1% formic acid as Buffer A and 80% acetonitrile plus 0.1% formic acid as Buffer B. Peptides were loaded onto a heated analytical column (Thermo Fisher, ES902; PepMap Neo Trap Cartridge C18 2 μm, 100 Å, 75 μm × 25 cm, 40 °C) via trap column (Thermo Fisher, 174500; Acclaim PepMap C18 5 μm, 100 Å, 300 μm × 5 cm nanoViper) and separated at 300 nl min −1 on a gradient of 3-25% Buffer B for 90 min, 25-40% Buffer B for 30 min, 40-95% Buffer B for 10 min and 95% Buffer B for 6 min. The Orbitrap and FAIMS were operated in positive-ion mode with a positive ion voltage of 2,100 V, ion transfer tube temperature of 300 °C, standard FAIMS resolution and compensation voltage of −45, −55 and −65 V with 4.2 ml min −1 carrier gas. Full-scan spectra were acquired in profile mode at a resolution of 120,000, with scan range of 400-1,600 m/z, 300% AGC target, automatically determined maximum injection time, intensity threshold of 2 × 10 4 , 2-6 charge state, dynamic exclusion of 60 s and mass tolerance of 10 ppm. MS2 spectra were generated in centroid mode using quadrupole isolation, an isolation window of 0.7 m/z, CID collision energy of 35 h, activation time of 10 ms, ion trap detection with a turbo scan rate, standard AGC target and maximum injection time set to automatic. Real-time search was performed using the ToxoDB GT1 protein database (release 49), including static methylthio (+45.9877 Da; C), static TMT6plex (+229.1629; K and any N termini) and variable oxidation (+15.9949; M). TMT SPS MS3 mode was used, enabling close-out, a maximum peptides per protein of five and a maximum search time of 40 ms. Scoring thresholds were set for Xcorr (2), dCn (0.1), precursor ppm (10) and charge state (2). MS3 spectra were generated in centroid mode, with a synchronous precursor selection of ten, mass spectrometry isolation window of 2 m/z, MS2 isolation window of 2 m/z, HCD collision energy of 55 at a resolution of 50,000, a mass range of 100-500 m/z and 200% AGC target for ten data-dependent scans.
Proteomic analysis. Raw files were analysed in Proteome Discoverer (v. 2.4) to generate peak lists and protein and peptide IDs using Sequest HT (Thermo Fisher) and the ToxoDB ME49 protein database (release 43). The following modifications were included in the search: dynamic oxidation (+15.995 Da; M), dynamic acetylation (+42.011 Da; N terminus), static TMT6plex (+229.163 Da; any N terminus), static TMT6plex (+229.163 Da; K) and static methylthio (+45.988 Da; C). Exported peptide and protein abundance files were loaded into R for further analysis.

Assessment of acute pathogenicity and brain cyst formation
Mice. Six-week-old female CD-1 mice (Charles River) were maintained at the Whitehead Institute for 3 weeks before being used in experiments. The mouse facilities were kept on a 12 h dark-light cycle under ambient temperature and humidity. All animal protocols were approved by the Institutional Animal Care and Use Committee at the Massachusetts Institute of Technology (0220-004-23).

Mouse infections.
Mice were infected by i.p. injection with 100 acute-stage parasites of either strain in 100 μl PBS (n = 10 each) or PBS alone (mock; n = 4). Their weights were recorded daily. Animals displaying excessive distress or morbidity were euthanized with CO 2 , followed by cervical dislocation. The surviving animals were killed 45 d post infection.
Brains were harvested from all mice at euthanasia. Brains from moribund animals were used in their entirety for cyst quantification.
For animals surviving to day 45, their brains were bisected at the time of harvest and one hemisphere was set aside for immunohistochemistry.

Ex vivo quantification of cysts.
Brains were dissected into 2 ml PBS and homogenized by repeated extrusion through an 18-gauge needle until uniformity was achieved. The homogenates were brought to 10 ml with PBS, pelleted by centrifugation (1,000g for 5 min) and the supernatant was discarded. The homogenates were resuspended in 700 μl PBS,the final volume was recorded and 100 μl was then combined with 900 μl of ice-cold 100% methanol and fixed at room temperature. After 5 min, the samples were centrifuged at 5,200g for 5 min (conditions used for all subsequent centrifugation steps), washed with 1 ml PBS, centrifuged again, and the supernatant was removed by aspiration. To label cysts, the pelleted samples were resuspended in 500 μl of primary staining solution (1:150 DBL-488 and 1:1,000 guinea pig anti-CDPK1 in PBS) and incubated overnight at 4 °C with rotation. The following day, the samples were centrifuged, washed with 1 ml PBS and incubated in secondary staining solution (anti-guinea pig-594, diluted 1:1,000 in PBS) with rotation at room temperature. After 1 h, the homogenates were washed again and resuspended in 1 ml PBS. For each sample, four 50-μl aliquots were plated into a clear-bottomed 96-well microplate and examined at ×20 magnification by fluorescence microscopy. Cysts were identified by double-positivity for both DBL and CDPK1. With the exception of those prepared from moribund mice, all samples were quantified blinded to allocation during experiments. Cyst burdens represent the mean of the four counts collected for each sample multiplied by the appropriate dilution factor.
Immunohistochemistry. Brain tissue was fixed in 10% formalin, embedded in paraffin and sectioned. Deparaffinized sections were hydrated and antigen was retrieved in 0.01 M sodium citrate buffer at pH 6.0, followed by blocking of endogenous peroxidase with 0.3% H 2 O 2 . Tissues were blocked with 2% goat serum (Invitrogen). T. gondii parasites were detected with rabbit anti-Toxoplasma antibody 73 (gift from F. Araujo, Palo Alto Medical Foundation; 1:100), followed by biotinylated goat anti-rabbit IgG (Vector Laboratories). ABC reagent (Vector Laboratories) was then applied and DAB substrate (Vector Laboratories) was added to visualize the T. gondii staining. The slides were counterstained with haematoxylin to visualize nuclei. Images were acquired at ×10 magnification using a Leica DM6000 wide-field fluorescence microscope.

RT-qPCR
Parasites were inoculated onto HFFs in T12.5 flasks for 4 h; the monolayers were then rinsed with PBS and returned to standard medium or treated to induce differentiation either by stress (Δbfd2 and Δbfd1::BFD1-TY; Fig. 4a) or 3 μM Shield-1 (DD-BFD1-TY and DD-BFD1-TYΔbfd2; Fig. 4e). After 48 h, the cultures were scraped, collected by centrifugation (1,000g for 10 min) and flash-frozen in liquid nitrogen. Total RNA was isolated using a Qiagen RNeasy plus micro kit, DNaseI-treated with a TURBO DNA-free kit (Thermo Fisher) and then quantified using a NanoDrop spectrophotometer for use in first-strand cDNA synthesis with SuperScriptIII (Thermo Fisher) and random hexamer priming. Dye-based quantitative PCR was performed using PowerUp SYBR Green master mix (Thermo Fisher) with 'No-RT' controls included for each sample. Before use, all qPCR primers (Supplementary Table 3) were verified for amplification efficiency in the 95-105% range and specificity by melt curve analysis. Reactions were run on a QuantStudio6 real-time PCR system (Applied Biosystems) using 'Standard' ramp speed with the following parameters: 2 min at 50 °C, 2 min at 95 °C, and 40 cycles of 15 s at 95 °C, followed by 1 min at 60 °C. The transcripts of interest were quantified by comparative Ct (ΔΔCt) using GCN5B as an internal control and normalizing to uninduced samples. Article https://doi.org/10.1038/s41564-023-01358-2

BFD2 RNA IP
Parasites were inoculated onto HFFs in 15-cm dishes for 4 h and then switched to alkaline stress or kept under standard conditions. After 48 h, the unstressed cultures were harvested by scraping and extrusion through a 27-gauge needle and 5-μm filter, pelleted by centrifugation (1,200g for 10 min) and frozen at −80 °C in complete Magna RIP lysis buffer (Millipore). The stressed samples were processed similarly after 72 h but with an additional extrusion through a 30-gauge needle before filtering. Approximately 2 × 10 7 parasites were harvested for each strain and condition. Unenriched 'input' samples were generated by reserving 10% of the starting lysate; the remaining volume was subjected to IP using the Magna RIP kit (Millipore) with 5 μg mouse anti-HA IgG (BioLegend, clone 16B12), coated onto protein A/G magnetic beads as described in the Magna RIP technical manual. RNA purified from both IP and input samples was concentrated by ethanol precipitation and resuspended in equivalent volumes of RNase-free water for use in downstream analyses.

Detection of BFD2-associated mRNA
Analysis of specific targets by RT-qPCR. Samples were treated with TURBO DNaseI (Thermo Fisher) and cDNA was generated from equivalent volumes of IP and input RNA using SuperScriptIII with random hexamer priming. Dye-based qPCR was performed as described earlier using primers from Supplementary Table 3. Enrichment of transcripts in the IP samples was calculated using the Percent Input method.
Unbiased discovery of BFD2 targets through next-generation sequencing. Input and IP samples from stressed HA-BFD2 parasites were poly(A)-selected using a NEBNext poly(A) mRNA magnetic isolation module (NEB) and libraries were generated using a SMARTer pico input stranded total RNA-seq kit (Takara Bio), excluding the step for mammalian ribosomal RNA depletion. The samples were run with 100 × 100 paired-end sequencing on the NovaSeq 6000 platform (Illumina). Reads were trimmed and mapped to the ME49 genome as described earlier.
To call BFD2-enriched messages, we adapted a previously described RNA IP and sequencing enrichment analysis pipeline 74 . For each transcript, raw read counts were used to calculate the read coverage (d), defined as the square root of the sum of squares for reads in both the IP and input control, that is, d = √(IP × IP + input × input). Only transcripts with more than five mapped reads in both the IP and input samples and d > 10 were included in the downstream analysis. To normalize for sequence depth, we converted raw reads to TPM for both the IP and input samples. This was then used to generate an enrichment ratio for each gene, defined as TPM IP /TPM input .
We performed Gaussian mixture modelling on log 2 -transformed enrichment ratios using the normalmixEM function in the R package mixtools 75 . Informed by previous studies 74 , model fitting was restrained to two distributions, assuming equal variance. Convergence was reached within 250 iterations, producing the Gaussian curves plotted in Fig. 6b. Posterior probabilities were then generated from the mixture model and used to calculate an odds ratio for each gene. An LOD ratio greater than zero indicates a higher probability of falling within the upper-rather than lower-distribution; thus, LOD > 0 was used as a threshold for enrichment. Note that several genes with evidence of BFD2 binding (that is, GCN5B and ASP5) do not meet this statistical threshold. Thus, we consider this a conservative cutoff for highly enriched mRNAs.

Statistical analysis
Information about biological replicates, the number of observations and the exact statistical tests performed is in the figure legends. For the RNA-seq analysis, adjusted P values were calculated using DESeq2, which uses the Wald test as a default. All other statistical tests were performed in Prism (GraphPad).

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
For BFD1, BFD2, TGME49_253790 and TGME49_224630, the subsets of genes identified as significantly affected by depletion are listed in Supplementary Table 1. The results of quantitative proteome profiling in Δbfd2 parasites are provided in Supplementary Table 2. Primers used for RT-qPCR analysis of BFD1, BFD2, GCN5B and ASP5 are available in Supplementary Table 3. The Ct values and analyses of the RT-qPCR experiments are provided in Supplementary Tables 4-6. Raw reads and analysis of RNA IP and sequencing for BFD2 are provided in Supplementary Table 7. Minimally processed results from all bulk RNA-seq performed in this study are provided in Supplementary  Table 8. The oligonucleotides used in this study for molecular cloning are in Supplementary Table 9. Unprocessed data from the transcriptional and proteome profiling described herein are available through Gene Expression Omnibus (GEO) and the Proteomics Identification Database (PRIDE), respectively, under the following accession numbers: GSE223819 (  H1 and H2; Fig. 1b) in dark grey. In each case a common gene-specific forward primer (P11-P16) was paired with reverse primers to either mNG (P17) or the respective endogenous 3′ UTR (P18-P23) downstream of the integration site. The expected PCR product size is given for each template and primer combination. Refer to Supplementary Table 9 for a complete list of the primer sequences. b, PCRs were performed on ME49/TIR1 gDNA (parental) as a control in addition to the respective tagged strain. Lanes are labelled with the reverse primer and gDNA template used in each reaction. For positive clones, bands were extracted and subjected to Sanger sequencing to verify in-frame integration of the tag. Fig. 3 | Genotyping and characterization of BFD2-deficient parasites. a, Δbfd2 clones were screened by PCR for replacement of the endogenous coding sequence with tdTomato using a common forward primer in the BFD2 5′ UTR (P31) and reverse primers binding to either BFD2 (P32) or the fluorescent reporter (P33). The diagram shows priming in the parental strain (top) and at the modified allele in Δbfd2 (bottom), with the expected product sizes indicated. The reverse primers and gDNA template used in each reaction are listed above the respective lane. b, Plaque assays after 16 d of undisturbed growth. c, Comparison of protein abundance in unstressed parental (Δbfd1::BFD1-TY) and Δbfd2 parasites. Quantitative proteomics identified a total of 29,806 unique peptides corresponding to 4,303 individual proteins. Significantly affected proteins (magenta) were defined as those meeting three criteria: (1) a minimum of two unique peptides, (2) absolute fold change > 2 and (3) P < 0.05. Differences are limited to canonical bradyzoite markers (CST1, LDH2 and SRS35A) or other developmentally regulated genes (TgSPT2) 76 . d,e, Effects of BFD1 or BFD2 deletion on the parasite transcriptome during infection of mouse primary neurons. Data are based on n = 3 independent infections with colour assigned based on log 2 (fold change) during conditional BFD1 expression 23 . Significantly affected genes (adjusted P < 0.05, Wald test in DESeq2) are indicated by larger point size. d, Differential expression analysis was performed for parasites lacking either factor, as compared with the parental strain. The number of genes meeting the cutoff for statistical significance is indicated. e, Comparison of the effects of BFD1 versus BFD2 deletion reveals a comparatively larger impact for the former. Pearson's correlation was performed on all significant points with a trend line fit by linear regression.

Extended Data
Article https://doi.org/10.1038/s41564-023-01358-2 Extended Data Fig. 5 | BFD2 deletion in a conditional BFD1 strain. a, Schematic of ligand-inducible BFD1. The DD-BFD1-TY strain was constructed previously 23 by integration at the HXGPRT locus in the Δbfd1 genetic background. A heterologous promoter (pTUB1) drives expression of the transgene but DD-BFD1-TY protein is only stabilized following treatment with Shield-1. b, Validation of BFD2 knockout in DD-BFD1-TY. Selected clones were screened using the same strategy described in Extended Data Fig. 3, verifying both loss of endogenous BFD2 and replacement with the fluorescence cassette. The gel (bottom) shows PCRs performed on gDNA from both the parental (DD-BFD1-TY) and DD-BFD1-TYΔbfd2 strains. The reverse primers used are listed over the respective well, with the expected product sizes indicated in the diagram (top).