key: cord-0853154-wciv8z5k authors: Rinchai, Darawan; Deola, Sara; Zoppoli, Gabriele; Ahamed Kabeer, Basirudeen Syed; Taleb, Sara; Pavlovski, Igor; Maacha, Selma; Gentilcore, Giusy; Toufiq, Mohammed; Mathew, Lisa; Liu, Li; Vempalli, Fazulur Rehaman; Mubarak, Ghada; Lorenz, Stephan; Sivieri, Irene; Cirmena, Gabriella; Dentone, Chiara; Cuccarolo, Paola; Giacobbe, Daniele Roberto; Baldi, Federico; Garbarino, Alberto; Cigolini, Benedetta; Cremonesi, Paolo; Bedognetti, Michele; Ballestrero, Alberto; Bassetti, Matteo; Hejblum, Boris P.; Augustine, Tracy; Panhuys, Nicholas Van; Thiebaut, Rodolphe; Branco, Ricardo; Chew, Tracey; Shojaei, Maryam; Short, Kirsty; Feng, Carl; Zughaier, Susu M.; Maria, Andrea De; Tang, Benjamin; Hssain, Ali Ait; Bedognetti, Davide; Grivel, Jean-Charles; Chaussabel, Damien title: High temporal resolution systems profiling reveals distinct patterns of interferon response after Covid-19 mRNA vaccination and SARS-CoV2 infection date: 2022-01-26 journal: bioRxiv DOI: 10.1101/2021.12.12.472257 sha: 74289bc99a90735923eab4eb05fd2b2ec4c57a5f doc_id: 853154 cord_uid: wciv8z5k Knowledge of the mechanisms underpinning the development of protective immunity conferred by mRNA vaccines is fragmentary. Here we investigated responses to COVID-19 mRNA vaccination via ultra-low-volume sampling and high-temporal-resolution transcriptome profiling (23 subjects across 22 timepoints, and with 117 COVID-19 patients used as comparators). There were marked differences in the timing and amplitude of the responses to the priming and booster doses. Notably, we identified two distinct interferon signatures. The first signature (A28/S1) was robustly induced both post-prime and post-boost and in both cases correlated with the subsequent development of antibody responses. In contrast, the second interferon signature (A28/S2) was robustly induced only post-boost, where it coincided with a transient inflammation peak. In COVID19 patients, a distinct phenotype dominated by A28/S2 was associated with longer duration of intensive care. In summary, high-temporal-resolution transcriptomic permitted the identification of post- vaccination phenotypes that are determinants of the course of COVID-19 disease. INTRODUCTION 73 COVID-19 vaccines are critical to the ongoing efforts to control the SARS-CoV-2 coronavirus 74 pandemic. To date, nine vaccines have received some form of approval for use in humans, 75 and phase III trials are ongoing for an additional 11 vaccines (1). Notable differences exist 76 among the vaccine products in terms of their design and the levels of protection they confer, 77 as well as the type, incidence, and severity of adverse events they may elicit. Gaining a 78 comprehensive understanding of the immunological factors underpinning the different 79 responses to various vaccines is a major endeavor. Yet, this knowledge is necessary for guiding 80 timely decisions to modulate vaccination protocols (e.g., the use of different types of vaccines 81 for the priming and booster doses). This information may also assist in matching of individuals 82 with the growing number of available vaccines based on their demographics, health status, 83 or any other relevant clinical/molecular phenotypes. 84 Blood transcriptome profiling measures the abundance of transcripts in whole blood and on 85 a system-wide scale. It was previously employed to comprehensively profile the immune 86 responses elicited by vaccines (2,3). Notably, this approach identified innate immune 87 signatures arising within hours after administering vaccines (4). In a recently published report, 88 Arunachalam et al. described the blood transcriptome profiles measured following the 89 administration of the BNT162b2 mRNA COVID-19 vaccine (5). They reported the presence of 90 an interferon (IFN) signature one day after the priming vaccination that was no longer 91 detectable on day 7. They further found a more comprehensive IFN/inflammatory signature 92 to be present 1 day after administering the booster dose. However, the sampling schedule 93 employed in this study was relatively sparse. And the sample collection time points 94 commonly selected in systems vaccinology studies are based on kinetics established for more 95 conventional vaccines -with sampling at days 1 and 7 often selected since they correspond 96 8 abundance of four modules was consistently increased across these time points, and all four 168 belonged to the module aggregate A28. Each "module aggregate" regroups sets of modules 169 that showed consistent abundance profiles across a reference set of 16 disease cohorts that 170 were employed for the construction of the BloodGen3 repertoire (see methods and (10) where modules occupy a fixed position and are arranged by aggregate. Each aggregate 178 occupyies a given row (Figure 2A) . Time-course gene-set enrichment analysis confirmed that 179 changes observed over time in four out of six A28 modules were significant. The response 180 profiles of the A28 modules showed a peak on day 2 post-vaccination. This was also visible on 181 a heatmap showing responses at each timepoint across individual subjects ( Figure 2B) . We 182 next examined whether this signature correlated with antibody responses measured 14-days 183 post-prime as well as at 14-days post-boost ( Figure 2C ). For this, correlation analyses were 184 run at the module level within Aggregate A28 using, as the endpoint, fold-changes in antibody 185 levels on days 7 and 14 post-prime and days 7 and 14 post-boost relative to the pre-186 vaccination baseline (immediately prior to the administration of the first dose of COVID-19 187 13 response, which was close to 50% of the constitutive transcripts on day 2 post-prime and 286 greater than 80% on day 1 post-boost. 287 We decided to then perform hierarchical clustering to identify subsets of modules within 288 the A28 aggregates that might group together based on patterns of transcript abundance 289 across all subjects and timepoints. Two sets of three modules each were, thus, identified 290 within the A28 aggregate. The first set comprised modules M8.3, M10.1, and M15.127 291 (referred to as A28/S1), and the second set comprised modules M16.64, M13.17, and M15.86 292 (referred to as A28/S2). Interestingly, we observed post-prime that, while modules in A28/S1 293 peaked on day 2, those belonging to A28/S2 peaked on day 1 ( Figure 5B ). Furthermore, 294 A28/S1 modules showed an extended peak post-boost, with day 2 levels being almost 295 identical to those of the day 1 peak, while A28/S1 modules peaked sharply on day 1, with 296 levels decreasing rapidly thereafter. These findings suggest that both sets of modules 297 measured distinct types of interferon response. Indeed, public datasets in which responses 298 to type 1 interferon were measured in-vivo indicated that A28/S1 modules are likely to 299 represent type 1 interferon responses (Figure 5B ), while we postulated that A28/S2 modules 300 might represent a type 2 interferon response. Modules forming the A28/S1 set comprise 301 some of the better recognized "canonical" interferon response genes, such as Oligoadenylate 302 Synthetase family members (OAS1, OAS2, OAS3, OASL), Interferon Induced Protein family 303 members (IFI6, IFI27, IFI35, IFI44, IFI44L), as well as Interferon Induced Protein With 304 Tetratricopeptide Repeats family members (IFIT1, IFIT3, IFIT5) (10). Modules forming the 305 A28/S2 set comprise instead most notably members of the Nuclear Antigen family members 306 SP100, SP110 and SP140, which are associated with interferon gamma signaling, as well as 307 transcription factors IRF9 and STAT2. Composition and functional annotations for A28 308 modules can be explored further at: https://prezi.com/view/E34MhxE5uKoZLWZ3KXjG/. 14 Finally, a strong association was found between the post-boost interferon signature and 310 the subsequent development of an antibody response. Indeed, positive correlations were 311 observed for all six A28 modules that reached significance on days 1, 2, and 3 post-boost. 312 Notably, this differed from the post-prime interferon response, for which significance was 313 reached only for four of the six modules and only on days 2 and 3. 314 Taken together, the high temporal resolution profiling results permitted the delineation 315 of distinct patterns of post-prime and post-boost interferon responses. The timing of the 316 responses observed at the individual module level contributed to the definition of the two 317 distinct sets of interferon modules. One set was associated with responses to type I interferon 318 in-vivo and dominated the post-prime response, with a peak on day 2. The post-boost 319 response showed a strong induction of both sets and also peaked on day 1. 320 321 Inflammation and erythroid cell signatures peak sharply on day 1 post-boost 322 We continued the dissection of the day 1 post-boost signature, focusing this time on 323 responses associated with inflammation and circulating erythroid cell precursors. 324 Aggregates A33 and A35, which are associated with inflammation, tended to decrease 325 from day 4 through day 6 post-prime but displayed instead a sharp and transient increase in 326 abundance post-boost. Indeed, a well-delineated response peak was observed on day 1 post-327 boost for both the A33 and A35 modules (Figure 6 ), but in contrast to the interferon response 328 (A28/S1), it did not extend beyond the first day. Three distinct response patterns were 329 identified via hierarchical clustering among the 21 modules that formed aggregate A35. The 330 "A35/S1" set comprised five modules, while "A35/S2" and "A35/S3" comprised ten and six 331 modules, respectively. The distinction between those three A35/inflammation module sets 332 was rather more subtle than was the case for the A28/interferon sets. Indeed, all three 333 15 module sets peaked on day 1 post-boost. Differences were rather apparent in the inflection 334 of changes measured on days 2 and 3 post-boost and in the "recovery phase", as abundances 335 appeared to dip below the baseline and progressively rise to reach pre-vaccination levels. The 336 underlying biological factors driving the grouping of the modules to those three distinct sets 337 could not be identified at this time. 338 Modules for three aggregates broadly associated with erythroid cell signatures also 339 displayed a sharp but transient increase in transcript abundance on day 1 post-boost. 340 However, the abundance tended to dip afterward, with a low peak on day 4 post-boost, 341 before recovering by day 7. Functionally, this signature was found to be most prominently 342 associated with immunosuppressive states, such as late-stage cancer or pharmacological 343 immunosuppression (16), which is consistent with published functional studies (18, 19) . We 344 also found such signatures were associated with more severe manifestations in babies 345 infected with Respiratory Syncytial Virus (RSV) (16). Moreover, erythroid precursors have 346 been recently associated with COVID most severe clinical outcomes (20). Finally, we did not 347 find evidence of an association between the day 1 post-boost inflammation or erythroid cell 348 signatures and the antibody responses. After the booster dose, the number of responsive modules peaked sharply on day 1, 353 then rapidly subsided beyond day 2, with the number of responsive modules on days 3, 4, 5, 354 and 6 being reduced to 8, 11, 3, and 2, respectively. Yet, changes within this later timeframe 355 are meaningful, as they specifically concern the set of five modules comprising aggregate A27, 356 which is associated with the presence of antibody-producing cells in the peripheral blood. 357 Three of the five A27 modules showed significant alterations after the booster dose (M16.60, 358 M13.32, M12.15) ( Figure 6 ). The proportion of differentially expressed transcripts in each 359 module was relatively modest (with an average of 15% at the peak of response), especially in 360 comparison with the interferon signatures described above (with an average of >80% for 361 some modules at the response peak). Yet, the trajectories of the five A27 modules were 362 relatively consistent, with only one of the modules (M15.110) showing a different pattern, 363 i.e., a peak on day 6, slightly above the levels observed on day 4. We also examined the 364 association of this post-boost plasmablast signature with the antibody response and found a 365 significant association starting from about day 3 and lasting until day 7 post-boost ( Figure 6) . 366 In summary, COVID-19 mRNA vaccination induced a plasmablast response that peaked on 367 day 4 post-vaccination. This was unexpected since such signatures typically are measured 368 around day 7 post-vaccine administration (e.g., in the case of influenza or pneumococcal 369 vaccines (6)). We were also able to demonstrate a logical association between this post-boost 370 plasmablast signature and the subsequent development of humoral immunity. 371 372 Patterns of interferon induction elicited by COVID-19 mRNA vaccines are also observed 373 among COVID-19 patients 374 Our work identified the interferon response as the most upstream factor associated with the 375 development of humoral immunity following COVID-19 mRNA vaccination. High-temporal 376 resolution profiling identified distinct patterns of interferon induction post-prime and post-377 boost and we next decided to determine whether similar response patterns could be 378 identified among patients with COVID-19 disease. 379 We relied for this on the original blood transcriptome data from the PREDICT-19 380 consortium Italian COVID-19 cohort comprising 77 patients with a wide spectrum of disease 381 severity. We used the response values for the six interferon modules from Aggregate A28 to 382 map individual COVID-19 patient samples along with post-vaccine samples on the same t-SNE 383 plot ( Figure 7A) . First, we confirmed that there was no apparent separation of the vaccination 384 and COVID-19 patient cohorts, and that batch correction was therefore not warranted before 385 proceeding with comparative analyses (Supplementary Figure 2) . This is consistent with the 386 results of meta-analyses we have previously conducted at the module level (16). To help with 387 the interpretation, k-means clustering was performed using the consolidated set of samples, 388 resulting in the formation of eight distinct clusters. Next, we examined the distribution of 389 samples from the vaccine and COVID-19 cohorts across the tSNE plot and among the eight 390 clusters. Timepoints at which an interferon response was detectable in vaccinated subjects 391 were of particular interest. Indeed, day 1 and day 2 post-prime samples (P1, P2), while 392 preferentially found in Clusters 1 and 5, appeared to be distributed across the entire t-SNE 393 plot. This is in contrast with day 1 and day 2 post-boost vaccination samples (B1, B2), which 394 were almost exclusively found in Cluster 5. A set of COVID-19 patients also co-localized in 395 Cluster 5, while others were found scattered across clusters, especially Clusters 1, 2, 6, and 3. 396 Interferon responses were detectable in all these clusters, but with important nuances. For 397 one, samples from Cluster 5 showed by far the most potent responses, with responses seen 398 in most cases across all six interferon modules, which was consistent with the post-boost 399 vaccine response ( Figure 7B ). In comparison, the response was less pronounced in samples 400 from Cluster 1, which was dominated by modules associated with type I interferon responses 401 (the A28/S1 set comprising M10.1, M8.3 and M15.127 described above). This pattern of 402 response was more consistent with the post-prime vaccine response. Signatures for samples 403 forming Clusters 2 and 6 were not well-defined and were in some cases absent, yet these 404 clusters also included COVID-19 patients. Samples forming Cluster 3 displayed a peculiar 405 18 signature, with an increase in the abundance of modules belonging to the A28/S2 set 406 (M15.64, M13.17, and M15.86) concomitantly with a decrease in modules forming A28/S1. 407 Among the samples forming this cluster, this pattern was most apparent for the COVID-19 408 patients. 409 Thus, we employed here the interferon responses observed post COVID-19 410 vaccination as a benchmark for the interpretation of COVID-19 patient signature. We were 411 able to establish that most COVID-19 patients display responses consistent with those found 412 post-vaccination, which, as established in this study, were associated with the development 413 of potent humoral responses. However, a subset of patients displayed patterns of interferon 414 response that are not typically seen in vaccinated individuals. It can thus be surmised that the 415 later patterns of interferon response might either be suboptimal or possibly even pathogenic. 416 The atypical interferon response signature observed in COVID-19 patients is associated 418 with a worse course of disease 419 The fact that some COVID-19 patients failed to display robust "post-vaccine-like" interferon 420 responses may be due to either a defective innate immune response, which may lead to 421 more severe disease course, or conversely to activation thresholds not being reached in 422 patients presented with milder disease. 423 Thus, we next examined patterns of interferon response in another original COVID-19 424 disease cohort, comprised exclusively of patients enrolled at the time of admission in the ICU 425 (the IMPROVISE cohort, which was also described above). As described above, we again 426 mapped individual COVID-19 patient samples along with post-vaccine samples on a t-SNE plot 427 based on similarities in the patterns of interferon responsiveness across the six A28 interferon 428 modules ( Figure 8A ). COVID-19 subjects were found to again be distributed throughout 429 multiple clusters. Patients who co-localized with day 1 post-boost vaccine samples tended to 430 have relatively short ICU stays (in Cluster 5 with potent A28/S1 and A28/S2 responses), and 431 only a few patients co-localized with day 2 post-prime samples in Cluster 3, which was 432 characterized by a more prominent A28/S1 signature compared with A28/S2. Furthermore, 433 distinct groups of patients in Clusters 1 and 6 displayed the peculiar pattern of interferon 434 response dominated by A28/S2 that was identified earlier among patients enrolled in the 435 PREDICT-19 cohort. Notably, patients from the IMPROVISE cohort displaying this pattern of 436 interferon response showed significantly lengthier stays in the ICU compared to patients 437 displaying patterns of interferon response that are consistent with those observed post-438 vaccination ( Figure 8B comparing left and right cluster: for length of hospital stay, t-test, p = 439 0.006 (**), mechanical ventilation days p =0.016 (*) and ICU stay p = 0.012(*)). 440 Thus, in a cohort of subjects uniformly presenting with severe disease, post-prime-like 441 patterns of interferon response dominated by A28/S1 were less prevalent. Post-boost-like 442 pattern of interferon response characterized by robust A28/S1 and A28/S2 signatures were 443 observed instead in most patients. A notable exception were patients presenting with 444 patterns of response dominated by A28/S2, not observed previously following vaccination 445 but which were found again in this second independent COVID-19 dataset. In this context we 446 could also establish that such response is associated with a worse disease course. This overall 447 supports the notion that patients harboring this signature may fail to mount an effective 448 immune response against SARS-CoV-2. Finally, we asked whether the A28/S2-dominated interferon response pattern associated with 455 worse disease outcomes in COVID-19 patients was also commonly found in other infectious 456 disease. 457 For this we first developed a standard definition of "Interferon Response 458 Transcriptional Phenotypes" (IRTPs): the two distinct signatures described above, A28/S1 and 459 A28/S2, were employed as "traits" for the definition of three main phenotypes observed 460 following vaccination and in response to SARS-CoV2 infection. 1) IRTP I encompassed A28/S1-461 dominated patterns of response: "A28/S1 ++ A28/S2 + ","A28/S1 ++ A28/S2 0 " and 462 "A28/S1 + A28S2 + " (see the method section for details). 2) IRTP II corresponded to a pattern of 463 interferon response characterized by the strong induction of both components: 464 A28/S1 ++ A28/S2 ++ . 3) IRTP III encompassed the A28/S2-dominated patterns of interferon 465 response: "A28/S1 -A28/S2 ++ ", "A28/S1 -A28/S2 + ", "A28/S1 0 A28/S2 ++ ", "A28/S1 0 A28/S2 + " and 466 "A28/S1 -A28/S2 0 ". These three IRTPs were in turn employed for the stratification of our 467 vaccination cohort at early time points following administration of the priming and booster 468 doses, as well as both of our COVID-19 cohorts and of several reference cohorts of patients 469 which we had generated as part of one of our earlier studies (10) Interferon Response Transcriptional Phenotype I (IRTP I), that we posit corresponds 474 to a response dominated by type 1 interferon (IFNa, IFNb), in absence of a substantial type 2 475 interferon (IFNg), was found in ±1/3 of the vaccinated subjects at peak response on day 2 476 21 post-prime ( Figure 8C : P2). It was however absent at peak response post-boost (B1). Similarly, 477 IRTP I was found among COVID-19 patients belonging to the PREDICT-19 cohort (although in 478 only about 10% of patients), but not among those belonging to the IMPROVISE cohort, who 479 presented with more severe disease. IRTP I was otherwise also found in ±10% of subjects 480 across most of our reference cohorts. However, as was the case of our severe COVID-19 481 cohort, it was absent in the comparator cohort comprised of patients with bacterial sepsis. In 482 the context of mRNA vaccination, IRTP II, which is characterized by the robust induction of 483 both A28/S1 and A28/S2 components, was observed following the booster dose in 95% of 484 samples profiled on day 1, which corresponds to the peak response. The priming dose of 485 Covid-19 mRNA vaccines was able to induce both components robustly but in only 48% of 486 samples at peak (day 2 post-prime). IRTP II was otherwise also prevalent in 487 which is consistent with our earlier observation. It was also found in most samples in the other 488 pathologies employed as comparators -except for RSV and bacterial sepsis (40% and 48%, 489 respectively). Interferon response transcriptional phenotype III (IRTP III), which is 490 characterized by an A28/S2-dominated response was observed only rarely post-COVID-19 491 mRNA vaccination. It was however prevalent among COVID-19 patients, with 25% and 22% of 492 subjects with this phenotype in the respectively. 493 However, it was not observed in patient with tuberculosis, influenza virus or HIV infection. 494 IRTP III is on the other hand found in 13% of patients with RSV infection and reached its peak 495 prevalence in patients with bacterial sepsis (36%). 496 In summary, those results show that in most instances both components of the 497 transcriptional interferon response can be robustly induced following or viral infection (i.e. corresponding to IRTP II). However, incomplete patterns of induction 499 can also be observed in some circumstances. We hypothesize that this may be due: 1) to 500 22 activation thresholds not being reached, in the case of IRTP I or 2) to subjects failing to mount 501 an effective interferon response, in the case of IRTP III, which in the context of SARS-CoV-2 502 infection appears to impact their ability to control the infection. Notably, besides 503 IRTP III phenotypes were only observed in a limited set of pathologies, including infection 504 caused by RSV, a virus that is known to interfere with the interferon response (21,22), and 505 bacterial sepsis that is characterized by a dysregulated host response to infection (23). 506 507 DISCUSSION 508 Relatively little is known about the types of in-vivo immune responses elicited by mRNA 509 vaccines in humans. To address this, we employed bulk blood transcriptomics to map the 510 immune changes taking place in-vivo after the administration of priming and booster doses 511 of COVID-19 vaccines in adult volunteers. We did so at a high-temporal resolution, collecting 512 small amounts of blood before and for nine consecutive days after the administration of the 513 priming and booster doses of COVID-19 mRNA vaccines. The use of blood transcriptomics 514 eliminated the need to choose a panel of analytes to measure vaccine responses, which is 515 one source of bias. The daily collection and profiling schemes adopted eliminated the need 516 to choose specific timepoints for measuring the response, thus eliminating a second source 517 of bias. 518 Profiling blood transcript abundance post-prime and -booster doses of COVID-19 519 mRNA vaccines at a high-temporal resolution revealed a well-orchestrated sequence of 520 immune events (Figure 9 ). The immune signatures elicited following the administration of the 521 two doses of mRNA vaccines differed drastically. Relatively modest changes were observed 522 post-prime that manifested primarily as the induction of interferon-response signatures that 523 were detectable over the first three days following the injection of the first dose. This was 524 23 followed by a more subtle response that could be attributed to the priming of the adaptive 525 response between days 4 and 6. Indeed, a decrease in the abundance of transcripts for 526 modules associated with inflammation was observed over these three days, which included 527 an increase in transcripts associated with plasma cells and T-cells on day 5. No further changes 528 were detected beyond day 6. After the booster dose, the plasmablast response was more 529 robust and peaked on day 4, but was not accompanied by a T-cell response peak as was the 530 case post-prime. Notably, in studies assessing blood transcriptional responses to vaccines, the 531 peak plasmablast response is typically observed on day 7, as it is, for instance, with influenza 532 and pneumococcal vaccines (6,14,15). As a result, sampling schedules in common use are 533 designed to capture changes on days 1, 7, and sometimes day 3, but would miss the peak of 534 the adaptive response to COVID-19 mRNA vaccines observed in our study. In addition to 535 eliminating potential blind spots, high-frequency sampling and profiling also permit the 536 precise resolution of signatures that show the complex kinetics of a response; for instance, 537 the erythroid cell signature peaks sharply post-boost and recedes well below baseline over 538 several days before recovering. The trajectory of this signature may be of significance in the 539 context of vaccination, as we recently described its association with immunosuppressive 540 states, such as late-stage cancer and maintenance therapy in liver transplant recipients (16). 541 In the same work, we found this signature to be strongly associated with the development of 542 a more severe disease in subjects with acute respiratory syncytial virus infection; and we 543 furthermore putatively associated this signature with populations of circulating erythroid 544 cells found to possess immunosuppressive properties (18). 545 Arunachalam et al. previously described the elicitation of qualitatively distinct innate 546 signatures on day 1 following the administration of priming and booster doses of COVID-19 547 mRNA vaccines, with the former inducing an interferon response and the latter a mixed 548 24 response that also presented an inflammatory component (5). Our findings are consistent 549 with these earlier observations and, employing a high-frequency sampling and profiling 550 protocol, permitted to further dissect those responses. Most notably, while interferon 551 responses appear a priori as the common denominator between the post-prime and post-552 boost responses, the temporal pattern of response that we observed indicates that these are, 553 in fact, qualitatively and quantitatively distinct. This was best evidenced by the differences in 554 the timing of the response peak, which corresponded to day 2 post-prime and day 1 post-555 boost. The kinetics of the response post-boost is, therefore, most consistent with what is 556 observed following injection of a single dose of influenza vaccine (6). Interestingly, a further 557 investigation of the patterns of response among the six modular components of the 558 interferon responses (module Aggregate A28) identified two distinct sets of modules. These 559 two sets of three modules each, A28/S1, and A28/S2, displayed distinct kinetics and 560 amplitude of response post-prime and post-boost. We have described, in an earlier report, 561 that distinct interferon modules could be employed to stratify patients with systemic lupus 562 erythematosus (24). Here we sought to specifically determine whether "post-prime-like" 563 patterns (i.e., dominated by A28/S1 -IRTP I) or "post-boost-like" patterns (i.e., with potent 564 induction of both components: A28/S1++, A28/S2++ -IRTP II) could be identified among 565 COVID-19 patients. Indeed, since those were associated with the subsequent development of 566 humoral immunity in the context of vaccination it may be surmised that it would also be the 567 case during the course of SARS-CoV-2 infection. This question was made particularly relevant 568 in the context of COVID-19 disease, since it has been reported that failure to induce interferon 569 responses is associated with worse disease outcomes (8, (25) (26) (27) . In the PREDICT-19 cohort, 570 composed by patients with predominantly mild or moderate pathology, both phenotypes 571 were indeed observed, along with a third "atypical" phenotype that was not observed post-572 25 vaccination. This latter phenotype is dominated instead by A28/S2, with A28/S1 abundance 573 low or even decreased (IRTP IIII). Notably, in a cohort of severe patients, both A28/S1++ 574 A28/S2++ ("post-boost-like" / IRTP II) and A28/S2>S1 ("atypical" / IRTP III) phenotypes were 575 also observed, with the latter being associated with extended lengths of stay in the ICU. 576 However, IRTP III did not appear to be preferentially associated with death in this setting, 577 which may be due to the supportive care provided to the patients. While, overall, our 578 observations support the notion that failure to mount robust interferon responses is 579 associated with a less favorable course of the disease, they also show that the response 580 elicited in these patients may be of a peculiar type, but is altogether not entirely defective 581 (i.e., with only one component. A28/S1, being primarily affected). One possibility is that this 582 peculiar response pattern may be associated with the presence of endogenously produced 583 autoantibodies that neutralize interferon, as has been previously described (27,28). The high 584 incidence of the IRTP IIII phenotype observed in patients with bacterial sepsis (about 1 in 3), 585 however suggests that other mechanisms may be at play. Taken together, it is not possible 586 for us to be conclusive on this point at this time and further investigations are thus warranted. 587 Other points remain to be elucidated. This includes the timing of the adaptive response 588 to mRNA vaccines, which appears to rise and peak several days earlier than what is normally 589 observed in responses to other vaccines (± 7 day peak). The priming mechanism underpinning 590 the robust polyfunctional response observed on day 1 post-boost remains to be determined 591 as well. And in particular, whether or not such a response, which would typically be 592 considered to be innate, is in fact antigen-specific. Interestingly, in that respect, the subjects 593 who were previously infected but recovered from COVID-19 did not display a noticeable day 594 1 inflammatory response, and their immune systems behaved like those of naïve individuals. 595 However, the number of recovered subjects was small, and the study was not designed to 596 26 directly address this question. Hence, further investigations will also be necessary. Notably, 597 the greater amplitude of responses observed post-boost and the presence of an inflammatory 598 component is also consistent with previous reports of the increase in the incidence of side 599 effects/discomfort following COVID-19 mRNA vaccine booster doses (29,30). 600 Thus, while this study contributes to a better understanding of drivers of mRNA vaccines 601 immunogenicity it can also serve as a resource to help inform the design of studies 602 investigating vaccine responses. Indeed, a decrease in sequencing costs provides laboratories 603 an opportunity to employ transcriptome profiling approaches in novel ways. One of them 604 being the implementation of high-temporal resolution profiling protocols. An advantage of 605 the delineation of transcriptome responses at high-temporal resolution is that it is doubly 606 unbiased, i.e., there is no need to select transcripts for inclusion in a panel because RNA 607 sequencing measures all transcript species present in a sample. Similarly, there is no need to 608 select specific timepoints for assessing the vaccine response, as all timepoints were profiled 609 within a given time frame. An obvious advantage of the approach is that it permits the 610 removal of potential blind spots and the detection of changes that may otherwise be missed 611 by more sparse sampling protocols. In addition to eliminating potential blind spots high-612 frequency profiling data helped resolve the vaccine response more precisely. This was the 613 case in our study of the interferon response, with the delineation of two distinct components 614 having been much more difficult if not for the resolution of peaks of response over the first 615 three days post first and second doses of vaccines. Some of the practical elements that may 616 contribute to making the routine implementation of the high-temporal resolution 617 transcriptomics approach viable include, as mentioned earlier, a substantial decrease in the 618 cost of RNA sequencing, especially 3'-biased methodologies. Along the same lines recent 619 publications showed, through down-sampling analysis, that much fewer deep reads than 620 27 usual are adequate for biomarker discovery projects, which could lead to further reductions 621 in the cost of RNA sequencing assays (31), with the lower costs permitting larger sample sizes 622 or, as in this case, a higher sampling frequency. Another consideration is the availability of 623 solutions for the in-home self-collection of samples. This is the case for the collection of RNA-624 stabilized blood with our custom method, which could be further improved. Novel solutions 625 are also being put forward that could permit the implementation of these methods at scale 626 (32). Finally, as we have shown, it is possible to implement the self-collection of samples for 627 serology profiling within a vaccinology study. 628 There were several limitations to our study. While the sample size was adequate for an 629 initial discovery phase, a larger study cohort would help to better resolve inter-individual 630 variations. The dataset we generated, however, has been made available for reuse, and it 631 should be possible to integrate and consolidate this dataset with those generated in follow-632 on studies by us and others (16). Follow-on studies would need to be purposedly designed to 633 formally address specific questions, for instance, comparing responses in individuals who had 634 previously been exposed to SARS-CoV-2 with those in naïve individuals. It would also be 635 interesting to compare responses elicited by the Pfizer/BioNTech and Moderna vaccines, 636 which was not possible in our study due to the small numbers of individuals that received the 637 Moderna vaccine. Indeed, although we hoped it would be possible to obtain more balanced 638 sample sizes for a more detailed comparison, the speed at which the vaccinations were rolled 639 out among our target population of healthcare workers meant we had very little control over 640 the number of volunteers that received the different types of vaccines or their status as naïve 641 or previously exposed individuals. It would also have been particularly interesting to enroll 642 patients from different age categories, especially the elderly population, but this again proved 643 impossible. 644 28 In conclusion, a several COVID-19 vaccines have already been approved for use in humans, 645 and an even greater number of them are currently in phase III trials (>20) (33 stabilizing solution aliquoted from a regular-sized tempus tube (designed for the collection of 712 3 ml of blood and containing 6 ml of solution; ThermoFisher, Waltham, MA, USA). This 713 method is described in detail in an earlier report (7), and the collection procedure is illustrated 714 in an uploaded video: https://www.youtube.com/watch?v=xnrXidwg83I. Blood was collected 715 prior to the vaccine being administered (day 0), on the same day, and daily thereafter over 716 the next 10 days. This protocol was followed for both the priming and booster doses. 717 For serology applications, 20 µl of blood was collected using a Mitra blood collection device 718 The presence of antibodies against selected Human Coronaviruses proteins in the serum was 730 measured with a home-built bead array based on carboxymethylated beads sets with six 731 distinct intensities of a UV-excitable dye. Each bead set was individually coupled to 3 SARS-732 CoV-2 proteins, envelope, nucleoprotein, Spike protein in its trimeric form-or its fragments, 733 and the S1 fragment of SARS-CoV S protein. Therefore, the complete array consisted of 6 734 antigens, including five SARS-CoV-2 antigens (Full Spike Trimer, Receptor Binding Domain, 735 Spike S1, Nucleoprotein, and Envelope), as well as the closely related SARS-CoV-S1 protein. 736 The binding of human antibodies to each viral antigen (bead set) is revealed with fluorescently 737 labeled isotype-specific mouse monoclonal or polyclonal antibodies. We measured total IgM, 738 total IgG, total IgA, as well as their individual isotypes, IgG1, IgG2, IgG3, IgA1, and IgA2, 32 reporting a total of 48 parameters per sample. The assays were performed on filter plates and 740 acquired on a BD-Symphony A5 using a high-throughput-sampler. An average of 300 beads 741 per region was acquired, and the median fluorescence intensity (MFI) for each isotype binding 742 was used for characterizing the antibody response. An antibody response index was 743 calculated as the ratio of the MFI of pooled negative blood controls collected prior to June 744 generate the raw counts. Raw expression data were normalized to size factor effects using R 762 package DESeq2. All downstream analyses were performed using R version 4.1 unless 763 33 otherwise specified. Global transcriptional differences between samples were assessed by 764 principal component analysis using the "prcomp" function. Transcriptome profiling data were 765 deposited, along with detailed sample information, into a public repository, the NCBI Gene 766 Expression Omnibus (GEO), with accession ID GSE190001 and BioProject ID: PRJNA785113 767 PREDICT-19 Cohort: Total RNA was isolated from whole blood lysate using the Tempus Spin 768 Isolation kit (Applied Biosystems) according to the manufacturer's instructions. Globin mRNA 769 was depleted from a portion of each total RNA sample using the GLOBINclear™-Human kit 770 (Thermo Fisher). Following the removal of globin transcripts transcriptome profiles were 771 generated via mRNA sequencing. Then mRNA-sequencing was performed using Illumina 772 HiSeq 4000 Technology (75 paired-end) with a read depth of 60M. Single samples were 773 sequenced across four lanes, resulting FASTQ files were merged by sample. All FASTQ passed 774 QC and were aligned to reference genome GRCh38 using STAR (2.6.1d). BAM files were 775 converted to a raw count's expression matrix using HTSeq (https://github.com/Sydney-776 Informatics-Hub/RNASeq-DE). Raw count data was normalized using DEseq2. The ensemble 777 IDs targeting multiple genes were collapsed (average), and a final data matrix gene was 778 generated for modular repertoire analysis. 779 780 Analyses were conducted using pre-defined gene sets. Specifically, we employed a fixed 782 repertoire of 382 transcriptional modules that were thoroughly functionally annotated, as 783 described in detail in a recent publication (10). Briefly, this repertoire of transcriptional 784 modules ("BloodGen3") was identified based on co-expression, as measured in a collection of 785 16 blood transcriptome datasets encompassing 985 individual transcriptome profiles. Sets of 786 co-expressed transcripts were derived from the analysis of a large weighted co-clustering 787 34 network. Downstream analysis results and visualizations were generated employing a custom 788 R package (35). "Module response" is defined as the percentage of constitutive transcripts 789 with a given abundance that was determined to be different between two study groups, or 790 for the same individual in comparison to a given baseline (in this study, pre-vaccination 791 abundance levels). The values, therefore, ranged from 100% (all constitutive transcripts 792 increased) to −100% (all constitutive transcripts decreased). Only the dominant trend (i.e., 793 increase or decrease in abundance over control/baseline) was retained for visualization 794 purposes on fingerprint grids or fingerprint heatmaps, with red indicating an increase and 795 blue a decrease in abundance. When performing group comparisons (e.g., cases vs controls 796 for the disease datasets used as reference), the p-value and false discovery rate cutoffs were 797 applied, which are mentioned in the figure legend. When performing longitudinal analyses, 798 the module response is determined by employing fixed fold-change and expression difference 799 cutoffs. Module response values obtained were used for data visualization. Significance was 800 determined for each module using the differential gene set enrichment function of the 801 dearseq R package (11). 802 803 Study cohorts were stratified based on patterns of interferon response for two distinct 805 interferon signatures, defined as A28/S1 (comprising modules M8.3, M10.1 and M15.127) 806 and A28/S2 (comprising modules M13.17, M15.64, M15.86). For this, phenotypes were 807 defined based on levels of response observed for these two "traits", as follows: 808 Percentage response of the six IFN modules were scored base on degree of response (% 809 response >= 50; score = 2, 0 < %response < 50; score =1 and (% response <= -50; score = -2, -810 50 < %response < 0 ; score =-1). Then the average scores of S1("M8.3", "M10.1" and 811 35 "M15.127") and S2 ("M13.17" , "M15.64" , "M15.86") and phenotypes were classified using 812 cutoff at S1/S2++ (avg score >=1), S1/S2+( 1< avg score < 0.33), S1/S20( 0.33 < avg score <= 813 0), and S1/S2 -(avg score < 0). The phenotypes were grouped as: 814 -"Interferon Response Transcriptional phenotypes I" = "IRTP I" = 815 "A28/S1++A28/S2+","A28/S1++A28/S20", "A28/S1+A28/S2+", 816 -"IRTP II" = A28/S1++A28/S2++", 817 -"IRTP III" = "A28/S1-A28/S2++","A28/S1-A28/S2+","A28/S10A28/S2++", 818 "A28/S10A28/S2+", "A28/S1-A28/S20" 819 -The "other " category encompassed the less prevalent phenotypes remaining = 820 "A28/S1+A28/S20", "A28/S10A8/S2+", "A8/S1+A28/S2", "A28/S10A28/S20", 821 "A28/S10A28/S2-", "A28/S1-A28/S2-", "A28/S1+A28/S2++" 822 Vaccine type and lot, and subjects' characteristics were recorded, including demographic, biometric data, blood group, underlying diseases, drugs usage, and previous COVID-19 disease. Every subject recorded and graded the symptoms that occurred after the first and second vaccinations doses, according to the NIH "DAIDS AE Grading The heatmap represents changes in abundance of antibodies specific to several SARS-CoV-2 antigens and control antigens relative to pre-vaccination levels. Red indicates a relative increase, and green indicates a relative decrease in abundance. Columns represent subjects arranged by timepoint and have colored tracks at the top indicating whether the subjects were previously infected with SARS-CoV-2 or not. The histogram above represents the average log2 fold-change over baseline for a given column. The rows represent antibody reactivities arranged by antigen specificity. The different rows represent the isotypes of reactive antibodies, arranged according to the color legend specified below the heat map. (C) Changes in antibody levels expressed as an "antibody index" are shown on the box plots, each corresponding to a given antibody type of a given specificity. Lines indicate changes for individuals previously infected with SARS-CoV-2 and who had recovered (in pink) and for naïve individuals (in green). Centerlines, box limits, and whiskers represent the median, interquartile range, and 1.5x interquartile range, respectively. Multiple pairwise tests (paired t-test) were performed comparing antibodies levels to baseline (D0). Asterisks: * represent p < .01, **represent p < .001, *** represent p < .0001. A. A. RNA stabilized blood The bar graph shows the cumulative module response at the various timepoints following the administration of the priming dose of the vaccine (noted P1-P14). The Y-axis values and numbers on the bars indicate the number of modules meeting the 15% response threshold (out of a total of 382 modules constituting the BloodGen3 repertoire, with percentage response corresponding to the proportion of transcripts predominantly increased or decreased compared to baseline using FDR < 0.1 as the cutoff to determine significance [DESeq2]). The number of modules for which abundance was predominantly increased is shown in red, and those for which abundance was predominantly decreased are shown in blue. The fingerprint grid plots represent the overall module responses on day 1 post-prime (P1) and day 2 postprime (P2). Modules from the BloodGen3 repertoire occupy fixed positions on the fingerprint grids. They are arranged as rows based on membership to module aggregates (rows A1 through A38). Changes compared to the pre-vaccination baseline are indicated on the grid by red and blue spots of varying color intensity, which indicate the "percentage response" for a given module. The color key at the top indicates the various functions attributed 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 se post−Prime and antibody to SARS.CoV2.S1. at Day 14 post−Prime C. IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 P1 P2 P3 P4 P5 P6 P7 P8 P9 P14 A28.1| M10. IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 se post−Prime and antibody to SARS.CoV2.S1. at Day 14 post−Prime to the modules that are represented on the grid. The response of the six modules comprising aggregate A28 is represented on a line graph that shows the proportion of responsive transcripts for each module across all the postprime timepoints. For each module, the statistical significance of the overall response was determined by timecourse gene set enrichment analysis. Four of the six A28 modules met significance thresholds FDR < 0.1 (M8.3: pvalue = 1.9-e4, FDR = 0.019, M10.1: p-value = 1.9-e4, FDR = 0.019, M15.127: p-value = 1.9-e4, FDR = 0.019, 727 and M15.86: p-value = 3.9-e4, FDR = 0.031) and all six A28 modules p < 0.05 (M13.17: p-value = 1.5-e3, FDR = 0.101 and M15.64: p-value = 0.044, FDR = 0.727). We also ascertained the significance of changes measured post-prime at the level of this module aggregate and at each time point (paired t-test comparing module response at each time point relative to the pre-vaccination baseline; * p<0.01, ** p<0.001, *** p<0.0001). (B) Heatmaps represent proportions of transcripts that changed within the six A28 modules at different timepoints and across different individuals compared to pre-vaccination baseline values. Red indicates that transcripts were predominantly increased over the baseline, and blue indicates that transcripts were predominantly decreased. Rows represent the six A28 modules arranged within an aggregate via hierarchical clustering. Columns represent samples grouped by timepoint and show profiles of individual subjects within each timepoint. (C) The heatmaps represent associations (Spearman correlation test) between antibody responses measured 14 days after administration of COVID-19 booster doses and transcriptional responses measured across nine consecutive days after the priming dose. The heatmap at the top provides the correlation coefficients across multiple days and for each day across multiple subjects, with rows corresponding to the six A28 interferon modules. The heatmap below shows the significance of the correlations shown on the heatmap directly above, with the same ordering of rows and columns. 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 LGALS1 Percentage response accounts for the proportion of transcripts for a given module showing differences in abundance post-prime (left) or post-boost (right) compared to baseline pre-vaccination levels. Changes in transcript abundance post-prime and post-boost for two distinct sets of interferon response modules that received the denomination A28/S1 and A28/S2 are plotted on separate graphs. For each module, statistical significance for the overall response was determined by time course gene set enrichment analysis. Significance was reported postprime in Figure 2 . Post-boost all six A28 modules met significance thresholds p < 0.001 and FDR < 0.001 (M8.3: p- • • • • • • • • • • • • • • • • • • • • • • • • • • • − A. A28/S1 *** *** *** * ** ** * * * *** *** *** * ** * ** * * 0 1 2 3 4 5 6 7 8 9 14 0 1 2 3 4 5 6 7 8 9 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 - value = 1.9-e4, FDR = 3.6-e4, M10.1: p-value = 1.9-e4, FDR = 3.6-e4, M13.17: p-value = 1.9-e4, FDR = 3.6-e4, M15.127: p-value = 1.9-e4, FDR = 3.6-e4, M15.64: p-value = 1.9-e4, FDR = 3.6-e4 and M15.86: p-value = 1.9-e4, FDR = 3.6-e4). In addition, we ascertained the significance of changes measured post-prime at the level of this module aggregate and at each time point (paired t-test comparing module response at each time point relative to the prevaccination baseline; * p<0.01, ** p<0.001, *** p<0.0001). The circle packing plots on the left show module responses at the individual transcript level for two public blood transcriptome datasets. The larger circle below indicates official symbols for the individual transcripts. It also highlights the modules included in A28/S2, shown directly on the right. The smaller circles above show changes in abundance of A28 transcripts for two public datasets. One study (GSE11342) measured blood transcriptional response in patients with Hepatitis C infection treated with alpha-interferon (23). The second study (GSE26104) measured transcriptional response in subjects with multiple sclerosis treated with beta-interferon (24). A red circle indicates a significant increase in the abundance of transcripts compared to the pre-treatment baseline (*|fold-change| > 1.5, FDR < 0.1). (B) Changes in abundance compared to baseline pre-vaccination levels are represented on a heatmap, with modules as rows and individual samples as columns. The modules are arranged by hierarchical clustering based on abundance patterns across samples. The samples are arranged by timepoints post-prime (top) and post-boost (bottom). (C) The heatmaps represent associations (Spearman correlation) between antibody responses measured 14 days after administration of COVID-19 booster doses and transcriptional responses measured across nine consecutive days after the booster dose. The heatmap on top provides the correlation coefficients across multiple days and for each day across multiple subjects, with rows corresponding to the six A28 interferon modules. The heatmap below shows the significance of the correlations shown on the heatmap on top, with the same ordering of rows and columns. The fingerprint grid plots show the modules that had changes compared with a fixed visualization and interpretation framework. Changes are shown for the day 1 post-boost timepoint (left) as well as day 5 (right) (percent response is determined based on statistical cutoff: DESeq2, FRD < 0.1). On the left grid, modules belonging to aggregates A35 (associated with inflammation) and A37 (associated with erythroid cells) are highlighted. The profiles of those modules are represented on the line graphs below, which show the average percentage responses of A35 and A37 modules across multiple timepoints. The percentage response for a given module is the proportion of its constitutive transcripts showing significant changes, ranging from 0% to 100% when transcripts were predominantly increased to 0% to −100% when transcripts were predominantly decreased. Each line represents the profile of the modules B1 B2 B3 B4 B5 B6 B7 B8 B9 B14 A27 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 B1 B2 B3 B IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 IgG2 IgG3 IgA1 IgA2 IgM IgG IgG1 Spearman correlation of module resonse post−Boost and antibody to SARS. constituting a given aggregate. Line graphs for A35 were split into three sets according to differences in clustering patterns (A35/S1, A35/S2, and A35/S3). On the right grid, modules belonging to aggregates A27 (associated with platelets) are highlighted. The corresponding line graph below represents the changes in abundance of A27 modules over time following administration of the second dose of vaccine. For each module, statistical significance for the overall response was determined by time course gene set enrichment analysis using the dearseq R package. For A35, 20 of 21 modules met significance thresholds (p-value < 0.05 and FDR < 0.01). It was also the case in 11 of 11 modules for A37 and 4 of 5 modules for A27 (Supplementary file 4) . In addition, we ascertained the significance of changes measured post-prime at the level of this module aggregate and at each time point (paired t-test comparing module response at each time point relative to the pre-vaccination baseline; * p<0.01, ** p<0.001, *** p<0.0001). (B). The heatmaps represent associations between antibody responses measured 14 days after administration of COVID-19 booster doses and transcriptional responses measured across nine consecutive days after the booster dose. Specifically, the heatmap at the top represents the correlation coefficients across multiple days and for each day across multiple subjects, with rows corresponding to the five A27 plasmablast modules. The heatmap below shows the significance of the correlations shown on the heatmap at the top, with the same order of rows and columns. . Samples from the consolidated cohorts were partitioned into 8 clusters via k-means clustering, the distribution of which is shown on the tSNE plot on the top right. B. Heatmaps show patterns of response for the six interferon response modules across the eight sample clusters. The red colors indicate that the abundance of transcripts for a given module is predominantly increased with the intensity representing the proportion of constitutive transcripts meeting a given threshold, which at the level of individual samples is a fixed fold change and difference cutoff (|Fold change| > 1.5, and |difference| > 10 in a given sample over its respective pre-vaccination baseline). The blue color denotes a predominant decrease in abundance of constitutive transcripts compared to the same individual's pre-vaccination baseline. Details are shown below for Clusters 3, 5, and 8 in separate heatmaps. (TP1, TP2, TP3, TP4 , all collected during ICU stay) are shown on the plot on the left. Samples from the consolidated cohorts were partitioned into 8 clusters via k-means clustering, the distribution of which is shown on the tSNE plot on the center. Length of ICU stay is shown on the tSNE plot on the right. Patterns of response for the six interferon response modules across the eight sample clusters are shown on a heatmap below. The red colors indicate that the abundance of transcripts for a given module is predominantly increased with the intensity representing the proportion of constitutive transcripts meeting a given threshold, which at the level of individual samples is a fixed (S1>S2) (S1++ S2++) (S1