There's a new double act in town. Blood-based, ultra-sensitive protein detection is teaming up with tissue-level spatial biology to give a deeper understanding of disease. As multi-omics begin to gain traction in informing diagnostic decisions, we asked expert Jorge Marques Signes about the key considerations for implementing the approach in the lab.
What unique clinical insights emerge when blood-based proteomics and tissue-level spatial biology are used together, rather than in isolation?
One of the biggest sources of confusion in biomarker testing is the assumption that tissue and blood are simply two ways of measuring the same thing. They are not. Each modality answers a different biological question, and each comes with very different implications for patients.
Blood-based testing is valuable because it is accessible. It can be performed early in the care journey, repeated over time, and integrated into routine clinical practice. That accessibility makes large-scale screening, early detection, and longitudinal monitoring possible.
Tissue testing serves a different purpose. Because biopsies are performed only when clinically justified and are typically collected infrequently, tissue provides a level of biological context that blood cannot. It reveals which cells express a target, how those cells are organized within the tumor microenvironment, and where a protein is localized within the cell.
That last part is underrated, and antibody-drug conjugates (ADCs) are the clearest example I know of. An ADC only works if the target is on the cell surface and the complex is internalized. So, how much protein is present in the tumor is not really the question.
The experience with TROP2 in non-small cell lung cancer illustrates this distinction. Conventional immunohistochemistry (IHC) scoring did not reliably identify patients who benefited from datopotamab deruxtecan. However, computational analysis measuring the proportion of TROP2 localized to the cell membrane relative to total cellular TROP2 showed predictive value. The tissue sample and antibody were the same; the difference was the biological question being asked.
You cannot get that answer from a concentration. Not from tissue, and not from plasma. So when people talk about combining the modalities, I do not think it is really about reconciling two numbers. Tissue is where you work out what the biology is. Blood is where you ask whether any part of it has a measurable correlate in circulation.
That is the direction I want the field to move in, and it is the harder problem. Take a tissue signature we already know means something, then find its surrogate in blood, so that what requires a biopsy today can eventually come from a blood draw.
It will not work for everything. A membrane ratio has no obvious plasma equivalent and I do not want to pretend otherwise. But for a good amount of biology, tissue tells you what to look for and blood is what makes it repeatable.
How are these integrated approaches already influencing diagnostic decision-making?
The reality is that blood and tissue-based approaches are at very different stages of clinical adoption, and it is important not to blur that distinction.
Blood-based biomarkers have already entered clinical practice. In 2024, the Alzheimer's Association published its first clinical practice guideline for blood biomarkers, focusing on performance standards rather than specific products. The framework is straightforward: highly sensitive tests can help rule out amyloid pathology, while tests that achieve both high sensitivity and specificity can support confirmation.
Several blood-based Alzheimer's tests have now received FDA clearance, and payer coverage is beginning to expand. In specialist memory clinics, these advances are already changing the diagnostic pathway by reducing the need for more invasive procedures such as lumbar punctures and amyloid PET imaging.
Tissue is a different situation, and I would not claim spatial proteomics is influencing neurodegenerative diagnosis today. We cannot biopsy the brain. All the tissue work is postmortem, and the methods that define the reference standard are conventional neuropathology – the established staging of amyloid and tau – not high-plex spatial. That distinction matters more than people allow. Every blood assay in this field is ultimately anchored to autopsy confirmation, and the studies validating plasma p-tau217 against neuropathologically confirmed cohorts are what let a clinician trust the number in front of them. It is an essential contribution. It is also an upstream one. While it shapes the assay, it does not sit in the patient’s workup.
Where spatial methods on brain-bank material are contributing is in explaining what the blood cannot. Two patients with very similar plasma profiles can follow different clinical courses, and a lot of that is co-pathology: TDP-43, alpha-synuclein, small vessel disease. It is common, an amyloid or tau blood marker does not see it, and you can only characterize it in tissue. That work is mapping which cells and regions actually contribute to the circulating signal, and where the current panel is blind. It will inform the next generation of blood tests. I would call it translational science with a diagnostic destination rather than diagnostics.
If the goal is to see blood and tissue genuinely informing the same clinical decision today, oncology offers a clearer example. Tumor tissue is routinely available, and clinicians already integrate tissue-based and circulating biomarkers to build a more complete picture of disease, treatment response, and patient management.
How can advances in pre-analytics remove some of the barriers to integrating multi-omic diagnostics into clinical workflows?
Tube composition, time to centrifugation, aliquot volume, freeze-thaw history; all of it moves an ultra-sensitive result. And on the tissue side, fixation time and block age do the same thing. From what we see working with customers, you resolve more variability by building the handling protocol into the phlebotomy and grossing workflow, so that the right thing happens by default, than by any change to the assay itself. It is an operational discipline problem more than a scientific one, and it is solvable.
There is a more interesting version of that answer, which is to remove the variables rather than control them. Some of the most encouraging recent work in this area has gone in a completely different direction: measuring p-tau217 and other neurodegeneration markers from a few drops of capillary blood taken by finger prick and dried onto a card, then shipped at ambient temperature. Correlation with venous plasma has been good, and the cards travel without cold chain. That is most of the pre-analytical problem I just described simply deleted. No centrifugation window, no freeze-thaw history, no cold shipping.
I would not oversell where it is. Recovery from dried spots is much lower than from venous plasma, accuracy still trails a properly handled venous sample, a meaningful share of collections fail, and the indeterminate zone is wider than most venous tests produce. It is a research method today and the groups doing the work say so themselves. But patients have collected usable samples on their own, and longitudinal sampling over months is being demonstrated. For a disease where you want to test early, broadly, and repeatedly, and where the people you most want to reach are the least likely to reach a specialist center, that direction matters more than the current performance gap suggests.
What else should labs consider before integrating multi-omic diagnostics into clinical workflows?
Normalization is an underappreciated barrier. A multi-omic result is not a measurement; it is several measurements coming off instruments that do not share units, dynamic range, batch structure, or failure modes. A concentration in picograms per milliliter and a cell density per square millimeter in an operator-defined region have no common scale and no common calibrator. You have to normalize before you can integrate, and there is nothing obvious to normalize against. Most groups run into this at the point of integration, which is far too late. The fix is unglamorous: decide the normalization strategy before the first sample is run. Consider bridging samples, shared controls, a defined batch structure, and locked analysis plan. Retrofitting it does not work.
It is also worth recognizing integration as an informatics and interpretation problem as much as an assay problem. Most pathology workflows are built around single-analyte IHC. Adding multiplex immunofluorescence or ultra-sensitive plasma testing means new validation, new instrumentation and, hardest of all to find, new interpretive expertise – all of it under existing turnaround pressure.
Sitting underneath that is the translation gap between discovery and clinical use. Research platforms are built for depth and plex; a clinical assay needs robustness and reproducibility, which usually means deliberately cutting the plex back to what has demonstrated value. That reduction step is where most multi-omic programs stall.
Standardization remains a recurring challenge across emerging diagnostics. Which aspects of multi-omics do you think are furthest from consensus, and what needs to happen next?
Consensus is weakest exactly where the modalities have to meet. Each one is standardizing on its own terms and nobody owns the join.
On the blood side, the live issue is calibration. Independent head-to-head comparisons of the commercial plasma p-tau217 assays have found real differences between them, with manufacturer-recommended cutoffs performing inconsistently when applied to cohorts the manufacturers did not develop them in. Other work in community populations has found broadly similar discrimination across assays, so I would put it carefully: several of these tests separate cases about equally well, but the absolute values and thresholds do not transfer between platforms.
That is a calibration problem, and those have known solutions. CSF amyloid went through exactly this and came out of it with certified reference materials and reference measurement procedures. Plasma p-tau does not have that infrastructure yet. Building it would be the highest-value standardization work anyone could fund right now.
On the spatial side, the gap is the reporting unit. Immune cell density, co-localization, margin infiltration; all of them depend on how the region of interest was drawn, and region selection is still substantially operator-dependent. Two laboratories running the same panel on the same block can report different values for entirely defensible reasons. That is harder than analytical imprecision, because there is no single right answer to converge on. Though I would push back on the idea that it is intractable.
The CIMAC-CIDC network under the Cancer Moonshot has shown you can get meaningful cross-site agreement on immune cell densities using harmonized clones and shared protocols. The engineering works. What is missing is agreement on which outputs to standardize in the first place.
Then there is the layer nobody owns. Suppose you standardized the plasma assay completely and the spatial assay completely. You would still have no agreed way to put them together into one interpretable result. There is no reference material spanning blood and tissue, no accepted framework for putting a concentration and a spatial density on the same footing, and no consensus on what an integrated report should even contain.
How would you prioritize the standardization gaps?
My order of priorities would be reference materials and reference methods for the circulating markers first; then minimum reporting standards for spatial assays, covering how regions are defined and how the analysis version is recorded; and only then a harmonized output format. The temptation is to skip to the third and build integration on unstandardized inputs. That is how you get results that publish well and do not reproduce.
What changes are still required to support broader clinical uptake?
When it comes to evidence, strong performance in a single research center is no longer enough. What laboratory directors want to see is reproducibility across institutions, across different instrument platforms, and across the types of specimens routinely found in clinical practice. That means proving performance not only on fresh, well-preserved samples, but also on archived tissue with varying fixation histories and years of storage behind it.
On regulation, the asymmetry between the two modalities is striking. Blood-based neurodegeneration markers have gone through 510(k) clearance with intended-use language that distinguishes rule-in from rule-out, and specialty care from primary care. Spatial assays have no comparable route in routine diagnostics. The precedent that exists is the tissue-based companion diagnostic developed alongside one specific therapy, which is narrow, expensive, and does not generalize.
Yet the factor that may have the greatest influence on adoption often receives the least attention: reimbursement. Blood-based Alzheimer's testing is beginning to gain commercial payer coverage, and notably, many of the emerging coverage frameworks are based on clinical performance and use case rather than a specific branded product. That model has the potential to accelerate adoption across the field.
Spatial technologies face a different reality. Most applications remain research-focused, with testing costs largely absorbed by laboratories or supported through research funding. As a result, adoption is concentrated in academic medical centers and specialized research environments.
What breaks that is studies designed for clinical utility rather than analytical performance. Show that the result changes a decision, and that the changed decision improves an outcome. Guidelines committees and payers have been explicit that this is what they want. Comparatively few multi-omic studies are built to answer it, because it is slower and more expensive than a correlation study. That choice gets made at protocol stage, and it largely determines whether an assay ever reaches patients.
Looking ahead five years, what do you think success will look like for multi-omic diagnostics?
My hope is that, five years from now, this is no longer considered newsworthy. That may sound modest, but it is actually the clearest measure of success. The goal is not novelty. The goal is routine clinical practice.
Five years is probably not enough for that across the board, so let me be concrete about what I would expect. Combined tissue and blood testing should be a routine orderable part of practice in a defined set of indications rather than a research add-on, with a few signatures validated well enough to sit inside guidelines.
In oncology, I would expect the tissue component of that to be multiplex immunofluorescence rather than sequential single-marker IHC, because the route from high-plex discovery to a defensible clinical assay now exists in practice. You use a high-plex platform, PhenoCycler-Fusion in our case, to find the signature. Then it is reduced to a robust automated assay on something built for clinical-trial scale, which for us is PhenoImager HT. That reduction step is the part the field has historically skipped, and it is the part that decides whether a spatial signature ever becomes a test rather than a paper.
I would also expect a shift from single timepoints to longitudinal monitoring. Tissue gives you the reference architecture, blood gives you the repeated measure, and following a patient across both is where the combination justifies its cost. If dried or capillary collection matures on the timeline it promises, the repeated measure becomes something the patient can provide without a clinic visit at all, which changes what longitudinal really means.
The clearest sign of maturity will be at the reporting layer, when the pathologist receives one integrated result to interpret instead of three datasets to reconcile. Computational integration has a genuine opportunity to prove itself there, with the caveat I would apply to all of this: it only works on standardized, calibrated inputs. An integration layer sitting on unharmonized assays will produce confident numbers that do not reproduce, and one visible failure of that kind would cost the field more time than going slowly ever would.
Realistically, five years means real depth in oncology and neurology rather than universal adoption. Those are the two fields where the biology and the assays are furthest along. The indications with clear decision impact will move first, and I would expect the rest to follow the evidence.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
